User’s Guide

®

Version 5.2.1

WinNonlin Version 5.2.1

WinNonlin® copyright ©1998-2008 Pharsight Corporation. All rights reserved. This software and documentation are owned by Pharsight Corporation, and may be used only as authorized in the license agreement controlling such use. No part of the software nor the accompanying documentation may be reproduced, transmitted, or translated, in any form or by any means, electronic, mechanical, manual, optical, or otherwise, except as expressly provided by the license agreement or with the prior written permission of Pharsight Corporation. This product may contain the following software that is provided to Pharsight Corporation under license: Actuate™ Formula One® copyright 1993-03 Actuate, Inc. All rights reserved. SentinelLM™ 6.0 copyright 1998 Rainbow Technologies, Inc. All rights reserved. Microsoft® XML Parser version 3.0 copyright 19982000 Microsoft Corporation. All rights reserved. IMSL® copyright 1993 Visual Numerics, Inc. All rights reserved. Information in the documentation is subject to change without notice and does not represent a commitment on the part of Pharsight Corporation. The documentation contains information proprietary to Pharsight Corporation and is for use by Pharsight Corporation customers only. Use of the information contained in the documentation for any purpose other than that for which it is intended is not authorized. NEITHER PHARSIGHT CORPORATION NOR ANY OF THE CONTRIBUTORS TO THIS DOCUMENT MAKES ANY REPRESENTATION OR WARRANTY, NOR SHALL ANY WARRANTY BE IMPLIED, AS TO THE COMPLETENESS, ACCURACY, OR USEFULNESS OF THE INFORMATION CONTAINED IN THIS DOCUMENT, NOR DO THEY ASSUME ANY RESPONSIBILITY FOR LIABILITY OR DAMAGE OF ANY KIND WHICH MAY RESULT FROM THE USE OF SUCH INFORMATION.

**Destination Control Statement
**

All technical data contained in the software and documentation are subject to the export control laws of the United States of America. Disclosure to nationals of other countries may violate such laws. It is the reader's responsibility to determine the applicable regulations and to comply with them.

**United States Government Rights
**

The software and documentation constitute “commercial computer software” and “commercial computer software documentation” as such terms are used in 48 CFR 12.212 (Sept. 1995). United States Government end users acquire the software under the following terms: (i) for acquisition by or on behalf of civilian agencies, consistent with the policy set forth in 48 CFR 12.212 (Sept. 1995); or (ii) for acquisition by or on behalf of units of the Department of Defense, consistent with the policies set forth in 48 CFR 227.7202-1 (June 1995) and 227.7202-3 (June 1995). The manufacturer is Pharsight Corporation, 321 E. Evelyn Ave., 3rd Floor, Mountain View, CA 94041.

Trademarks

PK Automation Application, PKS, Pharsight Knowledgebase Server and Trial Simulator are trademarks of Pharsight Corporation. Pharsight, WinNonlin, IVIVC Toolkit for WinNonlin, WinNonMix, Drug Model Explorer and DMX are registered trademarks of Pharsight Corporation. SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. SigmaPlot is a registered trademark of Systat Software, Inc. S-PLUS is a registered trademark of Insightful Corporation. NONMEM is a registered trademark of The Regents of the University of California. Actuate is a trademark and Formula One is a registered trademark of Actuate Corporation. SentinelLM is a trademark of Rainbow Technologies, Inc. Microsoft, the Internet Explorer logo, MS-DOS, the Office logo, PowerPoint, Visual C++, Visual Studio, Windows, the Windows logo, Windows NT, the Windows Start logo and XL design (the Microsoft Excel logo) are registered trademarks of Microsoft Corporation. Pentium and Pentium III are trademarks or registered trademarks of Intel Corporation. Adobe, Acrobat, Acrobat Reader and the Adobe PDF logo are registered trademarks of Adobe Systems Incorporated. IMSL is a registered trademark of Visual Numerics, Inc. Compaq Visual Fortran is trademark of Compaq Computer Corporation. All other brand or product names mentioned in this documentation are trademarks or registered trademarks of their respective companies or organizations.

Pharsight Corporation 321 E. Evelyn Ave., 3rd Floor, Mountain View, California 94041 Voice +1-650-314-3800 • Fax +1-650-314-3810 http://www.pharsight.com • support@pharsight.com

Contents

Chapter 1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 WinNonlin. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 WinNonlin editions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 In-vitro in-vivo correlation toolkit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 WinNonlin user documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Modeling and nonlinear regression. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Model fitting algorithms and features of WinNonlin . . . . . . . . . . . . . . . . 5 Summary of regression methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Pharsight Corporation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Scientific and consulting services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Pharsight contact information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Technical support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Licensing and product upgrades. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Customer feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Data Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13 WinNonlin windows and files. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 File types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 WinNonlin and WinNonMix file compatibility . . . . . . . . . . . . . . . . . . . 17 Hidden windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Workspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Printing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Page setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Page setup for workbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Page setup for text windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Page setup for charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Print preview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

iii

Chapter 2

Pharsight WinNonlin User’s Guide

Print . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Printing workbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Printing text objects and models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Print all . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Print all options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 WinNonlin options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Directories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Workbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Licensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Data dependencies and origins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 Saving dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 Refreshing results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 Unlocking derived data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Object origins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Data history worksheets and log files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 History worksheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 WinNonlin Log Viewer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Chapter 3 Workbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41 Editing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Inserting or deleting worksheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Inserting and deleting rows, columns or cells . . . . . . . . . . . . . . . . . . . . 43 Using formulas and functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Filling adjacent cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Formatting cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Clearing cell values and formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Undo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Find, find next and replace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Go to . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Column names and units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Units builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Data manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Freezing rows and columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Merge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Data exclusion and inclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 Re-including data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Transformations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

iv

Contents

Multiple transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Transform action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Multi-transformation filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Multi-transformation equation manager . . . . . . . . . . . . . . . . . . . . . . . . Rank transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Status code transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rule sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a rule set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 4

64 66 66 67 68 69 71 73

Data import and export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .75 Microsoft Word export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Word export document options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Word export chart options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 Word export workbook options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 SAS Transport file export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 NONMEM export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Data file structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Exporting to NONMEM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 Dependent variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 Other data items . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 Dose information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 Occasions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 SigmaPlot export. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 S-PLUS export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 ODBC import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Creating an import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Final settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 Creating a new connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 Loading a saved connection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 Queries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 Building a query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Schema filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Query builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 Loading a saved query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 Loading an existing import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 Refreshing an imported data set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 The custom query builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Select a builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Study selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

v

Pharsight WinNonlin User’s Guide

ODBC export. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Creating an export. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Export definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Building an export definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 Field selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Loading a saved export definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 Loading a saved export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 The Enterprise Software Development Kit . . . . . . . . . . . . . . . . . . . . . . . . . 107 Import file format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Usage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Export file format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Custom query builder interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 VB example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 Chapter 5 Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .115 Chart Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 X-Y variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 Scatter plots and bar charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 Histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Sort and group variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Error bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 Chart titles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Axis options. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Chart Designer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Frequently used chart formatting options . . . . . . . . . . . . . . . . . . . . . . . . . . 124 Changing the plot type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 Changing line style and markers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Changing axis scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 Creating a semi-log plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 Setting grid lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 Removing or adding walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 Chart templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 Copy and paste. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .131 Table Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Missing and excluded data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 WinNonlin table templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 Joining data sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Summary statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

Chapter 6

vi

Contents

Significant digits and column alignments. . . . . . . . . . . . . . . . . . . . . . . . . . Table formatting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Table definition files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating a table definition file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Loading a table definition file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 7

145 146 148 148 149

Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .151 Computing summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Output and computational formulae. . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Weighted summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 Output and computational formulae for weighted calculations . . . . . . 157 Noncompartmental Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . .159 NCA model selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 Model variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Urine models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Lambda Z or slope estimation settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 Lambda Z or slope range selection. . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 Calculation of Lambda Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Calculation of slopes for effect data. . . . . . . . . . . . . . . . . . . . . . . . . . . 163 Setting Lambda Z or PD slope ranges . . . . . . . . . . . . . . . . . . . . . . . . . 164 Partial areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 Therapeutic response. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Time deviations in steady-state data. . . . . . . . . . . . . . . . . . . . . . . . . . . 172 Model options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 Output options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 Weight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 NCA settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 Parameter names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 Computational rules for WinNonlin’s noncompartmental analysis engine 181 Data checking and pre-treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 AUC calculation and interpolation formulas . . . . . . . . . . . . . . . . . . . . 183 Partial areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184 Therapeutic response windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186 Sparse sampling computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

Chapter 8

vii

Pharsight WinNonlin User’s Guide

Chapter 9

Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .193 Loading the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 Model properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 Data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196 Model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 Dosing data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 Model options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 Output options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 Weight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 PK settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 Running the model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Stop modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Troubleshooting modeling problems . . . . . . . . . . . . . . . . . . . . . . . . . . 209 Errors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 Text (ASCII) output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 Workbook output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214 Chart output. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 User Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .217 ASCII user models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 Creating a new ASCII model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 User model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 Model structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220 ASCII model examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 WinNonlin variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 Temporary variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 The WinNonlin programming language . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233 Comparison operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Bioassay functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 Compiled user models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 Using FORTRAN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 Example using Digital (now Compaq) Visual FORTRAN 6.0 . . . . . . 237 Using C++ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 Example using Microsoft Visual C++ 6.0 . . . . . . . . . . . . . . . . . . . . . . 239

Chapter 10

viii

Contents

Chapter 11

Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .241 Weighted least squares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 Iterative reweighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 Reading weights from the data file . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 Scaling of weights. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 Weights in ASCII models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 The WinNonlin Toolbox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .245 Nonparametric superposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 Data and assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 Computation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 Performing nonparametric superposition . . . . . . . . . . . . . . . . . . . . . . . 248 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249 Semicompartmental modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 Data and assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 Computation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 Performing semicompartmental modeling . . . . . . . . . . . . . . . . . . . . . . 252 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 Crossover design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 Data and assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255 Descriptive analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256 Hypothesis testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Using the Crossover Design tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260 Numeric deconvolution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260 Deconvolution methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 Using deconvolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 Deconvolution variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 Deconvolution dosing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 Deconvolution settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273 Deconvolution output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 The IVIVC Toolkit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .277 Wagner-Nelson and Loo-Riegelman methods . . . . . . . . . . . . . . . . . . . . . . 278 Wagner-Nelson inputs and calculations . . . . . . . . . . . . . . . . . . . . . . . . 278 Loo-Riegelman inputs and calculations . . . . . . . . . . . . . . . . . . . . . . . . 279 Using the Wagner-Nelson or Loo-Riegelman method. . . . . . . . . . . . . 280 Data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282 Loo-Riegelman model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 283

ix

Chapter 12

Chapter 13

Pharsight WinNonlin User’s Guide

Lambda Z ranges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284 Output options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288 Title . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293 Data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 Convolution settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297 Convolution output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298 Convolution units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299 Levy plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300 Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 The IVIVC Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 Minimizing the IVIVC Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 Project Status. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 Saving and loading IVIVC projects . . . . . . . . . . . . . . . . . . . . . . . . . . . 305 In-vitro tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306 In-vivo tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307 Correlation tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310 ASCII user model for correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312 Prediction tab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Using the IVIVC Wizard with PKS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319 Chapter 14 Linear Mixed Effects Modeling . . . . . . . . . . . . . . . . . . . . . . . . . .323 An example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323 The linear mixed effects model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326 Construction of the X matrix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328 Linear combinations of elements of β. . . . . . . . . . . . . . . . . . . . . . . . . . 332 Determining estimability numerically . . . . . . . . . . . . . . . . . . . . . . . . . 333 Substitution of estimable functions for non-estimable functions . . . . . 334 Fixed effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336 Output parameterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336 Variance structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338 Repeated effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 Random effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 Least squares means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344

x

Contents

Contrasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Joint versus single contrasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nonestimability of contrasts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Degrees of freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Other options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix limits and constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fixed effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Variance structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Random coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Repeated measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contrasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Least squares means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix general options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Linear mixed effects text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Linear mixed effects workbook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tests of hypotheses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Akaike’s Information Criterion (AIC) . . . . . . . . . . . . . . . . . . . . . . . . . Schwarz’s Bayesian Criterion (SBC) . . . . . . . . . . . . . . . . . . . . . . . . . . Hessian eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Predicted values and residuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fixed effects parameter estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . Coefficients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Iteration history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Parameter key . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Computational details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Restricted maximum likelihood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Newton’s Algorithm for maximization of restricted likelihood. . . . . . Starting values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Convergence criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Satterthwaite’s approximation for degrees of freedom . . . . . . . . . . . . Comparing LinMix to ANOVA and SAS’s PROC MIXED . . . . . . . . . . . LinMix and ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LinMix and PROC MIXED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 15

346 347 348 349 349 349 350 350 352 353 355 356 356 358 359 360 361 361 361 362 363 363 364 364 364 365 365 365 366 366 368 369 369 370 372 372 373

Bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .375

xi

Pharsight WinNonlin User’s Guide

Introduction to bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375 Basic terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375 Bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376 The model and data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Linear model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Variance structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Missing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Variable name limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 Average bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379 Study designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379 Recommended models for average bioequivalence . . . . . . . . . . . . . . . 379 Least squares means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381 Classical and Westlake confidence intervals . . . . . . . . . . . . . . . . . . . . 383 Two one-sided T-tests. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384 Anderson-Hauck test. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386 Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386 Individual and population bioequivalence. . . . . . . . . . . . . . . . . . . . . . . . . . 387 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387 Bioequivalence criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388 Details of computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390 Bioequivalence Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395 Using average bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396 Fixed effects for average bioequivalence . . . . . . . . . . . . . . . . . . . . . . . 396 Random effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398 Repeated variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399 Bioequivalence options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400 Bioequivalence general options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401 Using population/individual bioequivalence . . . . . . . . . . . . . . . . . . . . 401 Fixed effects for population/individual bioequivalence . . . . . . . . . . . . 402 Bioequivalence options for population/individual bioequivalence . . . 402 Bioequivalence output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403 Average bioequivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403 Population/Individual bioequivalence. . . . . . . . . . . . . . . . . . . . . . . . . . 404 Ratio tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406 Chapter 16 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .409 Simulating responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409

xii

Contents

Simulation of a compiled library model . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating the input data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Setting up the model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Model options. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Running the simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparing dosing schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dosing regimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter 17

409 410 410 411 412 413 414 414 415 416

Scripting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .417 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417 Writing and editing a script. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418 Executing a script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420 Within WinNonlin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420 From the command line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420 From file manager or explorer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421 Scripting object properties and methods. . . . . . . . . . . . . . . . . . . . . . . . . . . 421 Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421 Engines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423 Script file components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426 Note on saving files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428 Processing multiple profiles in script files . . . . . . . . . . . . . . . . . . . . . . . . . 428 Special file formats. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430 Scripting the Pharsight Knowledgebase Server . . . . . . . . . . . . . . . . . . . . . 440 Using the Pharsight Knowledgebase Server . . . . . . . . . . . . . . .445 The PKS information structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446 Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446 Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447 Other documents. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448 Dosing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448 Dosing worksheet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448 Append dosing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449 Dosing dialog . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449 Accessing the PKS from WinNonlin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449 PKS sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450 Connecting to the PKS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451 Logging in . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452

xiii

Chapter 18

Pharsight WinNonlin User’s Guide

Creating a study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452 Study definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453 Study Creation Wizard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454 Subject identifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455 Subject ID attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455 Subject data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456 Subject data attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457 Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457 Sample data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458 Sample attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458 Dosing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459 Dosing attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460 Data Collection Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461 Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461 Study creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462 Loading a study or scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463 Select observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464 Select dose data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464 Scenario search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464 Creating scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465 Adding data columns to a scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . 466 Adding non-WinNonlin documents to a scenario . . . . . . . . . . . . . . . . 466 Optimizing PKS/WNL performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467 Change tracking and audit information . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470 Audit information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470 Update reason . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471 Name objects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472 Dosing information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472 Changing the dose regimen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473 Appending or updating a study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474 PKS libraries and object shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475 PKS share . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475 Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476 Loading a library from the PKS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477 Adding or updating a library in the PKS . . . . . . . . . . . . . . . . . . . . . . . 477 Library and share status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478 PKS Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479 WinNonlin and PKS differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480 Studies and scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481

xiv

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535 Model 104 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 530 Model 18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532 Pharmacodynamic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524 Model 9 . . . . . . . . . . . . . . . . 533 ASCII library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528 Model 15 . . . . . . . . . . . . . . . . 520 Model 2 . . . . . . . . . . . . . . . . . 535 Model 106 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528 Model 14 . . . . . 536 Model 108 . . . . . . . . . . . . . . . . . . . . . . . . . . . 533 Model 101 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534 Model 103 . . . . . 524 Model 10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536 xv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522 Model 7 . . . . . . . . . . . . . . . . . . . . 536 Model 107 . . . . . . 527 Model 13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519 Model 1 . . . . . . . . . . . . . . . . 515 Pharmacokinetic models . . . . . . . . . . . . . . . . . . 504 Noncompartmental analysis . 535 Model 105 . . . . . . . . . . . . . . . . 523 Model 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .503 Notation and conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522 Model 6 . . . . . . . . . . . . . . . . . . . . . . . . . . 506 Plasma or serum data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512 Sparse sampling (pre-clinical) data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522 Model 5 . . 519 ASCII models and files . . . . . . . . . . . . . . . .483 Appendix B WinNonlin Model Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525 Model 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534 Model 102 . 514 Drug effect data (model 220) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528 Model 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Contents Appendix A Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520 Model 3 . . . . . . . . . . . . . . . . . 521 Model 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505 NCA output parameters and their computation formulas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531 Model 19 . . . . . . . . . . . . . . . . . . . . . 526 Model 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . 507 Urine data . . . . . . . . . . . . . . . . . . 529 Model 17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544 Model 303 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536 Notation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545 ASCII library. . . . . . . . . . . 537 Linking PK and PD models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550 Appendix C Worksheet Functions . . .637 Compartmental modeling . 539 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546 Sigmoidal/dissolution models . . . 540 Model 51 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . Arrays and Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545 Linear models . . . 545 Simultaneous PK/PD link models . . . . . . . . . . . . 538 Indirect response models . . . . . . . . . . . . . . . . . . . . . . . . . . . 541 Michaelis-Menten models . . . . 537 Output parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547 Model 601 . . . . . . . . . . .557 Appendix E Appendix F Scripting Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652 xvi . . . . .Pharsight WinNonlin User’s Guide PK/PD link models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544 Model 304 . . . . . . . . . . . . . . 540 Model 53 . . . . . . . . . . . . . . . . . . 543 Model 302 . . . . . . . . . . . . . .551 Appendix D Commands. . . . . . . . . . . . . . . . 541 Model 54 . . . . . . . . . . . . . . 542 Model 301 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549 Model 604 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 548 Model 602 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540 Model 52 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .579 Warnings and Error Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549 Model 603 . . . . . . . . . . . . . . . . . . . . . . 545 Data . . . . . . 649 Table Wizard . . . . . . . . . . . . . . . . . . . . . . . . . 637 LinMix and Bioequivalence wizards. . . . . . . . . . . . . . . . . . . . . . . 541 Running an indirect response model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . 658 Pharmacokinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 660 Sparse data analysis methods . . . . . . . . . . . . . . . . . . . . . . . . . .657 Bioequivalence . . . . . . Warnings. IVIVC . . . . . . . . 657 In vitro-in vivo correlation and formulation science . . . . . . . . . . . . . . . . . . . . . . . . . 663 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Error messages . . . . . . Scripting language. . . . . . . . . . . . . . . . . . . .Contents Noncompartmental analysis . . . . . . . . . . . . . . . . . . . . . . 662 Statistics . . . . . . . . . . . . . . . . . . . . 653 653 653 654 655 Appendix G References . . . . . . . . . . . . 659 Regression and modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665 xvii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Pharsight WinNonlin User’s Guide xviii .

noncompartmental. WinNonlin editions WinNonlin is distributed in two editions. Pharsight's solution for secure storage. nonlinear modeling and the Pharsight suite of products and services under the following headings. The 1 . and in addition. WinNonlin Professional is made for scientists involved with nonlinear modeling. It supports the analyses and generates the figures. and listings required for regulatory submission. management and tracking of clinical and pre-clinical data. WinNonlin Professional and WinNonlin Enterprise. This chapter introduces WinNonlin. WinNonlin Enterprise provides all the features of WinNonlin Professional. shares data with ODBC-compliant databases. PK/PD link. pharmacodynamic. user documentation.Chapter 1 Introduction WinNonlin editions. nonlinear modeling and Pharsight contact information Welcome. • • • “WinNonlin” on page 1 “Modeling and nonlinear regression” on page 3 “Pharsight Corporation” on page 9 WinNonlin WinNonlin is a sophisticated tool for nonlinear modeling. and serves as a portal to the Pharsight® Knowledgebase Server™ (PKS). tables. It also supports creation of custom models to enable fitting and analysis of virtually any kind of data. WinNonlin includes a large library of pharmacokinetic. and indirect response models. It is particularly suited to pharmacokinetic and pharmacodynamic (PK/PD) modeling and noncompartmental analysis to evaluate data from bioavailability and clinical pharmacology studies.

and a step-by-step hardcopy with CD shipment test to confirm that installation is complete and functional.com/support/ If issues and work-arounds are discovered after product release. F1 key. noncompartmental analysis. information on the calculations performed by WinNonlin. See “References” on page 657 for a bibliography of references on the underlying theory. Step-by-step examples illustrating use of WinNonlin features in realistic scenarios.\WINNONLIN\USER DOCS Operating procedures for each software feature.pharsight. dissolution models. WinNonlin user documentation WinNonlin includes the following user documentation. Examples Guide Online help ..\WINNONLIN\USER DOCS and Installation and licensing instructions. an addendum to the release notes will be posted to support_sflogin. .1 WinNonlin User’s Guide combination of WinNonlin Enterprise and PKS supports compliance with the Food and Drug Administration’s (FDA) 21 CFR part 11 regulations. WinNonlin can be upgraded to include a suite of tools for analysis of in-vitro in-vivo correlations (IVIVC)..com/support/ Frequently asked questions and solutions for WinNonlin.php the customer support web site with any software patches.\WINNONLIN\USER DOCS Accessed from WinNonlin via the Operating procedures for the software features and referHelp menu. The IVIVC suite includes a variety of correlation models. http://www. deconvolution and plotting as well as a wizard to streamline setup of common IVIVC work flows. 2 . and appendices detailing WinNonlin functions.pharsight..... In-vitro in-vivo correlation toolkit With purchase of a special license. nor statistical analysis. WinNonlin user documentation Document Getting Started Guide User’s Guide Default location Description .\WINNONLIN\USER DOCS Addendum to release notes FAQ The manuals are not intended as a comprehensive text on nonlinear estimation. support_sflogin. or Help button ence material on WinNonlin’s computational engines.php Release notes . enhanced tools for convolution... Known issues and work-arounds in the current version. Table 1-1. http://www.

X. Most scientists have encountered linear models. the independent variable. Y = A1 + A2X1 + A3X2 + A4X3 + ε These examples are all “linear models” since the parameters appear only as coefficients of the independent variables. Models are used in the quantitative analysis of all types of data. at least one of the parameters appears as other than a coefficient. The simplest example is a line (linear regression): Y = A + BX + ε where Y is the dependent variable. biological. the slope. and with the power and availability of computers. as well as Analysis of Variance and Analysis of Covariance. mathematical models provide convenient and powerful ways of looking at data. A simple example is the decay curve: Y = Y0 exp(-BX) This model can be linearized by taking the logarithm of both sides. and to predict future results. Two other common examples are polynomials such as: Y = A1 + A2X + A3X2 + A4X3 + ε and multiple linear regression. In nonlinear models. finding a mathematical equation and a set of parameter values such that values predicted by the model are in some sense “close” to the observed values. 3 . and others. Note: WinNonlin’s LinMix module handles linear regression. but as written it is nonlinear. behavioral. to test hypotheses. Models can be used to help interpret data. The concern here is “fitting” models to data—that is.Introduction Modeling and nonlinear regression 1 Modeling and nonlinear regression The value of mathematical models is well recognized in all sciences— physical. models in which the dependent variable can be expressed as the sum of products of the independent variables and parameters. the intercept and B. A.

But it is also possible to fit data to the differential equations by numerical integration. Ratkowsky (1983) and Bates and Watts (1988) give a more detailed discussion of nonlinear regression theory. ed. In least squares fitting the “best” estimates are those that minimize the sum of the squared deviations between the observed values and the values predicted by the model. how does one “fit” the model to the data? Some criterion of “best fit” is needed. (1981) and Gabrielsson and Weiner (2001). two good references for the topic of general nonlinear modeling are Draper and Smith (1981) and Beck and Arnold (1977). and a set of data. but better estimates are obtained if the nonlinear form is used to model the observations (Endrenyi. With certain assumptions about the error structure of the data (no random effects). or nonlinear regression models. WinNonlin can fit models defined in terms of algebraic equations. and many have been suggested. There are many models that cannot be made linear by transformation. pages 304-305). All pharmacokinetic models derive from a set of basic differential equations.1 WinNonlin User’s Guide Another example is the Michaelis-Menten equation: V max C V = ---------------Km + C This example can also be linearized by writing it in terms of the inverses of V and C. Modeling in the context of kinetic analysis and pharmacokinetics is discussed in Endrenyi. (1981). the two are equivalent. This manual is not intended to be a comprehensive text on nonlinear estimation theory. differential equations or a combination of the two types of equations. but only two are much used. When the basic differential equations can be integrated to algebraic equations. A discussion of using nonlinear least squares to obtain maximum likelihood estimates can be found in Jennrich and Moore (1975). One such model is the sum of two or more exponentials. such as Y = A1 exp(-B1X) + A2 exp(-B2X) All such models are called nonlinear models. Given a model of the data. These are maximum likelihood and least squares. it is most efficient to fit the data to the integrated equations. However. The books by Bard (1974). 4 . ed.

estimates that result in a smaller sum of squared deviations. Model fitting algorithms and features of WinNonlin WinNonlin is a computer program for estimating the parameters in a nonlinear model. WinNonlin will compute values of the dependent variable and also estimate the variance-inflation factors for the estimated parameters. the more general method of maximum likelihood (ML) must be employed.Introduction Modeling and nonlinear regression 1 The sum of squared deviations can be written in terms of the observations and the model. well-known techniques can be used to solve these linear equations for the parameter estimates. That is. The program cannot determine the correctness of the model nor the value of any decisions or interpretations based on the model. In the case of linear models it is easy to compute the least squares estimates. WinNonMix fits models of this variety. A computer program such as WinNonlin is only a tool for the modeling process. the likelihood of the data is maximized with respect to the model parameters. This gives a set of linear equations with the estimates as unknowns. This method will not work for nonlinear models. such as population PK/PD models. i.e. That is. using the information contained in a set of observations. In the case of models with random effects. initial estimates of the parameters are made and then in some way modified to give better estimates. however. WinNonlin can also be used to simulate data from a model. One equates to zero the partial derivatives of the sum of squares with respect to the parameters. As such it can only take the data and model supplied by the researcher and find a “best fit” of the model to the data. because the system of equations that results by setting the partial derivatives to zero is a system of nonlinear equations without general solution methods. provide some information about the “goodness of fit” and about how well the parameters are estimated. Since it is not possible to know exactly what the minimum is. The iteration continues until hopefully the minimum (or least) sum of squares is reached. A more complete discussion of the theory underlying nonlinear regression can be found in the books cited previously. 5 . some stopping rule must be given at which point it is assumed that the method has converged to the minimum sum of squares. Consequently. In ML. given a set of parameter values and a set of values of the independent variables in the model.. any method for computing the least squares estimates in a nonlinear model must be an iterative procedure. The program does.

with the additional requirement that at every iteration the sum of squares must decrease. However. the Gauss-Newton algorithm with the modification proposed in Hartley (1961). To speed convergence. WinNonlin provides the “least squares” estimates of the model parameters. Many nonlinear estimation programs. with a choice of three algorithms for minimizing the sum of squared residuals: the simplex algorithm of Nelder and Mead (1965). and a Levenberg-type modification of the Gauss-Newton algorithm (Davies and Whitting (1972)). The Gauss-Newton algorithm uses a linear approximation to the model. its usefulness has formerly been limited by the extensive amount of computation it requires. if the data contain very little information about one or more of the parameters. The simplex method does not require the solution of a set of equations in its search and does not use any knowledge of the curvature of the sum of squares surface. including the original NONLIN. When the Nelder-Mead algorithm converges. WinNonlin avoids this difficulty by using singular value decomposition rather than matrix inversion to solve the system of linear equations. the adjusted parameter 6 . (See Kennedy and Gentle (1980). WinNonlin restarts the algorithm using the current estimates of the parameters as a “new” set of initial estimates and resetting the step sizes to their original values. As such it must solve a set of linear equations at each iteration. Much of the difficulty with this algorithm arises from singular or near-singular equations at some points in the parameter space. especially for those problems where some of the parameters may not be welldefined by the data and the sum of squares surface is complicated in the region of the minimum. the singular value decomposition algorithm will always find a solution to the system of linear equations.1 WinNonlin User’s Guide It is assumed that WinNonlin users have some knowledge of nonlinear regression. The parameter estimates that are obtained after the algorithm converges a second time are treated as the “final” estimates. The power and speed of current computers make it a more attractive choice. as discussed above. have found Hartley's modification to be very useful. The simplex algorithm is a very powerful minimization routine. WinNonlin uses the modification to the Gauss-Newton method proposed by Hartley (1961) and others.) One result is that the iterations do not get “hung up” at a singular point in the parameter space. Beck and Arnold (1977) compare various least squares algorithms and conclude that the Box-Kanemasu method (almost identical to Hartley's) is the best in many circumstances. This modification helps the algorithm locate the global minimum (as opposed to a local minimum) of the residual sum of squares for certain difficult estimation problems. but rather move on to points where the problem may be better defined. As indicated.

Although WinNonlin gives indications of this. Then the minimization algorithm may fail due to any number of numerical problems. In WinNonlin the partial derivatives required by the Gauss-Newton algorithm are approximated by difference equations. resulting in “constrained optimization. evidence that any nonlinear estimation problem is better or more easily solved by the use of the exact partial derivatives. At other times. (2) More importantly. points outside the bounds are assigned a very large value for the residual sum of squares. It is often desirable to set limits on the admissible parameter space.” For example. For more information on trust region methods.. One way to avoid this is by using a 'trust region' solution. In such a case it may be preferable to give up some properties of the unconstrained estimation in order to obtain parameter estimates that are physically realistic. al. why one may not want to use them. 1976). that is. For this reason. it is recommended that users always set limits on the parameters (the means of doing this is discussed in the modeling chapters of this document). This sends the algorithm back into the admissible parameter space. Note: The Gauss-Newton method with the Levenberg modification is WinNonlin’s default estimation method. if any.Introduction Modeling and nonlinear regression 1 vector may be so far from the least squares solution that the linearization of the model is no longer valid. the trust region methods obtain parameter estimates that are often meaningless because of their large variances. There are two reasons. however. it is possible for users to ignore this information and use the estimates as though they were really valid. Trust region methods tend to be very robust against ill-conditioned data sets. To fit those models that are defined by systems of differential equations. the model may contain a parameter that must be non-negative. The Levenberg and Marquardt algorithms are examples. 7 . Murray and Wright (1981) or Davies and Whitting (1972). and thus are not efficient with data sets and models that readily permit precise estimation of the parameters. but the data set may contain so much error that the actual least squares estimate is negative. In WinNonlin two different methods are used for bounding the parameter space. (1) They require more computation. the RKF45 numerical integration algorithm is used (Shampine et. setting reasonable limits on the parameter may prevent the algorithm from wandering off and getting lost. This algorithm is a 5th order Runge-Kutta method with variable step sizes. When the simplex method is used. a solution to the system of linear equations is not accepted unless it is sufficiently close to the parameter values at the current iteration. There is little. see Gill.

However. The model libraries and their uses are described in “WinNonlin Model Libraries” on page 503. Users can also create custom libraries. and the complexity of nonlinear estimation. Ill-conditioned indicates that for the given data set and model. Bounding the parameter space in this way is in effect a transformation of the parameter space. plus an additional utility library of models. ASCII library files corresponding to the same models as the built-in models. Reparameterization to make the models more linear also may help. (See Draper and Smith (1981) or Ratkowsky (1983) for discussions of reparameterization. as explained in “User Models” on page 217. or compiled. it often performs very well on ill-conditioned data sets. the user may want to try different bounds to see if the estimation is improved. To speed execution time. are also provided. WinNonlin is capable of estimating the parameters in a very large class of nonlinear models. However. two successive transformations are used to affect the bounding of the parameter space. Summary of regression methods Method 1: Nelder-Mead simplex Method 1 is a very powerful algorithm. many options and specifications may be supplied to the program. there isn't enough information contained in the data to precisely estimate all of the parameters in the model or that the sum of squares 8 .1 WinNonlin User’s Guide With the Gauss-Newton algorithms. Since it doesn't require the estimated variance-covariance matrix as part of its algorithm. the main WinNonlin library has been built in. Because of WinNonlin’s flexibility and generality. The second transformation uses the inverse of the normal probability function to go to an infinite parameter space. It must be pointed out that it may be possible to make the problem worse with this kind of transformation of the parameter space. but it is important enough that WinNonlin is supplied with a library of the most commonly used pharmacokinetic and pharmacodynamic models. experience suggests that this will rarely occur. The first transformation is from the bounded space as defined by the input limits to the unit hypercube. To make WinNonlin easy to use. Fitting pharmacokinetic and pharmacodynamic models is a special case of nonlinear estimation. This method of bounding the parameter space has worked extremely well with a large variety of models and data sets. many of the most commonly used options are set as WinNonlin defaults. It is well known that a reparameterization of the parameters will often make a problem more tractable.) When encountering a difficult estimation problem.

please visit Pharsight on the web at www. Pharsight Corporation is a public company headquartered in Mountain View. Method 2: Gauss-Newton with Levenberg and Hartley modification Method 2 is the default. When used to fit a system of differential equations. Although the procedure works well. including ill-conditioned data sets. It is not as robust as Methods 1 and 2 but is extremely fast. and also performs well on a wide class of problems.pharsight. Although the Gauss-Newton method does require the estimated variance-covariance matrix of the parameters. Trial Simulator™. it can be extremely slow. the Levenberg modification suppresses the magnitude of the change in parameter values from iteration to iteration to a reasonable amount. it can be slow. Method 3 is recommended when speed is a consideration or when maximum likelihood estimation or linear regression analysis is to be performed.pharsight. It is not recommended for fitting complex models. California. Pharsight Corporation helps companies make more informed and objective decisions that lead to more successful and predictable trial outcomes. Pharsight Corporation Pharsight Corporation provides software and services designed to improve the efficiency of drug development. This enables the method to perform well on ill-conditioned data sets. For further information.com. Method 3: Gauss-Newton with Hartley modification Method 3 is another Gauss-Newton method. Drug Model Explorer™ and the Pharsight Knowledgebase Server™ suite of products. More information is available on the web at www. including WinNonlin®.Introduction Pharsight Corporation 1 surface is highly curved. WinNonMix®. Products Pharsight offers a comprehensive suite of software products.com. The company's proprietary technology and world-class consulting expertise enable pharmaceutical and biotechnology companies to make more confident decisions in dealing with complex issues of drug development and product portfolio management. 9 .

Suite 205 Cary. pricing and availability. 10 . if applicable. please e-mail a complete description of the problem.com (fastest response time) Web: http://www.pharsight. On-site training is available by request. including copies of input data.com/training/.pharsight. Contact Pharsight for schedules. planned and delivered in accordance with specific needs and agreed-upon objectives. Consultants can facilitate: • • • • • Gathering and interpreting data Building drug and population models Guiding variability testing and trial optimization Addressing specific development program objectives Process automation Training Pharsight Corporation offers workshops and training courses at regular intervals.com/support/support_sflogin. model file and instructions to WinNonlin. phone. If further assistance is needed. E-mail: support@pharsight. The training schedule is also available on Pharsight’s corporate web site at www. please contact Pharsight technical support by e-mail or web (preferred). North Carolina 27518 For the most efficient service.php Phone/voice mail: 01-919-852-4620 Fax: +1-919-859-6871 Post: Pharsight Corporation 5625 Dillard Drive. as well as the ASCII output. Pharsight contact information Technical support Please consult the documentation (see “WinNonlin user documentation” on page 2) to address questions.1 WinNonlin User’s Guide Scientific and consulting services Our scientific consultants can design and optimize clinical trials and clinical development programs to meet an individual company's objectives. Pharsight Scientific Services are project-centered. detailing the error. fax or post.

com http://www.com +1-919-852-4620 11 . fax or through Pharsight’s customer support web site.com Telephone: +1-888-708-7444 (from US only) Fax: +1-650-314-3811 Request assistance with Pharsight software licensing by e-mail or telephone: E-mail: Telephone: Customer feedback Please submit requests for product enhancements and defect corrections by email. as follows.php +1-919-859-6871 licensing@pharsight.pharsight.com/support/support_sflogin. please contact Pharsight sales: E-mail: sales@pharsight.Introduction Pharsight contact information 1 Licensing and product upgrades For license renewals and product upgrades. E-mail: Web: Fax: support@pharsight.

1 WinNonlin User’s Guide 12 .

script. under the following topics. and modeling and analysis default settings. • See also: “Data import and export” on page 75. “Data dependencies and origins” on page 33: refreshing analysis output when source data change. See “Charts” on page 115 for instructions on creating and formatting plots. Chart windows contain plots. “Data history worksheets and log files” on page 37: tracking changes to data and operations in WinNonlin. See “Workbooks” on page 41. or modeling output. “Workspaces” on page 19: saving and reloading WinNonlin analyses. WinNonlin windows and files WinNonlin provides three window types for managing data objects: • • • Workbooks hold tabular data in one or more worksheets that support spreadsheet operations and formats. and unlocking output (removing dependency on source data). text and chart windows and their supported file types. • • • • • “WinNonlin windows and files” on page 13: WinNonlin workbook. Text windows contain ASCII text representing a model.Chapter 2 Data Management Working with WinNonlin data windows and files This chapter covers data handling in WinNonlin. “Printing” on page 21 “WinNonlin options” on page 28: default file locations. 13 . worksheet behaviors.

Choose Tools>Copy Workbook from the WinNonlin menus. analysis settings entered in the WinNonlin user interface can be saved to file.PRN.) or 1.STX) Pharsight Chart Object (*. New Chart will open the Chart Wizard to create the plot(s).DTA.PCO) Bitmap image (*. New Text or New Chart from the drop-down menu that appears. Supported file formats Window Workbooks Example Subject data Modeling output Dosing data Model parameters File formats: load/import ASCII (*.WMF) Chart windows 14 . Click the New (left-most) button on the WinNonlin tool bar and select New Workbook. (Before using New Chart.XPT. The new window opens.XLS) Pharsight Workbook Object (*. as noted in the last row of Table 2-1.WDO) Pharsight Chart Object (*.STX) WinNonlin Data Object (*.PWO) SAS® Transport (*.DTA) Microsoft Excel 5. 97 (*. *. Open the workbook in WinNonlin.WGO) File formats: save/export ASCII data (*. *.XPT.PCO) WinNonlin Graph Object (*. Choose File>New from the menus (Alt+F.DAT. The New dialog appears. Table 2-1. Select the appropriate radio button. 2000 (*.XLS) Pharsight Workbook Object (*. Any window can be hidden to reduce clutter in the WinNonlin parent window (see “Hidden windows” on page 18).PRN.JPG) Windows Metafile (*.2 WinNonlin User’s Guide To create a new window: 1.0/95. open the workbook window containing data to be plotted.PWO) SAS® Transport (*. 5. *. In addition.CSV) Microsoft Excel 4. *. File types Each window type can load and save the file types listed in Table 2-1. *. 2.0. 97.CSV) ASCII model parameters (*.BMP) Vector graphic (*. Click OK. To create a copy of an existing workbook: 1. 2. *. *.0/95. N). 3.DAT.

A Pharsight Chart Object (*.PMO. including values for the model parameters.PTO) 2 Text windows Modeling output ASCII models Scripts Notes ANOVA settings Model settings (Load/ Bioequivalence Save in mod.WMO) WinNonlin Command File (*. Supported file formats (continued) Window Example File formats: load/import ASCII text (*. the source ASCII model is updated to include any changes in WinNonlin.PMO) WinNonlin Model Object (*.WSC) WinNonlin Text Object (*.WDO. Pharsight objects (*.WTO.CMD) and script files (*. *. units.LIB) Rich text (*.PWO) saves all information in a workbook. as well as the graph attributes. A Pharsight Text Object (*.PWO. data variables.RTF) Pharsight Script (*.Data Management WinNonlin windows and files Table 2-1.PMO) Note: Excel 4. inclusions/exclusions and formatting. *. command files (*.WSC) are import-only. including all worksheets. constants (dosing). When an ASCII model is used with a Pharsight Model Object. or multi-chart windows. A Pharsight Model Object (*.PSC) Pharsight Text Object (*.PCO) saves all charts in a chart window. Pharsight chart and workbook objects can save multiple worksheets in a workbook. A Pharsight Workbook Object (*. *.WMO).PTO.PSC) Pharsight Text Object (*.WGO.PSC). sorting and units applied in WinNonlin. Non-Pharsight file formats may limit the information saved. *. as follows. including formatting and links to the graph data. data settings.LML) Pharsight Model Object (*.PCO) preserve all formatting. *.PTO) WinNonlin Script (*. Changes must be saved to a more recent Excel file type or Pharsight file (*.TXT) Library model(s) (*.0 files and WinNonlin object files (*.Model settings eling dialogs) LinMix or Bioequivalence Command File (*.PCO. data sort order. *.WTO) ANOVA Command File (*. *.PMO) contains the current model (the ASCII code or the number of the compiled model) and any associated data.PTO.RTF) Pharsight Script (*.ANV) LinMix or Bioequivalence Command File (*. and modeling options. *.TXT) Library model(s) (*.PWO.LML) Pharsight Model Object (*.LIB) Rich text (*. *.PTO) saves all text and formats in a text window. 15 .CMD) File formats: save/export ASCII text (*.

and multiple worksheets in a workbook. such as New = Old+0.). To convert non-numeric data to numeric data.” Some applications that read Metafiles will expect this information. . owing in part to the different sort order inherent in nonnumeric data. data inclusions/exclusions. then paste only the values (not the formula) to another column. Missing data If a data set includes missing values. non-numeric data may appear in workbook cells that should contain numeric data. This may be caused by leading or trailing blank spaces in some fields. However. Troubleshooting: loading tabular data Non-numeric data Occasionally. units and column headers are not saved. when transferring data from other packages. 16 . or units. Multiple worksheets cannot be saved to a single ASCII file. data exclusions.WMF) WinNonlin prompts whether to “Include placeable header information. Copy the new column. delete the other two columns.RTF) preserve text and formatting from text windows. then check that the character denoting missing values was specified when the file was opened (as set in the Options dialog accessed from the Open dialog). Last.BMP file.WMF. If you will open the chart in a Microsoft application. When a chart is saved to a . others will not. a snapshot of the plot that appears on-screen is saved. One indicator that the data are non-numeric is that non-numeric data will be left justified rather than centered in the cells. and the cells with the missing values do not have a gray background.2 WinNonlin User’s Guide ASCII data files save the text of the active worksheet or text window. etc. Saving a chart window as a Pharsight Chart Object saves all plots in the window and their attributes (formatting. Rich text files (*. create a new variable using a formula or function. such as Word or Excel.JPEG or . formatting. Non-numeric data may cause problems in WinNonlin calculations. Note: When a chart is saved as a Windows Metafile (*. Excel files preserve data. without any formatting. click Yes.

csv Excel 4.pto *.pwo *. If a number of blank columns or cells appear where they do not belong. check the Options (accessed from the Open dialog) to ensure that the correct delimiter is specified when loading the data. WinNonlin and WinNonMix file compatibility The following table outlines file types that can and cannot be shared between WinNonMix and WinNonlin. WinNonlin and WinNonMix file compatibility File Type ASCII Data Excel Data files WinNonlin *.dat.pto *.prn.lib *.pmo Not available *. *.0 (*.txt.pwo *.xls) Excel 97 (*. *.wgo Not available *.xls) Excel 5.csv Excel 4.dat. check the Treat Consecutive Delimiters as One option on the Options dialog. *. *. *. *.lml Compatibility Identical Identical WinNonMix *.stx *.0 (*.wdo *.xls) Excel 97 (*.wto *.0/95 (*.wmo *.xls) *.pmo *.wto *.rtf *. *. *.0/95 (*.wdo *.Data Management WinNonlin windows and files Column problems 2 If all of the data appear in the first column.lib *.dta. Table 2-2.pco *.wgo *.xls) *.anv Not available Text files Rich text files WinNonlin files: Data Text Graph Model Pharsight files: Data Text Chart Model SAS Transport files ANOVA command filesa LinMix command files Identical Identical Identical Identical Identical NA Identical Identical Identical Not compatible NA Identical NA 17 .xpt. *.rtf *.prn.pco *.dta.anv *.txt.xls) Excel 5.

*. *.cmd (import only) *.exp Not available a.dat. but hidden from the WinNonlin workspace and commands.PMO) include more information than WinNonlin model object files. WinNonMix model object files (*.lml *.dat. *. Hidden windows Windows of all types can be hidden so as not to clutter the parent window. *.dat.dta ASCII data: *. The Hide Window command leaves the active window open. Table Wizard. *.imp.pks Compatibility NA Not compatible Not compatible Not compatible NA NA Identical Identical NA NA Identical NA WinNonMix Not available *.psc. New *. *. and Chart Wizard.dta ASCII data: *. Hidden windows cannot be saved to files or workspaces.gtp *.dat.dta ASCII data: *. Descriptive Statistics.pmo ASCII data: *. 18 .dta Not available Not available *.2 WinNonlin User’s Guide Table 2-2.dta ASCII data: *. *.dta *.ANV files cannot be created. b.dat.imp.exp *.wsc *.tdf *.gtp *. *.pmo ASCII data: *.dat. WinNonlin and WinNonMix file compatibility (continued) File Type Bioequivalence command files Model object filesb Dosing files Model parameters Lambda Z range Partial areas Graph template Table definition files Command files Script files ODBC import and export files PKS files WinNonlin *. *. ANOVA command files can be imported into the Linear Mixed Effects Wizard or the Bioequivalence Wizard in WinNonlin. Hidden windows do not appear in list boxes (for modeling.tdf Not available Not available *. for example).

All workbooks.. If the workspace is new. Workspaces A workspace is a snapshot of all current dose and model settings. Check the window(s) to unhide and click OK.. To un-hide a window: 1. 19 . To save current work to a workspace: 1. text windows.PMO). which are named after the workspace. are saved to a Pharsight Model Object (*. V) from the WinNonlin menus. Note: Pharsight Model Object (*. Choose Window>Hide Window from the WinNonlin menus. The workspace file contains the relative paths to these objects.WSP). WinNonlin may prompt the user for the location of some source data files. if not previously-saved. Unsaved workbook data. are saved to a Pharsight Workbook Object (*.PMO) files may use absolute paths to data files. Choose Window>Unhide Windows. Choose File>Save Workspace (Alt+F. If a workspace containing model information is moved. such as modeling output. Open or select the window to be hidden. then loaded. to be reloaded in future WinNonlin sessions.Data Management Workspaces To hide a window: 2 1. This information can be saved to a workspace file (*. 2. from the WinNonlin menus. charts.PWO). and model settings that have not yet been saved to disk are implicitly saved to separate files. the Save Workspace dialog will appear. and all data object windows that are open in WinNonlin. Model settings. 2.

To save an existing workspace to a new workspace name or location: 1. To create a new. K) from the WinNonlin menus. 3. 3. and start a new. L) from the WinNonlin menus.2 WinNonlin User’s Guide 2. Limit the name to alphanumeric. dash and space characters. Navigate to the appropriate directory and enter a name for the workspace. In the Load Workspace dialog. 20 . Select File>Save Workspace As (Alt+F. WinNonlin will prompt the user to save any unsaved work. Click Save. To load a saved workspace: 1. Click Save. close all windows. Limit the name to alphanumeric. Choose File>Load Workspace (Alt+F. navigate to the appropriate directory and double-click the desired workspace. Choose File>New Workspace (Alt+F. empty workspace: 1. workspace. Navigate to the appropriate directory and enter a name for the workspace. empty. dash and space characters. The Save Workspace dialog appears. 2. W) from the WinNonlin menus. 2.

“Print preview” on page 25 to check print output before printing. The page setup dialog appropriate to the active object appears. “Print” on page 25 to print the current window. Page setup To set page settings for print output: 1. “Print all” on page 27 to print selected items from any open windows. G). See: • • • “Page setup for workbooks” “Page setup for text windows” on page 24 (including model windows) “Page setup for charts” on page 25 Page setup for workbooks Page setup options for workbook windows are specified on four tabs: Page Tab Orientation: Select Portrait (vertical) or Landscape (horizontal) page layout. 21 .Data Management Printing 2 Printing Customize print output using the following features: • • • • “Page setup” on page 21 to set formatting and layout options. Select Page Setup from the File menu (Alt+F.

including the header (top) and footer (bottom) margins.2 WinNonlin User’s Guide Scaling: The workbook can be scaled as a percentage of the normal size or select Fit to in order to make it fit on a certain number of pages. formatting and content for page headers and footers using the codes listed in the tables below and/or text to be printed verbatim. Any entry not preceded with the ampersand “&” will be reproduced as 22 . Paper Size: Select the size paper to use. in inches or centimeters. as selected under Units. Headers and Footers Tab Enter the alignment. Page Numbering: Select Automatic to start numbering pages at 1. Check Horizontally and/or Vertically to center the output on the page horizontally (across). Margins Tab Enter margins for the worksheet. vertically (up and down) or both. or check the Start with page number check box and enter the first page number.

Left-aligns the characters that follow. Special codes for workbook information &F &A &P &D &T &L &C &R &P + number &P . Centers the characters that follow. Prints the page number plus number. Table 2-3. Font codes &B &I &U &S &”fontname” &nn Uses a bold font. Strikeout font. Uses the specified font size . Prints the Current Date.number && Prints the Workbook Name. Table 2-4. Table 2-5. Prints the Worksheet Name. Uses the specified font. Codes must be space-separated. Right-aligns the characters that follow. Alignment codes must precede fontrelated codes.Data Management Printing 2 entered. Alignment codes &L &C &R Left-aligns the characters that follow. Prints the Current Time. Prints the page number minus number Prints an ampersand 23 .must be a two digit number. If the codes are entered in the wrong order WinNonlin may ignore some of them. Uses an italic font. Codes and text are centered unless &L or &R is specified. Special character codes for printing worksheet-specific information such as worksheet titles must appear last. Prints the Page Number. Underlines the header. Centers the characters that follow. Right-aligns the characters that follow.

Page setup for text windows Page setup options for text windows sets the page orientation and margins. Multiple rows or columns may be selected. Separate non-contiguous ranges with commas. Print Options: Set whether or not grid lines and row and column headers will print. Output can be specified to print in black and white. the title is printed at the top of each page. but they must be adjacent. Print Titles: Specify rows or columns for which titles should be printed on each page. Margins: enter the page margins in inches. Special codes for workbook information &N Prints the total number of pages in the document Sheet Tab Print Area: Enter the range of columns and rows to print using absolute or relative cell references. as absolute or relative cell references. Page Order: Specify the print order: Top to Bottom (left columns print before those farther right) or Left to Right (top rows print before the lower rows). so that items with shading or color do not use grayscale (gray “missing” cells will appear white). If a row is entered. the title is printed at the left edge of each page. 24 . Separate non-contiguous ranges with commas.2 WinNonlin User’s Guide Table 2-5. Orientation: Select Portrait (vertical) or Landscape (horizontal) page layout. If a column is selected.

Data Management Printing 2 Page setup for charts Page setup for charts sets chart size and the number of charts per page. Select Best Fit to scale the graph proportionally to fit the page (recommended). 2. depending on the Windows settings. Click the Print button in the WinNonlin tool bar. 4. Make sure that the object to print is the active window in WinNonlin. Select Layout for Printer to expand the scale of the graph to the size and layout of the selected paper orientation (similar to the Scaling function). Orientation: Select whether the graph is to be printed on the page vertically (Portrait) or horizontally (Landscape). 3. Multiple charts per page: If multiple charts are to be printed per page. to specify the top. A Print Preview window appears with the layout from the Page Setup dialog. Once this option is selected. 25 . as charts will be scaled to fit the selected number across and down the page. Layout: Select Screen Layout to print the graph at the size it displays on the screen. the controls for scaling and layout are disabled. 5. bottom. Print To print the active window using the default print settings: 1. Print preview To preview print output: 1. Margins: Enter a value in inches or centimeters. select this check box and select the format from the list box. left and right margins of the page. Click Close to close the window and return to WinNonlin. Click Next to scroll through multiple pages. Select File>Print Preview from the WinNonlin menus. Scaling: Select Actual Size to print the chart at its original size.

Click Preview to view the expected print output or OK or Print. Select the appropriate printer and paper size. Note: To select multiple sheets in a workbook. 2. Select the Print Range: • • The entire worksheet: Select All in the Print Range box. The Print dialog appears. change line colors to black. See also: “Printing workbooks” on page 26 and “Printing text objects and models” on page 26. If lines appear fuzzy in printed plots. select whether to print all worksheets (Entire Workbook) or only the currently selected worksheet(s) (Selected Sheet(s)).2 WinNonlin User’s Guide To print the active window: 1. 2. text). hold down the Ctrl key and click on each worksheet tab. Select the window to be printed. Enter the number of copies to print. use the Page setup dialog first in order to specify multiple charts per page and chart size. Selected Pages: Select Pages in the Print Range box and enter the page number(s) as a comma-separated list. Set options as appropriate for the object type (workbook. chart. 2. Choose File>Print (Alt+F. Printing workbooks 1. 26 . Note: When printing charts. 3. Printing text objects and models 1. P) from the WinNonlin menus. Note: Page numbers are printed for text objects only when the printing is initiated via the print button on the tool bar. 4. Use a hyphen to indicate ranges. Under Print What.

c. (See “Print all options” below. click the Print button in the Print All dialog. Choose File>Print All from the WinNonlin menus. b. Use the Options button to set up the page layout for charts.Data Management Printing 2 Print all To print a selection from the open windows: 1. 2. To print all open windows. Click Print. Print all options Multiple charts per page: Check this box and select the layout below to put more than one chart on a page. including all worksheets and charts. Click the “+” symbols to view individual windows. 3. with page breaks between sets of charts.) 6. 4. 27 . 5. worksheets or charts. To select specific items to print: a. Uncheck Select all objects. Check the box next to the items to print. Open the window(s) to print.

charts. Select the desired drive and directory.2 WinNonlin User’s Guide WinNonlin options WinNonlin provides options to set default file locations. chart labels and tables • disable/enable warning message for impending license expiration Tables Units Licensing Directories To set or change the default model library directories: 1. and/or text with or without page breaks • default orientation of the Final Parameters output table • default method for the calculation of AUC in noncompartmental analyses • treatment of group variables in the table generator • default use of page breaks in tables • display of units with column headers in worksheets. 28 . if it is not already open. 3. 4. Table 2-6. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. Choose Tools>Options from the WinNonlin menus and click on the Directories tab of the WinNonlin options dialog. WinNonlin options Category Directories Options • default directory for WinNonlin Library and User Library files Workbooks • whether to display formulas in worksheets • direction in which the Enter key moves the cursor in a worksheet • value used to denote missing data in worksheets Models • default output: workbooks. The picture of the folder opens. click the Browse button for that directory. To set the default file location for WinNonlin model library or user model library files. worksheet behaviors. Choose Tools>Options from the WinNonlin menus and open the Workbooks tab of the WinNonlin options dialog. 2. and modeling and analysis default settings as described in Table 2-6. Workbooks To set or change default workbook options: 1.

else. 3. 5. 4. 3. Models To set default model options: 1. Figure 2-1 and Figure 2-2 present an example of each layout. Choose the direction of cursor movement between worksheet cells when the Enter key is pressed: check Enter Moves Down to move the cursor down the column. enter it in the box labeled Missing Value. Check Page breaks for paginated output. Note that the formula bar at the top of each worksheet displays any formula in the current cell. in individual worksheet cells. Text: check to include text output listing model settings. This layout is a useful for performing additional analyses. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. 29 . Any cell containing this string will be grayed and treated as missing in (excluded from) modeling and analyses. fitting progress and final results. 2. the cursor will move to the right when the Enter key is pressed. To specify a character or string to represent missing data. such as descriptive statistics. Set the layout for the Final Parameter output worksheet: Checking Transpose Final Parameter Table will present each parameter in a separate column. A second worksheet called Non-transposed Final Parameters will present parameters in separate rows. rather than the values computed by formulas. Check the Show Formulas check box to display formulas. Choose Tools>Options from the WinNonlin menus and open the Models tab in the WinNonlin options dialog. on modeling output. Charts: check to include charts of model data and results. Use the check boxes to select the default form(s) of modeling output: • • • Workbook: check to include tabular results in a workbook.Data Management WinNonlin options 2 2.

2 WinNonlin User’s Guide Figure 2-1. using linear interpolation between log-transformed data points through Clast. Select the default NCA calculation method. This option is provided primarily for WinNonlin scripts that depend on the transposed format the Final Parameters worksheet. If the Transpose Final Parameters option is not checked. Linear trapezoidal (linear interpolation): linear trapezoidal rule. Example modeling output: un-transposed Final Parameters. Linear up/log down: linear interpolation between untransformed data up to Cmax. Linear trapezoidal (linear/log interpolation): trapezoidal rule with linear interpolation up to Cmax. and the second worksheet. Figure 2-2. the Final Parameters worksheet will use the untransposed format. See Section 3. for detailed discussion of AUC computation methods. 30 . log-linear interpolation between data points from Cmax. to be used in computing area under the curve (AUC) in noncompartmental analyses. Example modeling output: transposed Final Parameters.7 of Gabrielsson and Weiner (2001). 4. entitled Transposed Final Parameters. • • • • Linear/log trapezoidal: log-trapezoidal rule. and between log-transformed data from Cmax through Clast. will contain the transposed version. using linear interpolation between untransformed data through Clast. from among the methods below. extrapolated to Tinf.

check Place Page Break Data Outside Table Body. 3. Choose Tools>Options from the WinNonlin menus and select the Units tab of the Options dialog. Tables To specify default table options: 1. Units The display of units in workbooks. 4. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. Note that these settings can be overridden in individual tables and charts. Use the check box to set the default for creating page breaks when the value of the group variable(s) changes. Choose whether to have units displayed on chart labels and/or in tables. To specify default units display: 1. To print group variables outside and above the table. Licensing Be default. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. Units for specific output parameters can be changed for a modeling session (see “Model options” on page 203). 4. or choose File>Options from the Table Wizard menus. Select the Show Units with Column Headers check box to include units in column headings of output workbooks.Data Management WinNonlin options 2 5. 2. Choose Tools>Options from the WinNonlin menus and open the Tables tab. rather than as a column in the table body. 2. Click Apply to set these options and keep the dialog open or OK to keep these settings and close the WinNonlin options dialog. 3. charts and tables is set on this tab of the WinNonlin options dialog. The Licensing tab 31 . WinNonlin will generate a warning message upon launch if your license to use the software is within 30 days of expiration.

b. 3. To allow WinNonlin to produce a warning message on launch within 30 days or license expiration.) To disable or enable license expiration warnings: 1. select Disable. 5. The new setting will apply the next time you launch WinNonlin. Best Available: WinNonlin will attempt to secure an Enterprise license. or the best available from the network license server. select Enable under Warnings. choose the license type that WinNonlin should request from your license server (see “WinNonlin editions” on page 1): a.2 WinNonlin User’s Guide of the WinNonlin options dialog provides a control to disable. delay or enable this warning message. Click OK to apply your selection. 2. 2. The Licensing tab also provides a means for floating license users to specify a preference to obtain a license for WinNonlin Enterprise edition. Under Features. then. (See “WinNonlin editions” on page 1 for information about WinNonlin editions. Enterprise: WinNonlin will only accept Enterprise edition licenses. Choose Tools>Options from the WinNonlin menus and open the License tab of the Options dialog. 32 . To prevent WinNonlin from ever producing the warning. a Professional license. To prevent WinNonlin from producing the license warning for a set number of days (beginning on the day this selection is made). The new setting will apply the next time you launch WinNonlin. c. Choose Tools>Options from the WinNonlin menus and open the License tab of the Options dialog. select Postpone for under Warnings and select the appropriate number of days in the drop-down list. 4. To set a preference for type of floating license: 1. 3. Professional: WinNonlin will accept only Professional edition licenses. Professional. if no Enterprise license is available. Click OK to apply your selection.

Go to the main menu. 1. WinNonlin marks each derived product as out of date. 2. 2.Data Management Data dependencies and origins Installing an IVIVC license 2 Note: Installing an IVIVC License requires the same steps and listed descriptions as those for WinNonlin in the Getting Started Guide. by coloring it pink. follow the instructions in the WinNonlin Getting Started Guide.16. follow the instructions in the WinNonlin Getting Started Guide. You can refresh 33 . WinNonlin locks derived data products—including modeling or analysis output in workbooks. If the source data changes. Click OK. 5. Check the IVIVC Checkbox. pages 11-16. pages 12 and 13. 4. Open WinNonlin. Data dependencies and origins In order to protect the integrity of analysis results. To install a node license key code. 3. WinNonlin must be configured to recognize IVIVC. pages 14 . To install a floating license key code. 1. text windows. Click Tools > Options > Licensing tab. and charts—against changes. After the license is installed.

or. To view dependencies: 1. choose File>Dependencies from the WinNonlin menus. 2. and the model THEOPH_NCA. In the example below. 34 . The Parent View (below) shows the items’ source objects.2 WinNonlin User’s Guide the derived objects based on updated source data. make output independent of the source so that it is unlocked and editable.PWO.PWO (top item) provided the source data for the descriptive statistics results in the workbook THEOPH24_NCA_DSTATS. Its tree shows dependencies in reverse of the Parent View. the noncompartmental analysis output THEOPH24_NCA_OUTPUT. in a tree with source objects as branches. In this example. the modeling output workbook THEOPH24_NCA_OUTPUT. THEOPH_12VS24.PWO. with derived objects as branches.PWO is dependent on its source workbook.PMO. if desired. if any. The Child View shows objects that are derived from a selected item. Select an object from the drop-list at the top of the Dependency View dialog. After an analysis.

Click the Refresh button to view pedigree information for derived objects. See “Refreshing results” on page 35. Make the necessary edits to the source data and/or analysis settings. Click the Origins button to view pedigree information for derived objects.Data Management Data dependencies and origins 2 3. 4. the derived objects become out of date. and save the set to disk as a workspace (see “Workspaces” on page 19) or to the Pharsight Knowledgebase Server as a scenario (see “Using the Pharsight Knowledgebase Server” on page 445). create the desired output. keep all objects open. with dependencies intact (see “Data dependencies and origins” and “Saving dependencies” above). Refreshing results If source data or analysis settings are modified while their derived objects (output) remain open in WinNonlin. 2.” Note: Opening and using the OK button to close any modeling or analysis dialog marks derived objects as out of date regardless of changes to analysis settings. See “Object origins” on page 36. 4. To update output to reflect changes to source data or analysis settings: 1. each output window is shaded and its title bar includes: “Source Changed. Select an output item from the analysis to be refreshed at the top of the Dependency View dialog. Open all input and output in WinNonlin. or when they are saved together in a workspace or PKS scenario. To save objects with dependency information. 3. Choose File>Dependencies from the WinNonlin menus. 35 . Saving dependencies Dependencies persist only while related objects remain open in WinNonlin.

a change to data exclusions in THEOPH_12VS24. In the example above. After running an analysis.PWO and THEOPH24_NCA_OUTPUT. and any derived objects that lie between the source data and the selected item. as above. 2. will update that output and the NCA chart and workbook on which it depends. and statistics on the NCA workbook (THEOPH24_NCA_DSTATS. Selecting the descriptive statistics for refresh. Select the item to unlock at the top of the Dependency View dialog. Click the Refresh button to re-generate all output from that analysis.PWO).PCO). Selecting either NCA output object for refresh would update both NCA objects but not the descriptive statistics. Click the Detach button. making its data editable and removing object dependencies. Unlocking derived data Analysis and modeling output can be unlocked. Object origins The Origins button appears when an object that is dependent on any other object(s) is selected in the Dependencies View dialog. 3. Click this button to view a pedigree for each child object open in WinNonlin. including: 36 . and Refresh will no longer be available.PWO deprecated the NCA chart and workbook output (THEOPH24_NCA_OUTPUT. To unlock a derived product: 1. Changes to the source data will no longer mark this derived work as out of date.2 WinNonlin User’s Guide 5. choose File>Dependencies from the menus.

workbook name for source data For certain analyses. the WinNonlin Log Viewer. Two known cases that are not captured in the History page are filling a range of cells and moving a range of cells.DEP) when the source and derived objects are saved in a workspace. The sheet then logs edits to the workbook. A. The first line always notes creation of the workbook. that provides access to a log file that tracks all data operations performed in WinNonlin. e.Data Management Data history worksheets and log files 2 • • • • Object type and file or window name WinNonlin engine used to create the object. and indicates the number of worksheets. WinNonlin also includes a separate application. PKS operations. modeling or LinMix Source ID: identifier for any source data from which the object derives. This pedigree information is saved as an XML dependencies file (*. all data manipulations.. with the worksheet and. including application of units. if applicable. It cannot be deleted. History worksheet The History worksheet logs operations made on the workbook data. including sorts and transformations. See “WinNonlin Log Viewer” on page 39.g. a report of analysis settings and variables. Data history worksheets and log files WinNonlin provides two means to track data operations. and only the Annotations column can be edited (enter notes here). and the analysis operations listed in Table 2-7. 37 . This file is always saved as part of a PKS scenario. Each WinNonlin workbook includes a History worksheet that logs most operationsA on data from that workbook.

notification of any profile with all zero values for volume or concentration • For NCA.• Source data workbook and worksheet tistics • Missing/excluded values • For weighted analyses. and it occurs at time=0. if only 1 non-zero concentration is found for any profile. notification of weights < 0 (all values for that profile will be missing) • Notification of any floating point error. but no errors: “warnings generated on one or more profiles of data” • Success or failure of the LinMix/BioEq operation Merge • • • • • • • Source data workbook and worksheet #1 Missing/excluded values from source worksheet #1 Source data workbook and worksheet #2 Missing/excluded values from source worksheet #2 Sort variable used from each worksheet Include variables from each worksheet Success or failure of the Merge operation PK and Non• Source data workbook and worksheet compartmental • Missing/excluded values analyses (NCA) • Profiles with less than 2 usable records noted: “cannot be processed” • For urine NCA. and note that all values for that profile set to missing • Success or failure of the LinMix/BioEq operation LinMix and • Source data workbook and worksheet Bioequivalence • Missing/excluded values (BE) • In BE. and the profile number at which each occurred • If warnings generated. notification if any value(s) of the dependent variable <= 0 • Notification of any regressor/covariate found to be alphanumeric • Warnings or errors from the analysis. lambda_z exclusions (or the lack of any) for each profile • Modeling halted using Stop button • Success or failure of the modeling process Toolbox tools • Source data workbook and worksheet • Success or failure of the toolbox engine operation 38 . notification that all parameters set to Missing • For NCA. notification that some calculation(s) couldn't be made • If Ln transformation included.2 WinNonlin User’s Guide Table 2-7. Analysis operations logged in the History worksheet Operation Deconvolution Items logged • • • • Source data workbook and worksheet Missing/excluded values Notification of any failed or invalid profiles Success or failure of the deconvolution operation Descriptive Sta. if a least square mean is <0.

When the log file reaches its size limit. the file is purged to a backup file. including messages of success/failure 39 . choose Programs>Pharsight>WinNonlin 5.2>Log Viewer. not cell-by-cell values) • Changing WinNonlin registry settings • Transform operations • Specifying column units • Clearing units • PK/PD modeling errors • Cut/Copy/Paste operations • Include/Exclude data operations (include/ exclude parameters. • WinNonlin startup and finish • Open/save workbook • Replace/Remove data (replace/remove parameters. This log file size can be specified or changed using View>Options in the Log Viewer application menus. The application log contains entries for the following. the Log Viewer. WinNonlin writes this information to a separate log file. To run the WinNonlin Log Viewer: 1. From the Windows Start menu. not cell values) • Insert column/row • Delete column/row • Undo operations • Units conversions. for reviewing a record of operations performed in the application.Data Management Data history worksheets and log files 2 WinNonlin Log Viewer WinNonlin has a separate application.

2 WinNonlin User’s Guide 40 .

copied. Worksheet operations are covered in the following sections. • • • • “Editing data” on page 41: editing and formatting worksheet data using formulas. and/or pasted via the Windows Clipboard. “Transformations” on page 62: creating new workbook columns as functions of other columns. • “Inserting or deleting worksheets” on page 42 41 . and manipulation in WinNonlin worksheets Input and output data for WinNonlin analyses are stored and manipulated in Excel-compatible worksheets. Every workbook contains at minimum one data worksheet and a History worksheet. “Status code transformations” on page 69: replacing non-numeric codes with specified values. “Data manipulation” on page 56: sorting. Editing data WinNonlin worksheets support spreadsheet-like data entry and formatting. Worksheet editing operations are detailed in the following sections. merging workbooks. or imported (see “Data import and export” on page 75). functions.Chapter 3 Workbooks Data editing. copy and paste and fill-down. excluding and including data. shown as individual tabs within a workbook window. Numerical and text values may be entered directly into worksheet cells. Data can be typed or pasted into a new worksheet. formatting. and cut. which details many of the changes and operations made on the data (see “Data history worksheets and log files” on page 37). and transforming non-numeric values.

Click Yes to confirm deletion of the current worksheet. 3. Open the workbook to the insertion point for the new worksheet. 42 . Open the worksheet to be deleted. To add a new worksheet: 1. Select the worksheet to be named. Double click on the tab for the sheet. Choose Edit>Insert Sheet from the WinNonlin menus. 3. 2. To delete a worksheet: 1. These worksheets cannot be deleted. columns or cells” on page 43 “Using formulas and functions” on page 44 “Filling adjacent cells” on page 46 “Formatting cells” on page 46 “Clearing cell values and formats” on page 49 “Undo” on page 50 “Find. The new worksheet will be inserted in front of the currently-selected worksheet. Choose Edit>Delete Sheet from the WinNonlin menus. The Sheet Name dialog appears.3 WinNonlin User’s Guide • • • • • • • • “Inserting and deleting rows. To name a worksheet: 1. 2. find next and replace” on page 50 “Go to” on page 52 Inserting or deleting worksheets Every workbook must contain at least two sheets: a data worksheet and a worksheet labeled Data History (see “Data history worksheets and log files” on page 37). Other worksheets can be added and deleted. 2. Enter the name to be displayed on the tab and click OK. A new blank worksheet is inserted on top of the current worksheet.

select either Shift Cells Right or Shift Cells Down. Select Insert from the Edit menu (Alt+E. select the same number of rows or columns as the number to be inserted. I) or right-click and select Insert from the shortcut menu. Highlight a cell within the row. I) or right-click the selection and choose Insert from the shortcut menu. Select Delete from the Edit menu (Alt+E.Workbooks Editing data 3 Inserting and deleting rows. 3. columns or cells To insert a row or column: 1. To delete cells: 1. Note: To insert multiple row or columns quickly. A single cell is inserted. From the Delete dialog. 43 . 3. In the Insert dialog that appears. select how to move the surrounding cells. In the dialog that appears. 2. Highlight a row or column by clicking on the row or column label. select Entire Row or Entire Column. 2. 2. Select a cell with the mouse. 3. Select the cell(s) to be deleted. or 1. To insert a cell: 1. then choose Edit>Insert. I). Select Insert from the Edit menu (Alt+E. A row or column will be inserted above or left of the active row or column. D) or right-click the selection and choose Delete from the shortcut menu. Select Insert from the Edit menu (Alt+E. 2.

In contrast. Select Delete from the Edit menu (Alt+E. D) or right-click the selection and choose Delete from the shortcut menu. A1:B4 calls the first four cells in the first two columns (starting from the top left cell in the worksheet). Select the row(s) or column(s) to be deleted by highlighting the row or column label(s). A “$” makes a reference position absolute. using row number and column letter. Using formulas and functions Worksheet values may be entered as mathematical formulas. to create a new column of values equal to the difference between values in columns B and A. D). “=. to create a variable equal to the value in column B minus the value of cell A1. you would cell C1 = B1-$A1. To identify a range of cells. For example. or 1. rather than relative to the currently-selected cell. then use the fill-down feature to paste this formula to all cells in column C. C3 = B3A3. 2. etc. For example. C2 = B2-$A1. Pasting this formula down column C would cause C1 = B1-$A1. Select Delete from the Edit menu (Alt+E. See Table 3-1 for a list of valid operators. 2. enter the top-left and bottom-right cells. Cell references Formulas and functions refer to worksheet cells by location in the worksheet grid. you could set cell C1 = B1-A1. by beginning the cell’s value with the equals sign. Cell C2 would equal B2-A2.” This is useful for creating new columns as a function of other columns. separated by a colon. From the Delete dialog. Select a cell in each row or column to be deleted. C3 = B3-$A1.3 WinNonlin User’s Guide To delete rows or columns: 1. select Entire Row or Entire Column. 44 . A complete listing and syntax are provided in “Worksheet Functions” on page 551. for copy and paste operations. 3. A3 denotes the third cell in the first column. For example. WinNonlin worksheets also support an extensive list of functions. etc.

45 . Using the mouse. Press the Enter key to enter the formula into the cell.. select the first cell in the column of the new variable. Select the cell and position the cursor over the small black square at the lower right hand corner of this cell until the cursor changes to a small black cross. either type them or click on the desired cell (drag to select multiple cells) to insert the reference into the formula.Workbooks Editing data 3 Table 3-1. type the formula. including the operator and an expression representing the value. e. 2. =SQRT(A1) or =A1+5. Workbook operators Operator + / * ^ % = > < >= <= <> Description Addition Subtraction Division Multiplication Exponentiation Percentage Equal to Greater than Less than Greater than or equal to Less than or equal to Not equal to To create a new variable using a formula: 1. In the edit bar at the top of the worksheet. 4. 3. To enter cell references.g.

2. Filling adjacent cells A value or formula from one cell can be filled to adjacent cells up. or to the left or right. including the cells to be copied as the left-most cells in the selection.3 WinNonlin User’s Guide 5. To copy to adjacent cells using the mouse: 1. Select Edit>Fill>Down. 4. Select the cell to copy. Edit>Fill>Up. Press and hold the mouse button. The Format Cells dialog has five tabs for specifying the following. 3. Position the cursor over the small black square at the lower right hand corner of this cell until the cursor changes to a small black cross. Edit>Fill>Right or Edit>Fill>Left from the menus. Select the cells to be filled. Formatting cells Choose Data>Format (or click the Format tool bar button with a workbook in the active window) to apply formatting to the currently-selected worksheet cell(s). and drag the cursor down the column to copy the formula. 46 . and press the Enter key. 2. down. Press and hold the mouse button. If necessary. enter the value or formula to copy. To copy to adjacent cells in a selected block of cells: 1. and drag the cursor down and/or to the right over the target cells to copy into them.

The Format Cells dialog appears.Workbooks Editing data 3 • • • • • Note: Number format Alignment Font face and format Borders Fill color and pattern Data formatting is not saved in ASCII (text) files. Number format The Number tab of the Format Cells dialog sets the handling of numbers. The Sample field will display an example. click on the row or column header(s). Syntax in the Type box is as follows. and indicates whether thousands are separated by commas. To select whole rows or columns. • • • • 0 (zero) represents the a digit or exponent and decimal places. # represents digits to the thousands. 4. Select the cells to format. Type: Select a format here. Select Data>Format from the WinNonlin menus or click the Format tool bar button. 2. Symbols appearing after a semicolon indicate handling of negative numbers. An underscore “_” at the beginning of symbols indicates left-alignment of left parentheses. 47 . Click OK to apply the changes or Cancel to close the dialog without changes. To set the format for a cell or selection of cells: 1. An underscore “_” at the end inserts a space. if any. Category: Select a number category or All to show all types. Choose format settings from the tabs detailed below. in the selected cells. 3. to align positive values with negative values enclosed in parentheses. but not non-numeric entries.

but using scientific notation for numbers >= ten billion. Right: right-aligns cell contents. ?/? indicates display as fractions rather than decimals. s represents seconds. Choose Category = All and Type = General to display entries as they are typed in. Merge Cells: joins selected cells into a single cell that spans multiple columns and/or rows. or bottom of each cell. 48 . Center across cells: centers content in the left-most cell across a multicolumn selection. Fill: repeats cell contents enough time to fill the width of the cell. wraps text onto multiple lines to fit the cell width. d represents days. Vertical: aligns content at the top. assuming adjacent cells to the right are empty. Horizontal: alignment from left to right can take the following values: • • • • • • General: left-aligns text and right-aligns numbers.3 WinNonlin User’s Guide • • • • • • • • Note: * indicates that zero values will be replaced with a dash. Wrap Text: when text content is longer than the cell width allows. Center: centers cell contents. center. and merges cells. E indicates scientific notation. Left: left-aligns cell content. m represents minutes for times or months for dates. Alignment The Alignment tab of the Format Cells dialog sets alignment and line wrapping to the selected cells. h represents hours on a 24-hour clock. y represents years.

49 . blue lines between cells within the selection. First. see “Inserting and deleting rows. red line around the outermost edges of the block of selected cells. columns or cells” on page 43. but the cells remain in the grid. from cells: 1. To delete a cell from the grid. select a line style and color. the values and/or formats are deleted. and thin. The example below will put a thick. Borders Use this tab to create borders around the selected cell(s). but not formatting. color and format to the selected cell or cells. use the buttons or click between the marks in the diagram below to indicate border locations. To clear data values. Fill color and pattern Use this tab to set background color and pattern for the selected cell or cells. Clearing cell values and formats When cells are cleared. Select the cells to be cleared. Next.Workbooks Editing data 3 Font face and format Use this tab to assign a font face.

Select Edit>Clear>All from the menus. Undo will appear as a normal selection on the Edit and shortcut menus. U).1. 2. use the keyboard shortcut Ctrl+Z or right click in the window and select Undo from the shortcut menu. right-click in the worksheet and select Clear Values from the shortcut menu. To clear data values and formats from cells: 1. Undo The Undo capability allows the single. or press the Delete key. when editing workbooks. Select the cells to be cleared. Choose Edit>Clear>Values from the menus. 3.1 to 0. This will delete the information from the cells. Alternately. The following example will find concentration values from -0. but will not remove the cells. f or Ctrl+f). Select the appropriate options. charts or text windows. Choose Edit>Find from the WinNonlin menus (Alt+e. Find. When an operation has been performed that can be undone. Select Undo from the Edit menu (Alt+E. To clear formatting from cells: 1. find next and replace To locate specific data values within a worksheet: 1. Select the cells to be cleared. The Undo command is disabled if the most-recent action cannot be undone. Enter a numeric or text search string in the box labeled Find. below. most-recent action to be reversed. 2. To undo an operation: 1. Select Edit>Clear>Formats from the menus.3 WinNonlin User’s Guide 2. Alternately. 50 . 2.

– Less than or equal to the search string (<=). 4. Select Replace from the Edit menu (Alt+E. N). 51 . Search Area Allows searching over the entire data set. To repeat the designated search: 1. Operation This option searches for a value: – Equal to the search string (=). Click Find. Tolerance Finds values not different from the (numeric) search string by more than the tolerance (optional). The Replace dialog appears. the column Conc is one potential search area because the a cell in that column was selected before opening this dialog. To replace text or data in a grid: 1. the currently-selected column or a selection of cells. – Greater than the search string (>).Workbooks Editing data 3 Case Sensitive Checking this box causes only occurrences with the exact combination of uppercase and lowercase letters to be found. Select Find Next from the Edit menu (Alt+E. In this example. – Greater than or equal to the search string (>=). or – Not equal to the search string (<>). – Less than the search string (<). E).

enter: A3. The example above will replace all concentrations with a value of “BQL” with zero. Select any worksheet in the active workbook under Sheet. Enter a numeric or text replacement string in the box labeled With. with one addition: Remove Row Matching Criteria Check this box to remove any row that contains the search string within the search area. Note that. values of “bql” will not be replaced. 52 . Enter the grid location of the cell. Click Replace to review each occurrence before replacement or click Replace All to replace all occurrences without reviewing each. To move to a given cell using a cell reference: 1. Options are the same as for Find. The Goto dialog appears. 2. 3. since the Case Sensitive option is checked. 4. 5. Click Goto. 3. Choose Data>Goto from the menus (Alt+D. Go to The Go to function jumps to a specific cell in any worksheet within the current workbook. 4. above.3 WinNonlin User’s Guide 2. for the third cell (3) in the first column (A). Enter a numeric or text search string in the box labeled Replace. Select the appropriate options. For example. G).

Nonstandard units—units enclosed in braces—will be carried throughout any operations. otherwise. 5. The new column header appears in the worksheet. Note: WinNonlin does not allow spaces in column headers. If a data set is loaded that has column names with spaces WinNonlin will ask to substitute an underscore (_) for the spaces (and offer the opportunity to save the data set to a different name). in parentheses. Press Enter or click Close. Note that long column names can affect performance. 2. Each column in a worksheet can use different units. WinNonlin will display them in braces: {furlongs}. The units appear in the column header. Click the Edit Header button. Highlight a cell in the column to be named. 6. Highlight the column to be named. 4. 3. Type the name. Permissible units include the following. for multiple transformations. See “WinNonlin options” on page 28 for units display and default options.Workbooks Column names and units 3 Column names and units The Column Properties dialog sets column headers and units for the current worksheet or. the data cannot be used in analyses and models. Click the Column Properties tool bar button or double click a column header to open the Column Properties dialog. and will be used in data calculations. To specify column headers: 1. If the units are not standard or recognizable. 53 . but not used in any calculations. for the new worksheet that will contain the transformed data (see “Multiple transformation” on page 64). such as descriptive statistics.

To enter units in the Column Properties dialog: 1. Unit types Unit type Time day hour minute second millisecond microsecond nanosecond Mass gram mole pound IU Volume Liter Unit hr min s ms us ns g mol lb IU L Abbreviation day Prefixes can be used for units of mass and volume. Prefixes Prefix femto pico nano micro Abbreviation f p n u Prefix milli centi deci deca kilo Abbreviation m c d dk k Note: Units are not case-sensitive. 54 .3 WinNonlin User’s Guide Table 3-2. Table 3-3. 2. Select the column name in the column header field. Enter the units for that column in the New Units field.

The Column Properties dialog appears with the column selected. To convert units: 1. 4. See “Units builder” on page 55. Enter the new units in the Specify Units field. The Units Builder assists with units selection. The current units appear in the list box. The data in the column are converted to use the new units. 2. Select a cell in the column to operate on. 55 .PWO). Converting units Units can be converted automatically using the Column Properties dialog. Select a cell in the column to be converted. Click the Convert Units button. Units builder Units are edited in either the Column Properties dialog or the Units Builder. Note: Worksheet units can be saved only in Pharsight Workbook Objects (*. To clear the units associated with a column: 1.Workbooks Column names and units 3 3. Click the Specify Units button. Click Close to close the dialog. 3. 4. The Column Properties dialog appears with the column selected. Click Close to close the dialog. 4. Click the Clear Units button. Other file types lose unit assignments when the file is closed. Click the Column Properties tool bar button. Click the Column Properties tool bar button. The current units appears in the list box. 2. 3.

Click the Specify Units button to associate the units with the data column. 5. 2. Click OK to close the Units Builder. With the data open in WinNonlin. for example). click the Column Properties tool bar button or select Data>Column Properties from the WinNonlin menus. The appropriate abbreviation displays in the New Units field. 9. enter the units and operators in order. etc.3 WinNonlin User’s Guide To use the Units Builder: 1. 8. Prefixes (kilo. milli. The Units Builder dialog appears. or Volume) and the appropriate unit. 4. Data manipulation WinNonlin worksheet operations include the following: • • • • “Freezing rows and columns” on page 56 “Sorting” on page 57 “Merge” on page 57 “Data exclusion and inclusion” on page 60 Freezing rows and columns Freezing one or more rows and/or columns creates a non-scrolling region that remains on the screen while you scroll through the remaining data. 7. Mass. Click Close to close the Column Properties dialog. Once the unit (with appropriate prefix) displays correctly click the Add button to the right of that kind of unit. 56 . Select the column or row. 6. 3. or one cell at the outer-most row/column to freeze.) can be selected from the Prefix field. In the Column Properties dialog click the Units Builder button. Select the kind of unit (Time. To build units with operators (grams/ml. To freeze a row or column: 1.

Repeat for any other sort variables. The Sort dialog appears. Select a variable name and drag it to the Sort box. To merge data sets: 1. 4. Note: Exclusions are ignored when merging data sets. Sorting Sorting provides a quick means of ordering worksheet data. the sort variables are listed here. 2. Select the sort order (Ascending or Descending) and click OK. Open the worksheet and choose Tools>Sort from the WinNonlin menus or click the Sort button on the tool bar. 3. or highlight the variable then type the letter S while holding down the SHIFT key.Workbooks Data manipulation 3 2. If the data have been sorted previously. Choose Window>Freeze Panes from the WinNonlin menus. To unfreeze rows and/or columns: 1. 57 . Merge The Merge command joins selected data from any two worksheets into a new workbook. To sort the data in a worksheet: 1. Choose Window>Unfreeze Panes from the WinNonlin menus. Open the two data sets to merge in one or two workbook windows.

The Merge dialog appears. Drag the corresponding variable for data set 2 from its variables list to its sort variables list. It is divided into two areas. Check Carry along data for similar sort keys if you wish to fill-in empty cells for given sort level with the last-available value. on the right side of the dialog. For data set 1. These variables will appear in individual columns of the merged data set. drag variables to be copied to the output workbook from the Variable Collection box to the Include Variables box. If a secondary sort variable is needed. repeat. the name from data set 1 will be used as the column header in the merged data. Select the appropriate worksheet for each data set under Sheet1 and Sheet2. Be sure that the corresponding variables appear in the same order for each data set. drag a variable on which to join records from the data set 1 variable collection to its sort variables list. They need not match in data sets 1 and 2. 8. selected below. Select Tools>Merge from the WinNonlin menus. 5. 3.3 WinNonlin User’s Guide 2. Click Merge. An example is provided below. 7. For each data set. 58 . Select the workbook containing each data set from the pull-down menus for Dataset 1 and Dataset 2. 6. 4. The merged data records will be matched on values of the sort variables. The merged data set appears in a new workbook. If the names differ. one per data set. This applies in the case that one data set has more observations than the other for a given sort level.

1 0.35 0.76 Conc 2. Carry along data for like sort levels The following example shows the effect of selecting to carry along data for like sort levels when merging data.1 0. Merged data set without carry along for like sort levels Subject A A A A Time 0 5 5 10 0.76 Effect 0.1 0.27 0.98 1.95 59 . Merged data set with carry along for like sort levels Subject A A A A Time 0 5 5 10 Conc 2. Table 3-4.Workbooks Data manipulation 3 The settings in this dialog can be saved and re-loaded for later use by way of the Load and Save buttons.27 0. Note that the two datasets must be selected before loading settings.27 1. and data set 1 or 2 must contain the variables used in the file.35 0. the other two are Include Variables.95 Table 3-5.35 0.98 1. Subject and Time are sort variables. Data sets to merge Data set 1 Subject A A A Time 0 5 10 Conc 2.76 Data set 2 Subj A A A Time 0 5 5 Effect 0.98 1.95 Table 3-6.27 Effect 0.

3. all data are merged or transformed. For example. I. E. you can exclude all but the desired subject and refit the data. C)..3 WinNonlin User’s Guide Data exclusion and inclusion Specific cells. 2. The selected region is highlighted in red to indicate that it has been excluded. Excluding and including a selected range of cells To exclude a selected range of cells: 1. 2. E. To re-include a selected range of cells: 1. Select entire row(s) or column(s) by highlighting the row or column label(s). Select excluded cells to be re-included using the mouse. The Exclude dialog appears. Excluding and including data based on criteria To exclude data: 1. Exclusions are ignored in merges and transformations.. The data selections are made using the Exclude and Include commands on the Data menu. Select Data>Include>Selection from the WinNonlin menus Selection (Alt+D. Select Data>Exclude>Selection from the WinNonlin menus (Alt+D. 60 . S). Select Data>Exclude>Criteria. The selected region now is no longer marked in red. if only a single subject must be refit from a dataset holding many subjects. Note: Excluded cells are treated as missing data in analyses. rows and/or columns in a worksheet can be excluded from analyses. from the WinNonlin menus (Alt+D. 2. Enter the value(s) to search for in the Find field. Select entire row(s) or column(s) by highlighting the row or column label(s). Select the cells using the mouse. Select any desired options from those detailed below. S).

– greater than the search string (>). Case Sensitive Checking this box causes only occurrences with the exact combination of uppercase and lowercase letters to be found. To exclude each row containing the search value in the search area. – greater than or equal to the search string (>=). Click OK. or – not equal to the search string (<>). Tolerance A value not different from the search string by more than the tolerance will be treated as a match (applicable only when searching for numeric data). In the example below. or the selected cell(s). – less than or equal to the search string (<=). – less than the search string (<). Criteria Using this option search for a value: – equal to the search string (=). un-check Exclude Entire Row Matching Criteria. 61 . check this option. 4. Exclude Row Matching Criteria To exclude cells containing the search value. a single column. the column Time is shown as a potential search area because Time was the active column when the dialog was opened.Workbooks Data manipulation 3 Search Area Allows searching over the entire data set.

All data in the source column are transformed regardless of exclusions. then click OK. Select Data>Include>Criteria from the WinNonlin menus (Alt+D. See the examples below. for multiple transformations (see “Multiple transformation” on page 64). C). WinNonlin will use the first as baseline. Select Data>Reset Selections (Alt+D.3 WinNonlin User’s Guide Re-including data To re-include specific data: 1. To remove exclusions from the current worksheet: 1. which are carried into the new column. Other worksheets in the same workbook will not be affected. Select the inclusion criteria from those described for exclusions under “Data exclusion and inclusion” on page 60. 62 . In baseline transformations. WinNonlin will use the value corresponding to Time = 0 as baseline for each profile. If a profile has more than one value at time 0. or in the destination worksheet. Individual profiles are identified using one or more sort variables. N). 2. I. Transformations A transformation creates a new column in the current worksheet. Resetting selections Resetting the data selections clears all exclusions in the active worksheet. WinNonlin worksheets support the following transformation types: Natural log Log base 10 Square root Absolute value Change from baseline Percent change from baseline Ratio versus baseline X·Y X/Y X+Y X-Y X*Y ex 1/X Xn X*n X/n X+n X-n where X and Y are columns in the current worksheet and n is a number.

It will also sort the worksheet on the sort variables. Select the variables to be used from the pull-down menus for Column x. Select the position for the new variable in the worksheet: – Adjacent places the new variable to the right of the 'Column x' variable.Workbooks Transformations To transform a column. or No to calculate change from a single baseline value for the data set. enter a value for n in the box labeled n. Sort variables define unique profiles. The example below creates a new variable called Ln_Conc. For baseline transformations. Open the data and choose Tools>Transform from the WinNonlin menus or click the Transform tool bar button to open the Transform Data dialog. to compute change from baseline separately for each profile. creating a new column in the current worksheet: 3 1. Other data may require more than one variable. 3. Replace [New] with a column name for the new variable. 7. a single variable representing subject IDs suffices. In most cases. Click Yes to choose sort variables. 63 . 8. such as Subject and Period. If sort variables are included: In the Sort dialog. The example below creates a new column to the right of the BP column. – Append places the new variable after the last variable in the data set. and places it in a new column at the right of the existing worksheet columns. WinNonlin queries whether to specify sort variables. 4. Time and Column y. as needed for the transformation type. drag variables that identify individual profiles to the sort variables list. Select the desired transformation from the pull-down menu for Transform. 2. in the order presented here. 5. If required. WinNonlin will seek a baseline value (at time zero) for each unique combination of sort variable values. 6.

Click OK to complete the transformation and create the new variable. Supported operations include transformation of multiple columns. custom equations and rank transformations. The results appear in a new workbook.3 WinNonlin User’s Guide This example is from a crossover study. as well as search/replace/delete. data inclusion/exclusion. The transform will seek a baseline value at time zero for each unique combination of values for Subject ID and Formulation. You may specify column headers and units for the new worksheet as part of the multi-transform settings. 64 . Multiple transformation The multiple transformation feature creates a copy of a source worksheet with specific data operations applied. 9.

under Destination Workbook Name and Destination Sheet Name. Note: Use the Save button to capture all settings in XML format for later re-use. After adding transformations. To set up transformations. click the Add button and follow the instructions for the Transform action dialog. 4. To delete one or more transformation operations. 5. 2. then click the Delete button. To copy and transform data in an existing worksheet: 1. from the WinNonlin menus.Workbooks Transformations 3 All operations listed under Transforms in the Multi-Transform dialog will be applied to a copy of the source worksheet data. 3. Enter a name for the new workbook and worksheet. which will contain the results. 65 . use Add>Column Properties to set the column names and/or units for new columns. Select the workbook and worksheet containing the data to transform under Source Workbook and Source Sheet in the Multi-Transform dialog. 6. Open the data to transform in WinNonlin and choose Tools>Multi-Transform. check the box next to the transform(s)...

excludes. include/exclude or search/remove operation to the output of a Multiple transformation: 66 . equation manager” on page 67 Column Sets the column headings and units for the destina. removes or replaces it in the on page 66 destination sheet. Transform actions Action Description See: Add Filter Searches the source data for a specified value and “Multi-transformation filter” includes. To change an operation. “Multi-transformation using a formula or function. on page 53 Rank Creates a new column in the destination sheet as a “Rank transformations” on rank transformation of a source worksheet column. Click OK to create the new workbook.3 WinNonlin User’s Guide 7. 8. page 68 Transform Creates a new column in the destination sheet as a “Transformations” on transformation of a source worksheet column. select one of the following actions in the Transform Action dialog and click OK to proceed. Add Equation Creates a new column in the destination worksheet.“Column names and units” Properties tion worksheet. check the box next to that transformation (and no others) and click the Edit button. Transform action To add an operation to a Multiple transformation. page 62 Multi-transformation filter This dialog applies a search/replace. Table 3-7.

enter a value under Tolerance. 6. Check Case Sensitive to search for strings that match not only letters but also capitalization. Select the Search Area: the whole source worksheet or a specific column. Multi-transformation equation manager This dialog creates a new column in the destination worksheet of a Multiple transformation using a formula or function of columns in the source work- 67 . 2. For Replace operations. Enter the value or string to search for in the topmost field. To search for values within a range of a numeric search value. enter the value with which to replace the search string under With. – Include/Exclude: to (re-)include or exclude specified values (or rows containing them) in the destination worksheet. Select the appropriate operation(s) to find values equal to and/or less or greater than the search value. 5. Select the desired action under Operation: – Replace: to find a value (or range) and replace it with another. See also “Data exclusion and inclusion” on page 60.tolerance of the search value will be treated as matching the search value.Workbooks Transformations 3 To filter data values to appear in the destination worksheet: 1. Click OK to add the operation to the Multiple transformation. 7. Numeric values that are +/. – Remove: to delete rows containing specified values. 4. 3.

See: • • “Using formulas and functions” on page 44 for instructions on entering formulas and functions. Group By: WinNonlin will generate a separate set of ranking values for each unique value of the Group By variable. Rank transformations This dialog creates a new column in the destination worksheet of a Multiple transformation as a rank listing of a column from the source worksheet.3 WinNonlin User’s Guide sheet. Descending order ranks the highest value as 1 and increments by one for each smaller value. and increments by one for each larger value. and “Worksheet Functions” on page 551 for functions and syntax. Order: Ascending order ranks the lowest value as 1. using Add>Column Properties from the Multi-Transform dialog. Set the column name and units separately. Enter the formula or function in the field at the top of the dialog. 68 .

tabulation. Representing such a concentration as 0 (zero) is not always appropriate. Some common substitution rules for BQL samples are presented in the table below. One possible substitution rule for PK analyses is presented below: Table 3-9. In general. a substitution rule is defined by the list of possible nonnumeric representations for concentration and the values to be substituted for each. Standard laboratory procedures may require that such data be reported as a character string rather than a numerical value. and summarization are problematic when the data contain concentrations that fall below the lower limit of quantification (LLOQ) of the assay. as doing so can introduce a statistical bias or misrepresentation in some analyses.Workbooks Status code transformations 3 Status code transformations Data analysis. Common status code substitution rules Conditional substitution Use data for: Nonnumeric code BQL BQL BQL 0 "BQL LLOQ/2 0 " Unconditional substitution Before Tmax 0 After Tmax Missing First of consecutive after Tmax LLOQ/2 After first of consecutive after Tmax Missing PK analysis Summary statistics Listing of individual data Plotting of individual BQL data on log-scale Plotting of individual BQL data on linear scale The substitution rules are typically defined in standard operating procedures or method sheets for a given company or department. Possible substitutions for PK analysis Non-numeric code BQL Unconditional substitution Before Tmax After Tmax 0 Missing First of consecutive After first of conseafter Tmax cutive after Tmax LLOQ/2 Missing 69 . Different substitution rules are required depending on the intended use of the data. and substitution rules vary between companies and departments. Table 3-8. plotting. The rules may become more complex when one wishes to make a distinction between concentrations that are below the quantification limit (BQL) and below the detection limit (BDL).

Open the data set in WinNonlin. or identify an existing column containing the LLOQ for each observed concentration.3 WinNonlin User’s Guide Table 3-9.. The Status Code Transformation Wizard creates a new workbook based on an existing one. 70 . Choose Tools>Status Codes (e. Possible substitutions for PK analysis (continued) Non-numeric code NS ISV NA … Unconditional substitution Missing Missing Missing … Before Tmax After Tmax First of consecutive After first of conseafter Tmax cutive after Tmax WinNonlin provides a tool for transforming observation values from nonnumeric status codes to number values for use in analyses and plots. The user may specify an LLOQ and the new column header into which it will be placed. can be used in defining how status codes are transformed. Numeric concentration data and the non-numeric concentration status codes may be the same or different columns of the data set. The LLOQ.. in the WinNonlin menus to launch the Status Code Transformation Wizard. BQL). It transforms values as specified by user-defined rules. if any. The wizard also stores information about the lower limit of quantification (LLOQ) for one or more sampling assays. To select data to be transformed: 1.g. 2. by copying over selected columns..

Time Variable: column containing times for the data to be transformed. • 5. The substitutions may apply unconditionally across the data. as follows: • • • • • Sort Keys: each unique combination of sort key values defines one profile from which T-max will be calculated. Drag and drop columns from the Variables list to assign the variables to be included and that to be transformed. Each rule set will generate a new column of transformed observation data. The new columns will use the units (if any) of the original concentration data. Click Next to proceed to the Rule sets. Rule sets Up to three rule sets may be included in the Status code transformations. 4. or conditionally depending when the data point occurs relative to Tmax. 71 . a fixed LLOQ value. Status Code Variable: (optional) column containing the non-numeric status codes associated with each datum in the Concentration Variable (above). alternately. and whether two or more consecutive observations of the same code occur. A rule set includes a list of non-numeric codes and the value substitutions to be made when each code is encountered in the data. Concentration Variable: column containing the data to be transformed.Workbooks Status code transformations 3 At any point in this wizard. Carry Alongs: variables to be copied into the output data set. the settings and transformation rules may be loaded or saved from/to a file using the Load and Save buttons. LLOQ Variable: (optional) column containing the lower limit of quantification (LLOQ) associated with each measurement in the Concentration Variable. 3. set by checking Static Value and entering the value under LLOQ Value. will determine transformation of those concentration values. Make sure the appropriate data set is selected under Data Set.

72 . 3. To save a rule set to an external file. The unidentified codes will be listed in an alert dialog. Repeat to create up to three rule sets. Each rule set will generate a new data column in the output.3 WinNonlin User’s Guide To set up the status code transformation rules: 1. stored locally. Review the summary of the transformation settings. To save a rule set to the PKS as a system object. Use the Back button to make changes. 5. click Save. check PKS Load and click Save. 2. The wizard will detect the presence of unidentified non-numeric codes in the specified column. and click the Finish button to create the new workbook. and follow the instructions under “Creating a rule set” on page 73 to create a new rule set. Click Next to proceed to the final dialog of the wizard. 4. Click the Add button under Rule Sets.

the applied rule(s) will be stored with the scenario Creating a rule set Status code transformations can include as many as three rule sets. check PKS Load and click Load. Enter any descriptive text under Rule Set Description (optional). 73 .Workbooks Status code transformations 3 Note: If code-transformed data are saved to the PKS as a scenario. Unconditional Substitution: if the transformation is the same regardless of whether the status code occurs before or after Tmax. under NonNumeric Code. click Load. then enter the value to which the status code will be transformed here. Conditional Substitutions: if the transformation depends on the status code’s position relative to Tmax and/or whether the code is repeated in consecutive observations. Enter a name under Rule Set Name. enter the following: a. stored locally. 2. To import a rule set from an external file. To define a rule set: 1.) It is possible to load rule sets from a local file or the PKS. To load a rule set from the PKS. then leave Unconditional Substitution blank and enter all four of the following: – Before Tmax: value to assign if the status code occurs before Tmax. For each code. 3. b. This name will be the default name for the column generated by this rule. Enter each status code to be transformed in a separate row. (See “Status code transformations” on page 69. or is repeated.

this value will be assigned to the first of those contiguous observations. this value will be assigned to all but the first of those contiguous observations. See “Rule sets” on page 71. Yes indicates that when numeric values less than the LLOQ are encountered. – First of consecutive after Tmax: if the status code appears in two or more contiguous observations. and the variable LLOQ. 74 . One row or zero rows may be set to Yes. The LLOQ value is set in the first dialog of the Status Code Transformation Wizard. Syntax for the rules: The status rules are not case-sensitive. those values will be transformed as though they contained the status code in that row. after Tmax. if a LLOQ value or data column was assigned (see “Status code transformations” on page 69). above. The values entered may include numbers. operators. Click OK when the rule set is complete to return to the Status Code Transformation wizard. Use when < LLOQ: click to toggle between Yes and No. after Tmax. c. – After First of Consecutive after Tmax: if the status code appears in two or more contiguous observations. 4.3 WinNonlin User’s Guide – After Tmax: value to assign if an isolated (not more than one consecutive) status code occurs after Tmax.

“SigmaPlot export” on page 85: export data. (optional) run a script and (optional) return script output to WinNonlin. charts and worksheets into Word text. Oracle.) WinNonlin also exports data to several software packages. WinNonlin Enterprise includes a Software Development Kit (SDK) for building custom ODBC query tools. The Enterprise edition of WinNonlin can. in addition.” on page 14. (optional) run a script and (optional) return script output to WinNonlin. “SAS Transport file export” on page 78: save data in SAS-ready format. “NONMEM export” on page 80: export data and dosing to NONMEMcompatible files. “S-PLUS export” on page 86: export data. as follows. etc. • • • “ODBC import” on page 87 “ODBC export” on page 102 “The Enterprise Software Development Kit” on page 107 75 .) For example. and can re-import S-PLUS and Sigma Plot script output. WinNonlin Enterprise can read rows of PK data from an ODBC-compliant database. See the following for details. text and graphics file types for convenient interchange with other applications. and save analysis output back to a table in that database. (See Table 2-1.Chapter 4 Data import and export Sharing data with other applications WinNonlin opens and saves a variety of data.g. “Supported file formats. SAS. graphics and tables.. • • • • • “Microsoft Word export” on page 76: automatically transfer selected text. exchange data directly with Open Database Connectivity (ODBC) compliant databases (e.

It can also automatically export modeling or NCA output. open that document in Word. if any. Open the objects and choose File>Word Export from the WinNonlin menus. After completing model setup. uncheck Select all objects in the Word Export dialog as shown below. Worksheets are exported as Word tables. click the Start Modeling and Export to Word tool bar button. To select specific objects for export. open the worksheet. To export the active object to a new or existing Word document: 1. 76 . they cannot be edited in Word. To export modeling output to Word automatically: 1. create the selected WinNonlin output and export all output to Word. To export selected objects to a new Word document: 1. See “Using the Pharsight Knowledgebase Server” on page 445. WinNonlin will launch it and export the active window to a new document. 2. 2. If Word is not open. Microsoft Word export WinNonlin can export any or all open windows to a Microsoft Word document. Apply any chart formatting needed. chart or text window to export.4 WinNonlin User’s Guide WinNonlin’s Enterprise edition integrates with the Pharsight Knowledgebase Server to provide secure data storage. Click the Export to Word tool bar button. Charts and model objects export as Windows bitmaps. charts cannot be edited after export. or append the data to the end of the open document. a useful feature for automated analyses. In WinNonlin. thus. 3. WinNonlin will launch the analysis. To append data to an existing document. Text windows become text in Word’s “normal” paragraph style. management and security.

Checking a window or window type will export all objects within that window or window type. Orientation: Portrait (long side of the page is vertical) or Landscape (long side of the page is horizontal). Word export document options The following options apply to Microsoft Word export using File>Word Export in the WNL menus. 4.Data import and export Microsoft Word export 4 3. format. Click the Export button. Check the objects to export. below the chart picture. Text: adds a heading above the text that reads. 5. WinNonlin will launch Word.DOC file. as detailed under “Word export document options”. “Text -” followed by the file path and name (or window name. • • • Charts: adds the chart name and absolute path. and export the selected objects to a new . file name and most-recent save date for the chart. 77 . Add source line to objects: check this option to include text describing the source file and object name to each object exported as follows. Workbooks: adds the worksheet name and absolute path. headings and object numbering. if unsaved) in square brackets. file name and save date for the data. if needed. Click the + markers to expand the lists of windows and objects as needed. 6. Use the Options button to set options for page layout. below the data table.

Name the file and click Save. select Save As from the File menu. Chart format: export charts as Windows metafiles or bitmaps. Because SAS field names are more restrictive than WinNonlin column names. To export data to SAS (via SAS Transport files): 1. Word export workbook options Preserve formatting for tables: includes formatting for workbooks created in the WinNonlin Table Wizard. Select SAS Transport (*. The Save File dialog appears. Check this option and enter a number for the first chart to include “Figure N” above each chart. with page breaks between sets of charts. WinNonlin will request new names for any that would be invalid in SAS. Start figure numbers at N: adds figure numbers to charts. 3. This option can slow the export significantly. Each log chart appears below the original linear chart.4 WinNonlin User’s Guide Word export chart options Multiple charts per page: Check this box and select the layout below to put more than one chart on a page.xpt. 78 . Each worksheet is one table in the Word export. Add log charts: creates and exports a chart with a logarithmic Y-axis for each linear scatter chart. SAS Transport file export Any worksheet can be saved to a SAS Transport file. Check this option and enter a number for the first table to include the text “Table N” above each table.stx) in the Save as Type list. Add page break: forces a page break before and after each chart. *. Start table numbers at N: adds table numbers to exported worksheets. 2. With the desired worksheet active in WinNonlin.

5. You may also edit the label names and precision for exported values. If the Rename Columns dialog appears. 7. edit data set name. and are limited to alphanumeric characters and the underscore.Data import and export SAS Transport file export 4 4. or reload previously-saved settings. enter new names under Variable Name for any names that appear in red text. Enter the desired number of significant figures or decimal places under Precision. 6. Optionally.MAP). 79 . Use the Load and Save buttons to save these settings to an ASCII text mappings file (*. Digits) and decimal places. The new names must be no longer than 8 characters. Click OK to create the SAS Transport file. Scroll right and click on a cell under Precision Type to toggle between significant figures (Sig.

For example: # Variables in DV column # DEPC = 1 # DEPQ = 2 Thus. Data file structure The NONMEM data file contains two sections: the headers and data.g. Thus. the value in DV represents DEPQ. All NONMEM data are numeric. Abbreviations. 80 . Typically.” “B. e. The exported DV column can hold one or more dependent variable(s). Comments in the header can indicate how WinNonlin variable names map to those used in NONMEM. For example. whenever DVID is 2.DAT). Encoding. or “male” and “female” must be encoded as numeric values for use in NONMEM.4 WinNonlin User’s Guide NONMEM export WinNonlin provides a NONMEM Export function to save observation and (optional) dose data in a NONMEM-compatible data file (*. this encoding is explained in the header. NONMEM places a four-character restriction on names. NONMEM export can also generate a NONMEM control stream containing an outline of required commands or commands copied from a user-specified template. The following four items can appear in the header of a NONMEM data file exported from WinNonlin: 1. The header can include the DVID column mappings.. This will appear as a comment like that below: # ID TIME WT GEND AGE BP DV MDV DVID AMT RATE EVID 2. DVID. The DVID column disambiguates between dependent variables. The first line in the header contains the names of the columns in the data file. Names of columns. the header might contain: # Variable GEND # male = 1 # female = 2 3. “A. 4. discrete values stored as strings.” etc.

the dose rate. The event type: 0: An observation event (for DV) 1: A dosing event 2: An other-type event 3: A reset event 4: A reset-and-dose event 6 DVID 7 8 AMT RATE Required for PK. If more than one dependent variable (DV item) is exported. Required. -2 means that a zero order bolus for the rate is calculated. Nominal or actual time. A positive value indicates an infusion. one should always provide MDV. RATE = 0 for bolus dosing. 9 10 CMT EVID Usually provided. For infusions. (If DV is never missing. 81 . for simplicity. The RATE column shows a dot when no value is administered. Optional. The dependent variable’s values. then MDV can be skipped. Description Subject identifier. 2 for the second. User may include as many as are available. Frequently provided.Data import and export NONMEM export 4 NONMEM data may include the following columns. Always provided. The value of the independent variables. NONMEM columns Position Name 1 2 3 4 5 ID TIME independent variables DV MDV Presence Always provided. This form export should always include it. Table 4-1. but. -1 indicates a zero-order bolus for which NONMEM calculates the duration. Thus. this column disambiguates. this column will equal 1 for the first. There will be one column for each independent variable. and 3 for the third. The status of the dependent variable: 0: The row includes a valid value of DV 1: The row does not include an observation of DV or the value is missing. Indicates the compartment into which a dose was administered or from which DV was observed. if three DV items are chosen. The amount value of the dosing variable. The name DV is a fixed name for NONMEM.) Provided only when more than one DV item is present. Usually required.

Tab name Files Dependent variables Other data items Dose information Occasions Description Export file path and name. Comment character: Enter a character to precede and identify comments in the output file. Load a data set and a compiled compartmental PK. Choose File>NONMEM Export. The default is: #. If a control stream is output with the data. If a control stream is output with the data. Set the export data and settings using the following tabs. PK/PD link or indirect response model. Data File: Enter the name of the output data file (*. 4. Specify the dosing (if any). 82 . this string will appear in the $DATA statement. When all settings are complete. 3.4 WinNonlin User’s Guide Exporting to NONMEM To export data to NONMEM: 1. header information. then this string will appear in the $PROB statement. To be available for export. comment identifier Dependent variables and dependent variable ID column Independent variables Dosing variables and compartment to dose into (optional) Identifiers for within-subject occasions to be considered separately. 5. or use the Browse button to set it. Set up the model variables as detailed under “Modeling” on page 193. 2. It is useful to enter descriptive comments here to make it clear what is in the data file. Files Problem (optional): This string becomes the heading for the NONMEM export file.DAT) to be created. including the full (absolute) path. click OK to create the NONMEM file. and appears in the $INPUT COMMENT= command in the control stream. This character will precede the header rows. dose data must be either entered the Model Properties dialog (Dosing Regimen tab) or available on the Dosing worksheet of PKS study data (WinNonlin Enterprise only). as an additional means to explain variability 6. The Export to NONMEM dialog appears. Dose data for NONMEM export is drawn from the WinNonlin model.

CMT stands for compartment. 83 . Inhibits the creation of a new control stream. Specify the file name. and $DATA commands only. for the file to create (above) and the template file (below) or use the adjacent Browse button to locate each. Generate control stream from template copies a user-supplied control file template and inserts the appropriate $PROB. $INPUT. Other data items Use this tab in the NONMEM export dialog to identify the subject ID. DVID: If multiple dependent variables are included. Additional header comments will indicate which column (NONMEM abbreviation) is associated with each DVID value. the header comments in the NONMEM data file will indicate “DV=COL” where “COL” is the NONMEM abbreviation entered in this dialog box. or use the Browse button to set it. Generate skeleton control stream generates a skeletal control stream. If a single dependent variable is specified. • Dependent variables Specify one or more dependent variable(s) for the NONMEM export by checking the box under Select? for the appropriate columns. including full paths.Data import and export NONMEM export 4 Control stream: Select the appropriate option: • • Do not generate control stream: available for users who wish to write their own control stream. CMT must be a non-negative integer. All dependent variables must use numeric values. If more than one dependent variable is included. all will appear stacked in a single column identified in the headers as DV. and $DATA commands at the beginning. enter the compartment number here. and to set their names in the exported data. $INPUT. consisting of the $PROB. Enter a unique positive integer value for DVID for each dependent variable. time and other independent variables. including the full path. If the dependent variable is associated with a particular compartment of a PK model. Enter the names. usually less than 10. WinNonlin will create a dependent variable ID (DVID) column to disambiguate them.

one variable representing subject identifiers (required). See the NONMEM NM–TRAN documentation to determine which numbers to enter. CMT must be a non-negative integer less than 10. Treat as bolus. any number of independent variables (optional). Edit the values under NONMEM Abbreviation as necessary to read: a. Include no dose information. For infusions. Have NONMEM calculate rate (RATE=–2). b. Associate dose with a compartment: To dose to a specific PK compartment. this allows estimation of the duration parameter. The Dose Information tab of the NONMEM export sets options for exporting this dose information. numerals and/or underscore characters for each other independent variable. Case will be preserved as entered. Dose information Dose data for NONMEM export is drawn from the WinNonlin model. Check the box under Select? for: a. check this box then enter the compartment number under CMT.4 WinNonlin User’s Guide To identify independent variables for export: 1. In this case. 2. a unique name up to four letters. no dose data are exported. and must be specified in either the Model Properties dialog (Dosing Regimen tab) or via the Dosing worksheet included with PKS study data (WinNonlin Enterprise only). For infusions. These options include: • • • • • • Derive RATE from the data. All doses will be given a RATE of 0. 84 . TIME for the time variable. ID for the subject identifier. and c. b. one variable representing time (required) c. Each four-character abbreviation must be unique and cannot be a NONMEM reserved name. this allows estimation of the rate parameter. Have NONMEM calculate duration (RATE=–1).

The script must be open in a WinNonlin text window. Use occasions: check this box to generate an occasion identifier. At the beginning of an occasion. from the WinNonlin menus. The Occasions tab in the NONMEM export dialog exports a separate column to differentiate occasions (e. and select a method for creating occasions: • • list the time values at which new occasions start. All columns in the worksheet will be exported to a SigmaPlot notebook. if any. To automatically load script output. NONMEM will reset all compartment concentrations to 0. Choose File>Sigma Plot Export. Select the data to export under Workbook and Sheet. Enter a name for the new notebook under Note Book Name.. open or create it in a WinNonlin text window.. SigmaPlot export WinNonlin provides options to export worksheet data to SigmaPlot. 5. into WinNonlin: 85 . Note: SigmaPlot export requires installation of SigmaPlot 8 on the same machine as WinNonlin. choose its text window (or file path) from the list under Optional Script. 3. and (optionally) when the first occasion begins. 4. 6.. 2.Data import and export SigmaPlot export 4 Occasions Including occasional variation in a model can sometimes explain additional variability in the data. To include and run a script. from earliest to latest (do not include time 0). To include and run a script. To export data to SigmaPlot and (optional) run a script: 1. separate profiles within subjects). 7.g. or indicate how long each occasion lasts. with or without a script to run. if any. WinNonlin can automatically import script output. Open the data to export in a WinNonlin workbook.

2. 3. b. Click OK to export the data and execute the script. To include and run a script.. Note: S-PLUS export requires that S-PLUS version 6. do not perform any actions in WinNonlin until the script run completes and its output appears in WinNonlin. Check Wait for completion. To export data to S-PLUS and (optional) run a script: 1..4 WinNonlin User’s Guide a. WinNonlin can automatically import script output..” button to select the folder. All columns in the worksheet will be exported to an S-PLUS data frame. if any. 4. Select the data to export under Workbook and Sheet. open or create it in a WinNonlin text window.. 8. Open the data to export in a WinNonlin workbook. with or without a script to run. If you chose to return script output to WinNonlin. Choose File>S-PLUS Export.1 is installed on the same machine as WinNonlin. 86 . Enter the full path to the SigmaPlot output folder under Output or click the “. S-PLUS export WinNonlin provides options to export worksheet data to S-PLUS. from the WinNonlin menus.

If you chose to return script output to WinNonlin. into WinNonlin: a.IMP) contains all information required to draw data from a remote data source. See “The custom query builder” on page 97. b. then defining a query to draw specific fields and records from it.IMP).. This string can be cre87 . The script must be open in a WinNonlin text window. 7. including Watson version 6. Database connection: Both ODBC import and export use a connection string that defines the database type and location. Click OK to export the data and execute the script. ODBC import Use ODBC import to load data from an ODBC-compliant database directly to a WinNonlin worksheet. To include and run a script. 6. See also “ODBC export” on page 102.Data import and export ODBC import 4 5. See “Refreshing an imported data set” on page 97 for instructions. the ODBC import command provides the option to update the data to reflect the current database content. To automatically load script output. Enter the full path to the output folder under Output or click the “. for later re-use.0 DMLIMS. It includes two parts: 1. Once ODBC-imported data are loaded in WinNonlin. The connection and query settings can be saved to an import file (*. Check Wait for completion. if any.” button to select the folder. Instructions are provided under: • • “Creating an import” below “Loading an existing import” on page 96 WinNonlin also provides a Custom Query Builder for creation of dynamic link library (DLL) files that can access additional data source types. or import. Enter a name for the new data frame under Data Frame Name. An import involves establishing a connection to the database. if any.. 8. choose its text window (or file path) from the list under Optional Script. do not perform any actions in WinNonlin until the script run completes and its output appears in WinNonlin. Creating an import An ODBC import specification (optionally saved as *.

The query may be created in the ODBC Import wizard or loaded from previouslysaved import as described under “Queries” on page 92. Proceed to the Final settings dialog. “Loading a saved connection” on page 92 4. each can be loaded separately. Final settings The Final Settings dialog summarizes the ODBC import. One connection string can be re-used with any number of different queries. “Creating a new connection” on page 90 b. Its top section displays the connection string. or loaded from a previously-saved import or export as detailed under “Connection” on page 90. The lower section determines data formatting in WinNonlin. “Loading a saved query” on page 96 5. field(s) and records to import. “Building a query” on page 93 b. 2. Choose File>ODBC import from the WinNonlin menus. Create or load a query as detailed under: a. The middle section contains the SQL string for the query.4 WinNonlin User’s Guide ated by pointing to the database type and file. Create or load a connection string as detailed under: a. 2. Query: Once the database connection is established. 88 . Select Create a new import in the ODBC Import wizard and click Next. which can be edited before executing the import. an import requires a query to set which data table. WinNonlin translates the query into Structured Query Language (SQL). To create and save (optional) a new import: 1. 3.

Check Number of Records? provides a count of filtered records prior to the import. 3. Click Finish to import the selected data to a WinNonlin workbook. To save the import (*. and dates. and the option to return and edit the query if needed. In either case. so a password saved within is not secure. currency. Set Column Formats applies the appropriate number formats to columns containing times.Data import and export ODBC import 4 1. This allows scripting to use ODBC connectivity without end user intervention (see “FileType” on page 593). 89 . WinNonlin will then execute the import. Save Password: This option includes the database user name and password in the import file with the connection string.IMP). 4. Choose settings in this dialog (below). a dialog appears with the number of records returned by the filter. click Yes and save to the appropriate directory. and edit the SQL if desired. click No. Else. If Check Number of Records? was checked. Set Column Names adjusts the destination worksheet to have the same column names as the source fields. 2. • • • • Set Column Widths adjusts the destination worksheet's columns to match the widths of the fields in the database. CAUTION: The import file is not encrypted. Click OK to proceed.

and click Next. This establishes a link to the database. This connection string may be saved as part of an import or export settings file for re-use.4 WinNonlin User’s Guide WinNonlin can import up to 65. “Loading a saved connection” on page 92 extracts the connection string from a previously-saved import or export file. “Creating a new connection” on page 90 creates a new connection string. Each cell can hold up to 16KB of text. These are file-based data sources that can be shared among all users who have the same ODBC drivers installed. Creating a new connection To create a new connection string: 1. The File Data Source tab (below) lists all file Data Source Names (DSN’s) and subdirectories on the system. Select Create a new Import (or Export) in the ODBC Import (or Export) wizard. Choose File>ODBC Import (or Export) from the WinNonlin menus. Select Create a new connection and click Next. defining the database type and location. The ODBC drivers might impose additional limits. Connection The first step in an ODBC import or ODBC export is to establish a connection to a specific database. 90 . The data sources need not be dedicated to a single user or local to a specific computer. 3. 2. to be used with a new or saved query or export.000 rows into one worksheet. The import wizard is shown below. 4. Select a data source from one of the two tabs in the Select Data Source dialog.

The Machine Data Source tab (below) uses information stored in the machine registry. These are set up by the Database Administrator or Information Technology personnel. 5. Select a data source and click OK.Data import and export ODBC import 4 New data sources can be set up using the New button. enter the appropriate log in ID and password for the data source. Click OK to proceed with the import or export. 6. If a log in dialog appears. 91 . See the Microsoft Windows documentation for details. which provides the majority of the connection string information.

as detailed in the following sections. Select the import file (*. ODBC import requires a query specifying the data table. as detailed under “Queries” on page 92 for imports or “Export definition” on page 103 for export. An Open dialog appears. Select Create a new Import (or Export) in the first dialog and click Next. or an export definition to map WinNonlin data columns to database fields. field(s) and records to be imported. Select Load an existing connection and click Next.4 WinNonlin User’s Guide The next step is to set up a query to pull the desired data from the database. Proceed with the import or export. Loading a saved connection Once a connection string is saved in an import (*.IMP) or export (*. To load an existing database connection: 1. 5. 3. 2.IMP) or export definition file (*. 4.EXP) containing the connection string for the desired database and click Open.EXP) file. Choose File>ODBC Import or File>ODBC Export from the menus. Queries Once a database connection string is established (see “Connection” on page 90). • • “Building a query” on page 93 “Loading a saved query” on page 96 92 . See “Queries” on page 92 or “Export definition” on page 103. it can be loaded for use with different import queries or export definitions.

Data import and export ODBC import 4 Building a query A query notes the database table and fields (optional) to import to WinNonlin from an ODBC-compliant database (see “ODBC import” on page 87). These schema were created by your database administrators. Select Build a Query and click Next to proceed to “Schema filter” below. and load a list of the available tables. 3. the Schema Filter dialog appears. 2. select the schema to use in filtering the list of data tables then click OK to proceed to “Query builder” below. It can also include filters to extract specific records from the table. If the appropriate ODBC drivers are available. Choose File>ODBC Import from the WinNonlin menus. as detailed under “Creating a new connection” on page 90 and “Loading a saved connection” on page 92. If the Schema Filter dialog appears while building a query (“Building a query” on page 93). Load a new or saved connection string. To create a new query: 1. 93 . a metadata organizational tool. It provides a means to filter the available database tables by schema. Schema filter The WinNonlin Query Builder will access the database.

4 WinNonlin User’s Guide Query builder To select source tables and fields while building a query: 1. To filter the records. 2. click Next. click the Done button and proceed to step 7 below. 94 . or click Done to load all fields in the table and proceed to step 7 below. Select one or more fields to import. Click Next to select specific fields to be imported. select a table from which to import data. Begin the query setup as described under “Building a query” on page 93. When the Source Tables list appears. 3. 4. To import all the records from those fields.

Follow the instructions under “Final settings” on page 88 to complete the import. Enter a value in the Field Value field. Filters are applied in the order created. The Import: Final Settings dialog appears. 95 .Data import and export ODBC import 4 5. Click the Add button to set the filter. To create the filter: a. Click Done when the filters are complete. e. select the And or Or radio button and return to step a above. d. click the List Distinct Values button to select from among all existing values for that field. 6. 7. Choose an operator from that field's list box. c. If more than one criterion will be applied. b. Select a Field Name to filter on from that list box.

In the ODBC Import dialog select Load an existing Import and click Next. 96 . The default file type is *. respectively. In the ODBC Import: Query dialog select Load a query and click Next. 2. See “Final settings” on page 88 to complete the import and load the data into a WinNonlin workbook. Use File>ODBC Import from the WinNonlin menus to connect to a new or previously-saved database.IMP) contains the connection string and query to quickly re-load data from an ODBC database into WinNonlin. Choose File>ODBC Import from the WinNonlin menus. Loading an existing import An import setting file (*. 3. The ODBC Import: Final Settings dialog appears with the connection string and query loaded. Select the file with the desired connection string and click Open. any saved query can be loaded so long as it uses tables. To load an existing query: 1. It also provides a convenient template for a new ODBC import. as detailed under “Creating a new connection” on page 90 and “Loading a saved connection” on page 92. 2. fields and record types in the database being connected to. An Open dialog appears. 3. The ODBC Import: Final Settings dialog appears with all settings for that import loaded.IMP) file and click Open. Select the appropriate import (*.IMP. To load an existing import: 1.4 WinNonlin User’s Guide Loading a saved query Once a connection string (new or existing) has been established as part of Creating an import. An Open dialog appears.

Even if the workbook has been saved locally. 6.0: 1. When the settings are correct. those data can be refreshed to reflect the current the database source by choosing the Import ODBC command from the File menu. Further. Select File>ODBC Import from the menus. including Watson version 6. click the Finish button. Click Yes to refresh the data in the worksheet from the database. Refreshing an imported data set When data from an ODBC import are open in WinNonlin. 97 . The custom query builder The Custom Query Builder provides a means to build custom dynamic link library (DLL) files that access additional data source types. 5. The data from the database are then loaded into a WinNonlin worksheet. To use the custom query builder with Watson v.Data import and export ODBC import 4 4. Makes changes as needed in the Import: Final settings dialog. it can be re-opened and refreshed following this same procedure. WinNonlin Enterprise provides templates for creating the DLL files to access custom data sources. and using the Back button.0 DMLIMS.

98 .4 WinNonlin User’s Guide 2. 3. Select Use Custom Query Builder and click Next to proceed to “Select a builder” below. Enter the full directory path for the Watson templates or use the Browse button to locate the directory on your local machine or on a network server. To configure the Watson DMLIMS custom resource to ensure that data are imported correctly. In the Configuration dialog. 1. enter the name of the Watson LIMS tables owner. e. click Add.DLL. then double-click the appropriate DLL file. 4. WATSON_DMLIMS. 2.g. this dialog may be empty. If the Custom Query Builder has not been used before. To add resources to the Query Builder. Select a builder The Select a Builder dialog appears as part of the Custom Query Builder (see “The custom query builder” on page 97) to set the data resources to access.. select that item and click Edit.

etc. 99 . b. To load a saved query.g.WTS for Watson LIMS) and click Open. and alias for any custom database system (such as the Watson DMLINS) are set by the database administrator or IT personnel. Highlight Watson DMLIMS (or another custom query builder) in the Select a Builder dialog and click the OK button. 7. Request this information from that person or department. 1. click Yes. Enter user name. Use the Order Study By field to sort any category (e. Click OK to close the Configuration dialog. If no query has been saved for this builder. 6. The query is read and the file data is loaded into a worksheet. Study selection The Study Selection dialog lists the resources available in the data source for a custom query (see “The custom query builder” on page 97). password. click No and proceed with the following steps. The Open File dialog appears and provides the opportunity to select an existing query.) in order to find a particular listing more quickly. a. A log on dialog appears. Select a file (file type *.. Select a study by clicking on the row. password. Note: The user name.Data import and export ODBC import 4 5. study director. and click OK to proceed to the Study selection dialog. 8. and database alias. study number/ID. The import is complete. A dialog asks whether to load a query.

2. 3. 100 . If all available variables are desired. The Import Data Variables dialog appears. containing the list of variables available in the selected database. click the Select All button. drag each variable to the Selected Variables field. Click Next. contact your IT department. For a subset.4 WinNonlin User’s Guide Note: If no projects are listed in this dialog.

The Filter dialog appears. and field value. Then enter a name for the template. operator. Note: The fields are imported in the order shown in the Selected Variables list. A filter is defined by specifying one field name. To create a new template drag the variables for the template to the Selected Variables field and click the New button. A particular combination of variables can be saved as a template. The data can also be filtered further by clicking Next. Click Done if all the data in the selected fields are to be imported.Data import and export ODBC import 4 4. 101 . 5.

. 102 . See: • • “Creating an export” below “Loading a saved export” on page 107 Creating an export An ODBC export specification (optionally saved as *. the export requires a connection string defining the database type and location. This string is created in the ODBC Export wizard or loaded from a previously-saved import or export as detailed under “Connection” on page 90. Database connection: Both ODBC import and export can use the same connection string to define the database type and location. If Check # of rows returned was checked.4 WinNonlin User’s Guide 6. Either enter a value in the Value field or select from all specified values of that field. One connection string can be re-used with any number of different export definitions. It includes: 1. ODBC export WinNonlin Enterprise edition can export selected worksheet columns to specific fields within a table in any ODBC-compliant database. Export definition: Once the database connection is established. To set a filter: Select the field to filter on from the Field box.EXP) for later re-use. Click Done to load the data into a WinNonlin worksheet. Like ODBC import. 7. each can be loaded separately. Choose the operator below.g. an export requires a mapping of columns from the WinNonlin worksheet to fields in the target database table. It also requires an export definition to map columns in the source worksheet to fields in the target database table. Both the connection string and export definition can be saved to an export file (*. click the List Distinct Values button and select a value. Age > 12 and Gender = Male). Use the And/Or operators to create compound filters (e. Alternately.EXP) contains all information required to save data directly to an external database. Click the Add button to set the filter. WinNonlin will report a count of records returned. This export definition may be created in the ODBC Export wizard or loaded from a previously-saved export as shown under “Export definition” on page 103. 2.

“Loading a saved export definition” on page 106: One connection string can use any number of different export definitions. 2. • 103 . “Loading a saved export definition” on page 106 5. 3. WinNonlin will ask whether to save the export settings. “Building an export definition” on page 104 b. click No.Data import and export ODBC export To create and save (optional) a new export: 4 1. Choose File>ODBC export from the WinNonlin menus. Export definition An export definition maps WinNonlin worksheet columns to fields in one table of the target database for an ODBC export. Click Yes to create a new export (*. After the export completes. Make any adjustments needed in the Field selection dialog and click OK to complete the export. Create or load a connection string as detailed under: a. 6. • “Building an export definition” on page 104 specifies which data are to be exported. Create or load an export definition as detailed under: a. Select Create a new export in the ODBC Export wizard and click Next.EXP) file.EXP). to what tables and fields in the selected database. “Loading a saved connection” on page 92 4. Else. The settings can be saved to an export definition file (*. “Creating a new connection” on page 90 b.

To create a new ODBC export definition: 1. “Creating a new connection” on page 90 b. Create or load a connection string as detailed under: a. Open the set to export in WinNonlin and choose File>ODBC Export from the menus. As with the ODBC import. The ODBC Export dialog appears. to connect to the target database. the ODBC Export: Definition dialog appears. “Loading a saved connection” on page 92 After a connection string is set. if any have been set up by your IT personnel.4 WinNonlin User’s Guide Building an export definition Export definitions require that a connection string be loaded first. The Schema Filter dialog displays available schemas. 104 . 4. 2. 3. the schemas to use in selecting the data variables can be set. Select Create a new Export and click Next. Select Build a Definition and click Next.

it sets which data columns are exported to which database tables and fields. WinNonlin will ask whether to save the connection string and export definition to a new export file (*. The data type and field size are set by your Database Administrator. excluding any constants in fixed fields.Data import and export ODBC export 4 5. If the Schema Filter dialog appears. check the Fixed check box for that variable and enter the value under the field name. Using schema filters can speed up export operations using saved export files. This file contains all information needed to reproduce the same export. If this is a new export. Select the database table to receive the exported data under Table Name. To export one constant value for all records in a given field. Drag the variables from the Variables list to the appropriate database fields under Field Definitions. 2. select one item and click OK. Proceed to the Field selection dialog to complete the export. 4. 3.EXP). Field selection This dialog determines the core of an Export definition. This method can be used to differentiate one export from another. 6. Click OK to export the data to the database. 105 . 1.

WinNonlin will ask whether to save the export settings.EXP) can be re-used with any database connection. 8.4 WinNonlin User’s Guide Loading a saved export definition The field mappings in any ODBC export definition file (*. Make any adjustments needed in the Field selection dialog and click OK to complete the export. select the export file (*. “Loading a saved connection” on page 92 Once the connection string has been loaded.EXP) file. “Creating a new connection” on page 90 b. In the Open dialog. assuming the field and table names in the export file exactly match a table in the selected database. 6. if necessary. Create or load a connection string as detailed under: a. Else. Select Create a new Export and click Next. If the Schema Filter dialog appears. 7. Log on to the data source. If this is a new export. 4. Click Yes to create a new export (*. click No. use the Schema Filter to narrow the set of available data tables (optional) and click OK. The ODBC Export dialog appears. 5. 2. 3.EXP) and click the Open button. the ODBC Export: Definition dialog appears. To load an export definition: 1. Open the set to export in WinNonlin and choose File>ODBC Export from the menus. Select Load a Definition and click Next. 106 . 9.

the export settings.EXP). and click OK to export the data.EXP) and click Open. select Load an Export and click Next. An Open dialog appears. The Field Definitions are loaded automatically in the Field Selection dialog. This feature extracts both the connection string and the export definition from a saved export file (*. 3. System Administrators and Database Administrators who wish to create custom query builders for ODBC connectivity. Select the desired export definition file (*. can be reloaded for reuse with any data set containing the column names used in the export file. The ODBC Export: Export dialog appears.Data import and export The Enterprise Software Development Kit 4 Loading a saved export Once an ODBC export has been created and saved. including the connection string and/or the export definition. To load both a connection string (pointer to a specific database) and export definition (mappings). 107 . 5. 4. To load an existing export: 1. Make any needed edits to the export settings. Log on to the data source if needed. 2. Choose File>ODBC Export from the WinNonlin menus. The Enterprise Software Development Kit The Enterprise edition of WinNonlin includes a Software Development Kit (SDK) for Developers.

WSID=THUNDER. So. 6 lines contained in quotes. the creation of an import file for loading should leave the connect string blank. and let the application define the connect string.SERVER=thunder.Time1. potentially. an import file distributed by an IT department would.Time1. 1 2 3 4 5 6 BEGIN FILE (Note 6 lines) Connect string Query Set Column Formats (0 or 1:True or False) Set Column Names (0 or 1:True or False) Set Column Widths (0 or 1:True or False) Set Max RC (Always set to 1) END FILE Example: BEGIN FILE "DRIVER=SQL Server. since the connect string is dependent on the drivers installed on the machine.Network=DBMSSOCN." "Select Subject.APP=Visual Basic. this file is loaded.UID=dan. contain the following: BEGIN FILE "" "Select Subject. At query completion. 108 . When the end-user creates or loads a query.Conc from bguide1a where (Subject = '2 ')" "0" "0" "1" "1" END FILE This would allow WinNonlin Enterprise to define the connect string.Conc from bguide1a where (Subject = '2 ')" "0" "0" "1" "1" END FILE Usage Generally.4 WinNonlin User’s Guide Import file format The Import file is an ASCII file.

Data import and export The Enterprise Software Development Kit 4 the import could be saved to store the connect string generated by WinNonlin Enterprise. The new code will not compile unless all interfaces in the DLL are implemented. Current State: This ActiveX in-process server [QUERYBUILDERINTERFACES. BEGIN FILE 1 Connect String 2 Table Name (Table that data is exported to) 3 Worksheet Column for definition 1 4 Table Field for definition 1 5 Fixed Field Flag for definition 1 6 Worksheet Column for definition 2 7 Table Field for definition 2 8 Fixed Field Flag for definition 2 … 2+3*n-2) Worksheet Column for definition 1 2+3*n-1) Table Field for definition 1 2+3*n) Fixed Field Flag for definition 1 END FILE Custom query builder interface File: QUERYBUILDERINTERFACES. At this point a complete import file has been created. 109 . Export file format An Export file is an ASCII file that contains 2+3*n lines..e. If a field is fixed. where n is the number of column-field definitions. In some cases. When this DLL is registered on a machine. The interfaces: There are two interfaces: one accessed by WinNonlin (IQUERYBUILDER) and one accessed by the Custom Query Builder (ICONSUMER). the exported column in the worksheet is left blank. the developer can reference it and implement its interface. Each column-field definition is represented by three (3) consecutive lines. it may be necessary to code to a specific database API set to optimize performance when determining access rights and database structure. the table field that the column is exported to. and another line to determine if this field is fixed or not (i. a Flag. “True” or “False”). The definition consists of the column in the worksheet that is exported.DLL] is essentially a stubbed out API.DLL The customized query building feature allows site-specific and database-specific customization of query building.

When the resource manager registers the DLL. This interface is implemented by the Custom Query Builder. It is up to the developer of the Custom Query Builder whether to use this file is used as the help. Methods Method Name GetQuery GetError 110 Type Long String Function This tells the custom query builder to get the query (i. The return values are: <0 (Cancel). Table 4-2. This object is set by WinNonlin via IQueryBuilder. It is also stored in the registry when the Custom Query Builder is added to WinNonlin. Currently. QueryBuilderListName AppHelpFile String Name to be used in the list of available Custom Query Builders. This enables refreshing of the data source. String The version of the custom query builder. String This is used to set or get the help file name used by the Custom Query Builder. the name of the DLL and appends .e. This string is then passed to WinNonlin for audit purposes. =0 (No error). The Custom Query Builder could build its own string. build it). the help file is located in the same directory as the DLL. >0 (Error) Returns the error message if one exists. .HLP on to the end of the name to create the help file name to be used by the Custom Query Builder. String Final query string to be accessed once the custom query builder is done. It is stored in the registry when the Custom Query Builder is added to WinNonlin. Properties Property Name Type ConnectString QueryString QueryBuilderDescription QueryBuilderVersion Function String Connect string used by the generic ODBC query builder.4 WinNonlin User’s Guide IConsumer is the consumer of the imported data. The Custom Query Builder is responsible for importing data into the WinNonlin object by implementing the IConsumer interface. IQueryBuilder interface The interface comprises the properties and methods in Table 4-2 and Table 43. it takes the path of the DLL. String A description of the query builder. QueryBuilderExt String The extension used by the query builder for saving import files. This should point to the exact location of the help file. Table 4-3. It is not required to.

WorkSheet wrhSht WinNonlin calls this subroutine to set the consumer object in the Custom Query Builder. This information is stored in the registry by the WATSONLIMS. For the Integraware LIMS. Create a class named CQueryBuilder.The consumer object is an interface to the worksheet within WinNonlin. the base table must be defined.DLL. and attempts to Import data based on the contents of the Import File. Table 4-4. Methods (continued) Method Name Type Function 4 ImportData. Currently. The structure of the import file is up to the developer. >0 (Error) PopulateWorkSheet IConsumer interface The interface is accessed by the Custom Query Builder. This is also stored in the registry. Held for future use. This tells the Custom Query Builder to populate the worksheet object handed to it in the WorkSheet subrountine. This is used for scripting purposes. The extension of the file should be the same as returned in QueryBuilderExt. Cell (row As Long. Add a reference to QueryBuilderInterfaces. =0 (No error). the owner of the Watson tables must be defined if the log on ID is not the owner. Make it public. it is required that the class implement111 . highlight the Custom Query Builder on the Custom Query Builder list and click Edit. if necessary. For instance. IConsumer properties Property ColumnHeader(idx As Long) Units(idx As Long) Type String String Function This sets/gets the column header for the column.DLL 3. This is a hook to allow the Custom Query Builder Manager to configure the DLL. Implement QueryBuilderInterfaces. col As Long) Variant Sets/Gets the value for the given row and column. The return values are: <0 (Cancel). sumer Long This function is called after the GetQuery function and the WorkSheet subroutine.IQueryBuilder.ImportViaImportFile File as string ConfigureString CustomQueryBuilder This function takes an import file name (usually built and saved by the Custom Query Builder).DLL.Data import and export The Enterprise Software Development Kit Table 4-3. Not implemented yet. Create a new project of type ActiveX DLL. 2. VB example 1. for the WATSONLIMS. To access this method. As Icon.

Figure 4-1. stubbed code is generated. Option Explicit Implements QueryBuilderInterfaces. the object list has IQueryBuilder. If the selected property/method is not defined.IQueryBuilder 4. Since IQueryBuilder is implemented. Now each required method and property can be defined as illustrated below. The calling application needs to know the name of the class to create that implements the interface.4 WinNonlin User’s Guide ing the interface is named CQueryBuilder. and the other list has its properties and methods. 112 .

finish the code required to generate a valid SQL Query. Next. A Simple Object Diagram is shown to the right.Data import and export The Enterprise Software Development Kit 4 5. IQueryBuilder Properties: ConnectString QueryString QueryBuilderDescription QueryBuilderVersion QueryBuilerExt QueryBuilderListName AppHelpFile Methods: GetQuery GetError ConfigCutsomQueryBuilder ImportDataViaImportFile WorkSheet PopulateWorkSheet IConsumer Properties: Cell ColumnHeader Units 113 .

no data loaded rc= 0 R eturn to W inN onlin m ain processing 114 . no data loaded rc= 0 C all W orkS heet m ethod of C Q B w ith the IW krS ht as param eter C anceled. no data loaded rc< 0 C all G etQ uery on the C Q B rc> 0 S how error M essage. A ppH elpfile. C onnectS tring properties of the CQB C anceled. no data loaded rc< 0 C all P opulateW orksheet m ethod of the C Q B rc> 0 S how error M essage.4 WinNonlin User’s Guide T yp ica l u sa ge of th e C u stom Q ue ry B u ild e r (C Q B ) C reate object im plem enting IC onsum er: IW rkS ht C reate an Instance of the C ustom Q uery B uilder (C Q B ) w ith interface IQ ueryB uilder S et Q ueryS tring.

See “Chart Designer” on page 123. Chart Designer. Information on printing charts appears under “Printing” on page 21. Modeling output: Use the Model Options to include chart output from compartmental modeling or noncompartmental analysis. The Chart Designer supports additional plot types not available in the Chart Wizard. You can then change the plot type and formatting using the Chart Designer as detailed under “Chart Designer” on page 123. Modify type and formatting of existing charts in the Chart Designer. and other chart options WinNonlin provides high quality plots of input data and modeling or NCA output. See “Chart Wizard” on page 115. For additional plot types. Chart Wizard: Use this wizard to plot data in any worksheet as a XY plot. 1. 115 . Chart Wizard Use the Chart Wizard to plot worksheet data. create a plot and edit it in the Chart Designer (below). bar chart. There are three means of creating and editing plots. See “File types” on page 14 for details. 3. See the following for additional chart options.Chapter 5 Charts Using the Chart Wizard and Chart Designer to create and format plots. or histogram. • • • “Frequently used chart formatting options” on page 124 “Chart templates” on page 129 “Copy and paste” on page 130 WinNonlin can save charts to many different file formats. See “Output options” on page 204. 2.

Click the Next button to procede to the X-Y variables. The workbook must not be hidden. or histogram with the desired variables. Choose a chart type. bar chart. Multiple pairs of XY variables can be included. 4.5 WinNonlin User’s Guide To create a plot: 1. 3. X-Y variables The X-Y variables tab of the Chart Wizard sets the independent and dependent variables to be plotted. The worksheet must be the active worksheet in the workbook window. for overlay plots. These settings differ for Scatter plots and bar charts versus Histograms. 2. Open the worksheet(s) containing data to plot. Note: Additional chart types are available after a plot is created. See “Chart Designer” on page 123. 116 . The Chart Wizard appears. Click the Chart Wizard tool bar button or choose Tools>Chart Wizard from the WinNonlin menus. then use the Chart Designer to change the plot type. Make a scatter plot.

Charts Chart Wizard

5

**Scatter plots and bar charts
**

To select variables for a scatter plot or bar chart:

1. Make sure that the workbook to be plotted is displayed in the Data Set field (shown below). The columns in the active worksheet appear under Variable Collection. 2. To specify each X or Y variable, highlight a variable under Variable Collection and drag it to the X or Y Variable box.

3. To create an overlay chart, showing more than one pair of X and Y variables: a. If all variables are in one workbook, drag paired sets of X and Y variables to the appropriate variable boxes, so that pairs are listed side-by-side. b. If the variables come from more than one workbook, select a second workbook under Data Set to select X and Y variables from that data set. 4. (Scatter plots only) check Automatic Sorting to sort the data in ascending Xvalue order before drawing plot lines. This option would be inappropriate for a hysteresis curve. Figure 5-1 below provides an illustration of a simple data set plotted with and without automatic sorting.

117

5

WinNonlin User’s Guide

Figure 5-1. Plotting data with (middle) and without (right) automatic sorting

5. To create multiple plots or multiple series within a plot, click the Sorting tab and follow the directions under “Sort and group variables” on page 119. 6. (Scatter plots only) To set up error bars, open that tab of the dialog and see “Error bars” on page 120. 7. Click Next to proceed to the Chart titles. Histograms

To select variables for a histogram:

1. Make sure that the workbook to be plotted is displayed in the Data Set field. The columns in the active worksheet are listed in the Variable Collection list. 2. To specify the variable(s) for which frequencies will be displayed, highlight a variable under Variable Collection and drag it to the Y Variable box. Repeat to add variables on the same set of axes. 3. Enter the number of bars for the X axis (optional) under Number of Intervals.

118

Charts Chart Wizard

5

WinNonlin will determine the range of the Y variable, divide this range into the number of intervals, then count the number of observations that fall into each interval, i.e., the frequency for each interval. 4. To create multiple plots, click the Sorting tab and follow the directions under “Sort and group variables” on page 119. 5. Click Next to proceed to the Chart titles. Sort and group variables

To sort data into multiple plots, or group data series within a plot:

1. Open the Sorting tab of the Chart Wizard.

119

5

WinNonlin User’s Guide

2. Drag and drop variables from the Variable Collection list to the Sort or Group variables box: • • Sort variable(s): a separate plot (set of axes) will be created for each unique combination of sort variable values. Group variable: each plot will contain a separate line (scatter plot) or set of bars (bar charts) for each unique combination of group variable values.

The example illustrated above will generate one plot for each drug formulation, with an individual line (scatter) or set of bars (bar chart) for each subject. 3. (Scatter plots only) To set up error bars, open that tab of the dialog and see “Error bars” on page 120. 4. Click Next to proceed to the Chart titles. Error bars

To add error bars to a scatter plot:

1. On the Error Bars tab of the Chart Wizard and set the following: a. Up variable: error values in the upward direction (Y-axis increase). b. Down variable: error values in the downward direction (Y-axis decrease). c. Relative error bars: error values are expressed as absolute difference from the mean; that is, each error value is added to (up variable) or subtracted from (down variable) the corresponding datum. d. Absolute error bars: Error data are expressed as Y-axis co-ordinates for the top (up variable) or bottom (down variable) of each error bar.

120

Charts Chart Wizard

5

CAUTION: In overlay charts (more than one set of X-Y variables per chart), the selection

of absolute versus relative error bars applies to all data series in the chart. The example above uses relative error bars for a plot of average concentrations plus or minus the standard error at each time point. In this case, the standard error supplies both the Up and Down variables.

6 5 4 3 2 1 0 0 1 2 3 4 5 6

Mean = 3 Standard Error = 1 UP variable = Standard Error Down variable = Standard Error Relative Error Bars

Figure 5-2. Plotting mean and standard error using relative error bars

Absolute error bars are useful for plotting more than one value at each X-axis point, e.g. minimum, mean and maximum, as illustrated below.

6 5 4 3 2 1 0 0 1 2 3 4 5 6

Mean = 3 Minimum = 1 Up variable = blank Down variable = Minimum Absolute Error Bars

Figure 5-3. PLotting mean and minimum values using absolute error bars

2. Click Next to proceed to the Chart titles.

121

5

WinNonlin User’s Guide

Chart titles

1. To include a title at the top of each chart, enter it in the Title field in the final dialog of the Chart Wizard. 2. Select or enter the other options: • • • Axis titles by default use the variable names from the data set. To change an axis title, select and replace the text in the appropriate axis title field. Units: Enter or edit the units for the axis labels. See “Units” on page 31 to select whether column units are displayed as axis units by default. Sort key title: If sort variables are used the sort key title appears below the plot title to identify which value of the sort variable(s) appears in the plot, e.g., “Formulation = Tablet.” Edit the display name of each sort variable (in this case, Formulation) by selecting its number in the sort key field. Use group headers: Check this option to include the group variable name(s) in the chart legend, e.g., “Subject = DW.” Legend: When a chart includes more than one pair of X and Y variables, the Legend field provides a means to edit the text that will identify data from each data set in the chart legend.

• •

3. Click Finish to create the chart.

Axis options

Once a plot is created using the Chart Wizard or as part of modeling output, adjust the X and Y axes as follows. See also “Chart Designer” on page 123.

122

Charts Chart Designer To set axis options

5

1. Choose Chart>Axis Options from the WinNonlin menus to open the Axis Options dialog. 2. Select the appropriate options from the following. • • • Uniform Axis: If a sort key was used to identify multiple profiles, this optional applies the same axis range and divisions to all profiles. Logarithmic: Check this box to plot on a natural-log scale. Intervals: For a histogram, enter the number of intervals under Intervals.

3. Click OK.

Chart Designer

Use the Chart Designer to format and enhance charts created using the Chart Wizard (see “Chart Wizard” on page 115) or as part of modeling or NCA output. The Chart Designer provides extensive plot formatting options and a selection of additional chart types, including three-dimensional plots. See the Chart Designer online help system for detailed usage instructions.

To use the Chart Designer:

1. With a chart in the active window, choose Chart>Chart Designer from the WinNonlin menus, or double click on the chart or a specific plot element. Alternately, right-click on the chart and choose Chart Designer from the shortcut menu.

123

5

WinNonlin User’s Guide

The Chart Designer provides a tree view in the left pane. Click an + symbol to expand an element to view sub-elements. The tabs to the right change as different elements are selected in the tree. When an entire group of elements is selected from the tree view—all chart series, for example—the controls to the right apply to all of the selected items. 2. Select an item or item(s) to format from the tree view, and choose settings to the right. See the Chart Designer Help or “Frequently used chart formatting options” on page 124 for additional information. 3. Click Apply to execute the changes or OK to execute the changes and close the Chart Designer.

**Frequently used chart formatting options
**

This section highlights some commonly-used chart formatting options: • • • • • • “Changing the plot type” on page 124 “Changing line style and markers” on page 125 “Changing axis scaling” on page 125 “Creating a semi-log plot” on page 126 “Setting grid lines” on page 127 “Removing or adding walls” on page 128

**See also the Chart examples in the Examples Guide. Changing the plot type
**

To select the type of plot:

1. In the Chart Designer, click Chart at the top of the tree view. 2. Select the Type tab on the right side of the Chart Designer. 3. Select the 2D radio button for two-dimensional plots; 3D for three-dimensional. 4. Select the chart type from those available under Chart. See the Chart Designer Help for details and data requirements for each chart type.

124

Charts Frequently used chart formatting options

5

Changing line style and markers Chart colors, line styles, and markers (i.e., symbols) can be changed in the Chart Designer. The formatting may be applied to all data series or an individual series in the chart.

To format series lines:

1. To select a particular line for formatting, do one of the following. a. Double-click on that line in the chart. The Chart Designer opens to the appropriate tab, or b. With the chart in the active window choose Chart>Chart Designer from the WinNonlin menus. The Chart Designer opens. Then click Series in the tree view for all series, or use the + to open the list of series and select one. 2. Click the Lines tab.

3. To hide the line, uncheck the Show Series Line check box. 4. Otherwise, select the desired line style, width or color. 5. To set how line segments join at data points use the Join menu. 6. To specify how the ends of lines are displayed select from the menu for Caps. 7. Click Apply or OK. Changing axis scaling Charts use automatic scaling by default. As a result, axis scales may change when the chart window is resized. The axis scaling can be set to a fixed scale.

125

5

WinNonlin User’s Guide

To change the scale of an axis:

1. Double-click on the axis or choose Chart>Chart Designer from the WinNonlin menus. The Chart Designer opens. 2. Select the appropriate axis in the tree view pane. 3. To change the scale sizes use the Value Scale tab. 4. Make sure that the Automatic check box is unchecked to set a fixed scale. 5. Enter the minimum and maximum value to appear on the axis. 6. Enter the total number of major divisions. 7. Enter the number of minor divisions per major division.

This example creates an X axis that ranges from 0 to 4, with a major division at every multiple of 0.5, and a minor division at every quarter. 8. Click Apply or OK. Creating a semi-log plot A semi-log plot can be created quickly by using the Axis Options dialog or Chart Designer. Use the Chart Designer to set the log base.

To create a semi-log plot setting the Axis Options:

1. Select Axis Options from the Chart menu. The Axis Options dialog appears. 2. Select the Logarithmic check box for the Y-axis.

126

Charts Frequently used chart formatting options

5

3. Click Apply or OK.

To set the log base using the Chart Designer:

1. Open the chart and double-click on the Y axis or right-click on the chart and choose Chart Designer from the pop-up menu. 2. Select the Y axis in the tree view as shown below. 3. Open the Scale Type tab to the right. 4. Select the Logarithmic check box and enter the desired log base.

**5. Click Apply or OK. Setting grid lines
**

To remove the grid lines from a chart:

1. Open the chart and double-click on the Y axis or right-click on the chart and choose Chart Designer from the pop-up menu. 2. Select the X-axis in the tree view.

127

5

WinNonlin User’s Guide

3. On the Pens tab change the Major Grid Pen and Minor Grid Pen Style to the desired line style. Use NULL to show no grid lines. 4. Click Apply.

5. Repeat for the Y-axis. 6. Click Apply or OK. Removing or adding walls Chart walls form a border at the maximum X and Y values, enclosing the chart in a rectangle.

To set chart walls:

1. Double-click on the chart or right-click on the chart and choose Chart Designer from the pop-up menu. 2. Select Plot from the tree view menu. 3. Select the Walls tab.

128

Charts Chart templates

5

4. In the Pen group box, select the line style or change it to NULL for none. 5. Click Apply or OK.

Chart templates

WinNonlin provides the option to create chart templates containing customized settings, including chart type and formatting set in the Chart Designer.

To save a chart template:

1. Use the Chart Wizard or Chart Designer to create and format a plot with all the options to be saved in the template. 2. With the chart window selected, choose Chart>Save Template from the WinNonlin menus. 3. Set the file location and name. Save the file with the extension .GTP. 4. Click Save.

To load a graph template:

1. Create a chart with the desired content. The chart must contain the same number of X, Y, sort, group and error variables as the template. 2. With the chart selected, choose Chart>Load Template from the WinNonlin menus. 3. The Load Template dialog appears. Select the location and file that contains the graph template (*.GTP).

129

5

WinNonlin User’s Guide

4. Click Open. The graph is now updated with the formatting established in the template. Note: If a template that contains formatting for one data series is applied to a chart that contains multiple series, the template is applied only to the first series.

**Copy and paste
**

WinNonlin provides the capability to copy a graph paste it into another Windows application in Windows Metafile (*.WMF) format. Note: The Export to Word tool bar button exports the active object directly to Microsoft Word, and can be used to export an image of the current chart.

To copy and paste a chart:

1. Select the chart window containing the chart to copy. 2. Select Copy from the Edit menu. A copy of the chart image is placed on the clipboard in Windows Metafile (*.WMF) format. 3. Switch to the other application and (depending upon which application is being used), use the Paste or Paste Special function to paste the image. Note: In order to support cut and paste, the Windows Character Map is limited to the 255 characters noted in the Windows 98 ANSI version. Windows 2000 has multiple Character Mapping options (use Programs>Accessories>System Tools>Character Map in the Windows Start menu to view them). Windows: Western must be selected as the default option in order to cut and paste the characters supported in WinNonlin. Look to the lower, right corner of the Character Map to see if “Keystroke: Alt+XXXX” is within the 255 range for any special characters (characters not included on the computer keyboard).

130

Chapter 6 Tables Creating formatted tables in WinNonlin The Table Wizard generates a report. • • • • • “Table Wizard” on page 131 “Summary statistics” on page 144 “Significant digits and column alignments” on page 145 “Table formatting” on page 146 “Table definition files” on page 148 Table Wizard Follow these steps to create a new workbook containing a formatted table. with or without calculating summary statistics. 3. click on it again return to the Table Wizard. See the following for instructions and options. A sample of the selected template appears at the top of the Table Wizard. Make sure that the desired data are selected as the active worksheet in the workbook(s). 4. 131 . See “WinNonlin table templates” on page 134 for details and examples of each template. with flexible formatting options. Select a table template from those listed in Table 6-1.and analysis-ready table of worksheet data and (optional) summary statistics in a new WinNonlin workbook. 1. Open one or two data sets to provide table data. Nine table templates provide a choice of layouts and summary statistics. 2. Select Tools>Table Wizard from the WinNonlin menus or click the Table Wizard tool bar button to open the Table Wizard. Click on the sample to expand it.

Table templates Template Template 1 Template 2 Template 3 Template 4 Template 5 Template 6 Template 7 Template 8 Template 9 Description Summary statistics for variables in one data set. as shown for the variable Form below. Raw data and summary statistics from one data set by cross variable values. set the join variables as described under “Joining data sets” on page 143. For template 9. Note: Template 9 is available only when at least two workbooks are open. Summary statistics for one data set by cross variable values. select the workbook containing raw data in the left list. As template 6. For template 9. with columns ordered by cross variable within variable. Select the dataset to be used from the drop-down list near the lower left corner of the Wizard.17 1470.00 15.00 20. 7. group variables appear within the table. and the workbook containing computed values in the right list. 5. Form Capsule Subject 1 1 1 1 1 Time 0.00 10. demarcating the separate data groupings. each value of the group variable(s) can generate a page break. Raw data and summary statistics for one data set. Raw data from one data set. 132 . joining data from two workbooks.6 WinNonlin User’s Guide Table 6-1.00 2069.32 1914. Choose File>Options from the Table Wizard menus to set options for page breaks between group variable values (optional). See also “Tables” on page 31 for details on each table option. 6.00 Conc 0. Assign variables: Drag the variables to be included from the list at the bottom left to the appropriate boxes to the right. By default. As template 6.00 340.92 Alternately. As template 6.00 5. Raw data from one data set by cross variable values. with the group variable within the table or above it as shown below. with summary statistics in columns instead of rows.

32 1914. Click Create to create the table in a new WinNonlin workbook. 8. For example. click the button next to that variable. Set significant digits as described under “Significant digits and column alignments” on page 145.00 15. on the Workbook tab. indicating the page break settings. a page break will be inserted when the value of that group variable changes. Set formatting as detailed under “Table formatting” on page 146. As a result. (The representation of missing values is set in the Tools Options dialog. go to “Summary statistics” on page 144. 10.Tables Table Wizard Form= Capsule 6 Subject 1 1 1 1 1 Time 0. when N = 1 the SD cannot be calculated.92 A button will appear to the left of each group variable in the Table Wizard. Click Next. Click the Prev button to modify table settings if needed.00 5. Click Finish to create a preview of the table. that row will not appear in the table.00 340. 133 . 12.00 20. If an entire row is excluded.00 2069.00 Conc 0. the cell that would contain SD is blank in the resulting table.00 10. 13. 11. If the template includes statistics.17 1470. Missing and excluded data If a value needed for any statistical calculation is missing. If the button contains an X. 9. the table cell designated for the result will be left blank. To toggle the setting for a specific variable.) Excluded values in the data set are treated as missing in the calculation of summary statistics. See “Workbooks” on page 28.

The statistics are computed separately for each unique combination of values for the group variable(s). for each drug formulation. Template 1 This template computes and displays summary statistics for each column included under Variables.6 WinNonlin User’s Guide WinNonlin table templates The Table Wizard provides the following nine templates with different layouts and summary statistics usage. if any. median and maximum concentration and blood pressure for each subject. Template 9 joins data from two workbooks. 134 . This example computes the minimum.

This example lists concentration and blood pressure measurements for each subject.Tables Table Wizard 6 Template 2 This template lists raw data for each column selected under Variables. Data are sorted alphanumerically by group variable values. for each drug formulation. then by ID variable values. in the order listed. 135 . at each time in a crossover study. It does not generate summary statistics.

and summarizes those values for each subject and formulation. The example below lists concentrations and blood pressure measurements at each time. 136 . if any. and summarizes it for each unique combination of group variable values.6 WinNonlin User’s Guide Template 3 Template 3 adds summary statistics to template 2. It lists the raw data for each variable at each value of the ID variable(s).

137 . It computes separate statistics for each unique combination of group and cross variable values.Tables Table Wizard 6 Template 4 Like template 1. this template generates summary statistics and displays them without the raw data.

Subjects appear in separate table rows with each formulation in a separate set of columns. with cross variable values in adjacent columns.6 WinNonlin User’s Guide This example reports minimum. Template 5 This template reports raw data for each column in the Variables list. with the two formulations in adjacent sets of columns. A separate listing is included for each unique combination of values for the group and cross variables. 138 . This example reports raw concentration and blood pressure data for each subject at each time. median and maximum concentration and blood pressure for each subject and each drug formulation. at each ID variable value.

Tables Table Wizard 6 Template 6 This template adds computation of summary statistics to the listing of raw data as presented in template 5. Statistics are computed separately for each unique combination of group and cross variable values. 139 .

6 WinNonlin User’s Guide Template 7 This template presents the same information as template 6 with the columns re-ordered. all cross variable values are grouped together. In template 7. the regular variable columns are grouped together within values of the cross variable. In template 6. with one set for each regular variable. 140 .

within group and cross variable values.Tables Table Wizard 6 Template 8 Most templates present raw data in columns. Statistics are calculated for each ID variable value. 141 . Template 8 presents raw data in rows with summary statistics in the right-most columns.

It generates summary statistics for each unique combination of group and ID variable values. 142 . The two data sets are “joined” on these group and ID variable values. The template is designed to present raw data (aggregate variable) at different sample times (DS1 cross variable) and PK parameters (DS2 variables) in columns.6 WinNonlin User’s Guide Template 9 Template 9 joins data from the active worksheet in each of two workbooks. with a row for each subject (ID variable). For example. it can combine raw PK data from an input workbook with PK parameter estimates from a modeling output workbook.

” in the wizard The workbook holding PK parameters or other statistics must be selected from the right drop-down list. The Table Wizard joins the datasets by matching values for group and ID variables. one from each workbook. 143 ..Tables Table Wizard 6 The workbook containing raw data must be selected from the left drop-down list under “Please select datasets.. Joining data sets Template 9 merges two workbooks into one table. The group and ID variables are selected in matched pairs.

Repeat to set all needed pairs. 3. (See “Template 9” on page 142. then click Done to return to the Table Wizard. Click Add Join. (See “WinNonlin table templates” on page 134.6 WinNonlin User’s Guide To define a join: 1.) 144 . They need not have identical names. Select one variable from each data set to make a matched pair. In the Table Wizard. The Select Join Variables dialog appears. which are a part of most table templates. One set of statistics will be computed for each unique combination of group variable values. ID variables: Click Define Join for ID variables at the lower right and repeat the steps above to match variables that uniquely identify individual data profiles. 6. 5.) 2. to match group variables from the two data sets. Summary statistics The Table Wizard can compute summary statistics. select template 9 and select the two data sets that will provide table data. 4. but must represent the same entity with an overlapping set of values. Group variables: Click Define Join for group variables at the lower right. The values of these variables will be used to merge records from the two data sets.

Note: The statistics in the table will appear in the order listed here. To select all of the available summary statistics. click Next to set the Significant digits and column alignments. See “Output and computational formulae” on page 153 for details on each statistic. When finished. enter the desired percentage in the Confidence Interval box. This selection is made separately for each column and summary statistic in the table. Check the check box preceding each summary statistic to be included in the table. click All. Significant digits and column alignments The Table Wizard allows the user to select the number of decimal places or the number of significant digits to be used. 3. 145 . To include a confidence interval at a confidence level other than 95%.Tables Significant digits and column alignments To select the summary statistics to be included in a table: 6 1. 2.

borders or formatting. then enter the number of significant digits or decimal places to the right. including page numbers and footnotes. headings and page layout. click the button under Significant digits/Decimal places to select which you will enter. column headers. choose Decimal places and enter 0 as the number of places. Subject or Period. After the settings are complete. click the button under Alignment to select the alignment within table cells. 146 . Check Data Only at the top left corner of this dialog to create a table (worksheet) without page breaks.6 WinNonlin User’s Guide Significant digits: For each table column and summary statistic listed. Column alignments: For each column and summary statistic.g. click Next to proceed with Table formatting. e. Note: For variables with integer values. Table formatting The final dialog of the Table Wizard provides the means to format table cells.

Heading). To underline text for an item (e. 4.Tables Table formatting 6 1. select Page Footers under Item and check Page numbering below the footer text and enter the first page number under Start at. Landscape to orient it horizontally. To include page numbers at the bottom of each page. font size. options for units display appear at the bottom right. select Page Footers under Item and enter the text under Footer Text. Variable (Unit) to present units in parentheses after the variable name or Variable. Select an item to format from the pull-down list called Item.g. To include a table title on each page. Click the Set Font button to change the font. header or footer relative to the page. To include text at the bottom of each page. Check None to exclude units from the table. Check the appropriate box under Alignment to align the table title. 8. 2. 6. following the variable name. 147 . Insert line breaks by pressing the ENTER key. If the input workbook included units. or font style for that item. 5. Select the appropriate radio button in the Page Layout group box: Portrait to orient the long side of the page vertically. 3. Make adjustments and click OK. check the Underline box. select Page Headers under Item and type a title under Header Text. (Click the Prev button to set table cell alignments.) 7. Unit to present units after a comma. The Font dialog appears. 9..

To create and save a table definition file (*. 3.TDF) stores a user-defined table format. 4. Choose File>Save Definition As (Alt+F. To use a table definition file.6 WinNonlin User’s Guide Table definition files A table definition file (*.tdf): 1. Select a drive and directory and enter a descriptive file name. The table definition file can then be reused to quickly create tables with that format. Click OK. by creating a table with the desired formatting. The table definition file includes: • • • • • • Table template number Names and mappings of table variables (data columns) Summary statistics to be included (if applicable) Titles (up to five lines) Footers (up to five lines) Formatting: – Alignment and decimal places or significant figures for the data – Table and page formats from the final dialog of the Table Wizard Creating a table definition file Table definition files must be created in WinNonlin. Follow the instructions under “Table Wizard” on page 131 up to the table preview (just before clicking the Create button).TDF. The Save Current Settings dialog appears. variables. statistics (optional) and formatting. a data set must include the exact column names used in creating the definition file. using the file extension . to set up a table with the desired template. 148 . then saving the settings from within the Table Wizard. Table definition files are also useful for generating formatted tables via scripting. S) from the Table Wizard menus. 2.

149 . 5. The Table Wizard appears. 2. 3. as detailed under “Table Wizard” on page 131.Tables Table definition files 6 Loading a table definition file To load a table definition file (*. 4. Choose Tools>Table Wizard from the menus or click the Table Wizard button on the tool bar. Select the appropriate workbook. Open or create the worksheet containing data for the table. Make sure it uses the exact column names used in the Table Definition File. Specifications made via the table definition file will be imported into the Table Wizard. The table can be edited if needed.tdf) in the Table Wizard: 1. Select the appropriate table definition file and click Open. Choose File>Open Definition from the Table Wizard menus.

6 WinNonlin User’s Guide 150 .

2. to summarize modeling results. Computing summary statistics To set up descriptive statistics: 1. Separate statistics for subgroups are obtained through the use of sort variable(s). Open the input data worksheet. or to test for normal distribution of data. or click the Descriptive Statistics button on the tool bar. This feature is frequently used to create data to plot means and standard errors. The data set must be open and not hidden. 151 . Choose Tools>Descriptive Statistics from the WinNonlin menus. for preclinical summaries.Chapter 7 Descriptive Statistics Summary statistics and weighted summary statistics WinNonlin can compute summary statistics for variables in any workbook.

ninetieth. 5.7 WinNonlin User’s Guide The Descriptive Statistics dialog allows settings to be saved to an ASCII file for later re-use. Percentiles: Selecting the check box means that the first. ninety-fifth. 4. If output is combined onto one worksheet. Data points between 1. fifth.5 and 3 H-spreads from the hinges are marked by an o. Weight Variable: (Optional) See “Weighted summary statistics” on page 157. Select the options for the statistics: • • Confidence intervals: Enter the desired confidence interval in the % field. while pressing the Shift key. • • • 6. tenth. and the ninetyninth percentiles will be included in the output. the whiskers end at the maximum and minimum points. The whiskers end at the data points farthest from the median that are still within 1. 152 . or by highlighting the variable name then. typing the underlined letter in the name of the destination box. and the those outside 3 H-spreads are marked by x. Click Calculate. 3. A separate analysis is done for each unique combination of sort variable values. twenty-fifth. If there are fewer than five data points. Select sort variable(s) if desired. The box hinges are the 25th and 75th percentiles. Note: Select variables either by dragging the variable name to the appropriate box. seventy-fifth. through the Save Setup and Load Setup buttons. no units will appear with the output. The median is plotted as an asterisk (*). Box and Whiskers plot: The Box and Whiskers plot appears in a separate text window. See “Units” on page 153. Drag the variable(s) to be summarized to the Summary Variables box. Create Sheets for Summary Variables: When there are multiple summary variables the output can be sent to one or multiple worksheets in the output workbook. fiftieth. the data are graphed individually rather than as a box.5 times the H-spread (the distance between the hinges) of the hinges. If there are no outliers beyond this range.

Descriptive statistics output Statistic N Nmiss Nobs Mean SD Description Number of observations with non-missing data Number of observations with missing data Number of observations Arithmetic Average Standard Deviation: Variance SE Standard Error: SD ------N 153 .Descriptive Statistics Computing summary statistics 7 Units When summary statistics are calculated on a variable with units. Assuming that the variable summarized is x and has x-units specified. Nmiss. certain of the calculations will have units. CI Upper Var Everything else Units No units No units No units No units No units x-unit2 x-unit2 x-unit Output and computational formulae Summary statistics include the following computations. the units for the summary statistics are: Statistic N. CV% Geometric Mean Skewness Kurtosis Mean log. SD log Variance CI Lower Var. Table 7-1. Nobs CV%.

7 WinNonlin User’s Guide Table 7-1. Descriptive statistics output (continued) Statistic Variance Unbiased sample variance: N Description ∑ ( Xi – X ) N–1 Min Median Max Range CV% Geometric Mean Harmonic Mean Minimum value Median value Maximum value 2 i=1 ------------------------------- Range of values (maximum value minus minimum value) Coefficient of variation: (SD/Mean)*100 Nth root of the product of the N observations Reciprocal of the arithmetic mean of the reciprocals of the observations 154 .

+ … + ----⎞ ⎝x x x x x ⎠ 1 2 i–1 x+1 n (n – 1) = ----------------------------⎛ n ⎞ 1 1 ⎜ ---⎟ – --⎜∑ x⎟ x j⎠ i ⎝j = 1 Mean log SD Log CV% Geometric Mean Arithmetic average of the natural logs of the observations Standard deviation of the natural logs of the observations Coefficient of variation of the geometric mean: 2 exp ( ( SD_log ) ) – 1 × 100 SD Mean – t α ⁄ 2 ------N CI X% Lower Mean Lower limit of an X% confidence interval for the mean: 155 . Descriptive statistics output (continued) Statistic Pseudo SD Description 7 Jackknife estimate of the standard deviation of the harmonic mean..+ ---. For n points. xn the pseudo standard deviation is: n pseudo SD = where ( n – 1 ) ∑ ( Hi – H ) i=1 2 1 H = -.Descriptive Statistics Computing summary statistics Table 7-1.+ … + ---------.∑ H i n i=1 n and (n – 1) H i = ---------------------------------------------------------------------------------------------1..11 1 1⎛ ---.+ ----------. x1. ..

Otherwise. the quantile corresponding to the proportion p (0<p<1) is defined as: Q(p) = (1 . and χU2 cuts off an upper tail. (integer part) f = p*(n+1) .α)*100 is the percentage for the confidence interval. Coefficient of skewness: Skewness ∑ wi ( xi – xw ) ⁄ N -------------------------------------------------------3⁄2 2 [ ∑ wi ( xi – xw ) ⁄ N ] 3 156 .05. the empirical quantile is the smallest order statistic for j=0 or the largest order statistic for j=n. Note that for N>30. CI X% Lower Var Lower limit of an X% confidence interval for the variance: ( N – 1 ) × Variance ----------------------------------------------2 χU CI X% Upper Var Upper limit of an X% confidence interval for the variance: ( N – 1 ) × Variance ----------------------------------------------2 χL where χL2 and χU2 are from the χ2-distribution with N-1 degrees of freedom.7 WinNonlin User’s Guide Table 7-1. Descriptive statistics output (continued) Statistic Description CI X% Upper Mean Upper limit of an X% confidence interval for the mean: SD Mean + t α ⁄ 2 ------N where ( 1 – α ) × 100 is the percentage given for the confidence interval.j. of area α/2 where (1 . (fractional part) x(j) is the j-th order statistic. Thus. The above is used if 1 <= j < n. and t α ⁄ 2 is from the t-distribution with N-1 degrees of freedom.f) * x(j) + f * x(j+1) where j = int(p*(n+1)). the t-distribution is close to the normal distribution. χL2 cuts off a lower tail. For a sample size of n. P(ercentiles) The Pth percentile divides the distribution at a point such that P percent of the distribution are below this point. a 95% confidence level indicates that alpha = 0.

it is set to zero.Descriptive Statistics Weighted summary statistics Table 7-1. To create weighted summary statistics: 1. See “Computing summary statistics” on page 151. one for each sort variable. missing. using a column in the data set to provide weights. and the statistics below. Table 7-2. Output and computational formulae for weighted calculations The output for weighted summary statistics contains a column indicating the summary variable(s). Make all specifications as for un-weighted summary statistics. Weighted summary statistics output Statistic N Nmiss Nobs Description Number of non-missing observations (including those with weights = 0) Number of observations with missing data Number of observations (including observations with weights of 0) 157 . or excluded. 2. If a weight is non-numeric.– 3 2 2 [ ∑ wi ( xi – xw ) ⁄ N ] KS_pval Kolmogorov-Smirnov normality test p value. 4 Weighted summary statistics Summary statistics can be weighted. Descriptive statistics output (continued) Statistic Kurtosis Description Coefficient of excess (kurtosis): 7 ∑ wi ( xi – xw ) ⁄ N -------------------------------------------------. Drag the variable containing weights to the Weight Variable box (or highlight the variable to be used as the weight and press Shift+W).

– 3 2 2 [ ∑ wi ( xi – xw ) ⁄ n ] 4 158 . using weighted variance Upper limit of an X% confidence interval. using weighted mean and SD CI X% Lower Var CI X% Upper Var Skewness Lower limit of an X% confidence interval. using weighted variance Weighted coefficient of skewness: ∑ wi ( xi – xw ) ⁄ n ------------------------------------------------------3⁄2 2 [ ∑ wi ( xi – xw ) ⁄ n ] Kurtosis Weighted coefficient of excess (kurtosis): 3 ∑ wi ( xi – xw ) ⁄ n ------------------------------------------------.7 WinNonlin User’s Guide Table 7-2. using weighted mean and SD CI X% Upper Mean Upper limit of an X% confidence interval. Weighted Coefficient of variation: (Weighted SD/Weighted Mean)*100 CI X% Lower Mean Lower limit of an X% confidence interval. Weighted summary statistics output (continued) Statistic Mean Description Weighted Arithmetic average: ∑ wi × xi ---------------------∑ wi SD Weighted Standard Deviation: Weighted Variance SE Weighted Standard Error: Weighted SD -----------------------------N Variance Weighted Variance: ∑ wi ( xi – xw ) ----------------------------------N–1 CV% 2 where x w is the weighted mean.

and moments in richlysampled time-effect data. urine or PD data WinNonlin's noncompartmental analysis (NCA) engine computes derived measures from raw data. “Model options” on page 173 NCA computations and output parameters are covered under “Computational rules for WinNonlin’s noncompartmental analysis engine” on page 181 159 . “NCA model selection” on page 160 2.or sparsely-sampled data. using methods appropriate for either richly. “Partial areas” on page 167 5. for concentration data from blood/plasma (single dose and steady state) or urine (single dose only). “Therapeutic response” on page 168 6. These models support rich data sets as well as sparsely-sampled studies such as toxicokinetic studies. • Six PK models: extravascular. “Lambda Z or slope estimation settings” on page 162 4. One PD model for analysis of slope. “Dosing regimen” on page 171 7. WinNonlin provides the following noncompartmental analysis “models” for different types of input data. areas. • The following sections detail the steps to set up a WinNonlin NCA. intravascular (IV) infusion or IV bolus dosing. 1. “Model variables” on page 161 3. height.Chapter 8 Noncompartmental Analysis Areas and slopes for plasma.

Select the appropriate NCA model. 8. and click Finish to load the model. Click Next. Sparse data sets can include up to 250 subjects. Click Next. 160 . To analyze data with few observations per subject. The next step in NCA is setting Model variables. Start the PK/PD/NCA Analysis Wizard by clicking on that tool bar button or selecting that command from the Tools menu. 4. 6. 3. Note that the Model 200 is highlighted and selected by default. See “Sparse sampling computations” on page 188 for discussion. Open the worksheet containing data to be modeled. See “Noncompartmental analysis” on page 505 for details on the models and output parameters. and select it as the active worksheet (click on that worksheet to bring it to the front). 5. 7. Confirm that the correct model is selected. and “Sparse sampling (pre-clinical) data” on page 514 for additional parameters computed.8 WinNonlin User’s Guide NCA model selection To load a noncompartmental analysis model: 1. check Use Sparse Sampling. The model window will open with a sample graph. Select the Noncompartmental radio button. 2.

Click OK. 3. Lower times: the starting times for individual collection intervals. Carry Along Variables: data to be copied into the output workbook. subject ID and treatment in a crossover study. Volume: the volume of urine collected per time interval. e. observed values of interest. Drag and drop columns to assign variables to be modeled as detailed under “Model variables” on page 161. drug concentration in urine. • • • Concentration: the dependent variable. 2.e. After NCA model selection. drag and drop columns to assign variables to be modeled (see “Urine models” on page 161 for urine models): • • X: the independent variable. Drag and drop to specify the following (optional): • Sort Variable(s): categorical variable(s) identifying individual data profiles. • Note: The output data set will automatically include the original values of the sort variable(s) and variables being modeled (X and Y for plasma models or midpoint and rate for urine models).. 161 . generally time. choose Model>Data Variables from the WinNonlin menus to open the Data Variables dialog.g. 4. For sparse data sets.Noncompartmental Analysis Model variables 8 Model variables To specify data columns for each variable in a noncompartmental analysis: 1. in addition to the estimated parameters. drag and drop to assign the variable containing subject identifiers into the Subj_ID field. Lambda Z or slope estimation settings are the next step in NCA setup.. Y: the dependent variable. Urine models include the following variables. Urine models For urine models. WinNonlin computes the midpoint and excretion rate (amount eliminated per unit of time) for each collection interval. i. For plasma models. A separate analysis is done for each unique combination of sort variable values. 5.

the regression with the larger number of points is used. parameters for concentration data will be extrapolated to infinity. If λz can be estimated. etc. associated with the terminal elimination phase for concentration data. For each regression. WinNonlin repeats regressions using the last three points with non-zero concentrations. These measures will be estimated using linear or linear/log trapezoidal rule. last five. For concentration data During the analysis. then the last four points. Lambda Z or slope estimation settings WinNonlin will attempt to estimate the rate constant. Lambda Z or slope range selection If the user does not specify the data points to include in the calculation of λz or slopes. Points with a value of zero for the dependent variable are excluded.0001 of the largest adjusted R2 value. as selected in the Model Options (see “Model options” on page 173). and calculated from at least three data points. or slopes of up to two selected data ranges for drug effect data. yet does not choose disable curve stripping as described under “Setting Lambda Z or PD slope ranges” on page 164. λz. 162 .8 WinNonlin User’s Guide • Upper times: the ending times for individual collection intervals. but is within 0. WinNonlin estimates λz using the regression with the largest adjusted R2 and: • • If the adjusted R2 does not improve. an adjusted R2 is computed: 2 ) Adjusted R = 1 – ( 1 – R ) × ( n – 1 -----------------------------------------(n – 2) 2 where n is the number of data points in the regression and R2 is the square of the correlation coefficient. Points prior to Cmax or prior to the end of infusion are not used unless the user specifically requests that time range. then WinNonlin will determine the data points to include in λz or slope calculations as follows. NCA does not extrapolate beyond the observed data for drug effect. λz must be positive.

The actual slopes are reported.Noncompartmental Analysis Lambda Z or slope estimation settings 8 For sparse sampling data The Lambda Z ranges for sparse data show the mean concentration at each time (plasma or serum data) or mean rate for each interval (urine data). Data points for Slope1 are marked with “1” in workbook output and footnoted using “#” in text output. 163 . with linear or log regression. Calculation of slopes for effect data WinNonlin estimates slopes by performing a regression of the response values or their natural logarithm. If the user specifies the range only for Slope1. For drug effect data WinNonlin will compute the best-fitting slopes at the beginning of the data and at the end of the data using the same rules that are used for lambdaZ— roughly. It is the user’s responsibility to evaluate the appropriateness of the estimated value. then in addition to computing Slope1. they are not negated as for λz. The data points included in each slope are indicated on the Summary table of the output workbook and text for model 220. “*”. Calculation of Lambda Z Once the time points being used in the regression have been determined. then WinNonlin will also compute the best-fitting slope at the beginning of the data for Slope1. data points for Slope2 are labeled “2” and footnoted using an asterisk. WinNonlin will compute the best-fitting slope at the end of the data for Slope2. λz is defined as: λz = -1 (estimated slope) Note: Using this methodology. WinNonlin will almost always compute an estimate for λz. best adjusted R-square with at least three points. WinNonlin can estimate λz by performing a regression of the natural logarithm of the concentration values in this range on sampling time. If the user specifies the range only for Slope2. according to the user's choice.

AUClast. The instructions below are for the graph view. for model 220. (Individual profiles are defined by the sort variable(s) in “Model variables” on page 161). Setting Lambda Z or PD slope ranges To set the λz range for each PK profile. The user-specified range contains fewer than two non-missing observations. the Slopes tab of the Model Properties dialog (Model>Lambda Z Ranges or Model>Slopes).8 WinNonlin User’s Guide Limitations of Lambda Z and slope estimation It is not possible for WinNonlin to estimate λz or slope in the following cases. In these instances.) will still be reported and the software will issue a warning in the text output. Automatic range selection is activated. indicating that λz was not estimable. parameters that do not depend on λz (i. Tmax. • • • • • There are only two non-missing observations and the user requested automatic range selection.e. automatic range selection is activated. The Graph and Data buttons at the top right corner of the tab switch between a graphical or tabular interface to set time ranges for each data profile. etc. Cmax. and there are fewer than 3 points at or after the infusion stop time. For infusion data. or slopes to compute for each PD profile: 1. The time difference between the first and last data points in the estimation range used is approximately < 1e-10. Open the Lambda Z Ranges tab or. 164 . but there are fewer than 3 positive concentration values at or after the Cmax of the profile.

To turn off λz or slope estimation for all profiles. enter a start time that is greater than the last time point in a given profile. and an end time greater than the start time. select whether each slope should be computed using linear regression (Lin Reg) or regression on the log data (Log Reg). 165 .Noncompartmental Analysis Lambda Z or slope estimation settings 8 2. With this option checked: – λz or slopes will not be estimated. 3. Select the axis scale for the graphical representation below the lower left corner of the plots: Linear or Log Linear. check Disable Curve Stripping at the bottom left and skip to step 8. Note: To disable curve stripping for one or more individual profiles. 4. – Parameters that are extrapolated to infinity will not be calculated. For slopes.

Best Fit: Select this option to have WinNonlin determine the Lambda Z or Slope ranges as described under “Lambda Z or slope range selection” on page 162. the excluded data points will still be included in computation of AUC’s. etc. Data ranges (optional): For each profile. define the terminal elimination phase or slopes regions as follows. Use the right/left arrows just below the plot view to move between profiles. 166 . c. For slopes. Slope1 and Slope2. a. Select the last data point to be included: hold down the Shift key while left-clicking on a data point or enter a time value under End time. hold down the Ctrl key and left-click on that point.8 WinNonlin User’s Guide 5. Note: The units for Start and End time are those of observation times in the data set. Exclusions (optional): to exclude a data point from the calculations. User Time Range: Select the first data point to be included: left-click on a data point or enter a time value under Start time. use these steps to choose data ranges for up to two slopes. These exclusions apply only to Lambda Z or slope calculations. b. moments. 6. Repeat to reverse an exclusion.

by setting start and end times for each. The start and end times need not appear in the data set. 3. click OK. and the end time in the cell labeled AUC1 Upper. The time units are those of the input data set. if λz is estimable. click Apply to save the settings and move to another tab of the Model Properties dialog. Enter the start time in the cell labeled AUC1 Lower. 8. For another partial area. 167 . When done. 2. repeat this step for the next area (AUC2. and set values for each profile.Noncompartmental Analysis Partial areas 8 7. Open the Partial Areas tab of the Model Properties dialog (Model>Partial Areas or click the Partial Areas tool bar button). To specify the partial areas under the curve to be calculated: 1. and reload them for re-use. Up to 126 partial areas can be computed per profile. and can be after the last observed time. to exclude them. To save the settings and close the dialog. Partial area computations are optional. Use the arrow buttons directly below the plot to move between profiles. Partial areas The Partial Areas tab of the Model Properties dialog includes settings for computation of partial areas under the curve. simply leave the start and end times on this tab empty.). etc. Loading Lambda Z or slope ranges from an ASCII file The Load and Save buttons make it possible to store these time ranges to an external ASCII data file. See “Partial areas” on page 184 for additional information on partial area computations.

8 WinNonlin User’s Guide 4. and re-load it for later use. Parameters included depend on whether upper. For non-compartmental analysis of concentration data. Use the Save and Load buttons to save all settings in the Partial Areas tab to an ASCII data file. See “Therapeutic response windows” on page 186 for additional computation details. Repeat steps 2 and 3 for other profiles or use the fill-down method (drag bottom right corner of selected cells) to copy from one profile to those below. lower. or both limits are supplied. 5. Click OK when all settings are complete. 168 . Therapeutic response The Therapeutic Response tab of the Model Options dialog provides settings for computing times and areas during which observations fall above or below target values. WinNonlin will compute additional parameters noted in Table 8-1 below.

Noncompartmental Analysis Therapeutic response 8 Table 8-1. WinNonlin will make the computations detailed under “Drug effect data (model 220)” on page 515. Therapeutic window output for PK data Input values Parameters computed Lower and upper TimeLow TimeDuring TimeHigh AUCLow AUCDuring AUCHigh TimeInfDuring TimeInfHigh AUCInfLow AUCInfDuring AUCInfHigh TimeLow TimeHigh AUCLow AUCHigh TimeInfHigh AUCInfLow AUCInfHigh Time range for computation From dose time to last observation Description Total time below lower limit Total time between lower limit and upper limit Total time above upper limit AUC for time below lower limit AUC for time between lower limit and upper limit AUC for time above upper limit Total time between lower limit and upper limit Total time above upper limit AUC for time below lower limit AUC for time between lower limit and upper limit AUC for time above upper limit Total time below lower limit Total time above lower limit AUC for time below lower limit AUC for time above lower limit Total time (extrapolated to infinity) above lower limit AUC for time (extrapolated to infinity) below lower limit AUC for time (extrapolated to infinity) above lower limit Lower and upper From dose time to infinity using λz (if λz exists) Lower only From dose time to last observation Lower only From dose time to infinity using λz (if λz exists) For drug effect data (model 220). To compute time and AUC spent within a therapeutic concentration range: 1. Open the Therapeutic Response tab of the Model Properties dialog (Model>Therapeutic Response or click the Therapeutic Response tool bar button). 169 .

a value must be entered here. 2. Open the Therapeutic Response tab of the Model Properties dialog (Model>Therapeutic Response or click the Therapeutic Response tool bar button). if no baseline value is entered. Enter the lower and upper (optional) concentrations delimiting the target range and click OK. as baseline.) 170 . Baseline: baseline effect value for each profile. (See “Additional parameters for model 220 with threshold” on page 517. enter the following: a. In that case. if any. To set the baseline and (optional) threshold values for effect data (model 220): 1. Threshold (optional): effect value for use in calculating additional times and areas. Required unless dosing data are provided in a Dosing worksheet (as when data are loaded from PKS). For each profile. WinNonlin will use the effect value at dose time. If there is no data point at dose time. b.8 WinNonlin User’s Guide 2.

enter Current Units of mg and Dose Normalization of kg. Dose Normalization: If dose amount is normalized by subject bodyweight or body mass index. Enter dose information in the Dosing Regimen dialog. 2. if doses are in milligrams per kilogram of bodyweight. WinNonlin also assumes equal dosing intervals. 3. Choose Model>Dosing Regimen from menus or click the Dosing Regimen tool bar button to open the Dosing Regimen tab of the Model Properties. Dosing regimen Models 200 to 202 (Plasma) assume that data were taken after a single dose. WinNonlin will assume a time of zero. if Volume units were L. For example. as described below. dose normalization of kg would change those units to L/kg. For steady state. Current Units: Enter the units or select one item under Mass Prefix and Mass Unit then click the Add button. Click OK when finished. or read it from a worksheet as detailed in “Dosing data variables” on page 202. or after a final dose at steady state. Models 210 to 212 assume single-dose urine data. If dose time is not entered. This will affect output parameter units. For example. Dose normalization affects units for all volume and 171 . select the appropriate factor in this drop-list.Noncompartmental Analysis Dosing regimen 8 3. To enter dosing via the dosing regimen dialog: 1.

8 WinNonlin User’s Guide clearance parameters in NCA and PK models. as well as AUC/D in NCA plasma models. 7. enter the dose amount under Value for each row in which Constant = Dose. Excluded profiles appear in red (see “Data exclusion and inclusion” on page 60) 5. based on the tau value set above. 4. and Percent_Recovered in NCA urine models. Dose: For concentration data (models 200-212). For drug effect data (model 220): Using the same time scale and units as the input data set. enter the dose time under Value for each row in which Constant = Time of Last Dose. in most studies. Time deviations in steady-state data When using steady-state data. Click Apply to commit changes or OK to commit and close the dialog. enter the time of the most recent dose for each profile under Value for each profile. Use the Save or Load button to save or re-load dosing information using an external ASCII file. 172 . WinNonlin computes AUC from dose time to dose time + tau. 9. 6. See “Special file formats” on page 430 for file format. check Steady State and enter the dose interval for each row in which Constant = Tau. Time of last dose: If the dose was given at a time other than 0. Steady state: For steady state dosing (models 200-202 only). However. 8.

inclusion and precision for output parameters To set NCA model options: 1.Noncompartmental Analysis Model options 8 there will be sampling time deviations. Cmax. For steady state data. The Model Options dialog appears. That is. and format of final parameters worksheet • Default units for output parameters Parameter names • Names. Click Apply to set options or OK to set options and close the dialog. chart. 3. 2. but no later than dose time+tau. Click on a tab to select and adjust a group of options from those in Table 8-2. Model options Options for NCA output are divided into the following sections. ASCII and chart output • Method for computing area under the curve. the program will estimate the AUC based on the estimated concentration at 24 hours. with content as selected in the following NCA Model options. Cmin and Tmin are found using observations taken at or after the dose time. and/or ASCII text formats for modeling results • Uniform or weighted least squares • Text title for workbook. Output options After noncompartmental analysis is run. Table 8-2. if dose time = 0 and tau = 24. Select Model>Model Options from the WinNonlin menus or click the Model Options tool bar button. one or more new windows is generated. NCA model options Category Output options Weight Title NCA settings Units Options • Workbook. 173 . Tmax.083 hours. the last sample might be at 23. Note: Set default output options via Tools>Options in the WinNonlin menus.975 or 24. In this instance. and not the concentration at the actual observation time.

If it is not checked. Page Breaks: Check this box to include page breaks in the ASCII text output Output intermediate calculations: If this option is checked. predicted Y for the regression. AUMC and weight used for the regression See “Partial areas” on page 167. See “History worksheet” on page 37. but in tabular form. These worksheets present the same information as the text output. User defined settings History of actions done in and to the workbook. . 174 . The NCA workbook contains the worksheets summarized in Table 8-3. profiles with insufficient data will output missing parameter values. and for each of the sub-areas from partial area computations. containing the same information as the workbook output. the text output for the next analysis only will include values for iterations during estimation of λz or slopes. check this box to generate an NCA Chart window with a plot of observed versus predicted data for each profile. Summary of NCA worksheets Sheet name Final Parameters Dosing Summary Table Contents Estimates of the final parameters for each level of the sort variable The dosing regimen for the analysis The sort variables. Workbook: Check this option to generate an output workbook. profiles with all or many missing parameter estimates are excluded from the results. points included in the regression forλz or slopes. AUC. as shown below. Text: This option provides an ASCII text version of the analysis results.8 WinNonlin User’s Guide Exclude profiles with insufficient data: If this option is checked. X. residual for the regression. Partial Areas User Settings History Chart: If workbook output is selected. Modeling errors are also reported in this file. Table 8-3. Y. See “Data deficiencies that will result in missing values for PK parameters” on page 182 for a listing of cases that produce missing estimates and are excluded using this option.

175 . selections on the Weight tab of the Model options dialog apply to the regression that estimates λz or slopes.Noncompartmental Analysis Model options 8 Weight For NCA. Sparse sampling requires uniform weighting.

or use the Pre-selected scheme button to select the most common weighted least square options: 1/Y and 1/Y2. rather it is the relative proportions of the weights. and at the top of each chart in chart output. Select the Pre-selected Scheme radio button. If it is not. approxi- mately equal to 1/Yhat2. concentrations between 0 and 1 would have negative weights. Enter the title text below. on the User Settings worksheet in the workbook output. To use uniform weighting: 1. 176 . The title will be centered on the page. 2. Select Uniform Weighting from the pull-down list. rather than 1/Y. To include a title: 1. it already uses weighting implicitly. and therefore could not be included. To include a line break use Shift + Enter. Note: When weighting by 1/Y and using the Linear/Log trapezoidal rule. Title If included. if this were the case. one could argue that the weights should be 1/LogY. the title is displayed at the top of each printed page in ASCII text output.8 WinNonlin User’s Guide Note that it is not the weights themselves that are important. CAUTION: When a log-linear fit is done. However. the title will not display even if text is entered below. To use weighted least squares: 1. Select Observed to the Power n button and enter the value of n. Make sure that Title Text is checked on the Title tab of the Model options dialog. See “Weighting” on page 241 for discussion of weighting schemes. 2. It may include up to 5 lines of text.

given below. As above except when a partial area is selected that has an endpoint that is not in the data set. This method (method 2) is recommended for Model 220 (PD data). applied to each pair of consecutive points in the data set that have non-missing values. AUMC. In that case. given under “AUC calculation and interpolation formulas” on page 183. It uses the linear trapezoidal rule. If Cmax is not unique. and sums up these areas. WinNonlin inserts a final concentration value using: 1) the linear interpolation rule. is used to inject a concentration value for that endpoint. then the linear interpolation rule. and partial area computations. Linear Trapezoidal Method with Linear/Log Interpolation (method 4). • 177 .Noncompartmental Analysis Model options 8 NCA settings Calculation methods for AUC Four methods are available for the calculation of area under the curve. The default method is set on the Models tab of the WinNonlin options dialog (see “WinNonlin options” on page 28). • Linear Trapezoidal rule (Linear Interpolation) [default and recommended for effect data (model 220)]. or 2) logarithmic interpolation if the endpoint is after Cmax. The method chosen applies to all AUC. then the first maximum is used. in the NCA Settings tab of the Model options dialog. All methods reduce to the log trapezoidal rule and/or linear trapezoidal rule. If a partial area is selected that has an endpoint that is not in the data set. but differ with regard to when these rules are applied. or after C0 for bolus models (if C0 > Cmax).

entitled Transposed Final Parameters. and the logarithmic interpolation rule if the concentration is decreasing. rate or effect) is less than or equal to zero. will create a Final Parameters worksheet with one column for each parameter. the Final Parameters worksheet will present the untransposed format. the program defaults to the linear trapezoidal or interpolation rule for that point. WinNonlin converts values as necessary to display the output in the preferred units. and the Linear Trapezoidal (with Linear/Log Interpolation) methods all apply the same exceptions in area calculation and interpolation: If a Y value (concentration. If Cmax is not unique. the program defaults to the linear trapezoidal or interpolation rule. then the first maximum is used. Transpose Final Parameter Table This option. Note: The Linear/Log Trapezoidal. A second worksheet called Non-transposed Final Parameters will present one row for each parameter. and the second worksheet. Log interpolation is used to create points for partial areas after Cmax. This option is provided primarily for WinNonlin scripts that depend on the transposed format in the Final Parameters worksheet. if checked.8 WinNonlin User’s Guide • Linear/Log Trapezoidal Method (method 1). Units The Units tab of the Model options dialog (below) lists default units for output parameters and supports entry of preferred units. 178 . Uses the log trapezoidal rule (given under “AUC calculation and interpolation formulas” on page 183) after Cmax. will present the transposed version. if adjacent Y values are equal to each other. and linear interpolation is used otherwise. Similarly. and the logarithmic trapezoidal rule is used any time that the concentration data is decreasing. or after C0 for bolus (if C0 > Cmax). the Linear Up/Log Down. The linear trapezoidal rule is used any time that the concentration data is increasing. Linear Up/Log Down Method (method 3). Points for partial areas are injected using the linear interpolation rule if the surrounding points show that concentration is increasing. and linear trapezoidal otherwise. • See “AUC calculation and interpolation formulas” on page 183 for equations. or C0 for bolus models (if C0 > Cmax). If the Transpose Final Parameters option is not checked. See “Transpose final parameter table” on page 206 for example output from compartmental modeling.

Noncompartmental Analysis Model options To set a preferred unit: 8 1. Parameter names For noncompartmental analyses (NCA). Y-variable. WinNonlin provides the option to specify the variable names and precision for the final model parameters.g. and. PLASMAUNITS. 2.TXT. then load them back for later modeling. Use the Save and Load buttons to save preferred units as an ASCII text file. Case is preserved.TXT or URINEUNITS. Enter the unit under Preferred for the appropriate row. The units cannot contain space characters. dosing. Note: Preferred units can be set only after units are set for the X-variable. for plasma models.. 179 . Click Apply or OK. e.

NAM and MODELURINE. if it exists. Digits) for the parameter in that row. Each line contains: 180 . or nothing as the precision for that parameter. Sig digits: Click on a cell in this row to select whether to enter significant digits. Click on a cell under Include in Workbook to toggle between Yes (include that parameter) and No (exclude). Include in workbook: WinNonlin provides the option to include only a subset of final parameters in NCA workbook output. Precision: enter the number of significant digits or decimal places as (selected under Sig.NAM. The names cannot contain space characters. for example: MODEL202. MODEL202SS. All settings on the Parameter Names tab can be saved to a text file for later reuse by clicking the Save button (and loaded using the Load button). Case will be preserved. number of decimal places. As the parameters vary by model and with steady state dosing.8 WinNonlin User’s Guide Preferred name: Edit output parameter names as desired. save parameter information for specific models to different files. unless parameter names are set by company defaults (see “Default preferred parameter names” below). Default preferred parameter names Preferred parameter names can be set by a map file containing individual preferences or company standards. When WinNonlin opens the Parameter Names tab for an NCA model. The file lists WinNonlin NCA parameter names paired with preferred names. it will search its installation directory for an ASCII text file called DEFAULTNCAPARAMETERNAMES.MAP and load preferred parameter names from that file.NAM.

Precision where: ParameterName is WinNonlin’s default name for an NCA parameter.Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 ParameterName=PreferredName. then an integer indicating the number of significant digits or decimal places to use. Places” followed by a comma. Include equals “Yes” or “No” and indicates whether to include that parameter wherever NCA parameters are output. 181 . Include. Precision equals “Sig. • • • • • “Data checking and pre-treatment” on page 181 “AUC calculation and interpolation formulas” on page 183 “Partial areas” on page 184 and “Therapeutic response windows” on page 186 “Sparse sampling computations” on page 188 “NCA output parameters and their computation formulas” on page 506 Data checking and pre-treatment Prior to calculation of pharmacokinetic parameters. PreferredName is limited to alphanumeric characters and the underscore. Digits” or “Dec. Computational rules for WinNonlin’s noncompartmental analysis engine WinNonlin’s NCA computations are covered under the following headings. WinNonlin institutes a number of data-checking and pre-treatment procedures as follows.

8 WinNonlin User’s Guide Sorting Prior to analysis.: • Missing values: For plasma models. the lower time point is checked. then the first observed positive y-value will be used as an estimate for C0. that observation is excluded from the analysis. a concentration of zero. • Data exclusions NCA will automatically exclude unusable data points. if the lower time is greater than or equal to the upper time. or if the concentration or volume is negative.or Y-value is missing. for steady-state. i. the minimum observed during the dose interval. • • Data deficiencies that will result in missing values for PK parameters WinNonlin reports missing values for PK parameters in the following cases. If the regression yields a slope >= 0 or at least one of the first two y-values is 0. For urine models. For urine models. Data points for urine models: In addition to the above rules. records with missing values for any of the following are excluded: collection interval start or stop time. the data must be corrected before NCA can run. drug concentration. data within each profile is sorted in ascending time order.) Insertion of initial time points If a PK profile does not contain an observation at dose time (C0). Drug effect data: An effect value equal to the user-supplied baseline is inserted at dose time. If a weighting option was selected. IV bolus data: The program will perform a log-linear regression of first two data points to back-extrapolate C0. it is used in the regression. Dose time must be non-negative.e. if either the X. 182 . (A profile is identified by a unique combination of sort variable values. the associated record is excluded from the analysis. WinNonlin inserts a value using the following rules: • • Extravascular and infusion data: For single dose data. or collection volume. Data points preceding the dose time: If an observation time is earlier than the dose time.

No output is generated. AUCall. AUC calculation and interpolation formulas Linear trapezoidal rule t2 t1 AUC t2 t1 C1 + C2 = δt × ------------------2 AUMC t1 × C1 + t2 × C2 = δt × --------------------------------------2 183 . Amount_Recovered. AUClast. AUMClast all final parameters all final parameters All final parameters except Max_Rate. Percent_Recovered Bolus dosing and <=1 non-missing observation after dose time all final parameters Blood/plasma data with all zero values except one non-zero value at dosing time Urine data with all urine volumes equal to zero Urine data with all concentrations equal to zero Multiple observations at the same time point For NCA. it halts the NCA analysis and issues an error message.Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 Profile issue(s) Profile with no non-missing observations after dose time Blood/plasma data with no non-zero observation values Parameters reported as missing all final parameters All final parameters except Cmax. WinNonlin permits only one observation per profile at each time point. If it detects more than one.

Partial areas Rules for interpolation and extrapolation Partial areas within observed data ranges are calculated as described under “AUC calculation and interpolation formulas” on page 183 and “Calculation 184 . Note: If the logarithmic trapezoidal rule fails in an interval because C1 <= 0. then the linear trapezoidal rule will apply for that interval. or C1 = C2. C2 <= 0. then the linear interpolation rule will apply for that interval. Linear interpolation rule (to find C* at time t*) t* – t 1 C* = C 1 + -------------.× ( ln ( C 2 ) – ln ( C 1 ) )⎞ ⎝ ⎠ t2 – t1 Note: If the logarithmic interpolation rule fails in an interval because C1 <= 0.⎞ ⎛ C 2⎞ ln ⎝ -⎠ ln ⎝ ----.– δt × -----------------2 C2 ⎛ ----.⎠ C1 C1 where δt is (t2-t1).( C 2 – C 1 ) t2 – t1 Logarithmic interpolation rule t* – t 1 C* = exp ⎛ ln ( C 1 ) + -------------.⎞ ⎝ C 1⎠ AUMC t2 t1 t2 × C2 – t1 × C1 C2 – C1 2 = δt × --------------------------------------. or C1 = C2.8 WinNonlin User’s Guide Logarithmic trapezoidal rule AUC t2 t1 C2 – C1 = δt × -----------------C2 ln ⎛ ----. C2 <= 0.

then the log trapezoidal rule will be used. • • • Partial area boundaries The end time for the partial area must be greater than the start time. not “missing” or “BQL”) and Lambda Z is estimable. 185 . the corresponding Y value is interpolated between the first data point and C0.Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 methods for AUC” on page 177. because a logarithmic decline must be assumed in order to obtain the predicted values. If the start time for a partial area is before the last numeric observation and the end time is after the last numeric observation.) If a start or end time occurs after the last numeric observation (i. according to the AUC Calculation method selected in the Model Options dialog. then the log trapezoidal rule will be used for the area from the last observation time to the end time of the partial area.. (See “NCA settings” on page 177. the partial area will not be calculated. Both the start and end time for the partial area must be at or after the dosing time. then a linear or logarithmic interpolation is done to estimate the corresponding Y. The following rules apply if some portion of the partial area falls after the last valid observation: • If a start or end time falls before the first observation.e. Lambda Z is used to estimate the corresponding Y: Y = exp(alpha – Lambda-Z * t) = exp(alpha – Lambda-Z * tlast) * exp(–Lambda-Z * (t-tlast)) = (predicted concentration at tlast) * exp(–Lambda-Z * (t-tlast)) • • If a start or end time falls after the last numeric observation and Lambda Z is not estimable. If a start or end time falls within the range of the data but does not coincide with an observed data point. If both the start and end time for a partial area fall after the last numeric observation. where C0 is the minimum value between dose time and tau. C0=0 except in IV bolus models (model 201). where C0 is the dosing time intercept estimated by WinNonlin. and except for models 200 and 202 at steady state.

8 WinNonlin User’s Guide Therapeutic response windows For therapeutic windows. For example. the rectangle that is between the two boundaries is added to AUCDur and the piece that is left is added to AUCHgh. an interpolation is done to get the time value where the crossing occurred. AUCDur. This is equivalent to these formulae: AUCLow = AUCLow + ylower * (ti+1 – ti) AUCDur = AUCDur + (yupper – ylower) * (ti+1 – ti) AUCHgh = AUCHgh + AUC* – yupper * (ti+1 – ti) If there was one boundary crossing. TimeDur. The parts of this AUC* are added to the appropriate parameters AUCLow. ti+1) is computed following the user’s specified AUC calculation method as described above in section 3. The AUC for (ti. t*) is computed following the user’s specified AUC calculation method. for each pair of consecutive data points. The interpolation is done by either the linear rule or the log rule following the user’s AUC calculation method as described above in section 3. ti+1) and called the boundaries ylower and yupper. the rectangle that is under the lower boundary is added to AUCLow. if the concentration values for this interval occur above the upper boundary. and the program computes the time spent in each window determined by the boundaries and computes the area under the curve contained in each window. Call this AUC*. call this t*. or AUCHgh. it is determined if a boundary crossing occurred in that interval. including the inserted point at dosing time if there is one. To compute these values. i. depending on which window the concentration values are in.e. If no boundary crossing occurred. Call the pair of time values from the data set (ti. the difference ti+1– ti is added to the appropriate parameter TimeLow. The difference t* – ti is added to the appropriate time parameter 186 . the same rule is used as when inserting a point for partial areas.. one or two boundaries for concentration values can be given. To interpolate to get time t* in the interval (ti. or TimeHgh. ti+1) at which the concentration value is yw: Linear interpolation rule for time: t* = ti + [ (yw – yi) / (yi+1 – yi) ] * (ti+1 – ti) Log interpolation rule for time: t* = ti + [ (ln(yw) – ln(yi) ) / ( ln(yi+1) – ln(yi) ) ] * (ti+1 – ti) The AUC for (ti.

Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 and the parts of the AUC are added to the appropriate AUC parameters. If ylast is in the middle window for two boundaries: InfHgh values are the same as Hgh values. and the time and parts of AUC are added in for the interval (ti. Compute AUC* from tlast to t*. AUCInfLow is missing. the AUC rule after tlast is the log rule. ti+1). If the last concentration value in the data set is zero. Then the process is repeated for the interval (t*. t**). Then another interpolation is done from ti to ti+1 to get the time of the second crossing t** between (t*. unless an endpoint for the interval is negative or zero in which case the linear rule must be used. but now ylast is replaced with the following if Lambda Z is estimable: ylast = exp(alpha – lambdaZ * tlast). InfLow and InfDur parameters are missing. and the time and parts of AUC are added in for the interval (t*. AUCInfLow = AUCLow + ylast / lambdaZ. ti+1). If lambdaZ is not estimable. Otherwise. If there were two boundary crossings. This is equivalent to: t* = (1/lambdaZ) * (alpha – ln(ylower)) TimeInfDur = TimeDur + (t* – tlast) AUCInfDur = AUCDur + AUC* – ylower * (t* – tlast) AUCInfLow = AUCLow + ylower * (t* – tlast) + ylower / lambdaZ 187 . If lambdaZ is not estimable. The extrapolation rule for finding the time t* at which the extrapolated curve would cross the boundary concentration yw is: t* = (1/lambdaZ) * (alpha – ln(yw)). Add time from tlast to t* into TimeInfDur. add the rectangle that is under the lower boundary into AUCInfLow. For any of the AUC calculation methods. and add the piece that is left into AUCInfDur. ti+1). The extrapolated times and areas after the last data point are now considered for the therapeutic window parameters that are extrapolated to infinity. Add the extrapolated area in the lower window into AUCInfLow. (only if the last concentration was zero) If ylast is in the lower window: InfDur and InfHgh values are the same as Dur and Hgh values. Otherwise: Use the extrapolation rule to get t* at the lower boundary concentration. Then the times and parts of AUC are added in for the interval (t**. an interpolation is done to get t* as above for the first boundary crossing. the zero value was used in the above computation for the last interval. t*) as above.

The standard error of the data is also calculated for each unique time or midpoint value. It first calculates the mean concentration curve of the data. by taking the mean concentration value for each unique time value for plasma data. and add the piece that is left into AUCInfDur. and add the piece that is left into AUCInfHgh. it is recommended to use nominal time. for these analyses. Add time from t* to t** into TimeInfDur. add the rectangle that is under the lower boundary into AUCInfLow. 188 . Compute AUC** from t* to t**. add the rectangle that is in the middle window into AUCInfDur. the above procedure is followed and the InfDur parameters become the InfHgh parameters. rather than actual time. or the mean rate value for each unique midpoint for urine data. For this reason. Compute AUC* for tlast to t*. This is equivalent to: t* = (1/lambdaZ) * (alpha – ln(yupper)) t** = (1/lambdaZ) * (alpha – ln(ylower)) TimeInfHgh = TimeHgh + (t* – tlast) TimeInfDur = TimeDur + (t** – t*) AUCInfHgh = AUCHgh + AUC* – yupper * (t* – tlast) AUCInfDur = AUCDur + (yupper – ylower) * (t* – tlast) + AUC** – ylower * (t** – t*) AUCInfLow = AUCLow + ylower * (t* – tlast) + ylower * (t** – t*) + ylower / lambdaZ Sparse sampling computations The Noncompartmental Analysis (NCA) module provides special methods to analyze concentration data with few observations per subject.8 WinNonlin User’s Guide If ylast is in the top window when only one boundary is given. otherwise: Use the extrapolation rule to get t* at the upper boundary concentration value. add the rectangle that is under the lower boundary into AUCInfLow. If there are two boundaries and ylast is in the top window: If lambdaZ is not estimable. all extrapolated parameters are missing. The NCA treats this sparse data as a special case of plasma or urine concentration data. Add time from tlast to t* into TimeInfHgh. Use the extrapolation rule to get t** at the lower boundary concentration value. Add the extrapolated area in the lower window into AUCInfLow.

it will use the subject information to calculate standard errors that will account for any correlations in the data resulting from repeated sampling of individual animals. m – 1 ⎪ ⎩ ( tm – tm – 1 ) ⁄ 2 . i = m m = last observation time for AUCall. … . In addition. the sample standard error of the y-values at Tmax. Standard error of the mean AUC will be calculated as described in Nedelman and Jia (1998). For plasma data (models 200-202). i = 1 . and for the area under the mean concentration curve from dose time through the final observed time (AUCall). Standard error of the mean Cmax will be calculated as the sample standard deviation of the y-values at time Tmax divided by the square root of the number of observations at Tmax. Specifically: ˆ SE ( AUC ) = ˆ Var ( AUC ) Since AUC is calculated by the linear trapezoidal rule as a linear combination of the mean concentration values. or time of last measurable (positive) mean concentration for AUClast 189 . NCA will calculate all of the usual plasma or urine final parameters listed under “NCA output parameters and their computation formulas” on page 506. m ˆ AUC = where ∑ wi Ci i=0 C i = the sample mean at time i ⎧ ( t1 – t0 ) ⁄ 2 . and will account for any correlations in the data resulting from repeated sampling of individual animals. i = 0 ⎪ wi = ⎨ ( ti + 1 – ti – 1 ) ⁄ 2 . or equivalently. using a modification in Holder (2001).Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 Using the mean concentration curve. NCA will calculate the standard error for the mean concentration curve’s maximum value (Cmax).

which can be found as equation (A3) in Holder (2001): r ij 2 s ij = k = 1 ( r ij – 1 ) ∑ --------------------------------------------------------------r ij⎞ ⎛ r ij⎞ ⎛ + ⎝ 1 – ---.8 WinNonlin User’s Guide it follows that m ˆ Var ( AUC ) = where ∑ i=0 2 2 w i w j r ij s ij wi si ---------. The above equations can be computed from basic statistics. For cases where a non-zero value C0 must be inserted at dose time t0 to obtain better AUC estimates (see “Data checking and pre-treatment” on page 181). An added constant in AÛC will not change Var(AÛC). When computing the sample covariances in the above. the AUC estimate contains an additional constant term: C* = C0(t1-t0)/2. so SE_AUClast and SE_AUCall also will not change. Note that the inserted C0 is treated as a constant even when it must be esti- 190 . In other words. and appear as equation (7.vii) in Nedelman and Jia (1998). the standard errors are computed for Max_Rate. and for AURC_all.⎠ ri rj ( C ik – C i ) ( C jk – C j ) For urine models (models 210-212). the area under the mean rate curve through the final observed rate. NCA uses the unbiased sample covariance estimator. the maximum observed excretion rate. w0 is multiplied by C0.+ 2 ∑ --------------------ri ri rj i<j r ij = number of animals sampled at both times i and j r i = number of animals sampled at time i s i = sample variance of concentrations at time i s ij = sample covariance between concentrations C ik and C jk for all animals k that are sampled at both times i and j.0) is inserted.⎠ ⎝ 1 – ---. instead of being multiplied by 0 as occurs when the point (0.

so that the variances and covariances of those other data points are not duplicated in the Var(AÛC) computation. (Select as described under “NCA settings” on page 177. For the case in which rij = 1. The AUC’s must be calculated using one of the linear trapezoidal rules. then sij is set to 0 (zero). and ri = 1 or rj = 1.) 191 .Noncompartmental Analysis Computational rules for WinNonlin’s noncompartmental analysis engine 8 mated form other points in the data set.

8 WinNonlin User’s Guide 192 .

“Running the model” on page 209 5. ASCII models are human-readable and easily customized from within the WinNonlin interface. “Model properties” on page 196 3. To load a model: 1.DLL) files and ASCII text models (*. Custom compiled models can be created using FORTRAN or C++.Chapter 9 Modeling Model setup and output Setup and output for compartmental modeling in WinNonlin are covered under the following headings. “Model options” on page 203 4.LIB). “Loading the model” 2. “Output” on page 210 WinNonlin provides an extensive library of models. Compiled models provide faster performance. 193 . Open the data to which the model will be fit in a WinNonlin worksheet. detailed under “WinNonlin Model Libraries” on page 503. Loading the model The WinNonlin uses two types of models: compiled (machine language) dynamic link library (*. 1. Both model types can be used in scripts or loaded via the PK/PD/NCA Analysis Wizard. Custom models are covered under “User Models” on page 217.TXT or *.

Click the links in the table for a listing of models and model details.) Note: The image above shows the wizard as it appears for a user who has an IVIVC licence installed. Models to perform linear regression. Click the PK/PD/NCA Analysis Wizard button on the tool bar. The model will be associated with the active worksheet in that workbook. Select the type of model to load from those listed in Table 9-1. IV bolus. WinNonlin model types Model type Sigmoidal/dissolution models Linear models Noncompartmental analysis 194 Description Available only to users who have purchased and installed a WinNonlin IVIVC license. Geometric characterization of drug effect data or blood (plasma) or urine concentration data for IV infusion. or first-order input. or choose Tools>PK/PD/NCA Analysis Wizard from the WinNonlin menus. Other users will not see the IVIVC models. Dissolutions models to assist in evaluation of in vivo-in vitro correlations for formulation science. Table 9-1. Note: Hidden data sets are not available. 4. (See “Hidden windows” on page 18. Make sure that the correct workbook is selected at the top of the dialog. 3.9 WinNonlin User’s Guide 2. .

written and saved to an ASCII text file or compiled using C++ or FORTRAN. 8. Click Next. Compiled models run faster. for ASCII models. Legacy command files and WinNonlin Model Object (*. 6.WMO) files from prior WNL versions and Pharsight Model Object (*. specify whether to use the ASCII or a Compiled version. For compiled user models and new ASCII user models. ASCII models appear in a WinNonlin text window and are easily customized. and are available only in compiled form. Simultaneously fit PK and PD data.PMO) files can load ASCII or compiled user models. data and model settings. Click Next. For pharmacokinetic or pharmacodynamic models. These models can be based on any WinNonlin PK model. Select the model or. 7. Click Next. Four Michaelis-Menten (saturable elimination) models. Indirect response models Michaelis-Menten models Simultaneous PK/ PD link models User Models Command file or Model Object Note: Only one model may be loaded in WinNonlin at a time. WinNonlin model types (continued) Model type Pharmacokinetic models Pharmacodynamic models PK/PD link models Description 1 to 3 compartment models with 0th or 1st order absorption. 9 Any combination of WinNonlin’s compiled PK and PD models. 195 . load the model library file and enter the model number. with or without a lag time to the start of absorption. stored as ASCII libraries in WINNONLN\LIB.LIB. Each includes a different PK model and PD model 105. 5. Custom models. enter the model information as described under “User model parameters” on page 219.Modeling Loading the model Table 9-1. See also “Creating a new ASCII model” on page 218. treating PK parameters as fixed and generating concentrations at the effect sites for use by the PD models. 9. based on measured drug effect (inhibition or stimulation of response) and parameters defining either the build-up or dissipation of that effect. Four indirect response models. stored as ASCII library files in WINNONLN\LIB\MM. then loaded using this option. Eight variations on Emax and Sigmoid Emax models.

Model properties Once a model has been loaded. 6. Proceed to set the Model properties. WinNonlin performs a separate analysis for each unique combination of sort variable values. a crossover study might include Subject and Treatment as sort variables. 3. and drag it to the Y Variable field. To specify the roles of data columns in the model: 1. Choose Model>Data Variables from the WinNonlin menus or click the Data Variables button in the WinNonlin tool bar. Drag and drop to select one or more sort variables (optional) to identify individual data profiles. 2. For example. the following can be set: • • • • Data variables tell the model which data columns to use. Select the dependent variable. Select carry along variables (optional) to include in the analysis output. Dosing regimen or Dosing data variables set dose times and/or amounts. Click Finish to load the selected model. 196 . 7. For example. as applicable to the model. Simulations: No Y variable is needed for simulations. Model options select output formats. 4. Model parameters set parameter names and initial parameter estimates. with a description of the model. instead check the Simulate Data check box and enter units (optional) for the Y variable. computational algorithms and units. Study Number or Gender might be included as a carry-along variable to make it available for post-modeling analyses.9 WinNonlin User’s Guide 10. 5. Load the model as described under “Loading the model” on page 193. Data variables Modeling requires that the model variables be identified before other settings. Select the independent variable in the Variables list and drag it to the X Variable field. The Selection Summary dialog appears.

The FIXED command can be used for both modeling and simulating. Fixed parameters use a set value. WinNonlin Generated Initial Parameter Values with WinNonlin Bounds: WinNonlin will curve strip for initial estimates. The Model Parameters tab of the Model properties dialog appears. then produce boundaries on the model parameters for model fitting. Click OK and proceed to set the Model parameters. Enter the initial estimate for each parameter in the column labeled Initial. For IVIVC models only (IVIVC license required). with the second function. • User Supplied Initial Parameter Values. below. WinNonlin can compute initial estimates via curve stripping for single-dose compiled PK.Modeling Model properties 9 8. For all other situations. etc. so are no longer considered parameters. Note: Data corresponding to the first (sorted) value of the function variable will be used with the first function. some parameters may be given fixed values while others are estimated. PD. Statistics such as VIF are not computed for fixed parameters. Select Model>Model Parameters from the menus or click the Model Parameters tool bar button. Model parameters All iterative estimation procedures require initial estimates of the parameters. if applicable for the model. the user must supply initial estimates or provide boundaries to be used by WinNonlin in creating initial estimates via grid search. Specify the type of Parameter Calculation and Boundaries. and 2 for urine data. the second. Select the function variable. If curve stripping 197 • . The function variable might have the value 1 for plasma data. 2. PK/PD link. To set up the initial parameter estimates: 1. When user-supplied boundaries are used. This reduces the number of parameters in the model. every parameter requires both upper and lower bounds. the data set may include data on two compartments—plasma and urine. If a model fits multiple functions simultaneously. a function variable is required to tell WinNonlin which values of the Y variable apply to which function. and select WinNonlin or user-supplied bounds. linear or IVIVC models. 9. For example. See the note.

click in the cell in the Fixed or Estimated column and choose Fixed from the drop-down list. For IVIVC models. 4. 3. See also “Notes on initial parameter estimates and bounds” below. The structure of this file (PARMFILE) is described in the chapter on Scripting. Fixed parameters. 5. When it is checked. Note: Parameter boundaries not only provide a basis for grid searching initial parameter estimates. • WinNonlin Generated Initial Parameter Values with User-Supplied Bounds: WinNonlin will do curve stripping for initial estimates. and so forth for each successive sort level. This option is available when sort variables are included in the Data variables. Propagate Final Estimates. This can be valuable if the values become unrealistic or the model won't converge. The final parameter estimates from the first sort level provide the initial estimates for the second.9 WinNonlin User’s Guide fails. the program will terminate. To fix that parameter for all sort levels. To fix a parameter for a given sort level. 6. When all values are set. initial estimates must be provided for each level of the sort variable(s) unless Propagate Final Estimates is selected. Multiple sort levels If the data has one or more sort variables. but also limit the estimates during modeling. click Apply (to leave the dialog open and Proceed to the Dosing regimen) or OK (to close the dialog). 198 . some (but not all) parameters may be set to fixed values rather than estimated. The same type of parameter calculation and boundaries apply for all sort levels. The Save and Load buttons can be used to store and re-use parameter settings in an external ASCII text file. click the Fix all <Parameter Name> button at the lower right. as WinNonlin cannot grid search for initial estimates without user-supplied boundaries. Enter the fixed value in the adjacent cell. initial estimates and bounds are entered or calculated only for the first sort level. then grid search if that fails. See “Special file formats” on page 430.

With the Nelder-Mead option (METHOD 1). grid searching is requested). 2.e. even when WinNonlin is computing the initial estimates. For any model other than a compiled WinNonlin model. Note that this option has been added for convenience.. the grid will comprise 34 = 81 possible points. In this case. Notes on initial parameter estimates and bounds • • The use of boundaries (user or WinNonlin supplied) is recommended. 3. WinNonlin uses curve stripping only for single-dose data. Method 3 (Gauss-Newton (Hartley)) and Do not use bounds are recommended. boundaries are required. If the data are to be fit to a micro constant model. WinNonlin will perform curve stripping for the corresponding macro constant model then use the macro constants to compute the micro constants. boundaries are required. The grid used in grid searching includes three values for each parameter. Thus. Hold the cursor in the lower right-hand corner of this selection until the cursor turns into a small black + called a handle. but it may dramatically slow the parameter estimation process if the model is defined in terms of differential equations.Modeling Model properties To copy model parameter information from the first profile to the remaining profiles: 9 1. For linear regressions. If multiple dose data is fit to a WinNonlin library model and WinNonlin makes initial parameter estimates. When the Gauss-Newton method is used to estimate the parameters (METHODS 2 and 3). Release the mouse button. It is still recommended that the lower and upper boundaries be specified. Drag the handle over the target cells (to the bottom of the grid). then select (highlight) all cells for the first profile with the mouse. 4. the residual sum of squares is set to a large number if any 199 • • • • • . if four parameters are to be estimated. the boundaries are incorporated by applying a normit transform to the parameter space. if WinNonlin is requested to perform initial parameter estimates (i. The point on the grid associated with the smallest sum of squares is used as an initial estimate for the estimation algorithm. Enter values for the first profile. a grid search is performed to obtain initial estimates.

Specific models require the information described below. amount of the dose(s).) 2. For some models the number of constants depends upon the number of doses administered. 13. 3-7.) 8. The use of bounds with Nelder-Mead will. 14. 18 stripping dose # doses dose 1 time of dose 1 dose 2 time of dose 2 (etc. Dosing regimen WinNonlin uses model constants to store dosing information. including the number of doses. Model Number Constant 1 2 3 4 5 6 7 8 9 NCON (2*ND)+1 (3*ND)+1 (2*ND)+2 3 4 1. The use of bounds (either user-supplied or WinNonlin supplied) is always recommended with the Gauss-Newton methods. 200 . however. 16 bolus dose IV dose infusion length 17 stripping dose bolus dose IV dose infusion length time of dose 1 start time 1 time of dose 2 dose 2 For the table above. NCON = the total number of constants to be specified by the user. 12 # of doses dose 1 dose 2 (etc. keep the parameters within the bounds. 9. 11. WinNonlin dose information can be entered in the Regimen Dialog or read from worksheet columns using the Dosing Data Variables dialog (see “Dosing data variables” on page 202). and ND = number of administered doses. 19 # of doses dose 1 end time 1 start time 2 end time 2 (etc.9 WinNonlin User’s Guide of the parameter estimates go outside the specified constraints at any iteration. 10. Note that PD and IVIVC models do not require dosing. the use of bounds does not affect numerical stability when using the Nelder-Mead simplex option.) 15. • Unlike the linearization methods. which forces the parameters back within the specified bounds. and dosing time(s).

if doses are in milligrams per kilogram of bodyweight. Enter units for the dose under Current Units. select the appropriate factor in the Dose Normalization drop-list. If a sort variable has been specified. Include time values that fall within the data but beyond the last real dose for these phantom doses. For example. This will affect output parameter units. Select Model>Dosing Regimen from the WinNonlin menus or click the Dosing Regimen button in the tool bar. dose normalization of kg would change those units to L/kg. or select the unit prefix and unit below and click the Add button. For example. enter the largest number. The Dosing Regimen tab of the Model properties dialog appears. 4. enter the constants required for all levels of the sort variable(s). and enter doses of 0 for the extra doses. If dose amount is normalized by subject bodyweight or body mass index.Modeling Model properties To enter dosing data in the Dosing Regimen dialog: 9 1. Compiled models 2. if Volume units were L. enter Current Units of mg and Dose Normalization of kg. Note: If different profiles require different numbers of doses. Dose normalization affects units for all volume 201 . 3.

Dosing data variables Use this dialog to extract dosing information from worksheet columns. and set the model variables as described under “Data variables” on page 196. Note: Use the Save and Load buttons to save the dosing values and units to an external ASCII file and reload settings for re-use. Add Row. 3. Click OK or Apply to enter the dosing information.. and Percent_Recovered in NCA urine models. See also: • • “Dosing regimen” on page 200 for compartmental models “Dosing regimen” on page 171 for noncompartmental analysis To select columns containing dose information: 1. Choose Model>Dosing Data Variables. 2. Open the worksheet with dose information in WinNonlin. 1. as well as AUC/D in NCA plasma models. to size the grid appropriately for the number of constants required by the model. Load the data and model in WinNonlin. Enter each constant value for each profile in the column labeled Value. and Delete Row buttons. Check the ASCII text in the Model window to see how many and which constants to enter. 202 . The structure of this file is described under “Special file formats” on page 430 4. if needed. Un-check No Source. a row will be added for each profile. If the data set includes multiple profiles. 3. the Constant column of the Dosing Regimen dialog contains the names of elements of the CON array. ASCII models For ASCII models. instead of entering it manually in the dialog. 2..9 WinNonlin User’s Guide and clearance parameters in NCA and PK models. 4. Use the Insert Row. from the WinNonlin menus or click the Dosing Data Variables tool bar button.

respectively. 8. Model options The options for compartmental modeling output and computational algorithm appear on five tabs in the Model Options dialog. For noncompartmental analyses of steady-state data. 6. worksheet. and column containing the dose-interval values (Tau). 203 . Select the column containing dose amounts under Dose. 9. 7. Select the column containing dose times under Time. Select the workbook and worksheet containing the dose information under Workbook and Worksheet. Click OK. Select the workbook. check Steady State and open the Tau Mappings tab.Modeling Model options 9 5.

Select the desired options using the following tabs. The Output Options tab of the Model options dialog. Note: When attempting a difficult fit. Page Breaks: Checking this starts a new page at each profile in text output.9 WinNonlin User’s Guide To set compartmental modeling options: 1. see “Output” on page 210. Workbook: Tabular output in a workbook. overrides the defaults for the current modeling session. Select Model Options from the Model menu or click the Model Options tool bar button. Text: This provides an ASCII (text) version of the output. For the contents of the workbook output. below. Text title to appear at the top of modeling output (text and charts) Model fitting algorithm Preferred output units for parameter estimates 3. Output options The defaults output options are set in the WinNonlin Options dialog (see “WinNonlin options” on page 28). containing all modeling settings and output in one file. Click OK or Apply. as well as any messages or errors that occur during modeling. Check or un-check the following options to include or exclude the output. The Model Options dialog appears. 2. Output options Weight Title PK settings Units Select text. chart (plots) and/or workbook output forms Weighting of observation data. Chart: This option generates plots showing predicted values versus observed values for each level of the ID variable. use the ASCII output to check for errors. 204 .

scaling of the weights provides increased numerical stability. Iterative reweighting sets weights as the predicted value of Y to a specified power. However. as the weights of the observations are proportionally the same. Select from the options discussed below. 3. Programming statements can be used to define weights as part of a user defined model. 4. there are four ways to assign weights. See “Weighting” on page 241 for a detailed discussion of weighting schemes. Open the Weight tab of the Model options dialog (Model>Model Options). other than uniform weighting: 1. Predicted to the Power n: to use iterative reweighting. Pre-Selected Scheme: to select from among the most common schemes: – weight by 1/observed Y – weight by 1/observed Y2 – weight by 1/predicted Y (iterative reweighting) – weight by 1/predicted Y2 (iterative reweighting) Observed to the Power n: to use weighted least squares. the weights for the individual data values are scaled so that the sum of weights for each function is equal to the number of data values (with non-zero weights). Weighted least squares weights by a power of the observed Y.Modeling Model options 9 Weight For compartmental modeling. See “Scaling of weights” on page 242 for further discussion. To set the weighting scheme for a modeling session: 1. 2. Weights on File in Column: to read weights from a column on the data file. Input the value of n in the box provided. Include Weight as a variable on the data file allows specific weight values to be applied to each observed value. and click OK. This scaling has no effect on the model fitting. 205 . Input the value of n in the box provided. See “User Models” on page 217. 2. Scaling of Weights: When weights are read from the data file or weighted least squares is used.

For linear regressions. called Non-transposed Final Parameters.9 WinNonlin User’s Guide To to turn off scaling of weights. Title Titles can be displayed at the top of each chart and each page of ASCII text output. check the No Scaling of Weights check box. Transpose final parameter table If this option is checked. 206 . To include titles. The transposed form is particularly useful when creating final report tables or when summarizing the final parameters using descriptive statistics. The title also appears in workbook output. A second worksheet. called Transposed Final Parameters containing the transposed version. will present each parameter in a separate row. If this option is not checked. the Final Parameters output worksheet will show each parameter in a separate column. Unchecking the Title check box suppresses the titles without deleting the title text from the Model options dialog. See “Models” on page 29 for examples of transposed and non-transposed output. with the Final Parameters worksheet displaying non-transposed data and a second worksheet. the two worksheets will be reversed. check Title and enter the title text below. PK settings The PK Settings tab of the Model options dialog (Model>Model Options) provides control over the model fitting algorithm and related settings. only when it is printed. Minimization Select among the 3 minimization methods: – Method 1: Nelder-Mead simplex – Method 2: Gauss-Newton with Levenberg and Hartley modification – Method 3: Gauss-Newton with Hartley modification These methods are described in “Model fitting algorithms and features of WinNonlin” on page 5. use Method 3 without bounds. The use of bounds is especially recommended with Methods 2 and 3. Enter multiple lines using either CTRL + ENTER or SHIFT + ENTER to move to the next line.

data are generated for the end of infusion. The default for user models is the number of data points in the original data set.e. Convergence is achieved when the relative change in the residual sum of squares is less than the convergence criterion. the maximum is 20. the predicted data for the built-in PK models includes dosing times. 207 . in addition. When fitting or simulating multiple dose data.000..0001. this command is ignored and the number of predicted points is equal to the number of points in the data set. The minimum allowable value is 10. Convergence criterion Use this option to set the convergence criterion. Use this option to create a data set that will plot a smooth curve. For compiled models without differential equations the default value is 1000. This value may be increased only if the model has no more than one independent variable. concentrations are generated at dosing times.= -------------------------------------------------------∂P ( 1 ) Δ where Δ = the Increment multiplied by the parameter value. for infusion models. If the number of derivatives is greater than 0 (i. The default is 0. Note: To better reflect peaks (for IV dosing) and troughs (for extravascular. The program estimates these derivatives using a difference equation. in addition to the concentrations generated. differential equations are used). Number of predicted values This setting determines the number of points in the predicted data. For all three types.Modeling Model options 9 Increment for partial derivatives Nonlinear algorithms require derivatives of the models with respect to the parameters. IV infusion and IV bolus dosing). the predicted data plots may be much improved by increasing the number of predicted values. For example: ∂F F(P(1) + Δ) – F(P(1)) -------------.

Note: Linear or IVIVC models set preferred units for the independent (x) or dependent (y) variable. the default size is 4. all output will be use the initial estimates as the final parameters. Iterations Use this option to change the maximum number of iterations. Select the row with the appropriate parameter name. when computing the variance. Model size may range from 1 to 64. Select the Units tab of the Model options dialog. each iteration is really a reflection of the simplex. 2. For Method 1. For certain types of problems.9 WinNonlin User’s Guide Mean square The variance is normally a function of the weighted residual SS/df. Units Specify the preferred units for output of modeling or simulation. 208 . the default is 50. Alternately. the mean square needs to be set to a fixed value. or the Mean Square. For Method 1 (Nelder-Mead simplex). If the number of iterations is set to zero. To specify preferred units: 1. WinNonlin performs appropriate conversions as needed to report output in the preferred units. Y-variable and. The default units are displayed in the Default column. The input data worksheet columns for the X-variable. click the Units Builder button to construct and specify units in the Units Builder. Model size Use this option to alter the default memory requirements when using ASCII models only. 3. See “Linear models” on page 546 or “Sigmoidal/dissolution models” on page 547 for model details. where applicable. dosing must contain units to use this option. the default is 500. For Methods 2 and 3 (Gauss-Newton). Enter the units in the Preferred column for that row. Units for output parameters will be derived from these.

Troubleshooting modeling problems Occasionally when modeling. select the data set to be used for the modeling. Note: The Select Data for Modeling dialog will appear if the original data set selected for modeling has become un-associated with the model. If these types of problems occur. 209 .Modeling Running the model 9 4. Error messages may appear here as guidance in resolving the problem. and click OK. Stop modeling To stop modeling: 1. the only recourse is to obtain additional data. Check the history file associated with the PK Data Output window. apply the following steps in determining what may have gone wrong. Note that it is possible that a specified model and input values may be correct. check the Text (ASCII) output for error messages.TXT) file for later re-use. In case of an error the parameters can be reset to their original setting by clicking on the Reset to Defaults button. In this case. The preferred units can be saved to a (*. Running the model To start modeling: 1. or terminate normally and converge to unreasonable values of the parameter estimates. but there simply may not be enough information contained in the data to precisely estimate all of the parameters in the model. using the Save and Load buttons. If no PK Data Output window appears. WinNonlin will terminate abnormally. Click the Stop Modeling tool bar button. If that is the case. Select Model>Start Modeling from the menus or click the Start Modeling tool bar button. 5.

9 WinNonlin User’s Guide If the model was defined by the user (i. The data set may be ill-conditioned or the initial parameter estimates may be too far from the true values.via the Settings tab in the Model Options dialog box). In addition. warning messages may also be generated to alert the user that WinNonlin is having to work very hard to obtain the parameter estimates. This example uses a pharmacokinetic (PK) model. Carefully check the input data set and constants to make sure the values are correct. Errors When modeling is run. A plot of the data can be very helpful. Output One of the examples from the Examples Guide is used here to discuss the modeling output provided by WinNonlin. it will be written to the text output (if the user requested text output). search for the word ERROR in the text output. Also make sure that the model has not attempted any illegal arithmetic computations such as taking the logarithm of zero or the square root of a negative number. not one of the models in WinNonlin's built-in library).. try rerunning the problem and add LOWER and UPPER constraints. Note: To locate error messages. If an error occurs that does not prevent completion of modeling. check the Simulate check box in the Y Variable group box in the Data Variables dialog box. 210 . text output will be created. carefully check the model definition statements to make sure that there are no undefined or uninitialized variables. To do this. which may indicate that the model is ill-defined. Are the predicted values from the simulation reasonably close to the observed data? Often error messages occur because WinNonlin is having difficulty fitting the model to the data.Mead . error messages and warnings are written to the text output. Check the initial values of the parameters to see if they are reasonable. One way to do this is to rerun the problem as a simulation. If the problem recurs. If these messages occur. errors and warnings will be written there. try using METHOD 1 (Nelder . If an error occurs that prevents completion of modeling. whether or not the user requests it.e.

then the statements that define the model are also printed. including any errors that occurred during modeling. In addition. 211 . Arrays and Functions” on page 557 for an alphabetic catalog of WinNonlin commands.Modeling Output 9 Note: This section is meant to provide guidance and references to aid in the interpretation of modeling output. If the RANK is less than the number of parameters. this page shows the parameter values and the computed weighted sum of squared residuals at each iteration. modeling iterations and output parameters. then the problem is ill-conditioned. Text (ASCII) output The text output window contains a complete summary of the modeling for this PK example. That is. and other quantities. Model Commands The first page is a copy of the input commands used to run the model. the following two values are printed for each iteration: • RANK . If the matrix is of full rank. • • • “Text (ASCII) output” on page 211: text version of all model settings and output.Rank of the matrix of partial derivatives of the model parameters. After Example 1 is run. residuals.) If the model is defined in a user library. depending on the model run. “Chart output” on page 215: plots of observed and predicted data. and is not a replacement for a PK or statistics textbook. there is not enough information contained in the data to precisely estimate all of the parameters in the model. as well as several measures of fit. If a Gauss-Newton method is used (Methods 2 or 3). “Workbook output” on page 214: worksheets listing input data. depending which method is used for minimization. Minimization Process The second page varies. and any errors that occurred during modeling. up to three new child windows appear in the WinNonlin window. They are discussed in the following sections. (See “Commands. RANK = # of parameters.

When using Methods 2 and 3. The estimated standard errors and confidence intervals are only approximate. The nearer the model is to being linear. usually the PLANAR intervals will be wider than the UNIVARIATE intervals. the second page only shows the parameter values and the weighted sum of squares for the initial. If the simplex algorithm is used. or smallest. The confidence interval labeled PLANAR is obtained from the tangent planes to the joint confidence ellipsoid of all the parameter estimates. sum of squares) point. while the PLANAR confidence intervals take into account the correlations among the parameter estimates. then the estimation problem is very ill-conditioned. as they are based on a linearization of the nonlinear model. Final Parameters The third page of the output lists the parameter estimates. Many times a model will provide a good fit to the data in the sense that all of the deviations between observed and predicted values are small. The UNIVARIATE confidence intervals ignore the other parameters being estimated.9 WinNonlin User’s Guide • COND . The confidence interval labeled UNIVARIATE is the parameter estimate plus and minus the product of the estimated standard error and the appropriate value of the t-statistic. If the condition number gets to be very large. the better is the approximation. See Chapter 10 of Draper and Smith (1981) for a complete discussion of the true confidence intervals for parameter estimates in nonlinear models. using bounds (LOWER and UPPER constraints) often helps reduce the condition number. say greater than 10e6. final (best. and the next-to-best point. Both are 95% confidence intervals. but one or more parameter estimates will have standard errors that are large relative to the estimate.Condition number of the matrix of partial derivatives. 212 . and two confidence intervals based on this estimated standard error. the asymptotic estimated standard error of each estimate. COND is the square root of the ratio of the largest to the smallest eigenvalue of the matrix of partial derivatives. The estimated standard errors are valuable for indicating how much information about the parameters is contained in the data. For a complete discussion of the UNIVARIATE and PLANAR confidence intervals see page 95 of Draper and Smith (1981). Only in the very rare event that the estimates are completely independent (uncorrelated) will the two confidence intervals be equal.

to compute AUC from zero to the last time. the sum of weighted squared residuals.8 probably indicates serious problems with the data and/or model. The AUC is computed by the linear trapezoidal rule. AIC (Akaike (1978)) and SBC (Schwarz (1978)). calculated predicted values of the model function. in the sense that the Gauss-Newton algorithm will have trouble finding the minimum residual sum of squares. If the first time value is not zero.Modeling Output 9 Variance-Covariance Matrix.9. weights. and the correlation between observed and calculated function values. an estimate of the error standard deviation. Two measures which have been proposed for comparing competing models. area under the curve (AUC) is printed at the end of page five of the text output. Correlation Matrix. A large ratio between the largest and smallest eigenvalue may indicate that there are too many parameters in the model. for most problems it will be greater than 0. the standard deviations (S) of the calculated function values and standardized residuals. although the data may be well fit by the model. For discussion of the use of eigenvalues in modeling see Belsley. Also on page five are the sum of squared residuals. data sets that result in highly correlated estimates are often difficult to fit. and anything less than 0. it is usually not possible to drop one parameter from a nonlinear model. Residuals Page five of the output lists the observed data. High correlations among one or more pairs of the estimates indicate that the UNIVARIATE confidence limits are underestimates of the uncertainty in the parameter estimates and. Note that this AUC in the text output differs from the AUC on the 213 . Also. Runs of positive or negative deviations indicate non-random deviations from the model and are indicators of an incorrect model and/or choice of weights. then the eigenvalues will all be equal. and Eigenvalues Page four of the output shows the correlation matrix of the estimates and the eigenvalues of the linearized form of the model. Unlike the linear model case. Kuh and Welsch (1980). If the data have been fit to a WinNonlin library model for extravascular or constant infusion input. the correlation is not a particularly good measure of fit. If the parameter estimates are completely uncorrelated. Unfortunately. The eigenvalues are another indication of how well the data define the parameters. then WinNonlin generates a data point with time and concentration values of zero. the estimates may not be very reliable. residuals. are also printed on page five.

are printed on page six. for each level of the sort variables. and lower and upper bounds for each level of the sort variables. Iteration number. they represent about the third level of approximation and must be regarded with caution. Iteration number. CV%. The secondary parameter AUC is calculated from the model parameters. The dosing regimen specified for the modeling. and value for each parameter. Eigenvalues for each level of the sort variables. . estimates. along with their estimated asymptotic standard errors. secondary parameters that depend on dose use the first dose in their computation. weighted sum of squares. As such. univariate confidence intervals. Workbook output A PK Workbook window contains summary tables of the modeling data. initial values. standard error of the estimates. The standard errors of the secondary parameters are obtained by computing the linear term of a Taylor series expansion of the secondary parameters. A correlation matrix for the parameters. they. Dosing Correlation Matrix Eigenvalues Condition Numbers 214 . The secondary parameters are functions of the primary parameters whose estimates are given on page three of the output. Parameter names.9 WinNonlin User’s Guide Secondary Parameters tab of the tabular output. and planar confidence intervals for each level of the sort variables. rank and condition number for each sort level. Note: Note that the AUC values may not be meaningful if the data were obtained after multiple dosing or IV dosing (due to time 0 extrapolation problems). In the case of multiple-dose data. Summary of PK worksheets Item Initial Parameters Minimization Process Final Parameters Contents Parameter names. units. Table 9-2. These worksheets present the output in a form that can be used for reporting and further analyses. a summary of the information in the ASCII output. for each sort level. Secondary Parameters If there are any secondary parameters.

The sort variables. Diagnostics Differential Equations Partial Derivatives Predicted Data Secondary Parameters User Settings a. Secondary parameter name. Summary of PK worksheets (continued) Item Variance-Covariance Matrix Summary Tablea Contents 9 A variance-covariance matrix for the parameters. estimate of residual standard deviation (S) and degrees of freedom (DF). residual. sum of squared residuals (SSR). These output worksheets are presented together in one workbook window. for each sort level. estimate. Time and predicted Y for multiple time points. and maximum number of iterations allowed. Use the tabs at the bottom of the window to select the worksheet to use. Diagnostics for each function in the model and for the total: corrected sum of squared observations (CSS). The value of the partial derivatives for each parameter at each time point for each value of the sort variables. 215 . For indirect response models. and two measures of goodness of fit: the Akaike Information Criterion (AIC) and Schwarz Bayesian Criterion (SBC). transformed X. Y. If there are no statements to transform the data. convergence criterion. X. Select graph type using the tabs at the bottom of the window. predicted Y. transformed Y. for each sort level. the weighted correlation. weighted corrected sum of squared observations (WCSS). weighted sum of squared residuals (WSSR). Note: Worksheets produced will depend on the analysis type and model settings. and CV% for each sort level. minimization method. standard error of predicted Y. for each sort level. model number. also includes CP. Values of the differential equations at each time in the data set. also includes CP and Ce. weighting scheme. Chart output This PK analysis produces a chart window with five graphs for each level of the sort variable. For link models. standard error of the estimate. then X and Y will equal X(obs) and Y(obs). units. standardized residuals. weight. Title. the correlation between observed Y and predicted Y.Modeling Output Table 9-2.

b. b. c)/db. Each derivative is evaluated at the final parameter estimates. a. based on parameters a. Observed Y vs. Weighted Predicted Y vs. with the Observed Y overlaid on the plot. Partial Derivatives: plots the partial derivative of the model with respect to each parameter as a function of the x variable. then df(x. b. c)/da. 216 . a. Used for assessing the model fit. a. df(x.9 WinNonlin User’s Guide X vs. df(x. b. a. b. c)/dc are plotted versus x. X vs. Weighted Predicted Y: plots weighted predicted Y against observed Y. Weighted Residual Y: used to assess whether error distribution is appropriately modeled throughout the range of the data. If f(x. Weighted Residual Y: used to assess whether error distribution is appropriately modeled across the range of the X variable. Data taken at larger partial derivatives are more influential than those taken at smaller partial derivatives for the parameter of interest. Observed Y and Predicted Y: plots the predicted curves as a function of X. and c. c) is the model as a function of x. Scatter should lie close to the 45 degree line.

and used in Pharsight Model Objects and scripts. Compiling a model can significantly speed model fitting.DLL) file. a library with two models might contain the following. the ASCII and compiled models in the WinNonlin \LIB and \USERCOMP subdirectories serve as examples and templates. See also Gabrielsson and Weiner (2001).TXT) or model library files (*. For example. Either type can be loaded via WinNonlin’s PK/PD/NCA Analysis Wizard. especially for models with differential equations. In addition. Example user models appear in this chapter and the WinNonlin Examples Guide. Compiled user models are created and compiled in Fortran or C++ into a dynamic link library (*. Individual models may be up to 24 KB in size. One or more models can be saved together in an ASCII text file (*.LIB).TXT) or library (*. 217 .LIB). written in WinNonlin commands. or C++ In addition to its library of standard PK and PK/PD models. ASCII user models ASCII user models are custom models with or without differential equations. Within an ASCII model file. An ASCII library cannot exceed 30 KB.Chapter 10 User Models Creating custom models using WinNonlin command language. FORTRAN. and ends with an end-of-model (EOM) statement. WinNonlin supports two types of custom models: ASCII and compiled user models. and saved as text (*. each model begins with a MODEL n statement identifying the model by number. • • ASCII user models are written using WinNonlin commands and programming language.

Alternately. including a WinNonlin text window or Notepad. choose to load a User Model and Existing ASCII Model in the first two screens of the wizard. then enter the model number and the full file path as shown below. Details on model templates and structure appear under: • • • “Creating a new ASCII model” below “Model structure” on page 220 “ASCII model examples” on page 226 Creating a new ASCII model ASCII user models can be created in any text editor. copy the library (*. you would use the PK/PD/NCA Analysis Wizard (see “Loading the model” on page 193).LIB) file from the WinNonlin \LIB subdirectory and edit the copy as needed. To use a WinNonlin library model as a template. create a template with the appropriate number of parameters and equations using WinNonlin’s PK/PD/NCA Wizard as follows. 218 .10 User’s Guide WinNonlin MODEL 1 <model statements> EOM MODEL 2 <model statements> EOM To load the second model into WinNonlin.

so long as each uses a unique model number. Note: To ascertain this information for compiled user models. (See Table 10-1 on page 221.LIB file. Several models can be saved to one . To save the model. To complete the model parameters settings: 1. enter the model number. 5. WinNonlin’s PK/PD/NCA Analysis Wizard needs to know the number of algebraic and differential equations in the model.LIB. 6. User model parameters In order to create a new ASCII model template or use a compiled user model.DLL was compiled. 3. Select Create New ASCII Model and click Next. 7. Review the settings. 4. and copy and paste the new model text into the existing file. The Selection Summary dialog appears. 1. To do this. Enter the model information as detailed under “User model parameters”. This information is entered in the User Model Parameters dialog of the Wizard as follows. Click Next. commands and syntax detailed in this chapter and under “Commands. Arrays and Functions” on page 557. The default model library file name is USERASCI. It is often convenient to save under a new name that describes the model. look at the FORTRAN or C++ code from which the user model . Click Back to change settings or Finish to accept them. For ASCII models. Select User Model in the Model Types dialog and click Next. and the number and names of estimated and secondary parameters.) 219 . Edit as needed using the structure. WinNonlin will create a new text window containing a template for the model. Launch the PK/PD/NCA Analysis Wizard by clicking on the PK/PD/NCA Analysis Wizard tool bar button or choosing Tools>PK/PD/NCA Analysis Wizard from the WinNonlin menus.User Models ASCII user models 10 To create a new ASCII model from a template: 1. to identify the model within a library. save it as an ASCII file (*.LIB). 2. an integer between 1 and 32767. The ASCII Model Selection dialog appears. The Model Parameters dialog appears. open the library file in a text window.

3. Parameter names can include up to 8 alphanumeric characters. check Include differential equations and enter the number of differential equations. should be included within the model to reduce the amount of information that must be supplied to WinNonlin when using the model. Note: The maximum number of variables and operators in a model is 2000. use variables to represent model quantities that may vary across uses—e. Does the data set contain all the necessary information. In contrast. An ASCII model structures this information in blocks. Does the model change with values of the data or estimated parameters? 4. Enter the primary (estimated) and secondary (optional) parameters in the grid.g. or a combination of the two? Examples of algebraic nonlinear models Examples of differential equations WinNonlin Y = exp ( – a *t ) Y = a ⋅ cos ( t ) + b ⋅ sin ( t ) dz ⁄ dt = – k 1 t dz ⁄ dt = k 1 w – k 2 y 2. dosage—to make the model as general as possible. a system of differential equations. enter parameters in the order in which they appear in the model. Is the model expressible as an algebraic nonlinear function. or will the model require new variables that are functions of the input data? 3. 4. Model structure The following questions need to be considered in building custom models: 1. 220 . what are their initial values? Information that is fixed across modeling sessions. outlined in Table 10-1. Enter the number of algebraic functions in the model.. For compiled models. such as the number and names of model parameters. How many parameters and constants will be needed to define the model? 5. If the model includes differential equations.10 User’s Guide 2. If the model uses differential equations.

Useful for automated (scripted) analyses. call it in a script. Optional WinNonlin commands to load the model. Each block must end with an END statement. 221 . Optional commands to create global variables. FUNCTION block and EOM statement. Marks the beginning of the model. The model number. At minimum. Number each consecutively.. i. Required if differential equations are included. or store it in a multi-model library. The number must be an integer from 1 to 32767 and unique for each model in a library. It must begin with MODEL and end with EOM. either FUNCTION 1 or FUNC 1 signifies that Function 1 is being defined. N.User Models ASCII user models 10 Table 10-1. The function number is required even if there is only one FUNCTION block. Required.e. Statements defining the differential equations are included in this END block. Sets initial conditions for the system of differential equations. and begin analysis. Any block or command name can be represented by its first four characters or its full name. A function block must be included for each set of algebraic equations to be fit to data. is required to load the model via the PK/PD/NCA Wizard. The other blocks may appear in any order. alter model settings. Marks the end of the model. Variables that are defined within another block persist only within that block. TRANSFORM statements END COMMANDS statements END TEMPORARY statements END FUNCTION N statements END STARTING statements END DIFFERENTIAL Required if the model is a function of one or more differential equastatements tions. SECONDARY statements END EOM Optional statements to define parameters that are functions of the estimated parameters. Model structure Block: MODEL N Description Required. Required. Either REMARK or REMA indicates that the rest of that paragraph contains a note and is to be ignored by WinNonlin. variables that will be used in more than one model block. Optional statements to alter data values and/or create new variables before the model is fit. a model requires a MODEL statement. For example.

for examples that use the TRANSFORM block. if a data set contains the variables X8 and Y. Independent variables can also be derived or transformed here. and EXP9.PMO. See the user models in EXP3. Details appear under: • • • “WinNonlin variables” on page 231 “The WinNonlin programming language” on page 233 “Commands. prior to starting the estimation procedure. the output will include both the original variable and the transformed values. See also “Weights in ASCII models” on page 243. If this block transforms a variable from the input data set.PMO. New variables created in the transform block will not be included. it will override any weighting specified via the Data Variables dialog. and X8(Obs) and Y(Obs) (original values).10 User’s Guide Note: Comments can appear within or between any of the blocks. the blocks use WinNonlin programming statements and command language to define relationships between input data and parameter estimates. This block often defines the dependent variable (Y) or the weight (WT) as a function of other variables from the input data set. However. EXP4. and the following statements are included: X8 = SQRT(X8) Y=log(Y) then the output worksheet will include the columns: X8 and Y (transformed values). if it included: LOGY = log(Y) then the variable LOGY will not be included in the output worksheet.PMO. stored in the \EXAMPLES directory. Note: When WT is specified in a model file. using special variables that WinNonlin makes available to hold the model input and output. Each individual comment line must begin with REMA (or REMARK). For example. In general. 222 . Arrays and Functions” on page 557 WinNonlin TRANSFORM This block of statements is called once for each observation being modeled.

since the equations for both involve K10. 2 for FUNC 2. The NPARAMETERS command specifies the number of parameters to be estimated. that is. NFUNCTIONS specifies the number of equation sets associated with observations. one for urine. INITIAL. and saved in a Pharsight model object. its values must correspond to the FUNCTION numbers. The NCONSTANTS command sets the number of (dosing) constants in the model. When more than one function is defined (NFUN>1) it is important to define which observation data belong to each FUNCTION block. to define model-related values such as NPAR (number of parameters). when modeling both plasma and urine data for a one compartment IV bolus model it would be more efficient to model these simultaneously. etc. A function variable in a separate column indicates which observations belong with which function.e. This block cannot include WinNonlin programming statements. For example: 223 . i. The input data set should contain all observations (Y) in one column. within the same MODEL using multiple FUNCTION blocks. it may be more efficient to model the functions simultaneously. The NDERIVATIVES command tells WinNonlin the number of differential equations in the model. This variable is identified via WinNonlin’s Data Variables dialog. The equations for each compartment would appear in a separate FUNCTION block: one for plasma data. Each constant must be initialized via either the CONSTANTS command or the WinNonlin dosing interface. the constants should be specified at run time using WinNonlin’s dosing interface. and must include the number of values indicated by NPARAM.. the number of FUNCTION blocks. For example. the multiple dosing models in WinNonlin's library allow a variable number of doses per subject. The NSECONDARY command indicates the number of secondary parameters to be computed as functions of the estimated parameters.User Models ASCII user models 10 COMMANDS The COMMANDS block associates WinNonlin commands with a model. If a model will fit multiple functions that share one or more parameters. these items can be set using the WinNonlin interface. i. or using the FNUMBER N command.e. This block provides backward-compatibility with WinNonlin command files. where N is the column number. The variable K10 would be defined in a TEMPORARY block for use in both functions. 1 for FUNC 1. Instead of using a COMMAND block. Some models allow a variable number of constants. For example.. LOWER and UPPER set initial estimates and bounds. In such cases.

numbers or underscores. The DNAMES command may be used to define names for the columns in the input data. Variables that are used only within a given block can be defined within that block. but must begin with a letter. These names provide column labels in the output.693/K10 END TEMPORARY Statements in the optional TEMPORARY block define global variables. Variable names include up to eight (8) letters. COMM DNAMES ‘SUBJECT’. that is. and may be used in programming statements. variables that can be used by any block in the model.10 User’s Guide WinNonlin COMMANDS NSEC 1 END SECO S(1)=0. The PUNITS and SUNITS commands supply optional units for the primary and secondary parameters. commands name the primary and secondary parameters.’1/hr’ NSEC 1 SNAMES 'CL' SUNIT ‘ml/hr’ END SECO CL =0. and SNAMES. 224 . ‘FORM’. 'K1E' PUNIT ‘1/hr’. 'VOLUME'. ‘CONC’ NPARM 3 PNAMES 'K10'.693/K10 END The PNAMES. These names can then be used in the model definition statements. ‘TIME’. For user models. ‘ml’. units are not included in calculation but are carried along to the output.

F=Z(1)or F=Z(1)+3Z(2). e.. DZ(2).e.. a STARTING block is required to set initial values for each. starting values will equal the value of the dependent variable when X = 0. then X should also be assigned accordingly. Assign initial values to elements of the Z array as shown below. F=(D/V)exp(-K10t). such as C0 below. FUNC 1 F=A*EXP(-ALPHA*T) END FUNC 2 F=Z(1) END This section can also define weights by setting WT equal to the weight. If the initial values do not occur at X = 0. Note that the initial values could be estimated as parameters.User Models ASCII user models 10 FUNCTION Each set of equations that will be fit to a body of data requires a FUNCTION block. The reserved variable F should be assigned the function value as: • • an algebraic function. START Z(1)=C0 Z(2)=0 END DIFFERENTIAL This block defines a system of differential equations using the array: DZ(1).g. A DIFFERENTIAL block is required when NDERIVATIVES > 0. etc. e. In most instances. NDERIVATIVES > 0). or a function of one or more differential equations defined in one or more DIFFERENTIAL blocks.. Most models include only one. STARTING When a model includes differential equations (i.g. 225 .

Alternately.5)/α and -LN(0. though clearance is of greater interest. α and β are parameters to be estimated. use SNAMES as follows. For example. a model may be defined micro constants. interest is not in the model parameters per se. t denotes time and C(0) is constrained to be zero (i. S(2) the second. SECO REMARK Note that D and AUC must be defined above S(1) = D/AUC END Assign secondary parameter values to the S array. COMM NSEC 1 SNAME ‘CL’ END SECO CL = D/AUC END ASCII model examples Example 1 To fit a set of data to the following model: C(t) = Aexp(-αt) + Bexp(-βt) A.10 User’s Guide WinNonlin DIFF DZ(1)=-K12*Z(1) DZ(2)=K12*Z(1) . A + B = 0). etc. B is not estimated.e.5)/β 226 . This model will also estimate the half lives of α and β: -LN(0. The SECONDARY block addresses this. implying that B = -A. S(1) denotes the first secondary parameter. as C(0) = 0. but in functions of them.K20*Z(2) END SECONDARY In many instances..

and β. The above model will work fine.5)/P(3) END EOM The FUNCTION .User Models ASCII user models 10 The following statements define the model and secondary parameters: MODEL 1 FUNCTION 1 F=P(1)*EXP(-P(2)*X)-P(1)*EXP(-P(3)*X) END SECONDARY S(1)=-LOGE(*. S(1) and S(2) are the secondary parameters. 227 . The SECONDARY . α. to be estimated. However. as follows.END block defines function one (the only function).5)/P(2) S(2)=-LOGE(*. descriptive parameter names and comments could make the model easier to read. as a function of X (which is time in this example) and three parameters A. F denotes the concentration predicted by the model.END block defines secondary parameters (functions of the model parameters).

The PNAMES and SNAMES statements define the parameter and secondary parameter names that are used in the programming statements and the output. The following statements would appear after the MODEL statement: COMMANDS NPAR 2 NSEC 2 END Example 2 Consider a model characterized by the following differential equations. ‘BETA’ SNAMES ‘ALPHA_HL’. ‘ALPHA’.5)/BETA END EOM The above models are functionally equivalent.10 User’s Guide WinNonlin MODEL 1 remark: Example problem COMMAND NPARM 3 NSEC 2 PNAMES ‘A’. so they needn’t be set when loading the model to WinNonlin. ‘BETA_HL’ END FUNC 1 B=-A remark: Model is constrained so that C(0)=0 TIME=X E1=EXP(-ALPHA*TIME) E2=EXP(-BETA*TIME) F=A*E1 + B*E2 END remark: Secondary parameters SECO ALPHA_HL=-LOGE(*. In the second model. 228 . the COMMAND .5)/ALPHA BETA_HL=-LOGE(*. It is useful to also store NSEC and NPAR with the model.END block holds model commands.

= K 12 ⋅ Z ( 1 ) – K 20 ⋅ Z ( 2 ) dt with initial conditions Z1(0) = D and Z2(0) = 0. first use a PNAMES statement to assign meaningful names to the parameters and facilitate model definition. FUNC 1 F=Z(2) END The next set of statements completes the model specification. DIFF DZ(1) = -K12 * Z(1) DZ(2) = K12 * Z(1) .= – K 12 ⋅ Z ( 1 ) dt dZ 2 -------.K20 * Z(2) END EOM 229 . define the initial values for the differential equations. START Z(1) = D Z(2) = 0 END A FUNCTION block is required to define the model associated with the observed data. ‘D’ END Next. ‘K20’. MODEL 1 COMMAND NPARM 3 PNAMES ‘K12’. K20 and D. As in the previous example.User Models ASCII user models 10 dZ 1 -------. In this case Z2 represents plasma concentration. The parameters to be estimated are K12.

03 0. the statement: I = DTA(3) assigns I the value of the indicator variable. 0. The third is an indicator variable.0 2.0 3. Suppose there are two sets of data to fit simultaneously.5 1.7 0.0 .. Both data sets have the same model and parameters except that the second set requires a time lag.. 0 0 0 0.0 0. ‘ALPHA’. The model reads as follows.. use the DTA( ) variable.9 . Its value indicates to which set of data the observations belong. The second is concentration. ‘LAG’ END FUNC 1 I = DTA(3) IF I = 1 THEN :First Set of Data T=X F = A*EXP(-ALPHA*T) + B*EXP(-BETA*T) ELSE :Second Set of Data T = X-LAG F = A*EXP(-ALPHA*T) + B*EXP(-BETA*T) ENDIF END EOM 230 .5 1. MODEL 1 COMM NPAR 5 PNAMES ‘A’. ‘BETA’..10 User’s Guide Example 3 This example shows how to use the predefined variables to alter a model based on the current observation. Since the third column contains the indicator values.02 2 2 2 2 1 1 1 WinNonlin The first column is time.. Assume the following data: 0. To access its value during the model-fitting process. ‘B’.0 2.

DTA(N) This array contains values of all columns from the input data file (in the same order) corresponding to the current observation. using the FNUMBER command. the function with the time lag is fit. for a given observation. Thus. P(2). DTA(1) = value of 1st variable (column) DTA(2) = value of 2nd variable (column) etc. ‘BETA’. There is another way to fit these data. the second creates two. models that can be altered based on the current observation or value of the parameter estimates. ‘B’. etc. When I is not 1. P(1). The two functions can be defined separately. F 231 . These variables can also be accessed using DNAMES.User Models WinNonlin variables 10 When I = 1 the function without the time lag is fit. that is. This allows “adaptive” models—that is. MODEL 1 COMM NPAR 5 PNAMES ‘A’. The value of the function evaluated at the current observation. WinNonlin variables WinNonlin provides access to input data and parameter estimates during the model-fitting process through special variables. ‘LAG’ NFUNC 2 FNUM 3 END FUNC 1 T=X F = A*EXP(-ALPHA*T) + B*EXP(-BETA*T) END FUNC 2 T = X-LAG F = A*EXP(-ALPHA*T) + B*EXP(-BETA*T) END EOM These approaches produce the same parameter estimates but different output. and current values of the parameters to be estimated. The first creates one summary table. I = 2. ‘ALPHA’. one per function. in two FUNCTION blocks. listed below.

Subscripts are not allowed. or using DNAMES. END. This array contains the constants that have been input by the user (e. This array contains the values of the estimated secondary parameters. DTAARRAY(i. Values assigned to WT will override any weighting specified via the Data Variables dialog. CON. TIME(2) is invalid. Reserved variable names Variable names must not begin with: • • any WinNonlin keyword (e. X... YARRAY(j). NSEC. SECO. DTA. TRAN. DTIM. This array contains the current values of the parameters to be estimated. 232 . This array contains the current values of the differential equations evaluated at X.g. e. STEA. or EOM) any WinNonlin command or data array (e.. WTARRAY(j). These variables can also be accessed using PNAMES.10 User’s Guide X Value of the independent variable at which the functions or differential equations are to be evaluated. numbers and/or underscores and must start with a letter. OBJFN. OBJF. NVAR. 56. P. CARR. FUNC. dose). Z). XARRAY(j). This array contains the current integrated values corresponding to a system of differential equations.g. NOBS. F. NCAR.g. COMM. These are created as global variables in the Temporary block or as local variables within any block. DIFF. NDER. TEMP.j)..g. PAUC. the independent variables can be accessed via the DTA( ) array. P( ) and Z( ). NPARM.) If the model is a function of more than one independent variable. NTOT. NCON. WinNonlin P(N) CON(N) Z(N) DZ(N) S(N) WT Temporary variables Most models require other variables to define the model or parameters. NPAU. p. Weight assigned to the current observation. STAR. OBSN. These values are computed by WinNonlin (except when starting values for the system of differential equations are defined by the user). These variables can also be accessed using SNAMES. Variable names can include 1-8 letters. This value is identical to DTA(XNUM) (See the XNUM command.

*. Control is transferred here via a GOTO statement. It provides control statements. mathematical functions and loop control. Note that ** denotes exponentiation. and columns from the data file. secondary parameters.User Models The WinNonlin programming language 10 Aliases The commands PNAMES. control is transferred to the next ELSE. It is possible to define new variables for easier programming. The WinNonlin programming language WinNonlin provides a flexible programming language for ASCII user models. -. Valid mathematical operators are +. if expression is false. Statements GOTO Action Syntax Remarks Example LABEL Action Syntax Example IF THEN ELSE Action Transfers processing to a labeled statement. **. ELSE IF. Expressions are evaluated using standard operator precedence. IF I>4 THEN GOTO GREEN ELSE GOTO BLUE ENDIF Syntax Example 233 . statements in the IF block are executed. GREEN: If expression is true. SNAMES. constants. GOTO GREEN Denotes a label. and /. and DNAMES can define aliases for parameters. IF expression THEN statements ELSE statements Where expression is a logical expression. There should not be any blank spaces between GO and TO. GOTO label Where label is the statement label of an executable statement. or ENDIF statement at the same IF level. The label must end with a colon (:).

Var is always initialized to exp1.THEN. After executing the last round through the loop. DO I=1 TO 3 DO J=1 TO 5 statements NEXT NEXT IF statements can be nested up to 10 deep with the default model size (2). exp2 is an upper limit for the loop. exp2 is a lower limit for the loop. See IF .. exp3 is assumed to be 1. and up to 30 for model sizes 3 or greater.e. through the NEXT statement that ends the loop. IF statements can be nested up to 10 deep with the default model size (2). var has the value that caused the loop to exit (i. If exp3 is negative. STEP exp3 is optional. even if the loop is not executed (because the value of exp1 is beyond the limit placed by exp2).THEN. ELSE WinNonlin This statement is optional. If omitted. Terminates an IF block. it is beyond the limit set by exp2).10 User’s Guide Remarks ELSE Action Syntax Remarks Example ENDIF Action Syntax Remarks Example DO . If exp3 is positive. See IF .. exp2 is an expression whose value is the limit for the loop. exp1 is an expression whose value is the initial value of var. Marks the beginning of an ELSE block. Remarks Example 234 . exp3 is an expression whose value is the incrementor for the loop. and up to 30 for model sizes 3 or greater. An ELSE block consists of any executable statements between the ELSE statement and the next ENDIF statement at the same IF level. NEXT Action Syntax Each block of statements corresponding to each IF or ELSEIF statement must end with an ENDIF. Repeatedly executes the statements following the DO statement.. DO var=exp1 TO exp2 [STEP exp3] Where var is a variable to be incremented during the looping.

DEXP (X) ALOG (X). X) sine (X) cosine (X) Returns the maximum of X and Y. Y) MIN (X. DLOG10 (X) LOG10 (X) log base 10 of X: log10 (X) X 235 . DSQRT (X) SIN (X). WT. DLOG (X). DMIN1 (X. DATAN (X) ATAN2 (Y. Y). DABS (X) ATAN (X). DCOS (X) MAX (X. DSIN (X) COS (X). E. Y) ABS (X). Operator AND OR GT LT EQ GE LE > < = >= <= or greater than less than equal to greater than or equal to less than or equal to Definition and Functions Function Name LAG(var) INT(var) EXP (X). Var can be X. Returns the minimum of X and Y. DATAN2 (Y. DMAX1 (X. Returns the absolute value of X.User Models The WinNonlin programming language 10 Comparison operators Table 10-2.: INT(3. Y). LOGE (X) SQRT (X).9) = 3 ex natural log of X: ln(X) ALOG10 (X). X).G. tan-1(X) tan-1(Y/X) Result Returns the value of var at the last execution. or any temporary variable. Truncates var to an integer value. Y.

associated with the normit (NOR) and a sample size N. Bioassay functions Function name NORMIT (R. the model fitting process can be greatly accelerated by compiling the model into a 32-bit Windows dynamic link library (DLL) using either Compaq Visual FORTRAN 6. then p is set to 1/2N.0. This function returns the normit of p. Each model requires two files: 236 .10 User’s Guide Function names that are grouped together are interchangeable. both EXP(X) and DEXP(X) will produce the same result.N) This function returns the wt. WTNOR. For example. namely ex. where and LOGIT(P) LOGIT = Compiled user models For User Models defined in terms of several differential equations. If R=N.N) Result Let p denote R/N. it returns the value NOR such that WinNonlin Note: If R=0. That is. then p is set to 1-1/2N WTNORM (NOR.0 or Microsoft Visual C++ 6.

DLL for the above FORTRAN source file. Select Fortran Dynamic Link Library as the kind of product to create. The source code file (*. The files include: File F_UEXP6.DLL for the above FORTRAN source file. The source code file contains one subroutine specifying the model. Note. 2. Click OK. Choose File>New from the menus.C). 3.DLL is a reserved name in WinNonlin. F_UEXP15. 237 . that the project should not be named USRMOD.PMO). *. 5. Compiled .DLL Purpose FORTRAN source file for a system of two differential equations with data on both compartments (same model as EXAMPLES\EXP6.DLL Compiled . Enter a project name at the top left.F90 F_UEXP6. The compiled model. however.F90 or *. F_UEXP15. This subroutine must be named USRMOD( ). Two examples of the files required to build a DLL in FORTRAN have been installed to the \USERCOMP subdirectory of the WinNonlin installation directory. Either of these f90 files can be used as a template. 2. Example using Digital (now Compaq) Visual FORTRAN 6. 4. Open Visual Fortran. USRMOD. The same blocks are included.User Models Compiled user models 10 1. but implemented in FORTRAN or C++ instead of WinNonlin programming language.PMO ).F90 FORTRAN source file for a PK/PD link model (same model as EXAMPLES\EXP15. Using FORTRAN Creating a source file The format of this file is very similar to that of WinNonlin ASCII user models.0 To create a compiled user model in FORTRAN: 1.DLL.

In the next dialog select Empty dll application.F90 or USERCOMP\F_UEXP6.0. Choose Build>Build mymodel. 9. for this example: D:\MYPROJECTS\MYMODEL\RELEASE\MYMODEL.DLL.F90 (located in the WinNonlin installation directory) in a text editor. but the subroutine must be named USRMOD( ). Arrays and Functions” on page 557 14. To use one of the provided templates. Click OK to open a frame for the Fortran file. The same blocks are included. 11. Copy/paste the content into your new FORTRAN file and edit as needed. Enter the model code. Using C++ Creating a source file The format of this file is very similar to a WinNonlin ASCII model. 8. with a C++ implementation. 10. The .f90. Select Win32 Release and click OK. See the following for model structure and syntax: • • • “Model structure” on page 220 “WinNonlin variables” on page 231 “Commands.DLL is saved to the Release subdirectory in the current directory. 7. Note: The source file stored on disk can be assigned any name. 13. An example of this file is provided for Microsoft® Visual C++ 6. This example uses modelname.dll from the menus. This minimizes the size of the file. Enter a file name. 15. open USERCOMP\F_UEXP6. Choose Build>Set Active Configuration… from the menus.10 User’s Guide 6. Choose Project>Add to Project…>New from the Visual FORTRAN menus. Click OK. Click Finish. 238 . Select Fortran Free Format Source File. WinNonlin 12.

10. Select Win32 Dynamic Link Library. Click OK. 4. Enter a file name. 11. 239 .cpp extension. Either can be used as a template. These files include: File C_UEXP6. In the next dialog select An empty dll project. 9. 8. 5.c. 7. Press OK. Example using Microsoft Visual C++ 6. Select New from the File menu. 6. Note: It is important to use the . Select C++ Source File. run Example 15 in both the ASCII form and in the compiled form.DLL Purpose C source file for modeling example 6.DLL C_UEXP15. 6. Compiled . Two sets of files have been provided. 3.0.User Models Compiled user models 10 The files required to build a DLL in Visual C++ have been installed to the \USERCOMP subdirectory. Note: To demonstrate the benefits of compiling models.c file extension rather than the . On the Project menu select Add to Project…New.0 To compile a user model: 1. C source file for modeling example 15. Click Finish. Compiled . This example uses modelname.C C_UEXP6. Enter a project name. Open Microsoft® Visual C++ v.DLL file for modeling example 15. 2.DLL file for modeling example 6.C C_UEXP15.

This is not essential. 3. Select from the Build menu Set Active Configuration…Win32 Release... WinNonlin To alter this example for another model: 1. The DLL is built in the specified directory (D:\MYPROJECTS\MYMODEL\RELEASE\MYMODEL.10 User’s Guide 12. 240 . Use the templates provided (see table above). but it builds a smaller DLL file. etc. Replace “k12 = p[0]. .DLL) Note: Statements in the Visual C++ source file with // starting in the first column are comments. For each block. The source file stored on disk can be assigned any name. 2. but the subroutine must be called USRMOD( ).” with the parameters for the model as they would be stated in a Temporary block. 13. A frame opens for the program. Enter the code for the model. All integers should be declared as INT and all real numbers should be declared as DOUBLE. Select from the Build menu Build mymodel. 14. include the corresponding block from the template file.dll. C++ makes no assumptions about types of variables. other than the Command block.

WEIGHT = Yn. There are three ways to assign weights. and setting n = -0. Weighted least squares When using weighted least squares. Iterative reweighting: weight by a power of the predicted value of Y 3. WinNonlin weights each observation by the value of the dependent variable raised to the power of n. Weighted least squares: weight by a power of the observed value of Y 2.. selecting this option. If n has a negative value. For example. 1 ⁄ ( Y) .e. 241 . Weights are assigned after loading a model. i. then the corresponding weights are set to zero..5 instructs WinNonlin to weight the data by the square root of the reciprocal of observed Y. through the WinNonlin user interface: 1. programming statements may also be used to define weights as part of the model. via the Model Options dialog (see “Model options” on page 203 for compartmental models and “Model options” on page 173 for NCA). and one or more of the Y values are less than or equal to zero. i. other than uniform weighting.Chapter 11 Weighting Weighted nonlinear regression WinNonlin provides flexibility in performing weighted nonlinear regression and using weighted regression.e. Reading weights from the data file: include weight as a column in the data For a user ASCII model.

. The weights should be the reciprocal of the variance of the observations. Iterative reweighting differs from weighted least squares in that for weighted least squares the weights are fixed. and therefore the predicted values and the weights change at each iteration. where F is the predicted response. See “Scaling of weights” on page 242. selecting this option. As with Weighted least squares. ˆ 1 ⁄ ( Y) .5 instructs WinNonlin to weight the data by the square root of reciprocal of the predicted value of Y. the weights for the individual data values are scaled so that the sum of the weights for each function is equal to the number of data values (with nonzero weights). iterative reweighting can be used to obtain maximum likelihood estimates (see the article by Jennrich and Moore in the References). i.11 User’s Guide The program scales the weights such that the sum of the weights for each function equals the number of observations with non-zero weights. As with Weighted Least Squares. For iteratively reweighted least squares the parameters change at each iteration. For certain types of models. and one or more of the predicted Y values are less than or equal to zero. For example. Reading weights from the data file It is also possible to have a variable in the data file that has as its values the weights to use. WinNonlin Iterative reweighting Iterative Reweighting redefines the weights for each observation to be Fn. the program will scale the weights such that the sum of the weights for each function equals the number of observations with non-zero weights (see “Scaling of weights”). if N is negative. then the corresponding weights are set to zero.e. Scaling of weights When weights are read from the data file or when weighted least squares is used. and setting n = -0. 242 .

If the variable WT is defined with programming statements.0123 0.0051 is the same proportion as 1. or 1.0278/0.843 after rounding. Weights in ASCII models When writing a custom ASCII model. Consider the following example: 243 .Weighting Weights in ASCII models 11 The scaling of the weights has no effect on the model fitting as the weights of the observations are proportionally the same. weighting by 1/Y*Y).819 0.e. (0. The corresponding values are as shown in the third column.0051 0.g. it is possible to define the weights using programming statements at the same time the model is defined. Consider the following example: X 1 10 100 Sum Y 6 9 14 Y-2 0. WT is a special variable name in WinNonlin.0278 0.843/0.338..000 Suppose weighted least squares with the power -2 was specified (i.8426238. Each value of Y-2 is divided by the sum of the values in the third column. However. e. Note: Scaling can be turned off by selecting the No Scaling of Weights option on the Weight tab of the Model Options dialog or by using the NOSCALE command in a command file. Note that 0.0452254) * 3 = 1. and then multiplied by the number of observations. scaling of the weights provides increased numerical stability.0452 Weight assigned by WinNonlin 1.338 3.843 0.. this variable overrides any weighting specified in the WinNonlin Model Options dialog.0277778/.

(See “Iterative reweighting” on page 242. and 1/Y2 if X is less than or equal to 10. as shown in the following example: MODEL 1 FUNC 1 F = A*EXP(-ALPHA*X) . As the values of Y are constant across iterations. the corresponding weights are also constant.A*EXP(-BETA*X) WT = 1/(F*F) END (additional statements here) This is equivalent to specifying Predicted to power n = -2. it is also possible to use programming statements to iteratively reweight the observations. However.11 User’s Guide WinNonlin MODEL 1 TRANSFORM IF X > 10 THEN WT = 1/Y ELSE WT = 1/(Y*Y) ENDIF END (additional statements here) The above statements would set the weight of the observation equal to 1/Y if the value of X is greater than 10.) Each function to be fit can use a different weighting scheme. For example: MODEL 1 TEMP (additional statements here) END FUNC 1 (additional statements here) WT = (some function) END FUNC 2 (additional statements here) WT = (a different function) END (additional statements here) 244 .

Semicompartmental Modeling and Crossover Design The WinNonlin toolbox includes the following tools: • • • Nonparametric superposition: used to predict drug concentrations after multiple dosing at steady state.Chapter 12 The WinNonlin Toolbox Nonparametric Superposition. Semicompartmental modeling: estimates effect-site concentration given plasma concentration and a value for Ke0. WinNonlin’s nonparametric superposition function is used to predict drug concentrations after multiple dosing at steady state. calculates confidence intervals for treatment median and median difference between treatments. which does not assume any pharmacokinetic (PK) model. residual. and is based on noncompartmental results describing single dose data. This can be done by fitting the data to a compartmental model with some assumptions about the absorption rate of the drug. An alternative method is based on the principle of superposition. based on concentration data from a single dose. The predictions are based upon an accumulation ratio computed from the terminal slope (Lambda Z). and estimates relevance of direct. Numeric deconvolution: evaluating in vivo drug release and delivery based on data for a known drug input • Example are provided in the Examples Guide. Nonparametric superposition In pharmacokinetics it is often desirable to predict the drug concentration in blood or plasma after multiple doses. and period effects. The feature allows predictions from simple (the same dose given in a constant inter245 . Crossover design: performs nonparametric statistical tests.

(i=1. to characterize the drug absorption and elimination process. The former assumes that the effect of each dose can be separated from the effects of other doses. one must have a complete characterization of the concentration-time profile after a single dose. one may obtain the concentration C(t) at any time t through interpolation if t1 < t <tn. The results can be used to help design experiments or to predict outcomes of clinical trials when used in conjunction with the semicompartmental modeling function. or extrapolation if t > tn. then C(t) = β exp(-λzt) for t > tn. linear pharmacokinetics. after a single dose D.n). The extrapolation assumes a log-linear elimination process. and drug concentration. that is. In order to predict the drug concentration resulting from multiple doses. the time range included in the terminal phase is specified by the user. WinNonlin Computation method Given the concentration data points C(ti) at times ti.2. The drug concentration at any particular time during multiple dosing is then predicted by simply adding the concentration values as shown below. dosing. and linearity of the underlying pharmacokinetics.…. The required input data are the time. See “Performing nonparametric superposition” on page 248 for stepped instructions Data and assumptions Nonparametric Superposition assumes that each dose of a drug acts independently of every other dose. that the rate and extent of absorption and average systemic clearance are the same for each dosing interval. Note: User-defined terminal phases apply to all sort keys. dosing schedules and doses are the same for all sort keys.n).2. If 246 . assumes that changes in drug concentration will vary linearly with dose amount. (i=1. so that a change in dose during the multiple dosing regimen can be accommodated. the terminal phase in the plot of log(C(t)) versus t is approximately a straight line. Two assumptions about the data are required: independence of each dose effect. The slope λz and the intercept are estimated by least squares from the terminal phase.12 User’s Guide val) or complicated dosing schedules. If the absolute value of that line’s slope is λz and the intercept is ln(β). In addition. That is. and that linear pharmacokinetics apply. it is necessary to know C(ti) at sufficient time points ti.…. The latter.

j j = 1 .2 . The half life is ln(2)/λz. The total predicted concentration at time t will be: Conc ( t ) = ∑ Cj ( t ) . then steady state can be reached after sufficient time intervals. the steady state (ss) is reached: Css(t) = C1(t) + β exp[-λ(t+τ)]/[1-exp(-λτ)] To display the concentration curve at steady state.The WinNonlin Toolbox Nonparametric superposition 12 the user does not specify the terminal phase.…. so τ is greater than tn. estimates from the best linear fit (based on adjusted R-square as in noncompartmental analysis) will be used. t is relative to dose time) will be: Cn(t) = C1(t) + β exp[-λ(t+τ)] + β exp[-λ(t+2τ)] +…+ β exp[-λ(t+(n-1)τ)] = C1(t) + β exp[-λ(t+τ)]{1-exp[-λ(n-1)τ]} / [1-exp(-λτ)] As n→∞. The concentration due to dose Dj will be: Cj(t) = (Dj / D)∗C(t-τ j) where t is time since the first dose and C(t.m If the same dose is given at constant dosing intervals τ.e.m. Let the concentration of the first dose be C1(t) for 0 < t < τ. ti-1 < t < ti 247 .ti-1]. Then the concentration at time t after nth dose (i. j = 1. Linear interpolation is appropriate for log-transformed concentration data and is calculated by: C(t) = C(ti-1) + [C(ti)-C(ti-1)]*[t.. and each dose is administered after τ j time units from the first dose. For interpolation. and τ is sufficiently large that drug concentrations reflect the post-absorptive and post-distributive phase of the concentration-time profile.τ j) = 0 for t ≤ τ j.ti-1]/[ti. Suppose there are m additional doses Dj. WinNonlin assumes steady state is at ten times the half life. WinNonlin offers two methods: linear interpolation and log-interpolation.… .

and is evaluated by: C(t) = exp{log(C(ti-1))+[log(C(ti))-log(C(ti-1))]*(t-ti-1)/(ti-ti-1)}. 5. Make sure the appropriate data set appears under Dataset. ti-1 < t < ti For additional reading see Appendix E of Gibaldi and Perrier (1982). 6. A separate analysis is performed for each unique combination of sort variable values. 4. Performing nonparametric superposition Note: All quantities in the Nonparametric Superposition dialog should be entered in the units of the input data set. Concentration for First Dose: Drag the Concentration column to this field. If units appear in the input worksheet. The Nonparametric Superposition dialog appears. Number of output data points: Enter the number of predicted values to output.12 User’s Guide Log-interpolation is appropriate for original concentration data. Time for First Dose: Drag the Time column to this field. and its variables appear under Variables. and appear in the output. drag one or more columns to the sort variables field. Choose Tools>Nonparametric Superposition from the WinNonlin menus. WinNonlin To perform nonparametric superposition: 1. Open the data set providing the single-dose concentrations. those units are applied in nonparametric superposition. 2. Sort variables: If using sort variables to define individual data profiles. 248 . 3.

Enter the maintenance dose. a Nonparametric Superposition Workbook and a Nonparametric Superposition Chart. select Linear for log-transformed concentration data.The WinNonlin Toolbox Nonparametric superposition 12 7. d. c. Enter individual doses and dose times on the Variable Dosing tab. Output Nonparametric Superposition generates a workbook containing predicted concentrations and Lambda Z values. Enter time range for which to output predicted data. If a variable dosing schedule is selected. b. Define Terminal Phase: (Optional) To set the time range for Lambda Z calculations. WinNonlin checks for duplicate time values in the data and prompts for sort variables if they have not been specified. Select whether to display the chart at steady state or at a particular dose. 8. If a regular dosing schedule is selected. For a repeating dose regimen. 10. check Repeat every n time units and enter the repeat time. 11. The Concentrations worksheet contains times and predicted concentrations for each level of the sort variables. 9. 249 . n. WinNonlin performs the calculations and generates two new windows. enter doses and times for one cycle. Log for untransformed data. Regular dosing: For dosing at a regular time interval: a. Enter the value for Tau. the dosing interval. c. Variable Dosing: If the time between doses varies: a. in time units matching those of the time variable in the data set. Click Calculate. Method for computations: Select linear or logarithmic interpolation and extrapolation. Generally. Enter the initial dose in the Initial Dose field of the Regular Dosing tab. this output represents times and concentrations between two doses at steady state. and a chart window containing plots of predicted concentration over time for each sort level. check this check box and enter start and end times for the terminal elimination phase. the output includes the times and concentrations within the supplied output range. The Lambda Z worksheet contains the Lambda Z value and the half-life for each sort key. b. The workbook contains two worksheets: Concentrations and Lambda Z.

Stanski. Miller and Ham (1979)). See “Performing semicompartmental modeling” on page 252 for stepped instructions Data and assumptions Drug concentration in plasma Cp for each subject is measured at multiple time points after the drug is administrated. Vozeh. A scientist developing a PK/PD link model based upon simple IV bolus data can use this function to compute effect site concentrations from observed plasma concentrations following more complicated administration regimen without first modeling the data. The hysteresis loop collapses with the graph of effect versus effect-site concentrations.12 User’s Guide WinNonlin Semicompartmental modeling Semicompartmental modeling was proposed by Kowalski and Karim (1995) for modeling the temporal aspects of the pharmacokinetic-pharmacodynamic relationship of drugs. The results can then be compared to the original data sets to determine if the model suitably describes pharmacodynamic action after the more complicated regimen. Effect-site link model is used and the value of the equilibration rate constant keo that accounts for the lag between the Cp and the Ce curves is required. Computation method In the effect-site link model (Sheiner. This function should be used when a counterclockwise hysteresis is observed in the graph of effect versus plasma concentrations. The effect site concentration Ce is related to Cp by first-order disposition kinetics and can be obtained by solving the differential equation: 250 . This model was based on the effect-site link model of Sheiner. Vozeh. Stanski. To minimize the bias in estimating Ce. Miller and Ham (1979) to estimate effect-site concentration Ce by using a piecewise linear model for plasma concentration Cp rather than specifying a PK model for Cp. a hypothetical effect compartment was proposed to model the time lag between the PK and PD responses. WinNonlin’s semicompartmental modeling estimates effect-site concentrations for given times and plasma concentrations and an appropriate value of keo. the time points need to be adequately sampled such that accurate estimation of the AUC by noncompartmental methods can be obtained. The potential advantage of this approach is reducing the effect of model misspecification for Cp when the underlying PK model is unknown. A piecewise log-linear model is assumed for Cp.

the ‘log’ method uses log-linear piecewise PK model. WinNonlin provides three methods. use a piecewise linear model for Cp: Cp (t) = C p + λ j (t-t j-1) j–1 t j-1 < t < t j where λ j = ( C p – C p )/( t j – t j-1 ) and C p = Cp ( tj – 1 ) . One can also assume a log-linear piecewise PK model for Cp Cp (t) = C p exp[ . the default ‘log/linear’ method uses linear to Tmax and then log-linear after Tmax.λ j (t-t j-1)] j–1 here λ j = (ln C p .⎞ [ 1 – e e0 j j – 1 ] + λ j ( t j – t j – 1 ) ⎝ ⎠ k e0 The initial condition is Cp(0) = Ce(0) =0. which is usually given by compartmental PK models. The ‘linear’ method uses linear piecewise PK model. the ECp and E-Ce plot will be plotted.t j-1 ).The WinNonlin Toolbox Semicompartmental modeling 12 where keo is a known constant.ln C p )/( t j . one needs to know Cp as a function of time. and: j–1 j t j-1 < t < t j C ej = C ej – 1 e – k e0 ( t j – t j – 1 ) k e0 –λj ( tj – tj – 1 ) – k e0 ( t j – t j – 1 ) + C Pj – 1 ----------------. The Effect field is not used for the calculation of Ce but when it is provided. In order to solve this equation. assuming Ce (0) =Cp(0) = 0. j j–1 j–1 Using the above two equations can lead to a recursive equation for Ce: C ej = C ej – 1 e – k e0 ( t j – t j – 1 ) λj –k ( t – t ) + ⎛ C Pj – 1 – -----. 251 .[ e –e ] k e0 – λ j λ 1 and C e1 are estimated by a linear regression. Here.

at minimum. Keo: Enter the value for the equilibration rate constant. 9. 252 . Method for computations: Select Linear to use a linear piecewise PK model. Choose Tools>Semicompartmental modeling from the WinNonlin menus. Click Calculate. 3. Time: Drag the column containing times to this field. and its columns appear under Variables. Open a data set containing plasma concentrations and times. 5. This column is not used in estimating effect site concentration. 4. 7. Make sure the appropriate data set appears under Dataset. The Semicompartmental Modeling dialog appears.12 User’s Guide Performing semicompartmental modeling To perform semicompartmental modeling: WinNonlin 1. Sort variables: If using sort variables to define individual data profiles. Log for a log-linear piecewise PK model. or the default Log/linear for linear to Tmax and then log-linear after Tmax. 8. 2. 6. A separate analysis is performed for each unique combination of sort variable values. Effect: Drag a column containing drug effect data to this field in order to include concentration-effect plots in the output. Plasma Concentration: Drag the column containing blood concentrations here. drag one or more columns to the sort variables field. WinNonlin performs the computations and generates a new workbook and chart window.

The WinNonlin Toolbox Semicompartmental modeling 12 Output Workbook: WinNonlin creates a new workbook with the following data: any sort variables. Time. Ce (estimated effect site concentration). and Effect (if included). Conc. 253 . Examples are provided below. plots of Effect versus Cp and Effect versus Ce are also created. If Effect is included in the data set. Chart: The chart window contains plots of Cp versus Time and Ce versus Time. Note: Any units associated with the Time and Concentration columns in the input data are carried through to the semicompartmental modeling output.

In many situations the assumptions of variance homogeneity and normality. a separate plot appears for each. among other reasons.12 User’s Guide WinNonlin When the data set has multiple sort levels. it calculates confi254 . Crossover design The two-period crossover design is common for clinical trials in which each subject serves as his or her own control. This tool performs nonparametric statistical tests on variables such as Tmax and provides two categories of results. relied upon in parametric analyses. The nonparametric methods proposed by Koch (1972) can be used to analyze such data. WinNonlin’s Crossover Design tool performs statistical analysis of data arising from subjects when variance homogeneity and normality may not necessarily apply. may not be justified due to small sample size. First.

2.The WinNonlin Toolbox Crossover design 12 dence intervals for each treatment median. The data for one treatment must be listed first. along with the median of these n*m points. for example i=1 and k=1 is treatment T. it estimates the relevance of direct. See “Using the Crossover Design tool” on page 258 for stepped instructions. Suppose that n subjects are randomly assigned to the first sequence. The data are listed in columns 1 through 4 of Table 12-1 below. residual. subject j (j = 1.yij2. Y represents the outcome of the trial (for example Cmax or Tmax) and yijk is the value observed on sequence i (i=1. and column 5 gives the within-subject difference in Y between the two periods.. For “stacked” data. Test (T) and Reference (R).2). Crossover Design supports two data layouts: stacked in same column or separate columns. Finally. 255 .2).d2j)/2 are computed. all measurements appear in a single column..n. Similarly. TR. with one or more additional columns flagging which data belong to which treatment. Treatment is implied by sequence i and period k. Representation of the data Sequence (i) TR .…n+m) and period k (k = 1. and period effects. in which subjects are given treatment T first and then treatment R following an adequate washout period. i=1 and k=2 is treatment R.…. TR RT … RT Subject (j) 1 … n n+1 … n+m Yij1 y111 … y1n1 y2(n+1)1 … y2(n+m)1 Yij2 y112 … y1n2 y2(n+1)2 … y2(n+m)2 d2(n+m) d1n d2(n+1) Dij d11 dij = yij1 . Then the n*m crossover differences (d1i . then all the data for the other. n+1. It then calculates the median difference between treatments and the corresponding confidence intervals. m subjects are assigned to the RT sequence. Table 12-1. The alternative is to place data for each treatment in a separate column. Data and assumptions Consider a trial testing the effects of two treatments.

X(2). normal approximation is used: Xl = X(L).12 User’s Guide Descriptive analysis The median response and its confidence interval are calculated for Y of each treatment and the crossover difference between treatments.…. XN.…. where L = max ⎛ ⎜ i: ⎜ ⎝ ⎞ ⎛ N⎞ 2 – N < ( 1 – p ) ⁄ 2⎟ + 1 ∑ ⎝ i⎠ ⎟ ⎠ i=1 ⎞ ⎛ N⎞ 2 – N > ( 1 + p ) ⁄ 2⎟ – 1 ∑ ⎝ i⎠ ⎟ ⎠ i=1 U N N Xu = X(U).2 ⎠ ⎝2 2 ⎛ N⎞ N – int ⎜ N – z 1 + p -------⎟ ------------. X(N) be the order statistic of samples X1. X2. = X[(N+1)/2]. • For N <= 20. the exact probability for a binomial distribution is used: Xl = X(L). • For N > 20.2 ⎠ ⎝2 2 Xu = X(U). Let X(1). if N is even WinNonlin if N is odd The 100% confidence interval (CI) of Xm is defined as follows. 256 . The median Xm is defined as: Xm = [X(N/2)+X(N/2+1)]/2. C. where L = ⎛ N⎞ int ⎜ N – z 1 + p -------⎟ + 1 ------------. where U = where int(X) returns the largest integer less than or equal to X. where U = min ⎛ ⎜ i: ⎜ ⎝ The exact probability is ∑ ⎛ i⎞2 ⎝ ⎠ i=L N –N .

n+m. no period effect given no sequence effect. Thus.. ni=n for i=1. …. y2(n+1)k. 1980): [T . Hypothesis 4 can be tested using the bivariate Wilcoxon statistic on (Yij1. let Rijk equal: rank { yijk: y11k. For each period k. M = (n+m+1)/2 257 . 3.n(n+m+1)/2] / nm ( n + m + 1 ) ⁄ 12 ~N(0. hypothesis 2 can be tested using the Wilcoxon statistic on the difference D.1) Similarly. If R(Sl) is the rank of Sij in the whole sample.. the statistic to be used for testing hypothesis 4 is: L = (n+m-1)[nU1T*S-1*U1 + mU2T*S-1*U2] Where Ui is a 2x1 vector: [ R i1 -M. and ni=m for i=2. 2. Hypothesis 1 above can be tested using the Wilcoxon statistic on the sum S. no treatment and no sequence effect. no treatment effect given no sequence effect. R i2 -M]T. Yij2). and 4. hypothesis 3 can be tested using the Wilcoxon statistic on the crossover difference C.….The WinNonlin Toolbox Crossover design 12 Hypothesis testing Four hypotheses are of interest: 1. y2(n+m)k } The average rank for each sequence is ni R ik = ∑ Rijk ⁄ ni j=1 . The test statistic is: n n+m T = min ( ∑ Rl l=1 . where j=1.. The statistics are in the form described above. l = 1. y1nk. no sequence effect (or drug residual effect). ni. ∑ l = n+1 Rl ) The p-value is evaluated by using normal approximation (Conover. ….

Sequence: Drag the variable representing treatment sequence to this field. Subject: Drag the variable representing subject identifiers from the Variables list box to the Subject field. 4. 6. To use crossover design: 1. drag one or more columns to the sort variables field. Treatment data: Select whether the treatment data are in separate columns or stacked in one column. Note: See the WinNonlin Examples Guide for examples of crossover design. 3. Choose Tools>Crossover Design from the WinNonlin menus to open the Crossover Design dialog. or each treatment placed in a separate column. An example of each appears below. so the p-value can be evaluated. Make sure that the data set to use is open and active.12 User’s Guide S is the 2x2 covariance matrix: WinNonlin S = Σi Σj ( R ij1 – M ) 2 ( R ij1 – M ) ( R ij2 – M ) ( R ij2 – M ) 2 ( R ij1 – M ) ( R ij2 – M ) and L ~ chi-square(2). 2. Separate columns: 258 . Sort variables: If using sort variables in addition to subject IDs. 5. Each unique combination of sort variable values for each subject and treatment represents one data profile. treatment and sequence to define individual data profiles. Using the Crossover Design tool Crossover design supports two data formats: data for both treatments stacked in one column.

If response data for both treatments are stacked in a single column: a. Treatment: Drag the column the identifies each treatment to this field. b. Stacked in one column: 8. Click Calculate to perform the analysis. If data for each treatment appears in a separate column. drag the column containing response data for each treatment to the Treatment A and Treatment B fields as shown above. 259 .The WinNonlin Toolbox Crossover design 12 7. 9. Response: Drag the column containing response data to this field.

12 User’s Guide Output Crossover Design creates a new workbook with two worksheets. lower. as well as treatment and residual simultaneously. called Effects. The second worksheet. • 260 “Deconvolution methodology” . and 95%) for the treatment medians in the crossover data and the treatment difference median. and upper value parameters. It includes tests for an effect of sequence. WinNonlin Note: Units associated with the Response variable in the input data set units are carried through to the crossover design output. Deconvolution calculations and stepped usage instructions appear under the following headings. The output workbooks display units in the median. Numeric deconvolution Deconvolution evaluates in vivo drug release and delivery based on data for a known drug input. lists the test statistic and p-value associated with each of four tests. The first sheet lists the confidence intervals (at 80%. treatment and period. 90%.

all kinetic models defined in terms of linear combinations of linear mathematical operators (e. Depending upon the type of reference input information available. See “Numeric deconvolution” on page 260 for instructions on performing deconvolution. It is well recognized that classical linear compartmental kinetic models exhibit superposition linearity due to their origin in linear differential equations. including transdermal release and delivery. Similarly. the bioavailability is evaluated if the reference input is a vascular drug input.1 should operate in later versions. It considers other types of delivery. Linearity assumptions The methodology is based on linear system analysis with linearity defined in the general sense of the linear superposition principle.The WinNonlin Toolbox Numeric deconvolution 12 • Note: “Using deconvolution” on page 269 WinNonlin version 5. deconvolution. gastro-intestinal release is evaluated if the reference is an oral solution (oral bolus input).. etc. Deconvolution methodology Deconvolution is used to evaluate in vivo drug release and delivery when data from a known drug input are available. integration. gastro-intestinal release) or a composite form. but code to process deconvolution output may need modification to account for enhancements to the workbook output.g. Deconvolution provides automatic calculation of the drug input. It also allows the user to direct the analysis to investigate issues of special interest through different parameter settings and program input. Scripts written prior to WinNonlin 5. In this case. convolution. The Deconvolution feature is not limited in scope to drug delivery via the oral route. differentiation.1 established several enhancements to numeric deconvolution. the drug transport evaluated will be either a simple in vivo drug release (e.) constitute linear systems adhering to the superposition principle.g. typically consisting of an in vivo release followed by a drug delivery to the general systemic circulation. etc. However. Perhaps the most common application of deconvolution is in the evaluation of drug release and drug absorption from orally administered drug formulations. The scope and generality of the linear system approach thus ranges far beyond linear com261 . drug delivery from implanted devices.

a blood sampling) that can be reached by the drug molecule entering at A. The possible presence of the molecule at B at time t is a random variable due to the stochastic transport principles involved. It can be seen from this that the concentration of drug at the sampling site B at the arbitrary time t is proportional to the dose (N is proportional to the dose and g(t) does not depend on the dose due to the independence in the transport). To cut through all the complexity and objectively deal with the present problems.g. Imagine repeating this one molecule experiment an infinite number of times and observing the fraction of times that the molecule is at B at time t. Thus the probability of being at B for a molecule that enters A at time tA is g(t-tA). but now at time tA 262 .g. a bolus injection. For example.12 User’s Guide partmental models. Next. it is best to consider the kinetics of drug transport in a non-parametric manner and simply use stochastic transport principles in the analysis. such that the probability of a drug molecule's possible arrival at sampling site B is not significantly affected by any other drug molecule. physically structured modeling view of pharmacokinetics. The result is that the probability of being at point B at time t depends on the elapsed time since entry at A. Let B be a sampling point (a given sampling space or volume) anywhere in the body (e. e. Then the transport to sampling site B is both random and independent. In addition the probability of being at B at time t is the same and equals g(t) for all of the molecules. Let it be assumed that there is no significant interaction between the drug molecules.stochastic background WinNonlin The convolution/deconvolution principles applied to evaluate drug release and drug input (absorption) can be explained in terms of point A to point B stochastic transport principles. To fully appreciate the power and generality of the linear system approach. consider the entry of a single drug molecule at point A at time 0. Consider again the simultaneous entry of N molecules. This assumed property is called time-invariance. Convolution/deconvolution . Now let's further assume that the processes influencing transport from A to B are constant over time. For example. but is otherwise independent of the entry time. it is best to depart from the typical. That fraction represents the probability that a molecule is at point B given that it entered point A at time 0. in the present case the expected number of drug molecules to be found at sampling site B at time t (NB(t)) is equal to Ng(t). It is this combination of independence and equal probability that leads to the superposition property that in its most simple form reveals itself in terms of dose-proportionality. consider a simultaneous entry of a very large number (N) of drug molecules at time 0.. Denote this probability by the function g(t).

e... As a result the quantity g(t — tA)h(tA) is nonzero only over the range from 0 to t and the equation becomes g(t) = ∫0 g ( t – tA )h ( tA ) dtA t Suppose again that N molecules enter at point A but this time they enter at random times distributed according to h(t). As argued before. For typical applications drug input is restricted to non-negative times so that h(t) = 0 for t < 0.e. N B ( t ) = Ng ( t ) = N ∫ g ( t – t A )h ( t A ) dt A = 0 t ∫0 g ( t – tA )N'A ( tA ) dtA t Converting from number of molecules to mass units: MB ( t ) = ∫0 g ( t – tA )f ( tA ) dtA 263 t .e. This is equivalent to saying that the expected rate of entry at A is given by N’A (t) = Nh (t). t Pr ( t A ≥ t ) = ∫ h ( u ) du 0 The probability that such a molecule is at point B at time t is the average or expected value of g(t-tA) where the average is taken over the possible values of tA according to g(t) = ∫–∞ g ( t – tA )h ( tA ) dtA ∞ Causality dictates that g(t) must be 0 for any negative times.The WinNonlin Toolbox Numeric deconvolution 12 instead of 0. a molecule can't get to B before it enters A.e. the expected number of molecules at B at time t is the product of N and the probability of being at B at time t. i. It is instead a random quantity tA distributed according to some probability density function h(t). Suppose that the actual time at which the molecule enters point A is unknown. superposition. Combining this property with those of independent and equal probabilities. i.. i. results in an expected number of molecules at B at time t given by NB(t) = Ng(t-tA). i..

The function cδ (t) is denoted the unit impulse response (a..k. cδ is simply equal to the probability that a molecule entering point A at t = 0 is present in the sampling space at time t divided by the volume of that sample space. Tj + 1 – Tj t – Tj h j ( t ) = --------------------. The piecewise linear component is parameterized in terms of a sum of hattype wavelet basis functions.12 User’s Guide where MB(t) is the mass of drug at point B at time t and f(t) is the rate of entry in mass per unit time into point A at time t.a. The function representation WinNonlin models the input function as a piecewise linear “precursor” function fp(t) convolved with an exponential “dispersion” function fd(t). of drug at B is WinNonlin c(t) = ∫0 f ( t – u )cδ ( u ) du ≡ f ( t )*cδ ( t ) cδ ( t ) = g ( t ) ⁄ Vs t where “*” is used to denote the convolution operation and The above equation is the key convolution equation that forms the basis for the evaluation of the drug input rate. hj(t): fp ( t ) = ∑ xj hj ( t ) xj ≥ 0 t – Tj h j ( t ) = --------------------. Let Vs denote the volume of the sampling space (point B). Then the concentration. characteristic response or disposition function).. Tj + 1 – Tj 264 Tj < t < Tj + 1 Tj + 1 < t < Tj + 2 . The process of determining the input function f(t) is called deconvolution because it is required to deconvolve the convolution integral in order to extract the input function that is embedded in the convolution integral. f(t). The unit impulse response function (cδ ) provides the exact linkage between drug level response c(t) and the input rate function f(t). The former provides substantial flexibility whereas the latter provides smoothness. c(t).

The WinNonlin Toolbox Numeric deconvolution 12 hj ( t ) = 0 . f(t). Furthermore. The extra support point is well suited to accommodate an initial “burst” release commonly encountered for many formulations. according to the linear disposition assumption. 265 . The dispersion function fd(t) is defined as an exponential function: fd ( t ) = e ----------δ where δ is denoted the dispersion or smoothing parameter. such as stomach emptying. resulting from the above input function is. absorption window. The input function. The drug level profile. with the exception of the very first support point that is used to define the lag-time. pH changes. The wavelet support points (Tj) are constrained to coincide with the observation times. given by: c ( t ) = f p ( t )*f d ( t )*c d ( t ) where “*” is used to denote the convolution operation. Having just one support point prior to the first observation would limit this capacity. etc. which explains the scaling with the δ parameter. The hat-type wavelet representation enables discontinuous. one support point is injected halfway between the lag-time and the first observation. This point is simply included to create enough capacity for drug input prior to the first sampling. The dispersion function provides the smoothing of the input function that is expected from the stochastic transport principles governing the transport of the drug molecules from the site of release to subsequent entry to and mixing in the general systemic circulation. if any. finite duration drug releases to be considered together with other factors that result in discontinuous delivery. for the input function. is the convolution of the precursor function and the dispersion function: –t ⁄ δ f ( t ) = f p ( t )*f d ( t ) ≡ ∫ f p ( t – u )f d ( u ) du 0 t The general convolution form of the input function above is consistent with stochastic as well as deterministic transport principles. otherwise where Tj are the wavelet support points. The dispersion function is normalized to have a total integral (t = 0 to ∞ ) equal one. c(t).

The three steps are repeated until the objective function is optimized. including WinNonlin’s approach. Due to the stochastic transport principles involved. integral of squared second derivative). that is smoother than the precursor function. the drug level at time t is made up of a mixture of drug molecules that started their journey to the sampling site at different times and took different lengths of time to arrive there. The WinNonlin deconvolution method instead introduces the regularization directly into the input function through a convolution operation with the dispersion function. It is exactly this mixing in the convolution operation that provides the smoothing. The finer details (higher frequencies) are attenuated relative to the more slowly varying components (low frequency components). DTC methods differ basically in the way the input function is specified and the way the objective function is defined. the convolution of the precursor function with the dispersion function results in an input function. Consider. First. In essence. Thus. The purely residual-based DTC methods ignore any behavior of the calculated drug level response between the observations. The DTC method is an iterative procedure consisting of three steps. Third. fd(t). i. The more modern DTC methods. the convolution operation that leads to the drug level response c(t). consider both the residuals and other properties such as smoothness of the total fitted curve in the definition of the objective function. The convolution operation acts essentially as a low pass filter with respect to the filtering of the input information. The deconvolution method implemented in WinNonlin is novel in the way the regularization (smoothing) is implemented. f(t). Regularization methods in some other deconvolution methods are done through a penalty function approach that involves a measurement of the smoothness of the predicted drug level curve (e. the agreement between the observed data and the calculated drug level data is quantitatively evaluated according to some objective function. the drug level response at time t depends on a mixture of prior input. Second. The objective function may be based solely on weighted or unweighted residual values (observed–calculated drug levels). Thus. for example.12 User’s Guide Deconvolution through convolution methodology WinNonlin deconvolution uses the basic principle of deconvolution through convolution (DTC) to determine the input function.e. a smoothing due to the mixing operation inherent in the convolution operation. 266 WinNonlin . the new input function is convolved with cδ (t) to produce a calculated drug level response. the input function is adjusted by changing its parameter values. a convolution operation acts like a “washout of details”. WinNonlin allows the user to control the smoothing through the use of the smoothing parameter δ.g. Decreasing the value of the dispersion function parameter δ results in a decreasing degree of smoothing of the input function.

The WinNonlin Toolbox Numeric deconvolution 12

Similarly, larger values of δ provide more smoothing. As δ approaches 0, the dispersion function becomes equal to the so-called Dirac delta “function”, resulting in no change in the precursor function. The smoothing of the input function, f(t), provided by the dispersion function, fd(t), is carried forward to the drug level response in the subsequent convolution operation with the unit impulse response function (cδ ). Smoothing and fitting flexibility are inversely related. Too little smoothing (too small a δ value) will result in too much fitting flexibility that results in a “fitting to the error in the data.” In the most extreme case, the result is an exact fitting to the data. Conversely, too much smoothing (too large a δ value) results in too little flexibility so that the calculated response curve becomes too “stiff” or “stretched out” to follow the underlying true drug level response.

Cross validation principles

WinNonlin's deconvolution function determines the optimal smoothing without actually measuring the degree of smoothing. The degree of smoothing is not quantified but is controlled through the dispersion function to optimize the “consistency” between the data and the estimated drug level curve. Consistency is defined here according to the cross validation principles. Let rj denote the differences between the predicted and observed concentration at the j-th observation time when that observation is excluded from the data set. The optimal cross validation principle applied is defined as the condition that leads to the minimal predicted residual sum of squares (PRESS) value, where PRESS is defined as:

m

PRESS =

∑ rj

j=1

2

For a given value of the smoothing parameter δ, PRESS is a quadratic function of the wavelet scaling parameters x. Thus, with the non-negativity constraint x > 0, the minimization of PRESS for a given δ value is a quadratic programming problem that has a unique solution that can be determined numerically. Let PRESS (δ) denote such a solution. The optimal smoothing is then determined by finding the value of the smoothing parameter δ that minimizes PRESS (δ). This is a one variable optimization problem with an embedded quadratic-programming problem.

267

12 User’s Guide

User control over execution

WinNonlin

WinNonlin's deconvolution function permits the user to override the automatic smoothing by manual setting of the smoothing parameter. The user may also specify that no smoothing should be performed. In this case the input rate function consists of the precursor function alone. Besides controlling the degree of smoothing, the user may also influence the initial behavior of the estimated input time course. In particular the user may choose to constrain the initial input rate to 0 (f(0) = 0) and/or constrain the initial change in the input rate to 0 (f'(0) = 0). By default WinNonlin does not constrain either initial condition. Leaving f(0) unconstrained permits better characterization of formulations with rapid initial “burst release,” e.g., extended release dosage forms with an immediate release shell. This is done by optionally introducing a bolus component (or “integral boundary condition”) for the precursor function so the input function becomes: f(t) = (fp(t) + xδ(t)) * fd(t) = fp(t) * fd(t) + xδ fd(t) where the fp(t) * fd(t) is defined as before. The difference here is the superposition of the extra term xδ fd(t) that represents a particularly smooth component of the input. The magnitude of this component is determined by the scaling parameter xδ , which is determined in the same way as the other wavelet scaling parameters previously described. The estimation procedure is not constrained with respect to the relative magnitude of the two terms of the composite input function given above. Accordingly, the input can “collapse” to the “bolus component” xδ fd(t) and thus accommodate the simple first order input commonly experienced when dealing with drug solutions or rapid release dosage forms. A drug suspension in which a significant fraction of the drug may exist in solution should be well described by the composite input function option given above. The same may be the case for dual release formulations designed to rapidly release a portion of the drug initially and then release the remaining drug in a prolonged fashion. The prolonged release component will probably be more erratic and would be described better by the more flexible wavelet-based component fp(t) * fd(t) of the above dual input function. Constraining the initial rate of change to 0 (f'(0) = 0) introduces an initial lag in the increase of the input rate that is more continuous in behavior than the usual abrupt lag time. This constraint is obtained by constraining the initial value of the precursor function to 0 (fp(0) = 0). When such a constrained pre268

The WinNonlin Toolbox Numeric deconvolution 12

cursor is convolved with the dispersion function, the resulting input function has the desired constraint.

The extent of drug input

The extent of drug input in the Deconvolution module is presented in two ways. First, the amount of drug input is calculated as:

t

Amount input =

∫ f ( t ) dt

0

Second, the extent of drug input is given in terms of fraction input. If test doses are not supplied, the fraction is defined in a non-extrapolated way as the fraction of drug input at time t relative to the amount input at the last sample time, tend:

t t end

Fraction input =

∫ f ( t ) dt ⁄ ∫ f ( t ) dt

0 0

The above fraction input will by definition have a value of 1 at the last observation time, tend. This value should not be confused with the fraction input relative to either the dose or the total amount absorbed from a reference dosage form. If dosing is entered, as detailed under “Deconvolution dosing” on page 272, the fraction input is relative to the dose amount. Using deconvolution Numeric deconvolution is used to evaluate in vivo drug release and delivery when data from a known drug input are available. WinNonlin’s Deconvolution tool can estimate the cumulative amount and fraction absorbed over time for individual subjects, given PK profile data and dose per subject.

To use numeric deconvolution:

1. Obtain the unit impulse response function (UIR). For example, fit time and concentration data in WinNonlin using PK model 1, 8, or 18 (for N=1, 2, or 3, respectively), and adjust to reflect a single unit dose. The UIR must take the form:

269

12 User’s Guide

WinNonlin

N

cδ ( t ) =

∑ Aj e

j=1

–αi t

Since the UIR assumed the response from 1 dose unit, for model 1 (one compartment), take V (volume) from the 1-compartment IV model output and let A = 1/V. Make units corrections if needed (see below). For model 8 (two compartment), take the A and B from the model 8 output and divide by the Stripping Dose to get A1 and A2 for the UIR. Again, perform the units conversion if needed. Use model 18 (three compartment) similarly model 8. If the dose units differ from the concentration mass units in the deconvolution input dataset, take care in entering the A values: A values must use the units of the concentration data divided by the units of the dose used to determine the UIR. (See the example immediately below.) Alternately, use the Column Properties dialog to convert the concentration data to use the same mass units as the dose, or convert the dose amounts to use the concentration mass units. For example, suppose the oral concentrations use ng/mL, and a 1 mg dose was used in determining the unit impulse response. In the one-compartment case, A = 1/V, so the units for 'A’ in the model 1 output equal the inverse of the volume units. ‘A’ values must be converted to ng/mL/mg before they are used in the Deconvolution tool, i.e., A = 10^6/V. 2. Once the UIR is established and units adjustments are complete, open a data set containing individual PK profiles in a WinNonlin workbook. 3. Choose Tools>IVIVC>Deconvolution (IVIVC license) or Tools>Deconvolution (no IVIVC license) from the WinNonlin menus. 4. Define the variable mappings, dosing and model settings as detailed under “Deconvolution variables” below, “Deconvolution dosing” on page 272 and “Deconvolution settings” on page 273. 5. Click Calculate. Deconvolution generates a new chart and workbook, discussed under “Deconvolution output” below.

270

The WinNonlin Toolbox Numeric deconvolution 12

Deconvolution variables Numeric deconvolution requires a data set containing time-concentration data, with sort variables identifying individual profiles.

To select variables for numeric deconvolution:

1. Open the Variables tab of the Deconvolution dialog (Tools>IVIVC>Deconvolution (IVIVC license) or Tools>Deconvolution (no IVIVC license)).

2. Dataset: If needed, select the data set containing individual PK profiles. 3. Sort variables: If one or more variables contain values that differentiate individual data profiles, drag those columns to the Sort Variables field. 4. Time: Drag the variable representing time (or other independent variable) to this field. 5. Concentration: Drag the dependent variable to this field. 6. Use the Save Settings and/or Load Settings buttons to save or load all settings in the Deconvolution dialog via an external file. 7. Proceed to “Deconvolution dosing” below.

271

12 User’s Guide Deconvolution dosing Numeric deconvolution supports input of test dose information directly to the Deconvolution dialog, so that the fraction input can be computed relative to the administered dose. Dose information is optional; without it, deconvolution will run correctly, with the fraction input approaching a final value of 1. Dose time is assumed to be zero for all sort levels. If a dose value is entered for any sort level, values are required for all sort levels (all or none).

WinNonlin

To define dosing for numeric deconvolution:

1. Open the Dosing tab of the Deconvolution dialog (Tools>IVIVC>Deconvolution (IVIVC license) or Tools>Deconvolution (no IVIVC license)). 2. Dose units: (Optional) Enter units for all doses here. The same units must apply to the UIR and the test (oral) doses entered in this dialog. Dose units are applied only if the Time and Concentration columns in the data set have units. See Table 3-2 and Table 3-3 on page 54 for accepted units abbreviations. 3. Dose: Enter the dose amount for each sort level in the input data. 4. Use the Save Settings and/or Load Settings buttons to save or load all settings in the Deconvolution dialog via an external file. 5. Proceed to “Deconvolution settings” below.

272

The WinNonlin Toolbox Numeric deconvolution 12

**Deconvolution settings Numeric deconvolution in WinNonlin assumes a unit impulse response function (UIR) of the form:
**

N

cδ ( t ) =

∑ Aj e

j=1

–αi t

The analysis can use the observed times from the input data set, or in vivo times from an entirely separate workbook.

To define model settings for numeric deconvolution:

1. Open the Settings tab of the Deconvolution dialog (Tools>IVIVC>Deconvolution (IVIVC license) or Tools>Deconvolution (no IVIVC license)). 2. Number of exponential terms in the unit impulse response: Select the number of exponential terms (N in the equation above, with 1 < N < 9). 3. A and alpha: For each profile, enter (or copy/paste) values for each exponential term in the unit impulse response function, e.g., values obtained by fitting WinNonlin PK model 1, 8, or 18 (for N=1, 2, or 3, respectively).

273

12 User’s Guide

CAUTION: The A and alpha values must use the units shown in the A or alpha column

WinNonlin

headers (if any), to match the units used in the input data and dosing. In particular, A values must match the units of the concentration data. This may require conversion of A values if dose and concentration units differ. See “Using deconvolution” on page 269 for further discussion. 4. Smoothing: Select one of the following choices. • • • Automatic allows WinNonlin to find the optimal value for the dispersion parameter delta. None disables smoothing that comes from the dispersion function. If selected, the input function is the piecewise linear precursor function. User-Specified allows manual entry of the value of the dispersion parameter delta as the Smoothing Parameter, where delta > 0. Increasing delta increases the amount of smoothing.

5. Initial Rate is Zero: Check this option to constrain the estimated input rate to be zero at the initial time (lag time). 6. Initial Change in Rate is Zero: Check this option to constrain the derivative of the estimated input rate to be zero at the initial time (lag time) (requires initial rate of zero for automatic or user-specified smoothing). 7. Output data points: at the lower left area of the dialog, select the number and times for output data points by selecting from among the following options. a. Output from 0 to last time point: WinNonlin will generate output at even intervals from time 0 to the time of the final input data point for each profile, inclusive. Enter the total number of data points (<=1001) below. b. Output from X to Y: WinNonlin will generate output at even intervals from time X to time Y, inclusive. Enter the total number of data points (<=1001) below. c. Use observed times from input workbook: WinNonlin will generate output at each time point in the input data set for each profile. d. Use times from a workbook column: Select the workbook and time column. The worksheet containing the time values must be open in a WinNonlin workbook window, and can contain no more than 1001 times. 8. Click Calculate. Deconvolution generates a new chart and workbook, discussed under “Deconvolution output” below.

274

The WinNonlin Toolbox Numeric deconvolution 12

Deconvolution output WinNonlin numeric deconvolution generated workbook and chart output. The workbook output includes one workbook with the following six worksheets.

Table 12-2. Deconvolution workbook output Worksheet Values Content Time, input rate, cumulative amount (Cumul_Amt, using the dose units) and fraction input (Cumul_Amt / test dose or, if no test doses are given, then fraction input approaches 1) for each profile The smoothing parameter delta and absorption lag time for each profile Values for the UIR exponential terms for each profile

Parameters Exponential Terms Settings History Dosing

Fitted Values Predicted data for each profile Input settings for smoothing, number of output points and start and end times for output History of operations on the workbook. See “History worksheet” on page 37. If user-specified test doses were entered in the Deconvolution dosing dialog, a Dosing worksheet is included in the output, listing the doses and dose units.

Deconvolution generates one chart window with three charts for each profile: • • • Fitted Curve: observed time-concentration data vs. the predicted curve Input Rate: rate of drug input vs. time Cumulative Input: cumulative drug input vs. time

275

12 User’s Guide

WinNonlin

276

Chapter 13

**The IVIVC Toolkit
**

In vivo-in vitro correlation and validation tools for formulation development

Users who have a WinNonlin In Vivo-In Vitro Correlation IVIVC Toolkit License can access additional tools for evaluating correlations between in vitro drug properties, such as dissolution, and in vivo responses, such as fraction absorbed, in order to predict PK responses such as maximum concentration or area under the curve. WinNonlin IVIVC features are detailed under the following headings: • “Numeric deconvolution” on page 260: Users with or without an IVIVC Toolkit License can use deconvolution to evaluate drug absorption from time-concentration data, based on the unit impulse response. “Wagner-Nelson and Loo-Riegelman methods” on page 278: Estimate drug absorption assuming a one-compartment (Wagner-Nelson) or twocompartment (Loo-Riegelman) PK model. “Convolution” on page 293: Estimate or verify PK profiles based on the unit impulse response function and expected absorption profile. “Levy plots” on page 300: Comparing predicted and observed concentrations over time as a measure of correlation model fit. “Sigmoidal/dissolution models” on page 547: Sigmoidal models, including Hill, Weibull, Double Weibull and Makoid-Banakar, for modeling dissolution. “The IVIVC Wizard” on page 302: Structured environment for building, executing, saving and reloading common IVIVC work flows.

•

• • •

•

See the WinNonlin Examples Guide for examples demonstrating common IVIVC inquiries.

277

13 User’s Guide

WinNonlin

**Wagner-Nelson and Loo-Riegelman methods
**

Users with a WinNonlin IVIVC Toolkit License can access Wagner-Nelson and Loo-Riegelman methods to estimate fraction of drug absorbed. Wagner-Nelson inputs and calculations The Wagner-Nelson method estimates the fraction of drug absorbed over time, relative to the total amount to be absorbed, following the method described in Gibaldi and Perrier (1975) pages 130 to 133. It uses as a basis AUC values computed for each time point in your time-concentration data. Note: WinNonlin Wagner-Nelson computations assume single-dose PK data with a concentration value of 0 at dose time. If no concentration value exists at dose time, WinNonlin will insert a concentration value of 0. You can choose to calculate AUC using either the linear, linear/log, or linear up/log down trapezoidal rule. (See “Calculation methods for AUC” on page 177 and “AUC calculation and interpolation formulas” on page 183 for definitions and calculation methods.) Note that the two linear non-compartmental analysis (NCA) methods are equivalent, as no interpolation is used in the Wagner-Nelson method.) Similarly, you choose the method for computing Lambda Z: best fit, userspecified range, or user-specified value. If you specify a Lambda Z value, you may also enter an intercept or have WinNonlin compute it using the last positive concentration and associated time value: intercept = Lambda Z * tlast + ln(Clast) If you do not enter a value for AUCINF, it will be computed as for WinNonlin NCA: AUCINF = AUClast + Clast / Lambda Z You choose whether to use the observed or predicted value for Clast, where Clast_pred = exp(intercept - Lambda Z(tlast).) The cumulative amount absorbed at time t, normalized by the central compartment volume V1, is computed as: Cumul_Amt_Abs_V (t) = C(t) + Lambda Z * AUC(t)

278

The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13

Therefore, Cumul_Amt_Abs_V at t = infinity is: Cumul_Amt_Abs_V(inf) = Lambda Z * AUCINF The relative fraction absorbed at time t is then: Rel_Fraction_Abs(t) = Cumul_Amt_Abs_V(t) / Cumul_Amt_Abs_V(inf) The percent remaining at time t is computed as: Percent remaining (t) = 100 * (1 - Rel_Fraction_Abs(t) ) Data points with a missing value for either time or the concentration will be excluded from the analysis and will not appear in the output. Loo-Riegelman inputs and calculations Like Wagner-Nelson, the Loo-Riegelman method estimates the fraction of drug absorbed as a function of time, relative to the total amount to be absorbed, following the method described in Gibaldi and Perrier (1975) pages 136 to 142. It computes AUC values for each time point in your time-concentration data set, as in Wagner-Nelson. For Loo-Riegelman, you must provide estimates for K10, K12, and K21, as may be obtained by running WinNonlin PK model 7 on separate IV data. Note: WinNonlin Loo-Riegelman computations assume single-dose PK data with a concentration value of 0 at dose time. If no concentration value exists at dose time, WinNonlin will insert a concentration value of 0. You may supply a value for AUCINF or allow WinNonlin to compute it as for non-compartmental analysis: AUCINF = AUClast + Clast / Lambda Z You can choose to use the observed or predicted value for Clast, where Clast_pred = exp(intercept - Lambda Z(tlast). As in the Wagner-Nelson method, you choose the method for computing Lambda Z: best fit, user-specified range, or user-specified value. You may specify an intercept for the last case. If the intercept is not specified, WinNonlin will compute it using the last positive concentration and associated time value: intercept = Lambda Z * tlast + ln(Clast)

279

13 User’s Guide Note that the Loo-Riegelman method uses Lambda Z only to compute AUCINF; it uses the intercept only to compute Clast_pred. The cumulative amount absorbed at time t, normalized by the central compartment volume V1, is: Cumul_Amt_Abs_V(t) = Conc(t) + K10 * AUC(t) + Amt_Peripheral_V(t) The amount in the peripheral compartment at time t, normalized by V1 is computed sequentially as: Amt_Peripheral_V(t) = (Prior_Amt_Peripheral_V) * exp(-K21*δt) + ( K12 * Prior_Conc / K21) * (1 - exp(-K21*δt) + ( K12 * δt * δConc / 2 ) See Gibaldi and Perrier (1975) pages 138-139 for the derivation of this equation. Cumul_Amt_Abs_V at time infinity is: Cumul_Amt_Abs_V(inf) = K10 * AUCINF The fraction absorbed at time t is computed as: Rel_Fraction_Abs(t) = Cumul_Amt_Abs_V(t) / Cumul_Amt_Abs_V(inf) The percent remaining at time t is computed as: Percent remaining (t) = 100 * (1 - Rel_Fraction_Abs(t) ) Data points with a missing value for either time or the concentration will be excluded from the analysis and will not appear in the output. Using the Wagner-Nelson or Loo-Riegelman method

To use the Wagner-Nelson or Loo-Riegelman method:

WinNonlin

1. Open the data set containing individual PK profiles in a WinNonlin workbook. 2. For deconvolution using a one-compartment PK model: choose Tools>IVIVC Tools>Wagner-Nelson from the WinNonlin menus.

280

Define the variable mappings. 6. model properties and options as detailed under the following headings: • • • • • • • • “Data variables” on page 281 “Dosing regimen” on page 282 “Loo-Riegelman model parameters” on page 283 (Loo-Riegelman deconvolution only) “Lambda Z ranges” on page 284 “Output options” on page 287 “Weighting” on page 288 “Title” on page 290 “Settings” on page 291 5.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 3. representing the newly-loaded model. 3. loaded and refreshed as for any WinNonlin model. 281 . Data variables To define the data variables to use for the Wagner-Nelson or Loo-Riegelman method: 1. Run the analysis by choosing Model>Start Modeling from the WinNonlin menus. 4. 4. 2. Y Variable: Drag the dependent variable (concentration) to this field. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. X Variable: Drag the independent variable (generally. For deconvolution using a two-compartment PK model: choose Tools>IVIVC Tools>Loo-Riegelman from the WinNonlin menus. A model (text) window will appear. time) to this field. Choose Model>Data Variables from the WinNonlin menus. The model and settings may be saved. using the options on the File menu. 5. Sort Variables: Drag the variables needed to identify individual profiles to the Sort Variables list.

2. 4. Note: If no concentration value exists at dose time for a given profile. Value: For each profile. Use the Save button to save the dosing information for later re-use.13 User’s Guide 6. Click OK and proceed to set the Dosing regimen. WinNonlin will insert a concentration value of 0 at the specified dose time. WinNonlin 7. skip to “Loo-Riegelman model parameters” on page 283 for LooRiegelman or “Lambda Z ranges” on page 284 for Wagner-Nelson. To define the dosing: 1. enter the dose time. or use the Load button to load previously-saved dosing information from a file. Choose Model>Dosing Regimen from the WinNonlin menus. Carry Alongs: Drag variables to be included in the output workbook (but not used in calculations) to this list. and set the data variables (Model>Data Variables). the dosing should also have been loaded. 3. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. In that case. Dosing regimen If the data were loaded from PKS. 282 .

Loo-Riegelman model parameters The Loo-Riegelman method requires initial estimates for the model parameters describing the two-compartment model that fits the IV data. For Wagner-Nelson. 2. Value: For each profile. Choose Model>Model Parameters from the WinNonlin menus. To set up initial model parameters for the Loo-Riegelman method: 1.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 5. 6. and set the data variables (Model>Data Variables). skip to the Lambda Z ranges. Click OK. 3. Alternately. proceed to set the Loo-Riegelman model parameters. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. Parameter estimates may be obtained by running WinNonlin PK model 7 on the data prior to using the Loo-Riegelman tool. For Loo-Riegelman models. enter initial parameter estimates for each parameter. use the buttons at the bottom of the dialog to load the parameter 283 .

WinNonlin 4. Choose whether to enter Lambda Z information using a plot of each profile or by entering tabular data. and set the data variables (Model>Data Variables). 4. Choose Model>Lambda z Ranges from the WinNonlin menus. using the Graph and Data buttons at the top right. Click OK and proceed to set the Lambda Z ranges. Use the arrow buttons directly below the plot to move between profiles in the Plot view. Lambda Z ranges To set Lambda Z ranges for the Wagner-Nelson or Loo-Riegelman method: 1. 284 . 2. select Lambda Z and AUC settings as follows. 3.13 User’s Guide values from a PK output workbook obtained by running WinNonlin PK model 7 on the data prior to this analysis. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. For each profile.

6.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 5. If AUCINF is a user-specified value. then the other Lambda Z settings are unnecessary for the Loo-Riegelman method. the Lambda Z options apply only for the AUCINF computation. Note: If observed data does not extend significantly into the terminal phase (i. Use AUCINF_obs: WinNonlin will calculate the area under the curve as: AUCINF = AUClast + Clast / Lambda Z. Note that for the Loo-Riegelman method. absorption phase must be completed) then use of calculated AUCINF may significantly underestimate the actual AUCINF. User specified AUCINF: Enter a value for the area under the curve. where Clast is the last observed concentration. resulting in scaling errors in the computation of fraction absorbed. where Clast is the last predicted concentration. Lambda Z Calculation Method: Select one of the following options: 285 .e. Area Under the Curve: Select one of the following options for calculation of area under the curve to time = infinity: Use AUCINF_pred: WinNonlin will calculate the area under the curve as: AUCINF = AUClast + Clast / Lambda Z.

the excluded data points will be included in computation of AUC’s. WinNonlin will compute it as: intercept = Lambda Z * tlast + ln(Clast). User Specified Lz and Optional Intercept: (Available for AUC calculated as AUC_pred only. Since no specific points will be used in the computation of Lambda Z. Exclusions (optional): To exclude a data point from Lambda Z calculations. WinNonlin will determine the Lambda Z range as detailed under “Lambda Z or slope range selection” on page 162. WinNonlin 7. for the intercept. for multiple exclusions) for that profile in the Data view. the Lambda Z table in the output will give Predicted and Residual values for all Time and Conc values. 286 . Proceed to set the Output options. These exclusions apply only to Lambda Z calculations. c. If using the Plot view. Select the last data point to be included: Hold down the Shift key while left-clicking on a data point or enter a time value under End Time. hold down the Ctrl key and left-click on that point in the Plot view or enter the time value under Exclusions (or in the next column to the right. Select the first data point to be included: Left-click on a data point or enter a time value under Start Time in the Data view. If no intercept value is entered. Repeat to reverse an exclusion. When done.) Enter a value for Lambda Z and. optionally. d. click OK.13 User’s Guide Best fit: If this option is selected in the Plot view. To save the settings and close the dialog. User Time Range: Define the terminal elimination phase as follows: a. 8. Lambda_z. click Apply to save the settings and move to another tab of the Model Properties dialog. b. End Time and Lambda Z are empty in the Data view. or Start Time. Note: Any AUC. Intercept and K values entered must be greater than 0. select the axis scale for the graphical representation below the lower left corner of the plot: Linear or Log Linear.

Check the options to apply from among the following choices. These worksheets present the same information as the text output. Choose Model>Model Options from the WinNonlin menus and open the Output Options tab. 2. Note: Set default output options via Tools>Options in the WinNonlin menus. one or more new windows is generated. The workbook contains the worksheets summarized in Table 13-1. and set the data variables (Model>Data Variables). but in tabular form. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. Workbook: Check this option to generate an output workbook. with content as selected in the following options. After the analysis is run. 3.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 Output options To define the output options for the Wagner-Nelson or Loo-Riegelman method: 1. 287 .

If Lambda Z is user-specified.13 User’s Guide . containing the same information as the workbook output. and relative fraction absorbed. Weighting The weighting options for the Loo-Riegelman and Wagner-Nelson methods apply to the calculation of Lambda Z (see “Lambda Z ranges” on page 284). including time. and a plot of the Lambda Z fit (excluded for Loo-Riegelman with user-specified AUCINF). and set the data variables (Model>Data Variables). Details for fitting Lambda Z. Text: This option provides an ASCII text version of the analysis results. Loo-Riegelman method only. Profile estimates for each level of the sort variable. Summary of output worksheets Sheet name Wagner-Nelson Contents Wagner-Nelson method only. Loo-Riegelman Lambda Z Fit User Settings History Chart: If workbook output is selected. check this item to generate a Chart window with two plots per profile: a linear plot of Relative_Fraction_Absorbed vs. the latter plot is excluded. Profile estimates for each level of the sort variable. WinNonlin Table 13-1. Units are those of the input data. cumulative AUC and cumulative amount and relative fraction absorbed. Page Breaks: Check this box to include page breaks in the ASCII text output. Modeling errors are also reported in this file. concentration. Not included for Loo-Riegelman with userspecified AUCINF. concentration and cumulative AUC. including time. Click OK and proceed to set the Weighting. “History worksheet” on page 37. To define the weighting scheme to use in fitting Lambda Z: 1. 288 . User defined settings History of actions done in and to the workbook. Units are those of the input data. amount in the peripheral compartment. amount absorbed. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. 4. Time.

rather it is the relative proportions of the weights. 289 . concentrations between 0 and 1 would have negative weights. Select Uniform Weighting from the pull-down list. 3. rather than 1/Y. To use uniform weighting: 1. To use weighted least squares: 1. However. See “Weighting” on page 241 for discussion of weighting schemes. Note that it is not the weights themselves that are important. it already uses weighting implicitly.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 2. Choose Model>Model Options from the WinNonlin menus and open the Weight tab. Select Observed to the Power n button and enter the value of n. and therefore could not be included. Select the desired weighting scheme. Common schemes are detailed below. 2. CAUTION: When a log-linear fit is done. if this were the case. one could argue that the weights should be 1/LogY. Select the Pre-selected Scheme radio button. 2. approxi- mately equal to 1/Yhat2. or use the Pre-selected scheme button to select the most common weighted least square options: 1/Y and 1/Y2. When weighting by 1/Y and using the Linear/Log trapezoidal rule.

4. To define and include the title: 1. on the User Settings worksheet in the workbook output. 3. 2. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. The title will be centered on the page. Click OK and proceed to set the Title. and at the top of each chart in chart output. Enter the title text below. Choose Model>Model Options from the WinNonlin menus and open the Title tab. Make sure that Title Text is checked. the title is displayed at the top of each printed page in ASCII text output. Title If included. 290 . the title will not display even if text is entered below. To include a line break use Shift + Enter. and set the data variables (Model>Data Variables). If it is not.13 User’s Guide WinNonlin 3. It may include up to 5 lines of text.

Settings To set the AUC calculation method for Wagner-Nelson or Loo-Riegelman: 1.The IVIVC Toolkit Wagner-Nelson and Loo-Riegelman methods 13 5. Load the data and model as detailed under “Using the Wagner-Nelson or LooRiegelman method” on page 280. 291 . Click OK and proceed to the IVIVC Settings tab. and set the data variables (Model>Data Variables). Choose Model>Model Options from the WinNonlin menus and open the Settings tab. 2.

This method (method 2) uses the linear trapezoidal rule. Linear Trapezoidal rule (Linear Interpolation). rate or effect) is less than or equal to zero. Linear Trapezoidal Method with Linear/Log Interpolation (method 4). Linear Up/Log Down Method (method 3). the Linear Up/Log Down. Calculation Method: Select the method for computing area under the curve. The linear trapezoidal rule is used any time that the concentration data is increasing. All methods reduce to the log trapezoidal rule and/or linear trapezoidal rule. The Linear/Log Trapezoidal. since no interpolation is done. Uses the log trapezoidal rule (given under “AUC calculation and interpolation formulas” on page 183) after Cmax. and the logarithmic trapezoidal rule is used any time that the concentration data is decreasing.13 User’s Guide 3. 292 . and linear trapezoidal otherwise. and the Linear Trapezoidal (with Linear/Log Interpolation) methods all apply the same exceptions in area calculation and interpolation: If a Y value (concentration. since interpolation is not used in Loo-Riegelman and Wagner-Nelson methods. or after C0 for bolus (if C0 > Cmax). if adjacent Y values are equal to each other. Note: 4. given under “AUC calculation and interpolation formulas” on page 183. Four methods are available. the program defaults to the linear trapezoidal or interpolation rule. WinNonlin • • • Note: Methods 1 and 4 are the same for Wagner-Nelson and Loo-Riegelman. The method chosen applies to all AUCINF computations. and sums up these areas. the program defaults to the linear trapezoidal or interpolation rule for that point. Click OK. applied to each pair of consecutive points in the data set that have non-missing values. If Cmax is not unique. • Linear/Log Trapezoidal Method (method 1). but differ with regard to when these rules are applied. Similarly. then the first maximum is used. Equivalent to Linear Trapezoidal rule.

and use the derivatives in the convolution.e... then the derivatives are piecewise constant. To specify a polyexponential. 2. for input rate data. or the derivative of this spline function.g.g. otherwise where N is restricted to be 1 or 2. you must supply the polyexponential coefficients. e. in any combination. i. you must supply a dataset to be fit with the splines. 293 . You can choose to convolve either this spline function. m = 0. the Convolution tool will find the spline with the requested degree of polynomial splines. t j + 1 ). e. For the first case. tε ( t j... if cumulative input data is fit with linear splines. the A's and alpha's. they must take the form: N i–1 f(t) = ∑ bij ( t – tj ) i=1 . Polyexponentials must take the form: N f(t) = ∑ Ai e –( αi ⋅ t ) i=1 where N <= 9 and t >= 0.e. j = 1.The IVIVC Toolkit Convolution 13 Convolution The WinNonlin Convolution tool supports analytic evaluation of the convolution integral. …. F ( t ) ⋅ g )(t) = ∫0 f ( t – u )g ( u ) du t where f(t) and g(t) are constrained to be either polyexponentials or piecewise polynomials (splines). Splines must be piecewise constant or linear splines. so the piecewise constant function is convolved with the user's other specified function. the Convolution tool will fit the data with the requested degree of polynomial splines. For the derivative of a spline. For a spline function. differentiate the splines to obtain polynomial splines of one degree lower. i. and will use these polynomial splines in the convolution. for cumulative input data (so the derivative is input rate data). For example.

Open the data set(s) to be convolved in a WinNonlin workbook. For t ε (ti.13 User’s Guide Linear splines are the lines that connect the (x.ti). This option is not available for cumulative input data since the derivative will be zero. Choose Tools>IVIVC Tools>Convolution from the WinNonlin menus. and set the degree of polynomial splines (Linear or Constant). Select Use derivative of the fitted function to fit the data with the derivative of the piecewise polynomial (first of two polynomials only). y = yi + mi (t . Select Use this fitted function to fit the data with the piecewise polynomial. 2. • • 294 . where mi is the slope (yi+1 . ti+1). as for cumulative input data.9 sets of coefficient (“A’s”) and exponent (“alphas”) values per subject.y) points. none of the alpha values for the first polyexponential can be the same as any of the alpha values for the second polyexponential. The first dialog defines the functions f(t) and g(t) described above. For input rate data. you may select a dataset containing 1 . as for input rate data. each piecewise constant spline will have the value that is the average of the y-values of the endpoints. Note that 'linear splines' are required for cumulative input data. The Convolution Wizard appears.yi) / (ti+1 . See the above discussion for detail on function types and data requirements. The derivative of a linear spline function is a piecewise constant function. with the constants being the slopes mi. 3. Polyexponential: Optionally. either could represent the input function or unit impulse response function. Select the appropriate function type and settings for the Input function and Unit Impulse Response (UIR) function. Note that f(t) and g(t) are interchangeable. If two polyexponentials are to be convolved. • Splines: Convolution will find the polynomial coefficients. To use convolution: WinNonlin 1.ti). Select the workbook containing time concentration data to be fit under Required Dataset.

if you selected two polyexponentials without optional input data sets. The example below shows the setup for one polynomial and one polyexponential with an optional input data set. Data variables The options in this dialog vary.The IVIVC Toolkit Convolution 13 4. described under “Convolution” on page 293. Use the Save or Load button to save or load convolution settings to/from an external file. 295 . depending on the function types selected in the Convolution Function dialog. 2) WinNonlin will use the cross-product of any unmatched sort keys. Click Next to proceed to “Data variables” or. to proceed to “Convolution settings”. 5. Note: When both functions use an input data set: 1) the data will be joined on any sort keys that have identical names in both data sets.

they are not used in calculations. Choose the number of coefficient/exponent pairs under Number of exponential terms. See the above note about sort keys.13 User’s Guide To set up a polynomial function (splines): WinNonlin 1. Time: Drag the independent variable to this field. See the note about sort keys. 296 . Note that the data set may contain no more than one time record for each profile (unique combination of Sort variable values). To set up a polyexponential function that uses a dataset for exponentials: 1. These columns will be carried as-is to the results. 2. 4. 2. 3. Carry Alongs: Drag data columns to include in the output to this field. Sort Variables: Drag the variables needed to identify individual profiles to this field. Cumulative Input: Drag the dependent variable to this field. Sort Variables: Drag variables needed to identify individual profiles to this field. above.

Select the time option: • Use default times: (not available for two polyexponentials) Output will include time points from 0 (zero) to one of the following: – For two polynomials. 297 . respectively. twice the last time value in the data set • Output from X to Y: Output will spread the number of data points evenly across times from X to Y. as must all Alphas. in the same row. inclusive. To set the number and times of output data points (optional): 1. described under “Convolution” on page 293. The example below shows the setup for one polynomial and one polyexponential without an optional input data set. they are not used in calculations. 4. Click Next to proceed to “Convolution settings”. These columns will be carried as-is to the results. depending on the function types selected in the Convolution Function dialog. All A’s must use the same units. the last time value in one of the data sets plus the last time value in the other dataset – For one polynomial.The IVIVC Toolkit Convolution 13 3. Convolution settings The options in this dialog vary. Carry Alongs: Drag data columns to include in the output to this field. 5. Terms Variables: Drag each coefficient and exponent pair to the A and Alpha fields. Enter the total number of output data points under Number of output data points. 2.

Convolution output Convolution generates workbook and chart output. 298 . A’s and Alphas: Enter values for each coefficient and exponent pair together in the same row. Use the abbreviations detailed in Table 3-2 on page 54. Click Calculate to run the convolution. If the A’s and alphas are drawn from a workbook. Choose the number of coefficient/exponent pairs to enter under Maximum number of exponential terms. Note that alpha units must equal 1/(Time units).13 User’s Guide WinNonlin To set up a polyexponential function that did not use a dataset for exponentials: 1. their column units will apply. 2. The output is described under “Convolution output”. The chart output includes one plot of the convolved data over time for each level of the sort variables. The units can be blank (no units). The workbook contains the worksheets listed in Table 13-2. 4. 3. Units in the output are described under “Convolution units” on page 299. enter units for the coefficients (A) and exponents (alphas) to the right. Optionally.

then the convolved data has units: ted function" Time units * Input Rate units * Y units The two Time columns must use the same units (if any). If all input Splines and splines.) 299 .If the Time and Input Rate columns have units. Convolution units Functions Units in output Splines and polyexpo. If all input with option "use this fit. as follows. “History worksheet” on page 37. with option "use provided units for the A's.If the Time and Cumulative Input columns have units. then the convolved data have units: derivative of this fitted Cumulative Input units * A units function" Splines and splines. and if the alpha units are 1 over a time unit. then the convolved data have the following this fitted function" units. and you have nential. with option "use vided units for the A's.columns have units. The two Time columns must use the same units (if any).) * A units (2nd poly. Time units * Input Rate units * A units Splines and polyexpo. If alpha and A units are provided.The IVIVC Toolkit Convolution 13 Table 13-2. Table 13-3. then the convolved data has units: Time units from alpha units * A units (1st poly. all A units must also be the same. then the convolved data has units: tive of this fitted funcCumulative Input units * Y units tion" Polyexponential and polyexponential All alpha units must be the same. with option "use deriva. and you have pronential.columns have units. Note that the alpha units must equal 1/(Time units). Convolution workbook output Worksheet name Convolution Settings History Content Two columns: Time and the convolved data points for each level of the sort variables Settings and input workbook names for the convolution Record of operations on the workbook. Convolution units The units in convolution output differ with the functions used.

WinNonlin will group data by formulation within each Levy Chart. drag the variable containing sample data from the Variables list to the Values field. The plots include a linear fit of the data points with intersection at origin. 2. whichever is smaller. WinNonlin will use the cross-product of sort values to perform the plotting and modeling. 6.13 User’s Guide WinNonlin Levy plots Users with a WinNonlin IVIVC Toolkit License can create Levy plots to compare in-vivo absorption versus in-vitro dissolution data visually. if any. Each 300 . 4. For nonmatching sort variables. 3. in the corresponding Dataset fields. For sort variables that exist in both data sets (identical column name). Choose Tools>IVIVC Tools>Levy Plots.. and the correlation (R2) value and the slope for the model line. as well as a line at unity. drag the variable containing sample times from the Variables list to the Independent field. Sort Variables (optional): Drag any variables required to identify separate plots from the Variables list to the Sort Variables field for one or both data sets. fraction absorbed or dissolved). 5. must match in the two data sets. drag the variable identifying different drug formulations from the Variables list to the Formulation field.. WinNonlin will include only those sort values that exist in both data sets. one containing in-vivo and another containing in-vitro data. Independent: For each data set. Open the in-vivo and in-vitro data sets in separate WinNonlin workbooks. from the WinNonlin menus to open the Levy Plots dialog. Dataset: Select the appropriate data sets. Values: For each data set. as time-versus-time plots at matched values for absorption and dissolution. To create one or more Levy plots: 1. 7. Matched time pairs will be sampled from the data within the range from zero to the maximum observed dissolution or deconvolved absorption value. using linear interpolation to calculate data points as needed. WinNonlin uses these data sets to plot in-vivo versus in-vitro times at user-specified Y values (typically. one each for the in-vivo (vertical axis) and in-vitro (horizontal axis) profiles. The two data sets may contain the same and/or different sort variables. Levy plots require two separate workbooks. Units for these variables. Formulation (optional): For each data set.

To enter more than one title line. 10. 10. Data for formulation values that appear in only one data set will not appear in the chart or workbook output (for a given profile. use Ctrl+Enter to insert line breaks. a. The first value will appear at Y=0.. Note: When a Y value to be plotted exists in the data set. T(in-vivo)). drag any additional variables to include in the output workbook from the Variables list to the Carry Alongs field. These variables are not used in the plots. up to the end of the input data. WinNonlin will use only those formulation values that appear in both data sets.02) rather than percents (e. WinNonlin will multiply the values by 100 before plotting.The IVIVC Toolkit Levy plots 13 formulation will appear in a different color with a corresponding entry in the plot legend. Values to Match: WinNonlin will plot in-vivo versus in-vitro times corresponding to the Y values selected here. 0. If none of the formulation values match between data sets for any sort key. Plot Values at every N: Select this radio button to plot interpolated time values at regular intervals across the sample data.g. edit the chart title and axis labels in the appropriate fields at the bottom of the dialog box. Note: The summary line on the Levy plot is fit to all formulations combined. Plot at these values: Select this radio button and enter a list of Y values to plot interpolated times at those values. d. For example. 2). etc. Plot labels: If desired.g. 20. Enter the interval to the right. WinNonlin will use the data. if sort keys are used).. b. an interval value of 10 would generate plot values at Y = 0. Show datapoint labels: Check this box to display the numeric time values with each data point in the charts as (T(in-vitro). c. Datapoints are fractional: Check this box if the data are expressed as fractions (e. all other values are computed using simple linear interpolation. Carry Alongs (optional): For each data set. 8. then at the specified interval up to the end of the sample data range. The same selections apply across all sort levels. in the units of the Y variables. WinNonlin will not plot any points that would be extrapolated from one of the data sets. 9. no output will appear at all. 301 . The maximum Y value shown will be lesser of the maximum observed fraction dissolved and the calculated fraction absorbed. 30.

The workbook contains the following worksheets: Table 13-4. Predicted data based on linear regression. Output from fitting model 502. Vivo_Time. including: sort variables. plus a feature to automatically estimate the unit impulse response (UIR) for one or more profiles given an oral. validate it and use it to make predictions.13 User’s Guide 11. Any set of operations set up in the Wizard can be saved and reloaded as an IVIVC project (*. History of operations on the workbook. and Carrys Along variables (if included). click the Create Chart button to generate the Levy plots and accompanying workbook. including: sort variables (if included) and for each unique combination of sort variable values: Formulation (if included). Output The Levy Plot tool outputs one workbook and one plot window. Matched_Y. including estimates and statistics for the Yintercept (INT) and slope. including the 302 . and for each unique combination of sort variable values: X. The Wizard supports setup and running of all or a subset of operations required to create an IVIVC. See “History worksheet” on page 37. It provides workflow support for complete. 12. Use the Load Settings button to load values from such a file. WinNonlin Predicted Values Parameters History The IVIVC Wizard The IVIVC Wizard provides an organized environment in which to build and manipulate in-vitro in-vivo correlation (IVIVC) models using the underlying functionality described under “The IVIVC Toolkit” on page 277. if any. IV or IR formulation. two-stage IVIVC development. When all settings are complete. Y and Carry Along variables (if included). Use the Save Settings button to save entries in the dialog to an external file for later use. Levy workbook output Worksheet Plotted Values Content Values used in the plots. Vitro_Time. including interpolated and original data values (where interpolation wasn’t necessary).IVC). The plot window contains one Levy plot for each level of the sort variables.

one workbook for dissolution data. Close any models that are open in WinNonlin. Choose Tools>IVIVC>IVIVC Wizard. See the WinNonlin Examples Guide for examples demonstrating a full IVIVC workflow. set your monitor resolution to a minimum setting of 1024 x 768. • “In-vitro tab” on page 306 303 . Note: To maximize IVIVC Wizard operation. Individual functions are run from the separate tabs of the IVIVC Wizard as detailed in the following sections..g. with one profile per formulation.. and another with PK profiles. To use the IVIVC Wizard: 1. either individual or average. from the WinNonlin menus to launch the IVIVC Wizard. Note that in-vitro and in-vivo data must appear in separate workbooks. e.The IVIVC Toolkit The IVIVC Wizard 13 related data sets. Open the necessary data set(s) in WinNonlin workbook windows. You must have a valid WinNonlin license and a valid WinNonlin IVIVC Toolkit License installed in order to access this menu. 2. 3.

the IVIVC Wizard remains in the Windows foreground so long as it is open and WinNonlin is not minimized. Minimizing the IVIVC Wizard To avoid potential issues with other WinNonlin dialogs. The Dependencies dialog will not work for IVIVC Wizard operations and output. testing and using and IVIVC. Each item appears next to a square that can take the following colors and meanings. Project Status The right hand pane of the wizard shows the status of steps that may be completed in the process of creating. but not up to date due to changes to the data in WinNonlin or the settings in the IVIVC Wizard. minimize the main WinNonlin window using the third button from the right end of the WinNonlin title bar. 304 . Process is ready to run. Table 13-5. reloading and re-running the IVIVC project. Process is not ready to run. Use the buttons next to the items to view related data or settings. one or more required settings or data items is not available. To minimize the IVIVC Wizard (and leave it open). Project status color codes Color Green Yellow Red Meaning Process has been run successfully and is up to date.13 User’s Guide • • • • Note: “In-vivo tab” on page 307 “Correlation tab” on page 310 “Prediction tab” on page 313 “Options” on page 315 WinNonlin Output generated from within the IVIVC wizard should be refreshed by saving. Note that there is no requirement to complete all items.

to the PKS. If any windows are open in WinNonlin. Click Yes to proceed. be sure to move all files (. 4. To save an IVIVC project to your hard drive: 1. It will not be possible to load a saved . To load a saved IVIVC project: 1..ivc + workbooks. In the Save Project dialog. Note: Use PKS>Save or PKS>Save As to save an IVIVC project and. output. Set up the desired project settings in the IVIVC Wizard either manually or by loading a previously-saved project. all the WinNonlin objects associated with the IVIVC project will be saved to the same directory. Click the Save Project. Run any analyses for which you would like to save data as part of the project. The WinNonlin objects associated with the IVIVC project can be easily recognized because they have had the IVIVC project name prefix associated with their filenames. 4. If the project files must be moved to another location.ivc file. 305 . 3.. Keep the wizard open.ivc file if any of the associated files are not present in the same directory. Click the Load Project button at the top right corner of the wizard. Click Open to load the project into the IVIVC Wizard. optionally.. from the WinNonlin menus. Choose Tools>IVIVC>IVIVC Wizard. 3..The IVIVC Toolkit The IVIVC Wizard 13 Saving and loading IVIVC projects You may save any collection of settings. with or without the file extension (*. text. Locate and select the settings (*. the Wizard will warn you that those objects will be closed. Note that any open windows must be closed in WinNonlin before an IVIVC project can be loaded.IVC). navigate to an appropriate directory and enter any Windows-compatible filename. 2. button at the top right corner of the wizard. 2. Note: Once a user saves an . charts) together. input data and output in the IVIVC Wizard as an IVIVC project.IVC) file. Click Save.

dissolution and formulation identifiers. from among those open in WinNonlin. Variables: Drag column names from the Variables list to the appropriate field on the right to identify those containing time. Sheet: Select the worksheet containing the percent dissolution data. as fraction dissolved over time.13 User’s Guide In-vitro tab The IVIVC Wizard. test and reference (target) formulations. In-Vitro tab includes settings to identify your dissolution data. WinNonlin Dissolution data selection Dataset: Select the workbook containing your dissolution data. and to fit a dissolution model to smooth the data if desired. Dissolution data partitioning Drag and drop a formulation to one or more of the following to identify: 306 .

In-Vivo tab includes settings to identify your time-concentration or absorption data. you may wish to fix certain parameters to known values. Target batch (optional): Formulation to be used as the comparator in predictions. See “Model parameters” on page 197 for instructions. in order to provide smoothing you may fit the dissolution data to a sigmoidal dissolution model. via the Predictions tab. optionally based on observed Y or predicted Y (Yhat) values. For example. Click Fit Dissolution Data to run the model or linear interpolation and generate output. Weighting: Select how to weight data during modeling. to use linear interpolation on the raw data instead of modeling. 307 . In-vivo tab The IVIVC Wizard.) External validation (optional): Formulation(s) to be used in model validation.1. the asymptotic fraction dissolved. Use this tab to fit a unit impulse response (UIR) function to reference concentration data.1.) Dissolution model Optionally. then set the following. Depending on the model. Finf. (See “Prediction tab” on page 313. check Use raw data above the models.The IVIVC Toolkit The IVIVC Wizard 13 Internal validation (required): Formulation(s) to be used in building the IVIVC model(s) and for validation on the Correlation tab. (See “Correlation tab” on page 310.2. Select one of the IVIVC models described under “Sigmoidal/dissolution models” on page 547. can be fixed to 1. Alternately. and/or to deconvolve an existing UIR and concentration data in order to estimate fraction of drug absorbed over time.0 to prevent incomplete measurements from biasing other parameter estimates. formulations and dosages.3. See also “Weight” on page 205. Model parameters: 1.

The reference formulation must use all oral. concentration (Values) or fraction absorbed (Absorption). IV bolus or IV infusion input. time (Independent). Formulation information Reference formulation: Select the identifier for the formulation to be used in computing the UIR (and also in building IVIVC correlation models). The IVIVC Wizard does not support combinations of inputs for UIR calculation. IV bolus or IV infusion.13 User’s Guide WinNonlin PK data selection Dataset: Select the workbook containing your time-concentration or timeabsorption data from among those open in WinNonlin. and drug formulation identifiers (Formulation). 308 . Reference data type: Identify whether the reference input was oral. Variables: Drag column names from the Variables list to the appropriate field on the right to identify those identifying individual profiles (Sort variables). Sheet: Select the worksheet containing the data.

(See “AIC” on page 483 and “WRSS” on page 500. using the doses entered into the Wizard and the UIR’s just generated.The IVIVC Toolkit The IVIVC Wizard 13 Dose units: Optionally. check this box to generate the polyexponential describing the IV bolus PK. respectively. the Wizard can compute UIR’s for all subjects. Unit impulse response Based on the reference formulation you select. It will generate estimates for fraction absorbed at 200 time points. It will fit a polyexponential function to each profile and choose the one with the best fit based on Akaike Information Criterion or Weighted SSR. convolved with first-order absorption. In the Fa-avg. the A* and alpha* columns on the Transposed Final Parameters sheet are editable. in the units identified above. Enter the dosage amount for each formulation in the table below. enter the dose units. The absorption is mathematically separated from the decay. Then check Data already deconvolved and averaged to make two of the output workbooks editable. the Time and Mean columns are editible. This assumes that absorption is truly first order and fits a model that is n-compartment polyexponential. you must still follow the above procedure to generate the output workbooks in their appropriate format. The IVIVC Wizard will use the patient and formulation sort keys specified to determine individual profiles to deconvolve.) Include time lag: For an oral reference formulation. If you have already determined the UIR and (average) Fraction Absorbed. For an infusion. the UIR will simply be a polyexponential model of the reference data. enter the infusion time for each formulation. In UIR.pwo. If this option is not checked.pwo workbook. to compute fraction absorbed over time. check this box to include a lag time parameter in the UIR model. click Deconvolve to perform numeric deconvolution. You can paste in the values that you prefer to use. Strip Ka: For an oral reference formulation. Click Generate UIR when all settings are ready to fit the unit impulse response function(s) and generate related charts and workbook output. as chosen under Model Selection. 309 . Then.

Correlation tab supports creation and validation of an IVIVC correlation model for dissolution data defined as “internal” on the InVitro tab. See “Invitro tab” on page 306 and “In-vivo tab” on page 307.13 User’s Guide Correlation tab The IVIVC Wizard. To validate the correlation. The correlation will be built using those formulations identified as “internal” (or “reference”) on the In-Vitro and In-Vivo tabs. To do so. and compare to actual values from the in-vivo data. the Wizard can estimate AUC’s and maximum concentrations based on the model predictions. and absorption data set up or estimated via the In-Vivo tab. then run noncompartmental analysis on that predicted PK data and the observed in-vivo data. It will fit the selected correlation model to absorption data using the dissolution profiles from the In-Vitro tab as inputs to the correlation function. WinNonlin 310 . it will convolve predicted fractions absorbed and UIR’s from the other tabs.

The IVIVC Toolkit The IVIVC Wizard 13 Correlation model Select the correlation model to use to fit fraction absorbed (Fabs). Fabs vs Fdiss: Fraction absorbed and fraction dissolved over time. Model parameters: See “Model parameters” on page 197. (See “Calculation methods for AUC” on page 177 and “AUC calculation and interpolation formulas” on page 183 for definitions and calculation methods.) Tcutoff (optional): If using a library model. or click the Edit User Model button to write your own in the WinNonlin modeling language. Calculation method: You can choose to calculate AUC using either the linear. or linear up/log down trapezoidal rule. where Clast is the last predicted concentration (AUCINF). 311 . Plots Check the plots to include in the output: Tvitro vs Tvivo: Levy plot of in-vitro versus in-vivo times for a given fraction absorbed or dissolved. Click Run Validation to generate a table of prediction errors for AUC and Cmax: %_PE = 100* (Predicted-Observed) / Observed. Validation AUC: Select which computation of area under the curve to use in the validation: from dose time through the last positive concentration value (AUClast). Click Generate Plots to create the plots. linear/log. AUClast + Clast/ Lambda Z. Click Build Correlation to run the IVIVC model. (See “ASCII user model for correlations” on page 312.) Average: Select whether to compute averages as the mean or geometric mean. where Clast is the last observed concentration (AUCINF_obs). from dose time through the final observation time (AUCall). or AUClast + Clast/Lambda Z. enter the time after which the parameter Diss should stop increasing.

and enter its initial value below. The first and last values in the dissolution data are used when T is respectively before or after the time range of the dissolution data. which is called T. The wizard will translate this variable into the DTA array that is part of the WNL modeling language.This function is used to interpolate dissolution data.This variable contains the dose amount for each formulation. Enter the number of model parameters in the topmost field near the left side of this dialog. and X is then scaled or shifted by model parameters to yield an in vitro time.” the average is taken over the absolute values of %_PE for the formulations specified as Internal. No other statements are required. ASCII user model for correlations A custom model for an IVIVC correlation (see “Correlation tab” on page 310) needs to include only the parameter definitions and model equations. IVINTERP is then used to interpolate the dissolution data to find the dissolution at time T. DOSEVIVO . and the following keywords that are specific to IVIVC. Use the syntax of the WinNonlin modeling language detailed under “User Models” on page 217. 312 .13 User’s Guide For “Avg Internal. WinNonlin IVINTERP() . Edit the statements to the right to reflect your model. A common usage is for X to be the in vivo time.

Prediction tab generates a table of prediction errors for AUC and Cmax by comparing predicted data for test formulations on the Prediction tab to the target formulation identified on the In-Vitro tab.The IVIVC Toolkit The IVIVC Wizard 13 Prediction tab The IVIVC Wizard. 313 .

Sheet: Select the worksheet containing the percent dissolution data. in the units identified above. 314 . Variables: Drag column names from the Variables list to the appropriate field on the right to identify those containing time. Formulation information Dose units: Optionally. enter the dose units for the formulations. from among those open in WinNonlin.13 User’s Guide WinNonlin Dissolution data selection Dataset: Select the workbook containing your dissolution data. Enter the dosage amount for each formulation in the table below. as fraction dissolved over time. fraction dissolved and formulation identifiers.

0 to prevent incomplete measurements from biasing other parameter estimates. then set the following. you may wish to fix certain parameters to known values.1. Prediction Click Predict PK to generate the table of prediction errors for AUC and Cmax.The IVIVC Toolkit The IVIVC Wizard 13 Dissolution model Optionally. optionally based on observed Y or predicted Y (Yhat) values. the asymptotic fraction dissolved. check Use raw data above the models.1. Depending on the model. Finf. See also “Weight” on page 205. to use linear interpolation on the raw data instead of modeling.2. Options tab includes settings that control how operations on the other tabs will behave. Select one of the IVIVC models described under “Sigmoidal/dissolution models” on page 547.3. For example. Model parameters: 1. comparing test formulations to the target formulation identified on the In-Vivo tab. Weighting: Select how to weight data during modeling. can be fixed to 1. 315 . Options The IVIVC Wizard. Alternately. in order to provide smoothing you may fit the dissolution data to a sigmoidal dissolution model. See “Model parameters” on page 197 for instructions. Click Fit Dissolution Data to run the model or linear interpolation and generate output.

using fitted model selected on “InVitro” tab Object Types Workbook Text Chart 316 .Fit Dissolution Data tion data points. Output objects Select whether to hide specific output windows generated by the IVIVC Wizard in the WinNonlin window. See also “Hidden windows” on page 18. IVIVC Wizard Output Objects File Name “Dissolution Out” Description Process Simulated dissolu.13 User’s Guide WinNonlin Modeling runtime options Enter the number of predicted values to be generated by modeling operations initiated in the IVIVC Wizard and the maximum allowed number of iterations during model fitting.

Deconvolve put – fraction absorbed vs.The IVIVC Toolkit The IVIVC Wizard 13 ”Deconvolution Out” Deconvolution out. time for each profile Unit Impulse Generate UIR Response (UIR) modeling output Average fraction absorbed by formulation Fraction absorbed data going into the correlation model fitting Fraction dissolved data going into the correlation model fitting Correlation modeling output Input vivo data to correlation model simulation for validation step Input vitro data to correlation model simulation for validation step Simulated correlation output using “internal” dissolution profiles Validation convolution output – individual profiles of UIRs convolved with simulated correlation output Deconvolve Workbook Chart “UIR” “Fa-avg” Workbook Text Chart Workbook “Corr Model Vivo In” Build Correlation Workbook “Corr Model Vitro In” Build Correlation Workbook “Corr Model Out” Build Correlation “Corr Model Sim Vivo Validate Correlation Workbook Text Chart Workbook “Corr Model Sim Vitro” Validate Correlation Workbook “Val Corr Sim Out” Validate Correlation Workbook Chart “Val Conv Out” Validate Correlation Workbook Chart 317 .

with PE computations.13 User’s Guide Validation convolution output sampled at in vivo times.pwo] “Levy Plot Out” Levy plot using [Fa-avg] and modeled dissolution data “Corr Plot Out” Correlation plot using [Fa-avg] and modeled dissolution data “Avg PE Stats Out” Average absolute prediction error statistics for ‘internal’ formulations “Val Stats Out” NCA outputs averaged by formulation for observed and predicted data. using fitted model selected on “Prediction” tab “Pred Corr Sim Input vivo data to Vivo” correlation model simulation for prediction step “Pred Corr Sim Input vitro data to Vitro” correlation model simulation for prediction step “Val PK NCA In” Validate Correlation Workbook WinNonlin Validate Correlation Validate Correlation Workbook Workbook Generate Plots Chart Generate Plots Chart Validate Correlation Workbook Validate Correlation Workbook Predict PK Workbook Text Chart Predict PK Workbook Predict PK Workbook 318 . “Pred Dissolution Simulated dissoluOut” tion data points.goes into NCA “Val Baseline NCA NCA results from Out” in vivo dataset “Val PK NCA Out” NCA results from [Val PK NCA In.

Load the Scenario. for input to NCA “Pred PK NCA NCA results for Out” predicted profiles “Pred Conv Out” Prediction convolution output – individual profiles of UIRs convolved with simulated correlation output “Pred Stats Out” NCA outputs averaged by formulation for target and predicted data.The IVIVC Toolkit Using the IVIVC Wizard with PKS 13 “Pred Corr Sim Out” Simulated fraction absorbed output from correlation model “Pred Target NCA” NCA results from in vivo dataset for ‘Target’ formulation “Pred PK NCA In” Predicted concentration profiles. with PE computations. 1. 319 . 2. Predict PK Workbook Chart Predict PK Workbook Predict PK Workbook Predict PK Predict PK Workbook Workbook Chart Predict PK Workbook Using the IVIVC Wizard with PKS Initiating a New IVIVC Project in an existing PKS Scenario Note: Use these steps when you have an existing PKS Scenario that you want to apply an IVIVC analysis to. in vitro. 3. and (optionally) test in vitro data. Create relevant workbooks for in vivo. Proceed to use the IVIVC Wizard. sampled at in vivo times. Save per instructions below.

ivc) from “Other Documents. Save the IVIVC project to disk. Save per instructions below. 1. 1. Open the IVIVC Wizard. 3. in step 2 you can use the file explorer to drag and drop all files (except the .13 User’s Guide Loading a IVIVC Project from PKS WinNonlin Note: Use these steps to load an IVIVC project that has been properly saved to a PKS scenario. 1. Extract the IVIVC project file (filename . 4. Proceed to use the IVIVC Wizard. Saving an IVIVC Project to PKS Note: Use these steps to save an IVIVC project to a PKS Scenario. Merging an Existing IVIVC Project into PKS Note: Use these steps when you have an existing IVIVC project on your local disk and would like to merge the work into an existing PKS Scenario.” 3. Load the Scenario. 2. Load the IVIVC project file (filename . Proceed to use the IVIVC Wizard.ivc) from the disk location it was extracted to. Open the IVIVC Wizard. Then save the project to a clean directory. This way. 2. 5. 320 . 5. text. 4. TIP: Before beginning the merging instructions. Save per instructions below. open the IVIVC project in the IVIVC Wizard.ivc file) into WNL to open them.ivc) from its disk location. and chart objects) from their disk locations. Open the IVIVC project file (filename . Load the Scenario. Open all IVIVC project files (workbook.

Click on PKS > Other Documents. Add the *.ivc file that you just saved. 3.The IVIVC Toolkit Using the IVIVC Wizard with PKS 13 2. 5. 321 . 4. Close the IVIVC Wizard. Click on the PKS > Save or PKS > Save As to save the PKS scenario.

13 User’s Guide WinNonlin 322 .

It can analyze regression and covariance models. including unbalanced designs. “LinMix output” on page 361 and “Computational details” on page 366 cover LinMix calculations and output parameters.Chapter 14 Linear Mixed Effects Modeling ANOVA. An investigator wants to estimate the difference between two treatments for blood pressure. • • • • An example The following example is used as a framework for discussing the functions and computations of the Linear Mixed Effects Modeling wizard. “LinMix Wizard” on page 350 provides stepped instructions. this chapter starts by creating a statistical model for a common experimental design that utilizes several error terms. labeled A and B. as well as classical ANOVA modeling terms. “The linear mixed effects model” on page 326 discusses LinMix general theory in more detail. each subject 323 . It is a statistical analysis system for analysis of variance for crossover and parallel studies. “Comparing LinMix to ANOVA and SAS’s PROC MIXED” on page 372 explains similarities and differences among three analysis packages. Because responses are fairly heterogeneous among people. and regression models The Linear Mixed Effects function (LinMix) performs analyses using linear mixed effects models. and can calculate both sequential and partial tests. which should increase the power of the study. LinMix is discussed in the following sections. the investigator decided that each study subject will be his own control. • “An example”: Since linear mixed effects modeling is most easily introduced using an example. ANCOVA. Therefore.

εijkm is assumed to be normally distributed with mean 0 and variance σ2 > 0. The mixed effect model allows one to model the mean of the response as well as its variance. Hence. Since the treatment order might affect the outcome.14 User’s Guide will be exposed to treatments A and B. assume that subject effect is normally distributed with mean 0 and variance Var(Sk(j)) ≥ 0. note that this crossover study has two periods. Therefore.τ2 in all people with similar inclusion criteria as the study subjects. For the analysis. it is desirable to treat subject as a random effect. the mean is: E{yijkm} = μ + πi + δj + τq (2) (1) WinNonlin 324 . The model will be of the form: yijkm = μ + πi + δj + τq + Sk(j) + εijkm where: i = period index (1 or 2) j = sequence index (AB or BA) k = subject within sequence group index (1 … nj) m = observation within subject index (1. the purpose of the study Sk(j) = effect due to subject εijkm = random variation Typically. 2) μ = intercept πi = effect due to period δj = effect due to sequence grouping τq = effect due to treatment. half the subjects will receive A first. The purpose of the study is to estimate τ1 . Note that in model (1). 2) q = treatment index (1. This mixing of the regression model and the random effects is known as a mixed effect model. 1 and 2. and the other half will receive B first. and two sequence groups AB and BA. Study subjects are assumed to be randomly selected from the population at large.

(3) E{yijkm}is known as the fixed effects of the model.24 11.43 10. The data are shown in Table 14-1.67 325 .63 9.70 10. since these terms represent random variables whose variances are to be estimated.2 7.6 9. while Sk(j) + εijkm constitutes the random effects.91 8.88 9. A model with both is a mixed effect model.13 9. This is appropriate since the fixed effects model the mean of the response.Linear Mixed Effects Modeling An example 14 and the variance is given by: Var{yijkm} = Var{Sk(j)}+ σ2.63 9.31 9.02 7. The fixed and random effects of the model are entered on different pages of the wizard. Data for the example Subject 1 2 3 4 5 6 7 8 9 10 11 12 7 8 9 10 11 Sequence AB AB AB AB AB AB BA BA BA BA BA BA BA BA BA BA BA Period 1 1 1 1 1 1 2 2 2 2 2 2 1 1 1 1 1 Treatment A A A A A A A A A A A A B B B B B Response 9.1 9. A model without the fixed effects is known as a random effect model. Table 14-1.15 9. while the random effects model the variance of the response. A model with only the residual variance is known as a fixed effect model.23 9.82 11. The LinMix wizard is used to analyze these data.

14 User’s Guide Table 14-1.31 WinNonlin LinMix can fit linear models to Gaussian data.91 8. Data for the example (continued) Subject 12 1 2 3 4 5 6 Sequence BA AB AB AB AB AB AB Period 1 2 2 2 2 2 2 Treatment B B B B B B B Response 11. Independent variables can be continuous or discrete.73 9.28 11. The mean and variance structures can be simple or complex.77 8. and no distributional assumption is made about them.” The linear mixed effects model WinNonlin’s Linear Mixed Effects Modeling feature is based on the general linear mixed effects model. which can be represented in matrix notation as: y = Xβ + Zγ + ε 326 .34 8. This model provides a context for “The linear mixed effects model.72 11.

then the R matrix is determined by the variance structure specified for the repeated model. If the model includes a repeated specification. the random effects tabs in the Variance Structure dialog specifies the model terms for constructing Z and the structure of G. γ and ε are uncorrelated. normally distributed with mean 0 and variance G. i. one can show that the variance of y is: V = ZGZ + R If Z ≠ 0 . the model includes one or more random effects. In the LinMix wizard. Listing all the parameters in a vector. It is instructive to see how a specific example fits into this framework. If the model doesn’t include a repeated specification. β is a vector of unknown (fixed effect) parameters. then R is the residual variance—i. γ is a random unobserved vector (the vector of random-effects parameters).e. and the repeated tab specifies the structure of R. The fixed effects model is shown in equation 2 under “An example” on page 323. Z is a matrix of known constants. the design matrix for γ. if R = σ2I and Z = 0. Using the assumptions for the linear mixed effects model. one obtains: T 327 .. R = σ2I. the G matrix is determined by the variance structures specified for the random models.. LinMix breaks the model into the following areas: • • • Note: Fixed effects Variance structure Least squares means • • Contrasts Estimates In the linear mixed effects model. then the LinMix model reduces to the ANOVA model.e. and ε is a random vector.Linear Mixed Effects Modeling The linear mixed effects model 14 where: X is a matrix of known constants and the design matrix for β. normally distributed with mean 0 and variance R. the fixed effects dialog specifies the model terms for constructing X.

LinMix takes this sum of terms and constructs the X matrix. A user specifies a model by giving a sum of terms. below. the first column of X contains all ones. The collection can be directly entered into γ.14 User’s Guide βT = [μ. The Z matrix will consist of 1’s and 0’s. if A and B are variables in the data set: A + B + A*B. This is demonstrated using two classification variables. The way this is done depends upon the nature of the variables A and B. Wizard Page Fixed Effects Specifies Tabs on page Specifies X matrix Variance Structure Specifies Z matrix Random tabs (1 or more) Repeated tab (exactly 1) Creates Generates β Generates G = var(γ) Generates R = var(ε) WinNonlin Construction of the X matrix A term is a variable or a product of variables. δ2. j) will be 1 when observation i is from subject j. check No Intercept in the Fixed Effects dialog of the wizard. The No Intercept option By default. regressor/covariate or classification. π2. and 0 otherwise. To exclude it. 328 . The details will be discussed later. Element (i. A summary of the wizard pages and the part of the model they generate is listed below. In this model. The variance of γ is of the form G ≡ Var(γ) = Var{Sk(j)} In(1)+n(2) where Ib represents a b×b identity matrix. Drug and Form. there is a corresponding column in the X matrix. For each subject. τ1. τ2] For each element of β. π1. The residual variance is R = Var{εijkm} = σ2 I2[n(1)+n(2)] The parameter to be estimated for R is σ2. The parameter to be estimated for G is Var{Sk(j)}. there is one Sk(j). The rules for translating terms to columns of the X matrix depend on the variable types. The LinMix wizard references this column as the intercept. the X matrix is composed entirely of 1’s and 0’s. δ1. For example. The exact method of constructing it is discussed under “Construction of the X matrix”.

consider the fifth row of Table 14-3. The association between levels and codes is determined by the numerical order for converted numerical variables and by the character collating sequence for converted character variables. The product of two continuous variables is the ordinary product within rows. A term containing one continuous variable produces a single column in the X matrix. Table 14-3 shows the product of Drug and Age. …. Table 14-5 gives their product. Table 14-4 shows two classification and indicator variables. Table 14-3 demonstrates this for the variable Drug. The product of a classification variable with a continuous variable is obtained by multiplying the continuous value by each column of the indicator variables in each row. For example. j)th partition. Variables in sample data set Drug Name DrugA DrugA DrugA DrugA DrugB A. 329 . Age and Weight. For example.…. an indicator variable is placed in the X matrix for each level of the variable. The product of Drug and Age is: [0 1 0] ⊗ [45] = [ 0 45 0]. An integer value is associated with every level of a classification variable. The product operations described in the preceding paragraph can be succinctly defined using the Kronecker productA.Linear Mixed Effects Modeling The linear mixed effects model 14 and two continuous variables. i=1. The product of Drug and Form is: [0 1 0] ⊗ [1 0 0 0] = [0 0 0 0 1 0 0 0 0 0 0 0] The result is the fifth row of Table 14-5. in Table 14-2. Then the Kronecker product A ⊗ B = (aij B) is an mp × nq matrix expressible as a partitioned matrix with aij B as the (i. Table 14-2. Now consider the fifth row of Table 14-4. The model specification allows products of variables. m and j=1. 0 0 0 0 1 Value Name capsule drops syrup tablet capsule 0 1 2 3 0 Form Value 48 34 27 23 45 120 124 123 172 200 Age Weight Let A = (aij) be a m×n matrix and let B = (bij) be a p×q matrix. When a single classification variable is specified for a model.n.

14 User’s Guide Table 14-2. Indicator variables and a product of a classification variable with a continuous variable Drug Name DrugA DrugA DrugA DrugA DrugB DrugB DrugB DrugB DrugC DrugC DrugC DrugC 0 0 0 0 1 1 1 1 2 2 2 2 Value 1 1 1 1 0 0 0 0 0 0 0 0 Indicator variables DrugA DrugB 0 0 0 0 1 1 1 1 0 0 0 0 DrugC 0 0 0 0 0 0 0 0 1 1 1 1 48 34 27 23 45 26 41 32 38 17 32 23 Age DrugA 48 34 27 23 0 0 0 0 0 0 0 0 Drug*Age DrugB 0 0 0 0 45 26 41 32 0 0 0 0 DrugC 0 0 0 0 0 0 0 0 38 17 32 23 330 . Variables in sample data set (continued) Drug Name DrugB DrugB DrugB DrugC DrugC DrugC DrugC 1 1 1 2 2 2 2 Value Name drops syrup tablet capsule drops syrup tablet 1 2 3 0 1 2 3 Form Value 26 41 32 38 17 32 23 150 120 130 190 125 175 150 Age Weight WinNonlin Table 14-3.

Linear Mixed Effects Modeling The linear mixed effects model 14 Table 14-4. capsule drops syrup tablet 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 1 0 0 0 1 Name Val. DrugA DrugB DrugC capsule 0 capsule 0 Table 14-5. Two classification variables and their indicator variables Drug DrugA 0 DrugA 0 DrugA 0 DrugA 0 DrugB 1 DrugB 1 DrugB 1 DrugB 1 DrugC 2 DrugC 2 DrugC 2 DrugC 2 1 1 1 1 0 0 0 0 0 0 0 0 Indicators 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 Form Name drops syrup tablet drops syrup tablet drops syrup tablet capsule 0 1 2 3 1 2 3 1 2 3 1 0 0 0 1 0 0 0 1 0 0 0 Indicators Val. Product of the two classification variables Drug DrugA DrugA DrugA DrugA DrugB DrugB DrugB DrugB DrugC DrugC Form capsule drops syrup tablet capsule drops syrup tablet capsule drops 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 Indicator 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 331 .

The range of the subscript j starts with 0 if the intercept is in the model and starts with 1 otherwise. The expression: (5) Σljβj j with the lj constant is called a linear combination of the elements of β. Product of the two classification variables (continued) Drug DrugC DrugC Form syrup tablet 0 0 0 0 0 0 0 0 0 0 Indicator 0 0 0 0 0 0 0 0 0 0 1 0 0 1 WinNonlin Linear combinations of elements of β Most of the inference done by LinMix is done with respect to linear combinations of elements of β. A linear combination of β is said to be estimable if it can be expressed as a linear combination of expected responses. Most common functions of the parameters. The X matrix has a column of ones in the first position followed by three treatment indicators. The second form is specific to model (4) above. Linear combinations of β are entered in LinMix through the wizard’s Estimates and Contrasts pages.τ3. as an example. can be generated by choosing appropriate values of lj. τ1 . and τ2. the model where the expected value of the response of the jth subject on the ith treatment is: μ + τi (4) where μ is a common parameter and μ+τi is the expected value of the ith treatment. Consider. such as μ + τ1. The β vector is: β = [β0 β1 β2 β3] = [μ τ1 τ2 τ3] The first form is for the general representation of a model. Suppose there are three treatments.14 User’s Guide Table 14-5. In the context of 332 .

any linear combination involving only τs is estimable if. c2. where the sum of the coefficients is zero. is called a contrast. The linear combination τ2 is not a contrast of the τ’s and thus is not estimable.τ3 is a contrast of the τ’s and thus is estimable. lj = cj for j=1. If there is more than one z. The linear combination is estimable if: l Z ------------------------. A linear combination of effects. In general. Determining estimability numerically The X matrix for the model in equation (4) has three distinct rows. l1 + l2 + l3 = 0. Note that c1=1. and c3. Also shown in Table 14-6 is a vector (call it z) that is a basis of the space orthogonal to the rows of X. The linear combination τ1 . and c3=0 generates μ + τ1.3. and tol is the Estimability Tolerance.< tol T l ∞ Z Z T (8) where || • ||∞ is the largest absolute value of the vector. a linear combination of the elements of β is estimable if it can be generated by choosing c1. The linear combination 5 in vector notation is lTb and is estimable if lTΖ = 0. and these are shown in Table 14-6. 333 . c2=0. Also. For example. Allow for some rounding error in the computation of z. so μ + τ1 is estimable. the number of rows in the basis of the space orthogonal to the rows of X is equal the number of linear dependencies among the columns of X.2. A rearrangement of expression (6) is: (c1 + c2 + c3)μ + c1τ1 + c2τ2 + c3τ3. c1(μ + τ1) + c2(μ + τ2) + c3(μ + τ3). (7) (6) The form in (7) makes some things more obvious. and only if. Let l be a vector whose elements are the lj of expression (5). then (8) must be satisfied for each z.Linear Mixed Effects Modeling The linear mixed effects model 14 model 4.

j is the effect of the j-th level of a factor B.) The regression coefficient is 1. regress row 10 on row 5. Now calculate residuals: (row 9) . LinMix tries to be helpful. regress row 9 on row 3. Consider the model: yijk = μ + αi + βj + (αβ)ij + εijk (9) where μ is the over-all mean. so this will be used as an example. so there is no need to compute residuals.1× (row 3). i = 1. The regression coefficient is 0. Within the α columns. 2. i. This section is to describe how the substitution is made.. The canonical form of the QR factorization of X is displayed in rows 0 through 8 of Table 14-7. and if a user attempts to estimate an non-estimable function. row 9 is already 0 so no operation is done. and the next operation applies to row 10. σ2). αi is the effect of the i-th level of a factor A. (αβ)ij is the effect of the interaction of the i-th level of A with the j-th level of B. and the εijk are independently distributed N(0. Notice is given in the text output when the substitution is made. The most common attempt to estimate a non-estimable function of the parameters is probably trying to estimate a contrast of main effects in a model with interaction. Within the αβ columns. The numbers in the β columns of row 10 are zero. The result is shown in row 10. The coefficients to generate α1 . 2. and it is the user's responsibility to check the coefficients to see if they are reasonable. row 10 is a deviation of row 9 from 334 . At this point.).α2 are shown in row 9. The process is a sequence of regressions followed by residual computations. (Row 4 is all zeros and is ignored. not just the α columns. The basis of the space orthogonal to the rows of X Elements of β Distinct rows of X μ 1 1 1 Basis of orthogonal space -1 τ1 1 0 0 1 τ2 0 1 0 1 τ3 0 0 1 1 Substitution of estimable functions for non-estimable functions It is not always easy for a user to determine if a linear combination is estimable or not. so skip to the αβ columns. In the μ column. LinMix will substitute a “nearby” function that is estimable. This operation goes across the entire matrix. j = 1.e.14 User’s Guide WinNonlin Table 14-6. (Rows 6 through 8 are all zeros and are ignored.

5 0. 2.5 1 α1 0.5 Final result 0 335 . The QR factorization of X scaled to canonical form μ 0 1 2 3 4 5 6 7 8 9 10 0 0 1 0 1 -1 0 -1 0 0 0 0 0 0 0 -0. Partition the QR factorization of X vertically and horizontally corresponding to the terms in X.5 0 0. The result is the expected value of the difference between the marginal A means.5 1 α2 0.5 0 0 1 β2 0.5 -0.5 0 1 0.5 0 0 -1 0 (αβ)11 (αβ)12 (αβ)21 (αβ)22 0. Table 14-7.5 0 1 0 0 0 0 0.25 -0. For each vertical partition.5 0 -1 0 0 0.5 0 -0.25 -0. do the following: 1. This section provides a general description of the process just demonstrated on a specific model and linear combination of elements of β. do a least squares fit) the work row on the rows in the diagonal block of the QR factorization.5 -1 0 β1 0.e. 3.5 0 0.5 0 -0. Subtract the deviation (row 10) from the original (row 9) to get the estimable function displayed in the last row.25 0.5 0. Subtract the final values in the work row from the original coefficients to obtain coefficients of an estimable linear combination.Linear Mixed Effects Modeling The linear mixed effects model 14 an estimable function. from left to right.5 0 0.5 0 -0. Calculate the residuals from this fit across the entire matrix.. Overwrite the work row with these residuals.5 0 -1 0 0.5 -0.25 0. Put the coefficients of a linear combination in a work row. This is likely what the user had in mind. Regress (i.

Then column j provides no additional information about the response y beyond that of the columns that come before. In this example. then each parameter can be uniquely estimated. X2. To solve this issue. X1. X0 and X1 are clearly linearly independent. it seems sensible to not use the column in the model fitting. Suppose column j is in the span of columns 0. Output parameterization Suppose X is n×p. 336 . where n is the number of observations and p is the number of columns. the number typically assigned in ANOVA. since X3 = X0 . one must impose additional constraints on the estimator of β. If rank(X) < p. 1. and X3 correspond to the three levels of the treatment effect. The degrees of freedom for an effect is the number of columns remaining after this process. Hence β3 would be set to zero. j-1. its parameter estimate is set to 0. as is X2 linearly independent of X0 and X1. consider a simple one-way ANOVA model. treatment effect has 2 degrees of freedom. …. In addition. then there are infinitely many solutions. See “Predicted values and residuals” on page 364 for computation details. If X has rank p. the model can include interaction and nested terms. denoted rank(X). When this happens. X1.14 User’s Guide WinNonlin Fixed effects The specification of the fixed effects (as defined in “The linear mixed effects model” on page 326) can contain both classification variables and continuous regressors. As an example.X1 -X2. The design matrix is X0 1 1 1 1 1 1 X1 1 0 0 1 0 0 X2 0 1 0 0 1 0 X3 0 0 1 0 0 1 X0 is the intercept column. But X3 is not linearly independent of X0. In this case. with three levels. and X2.

Transformations are applied to the response before model fitting. taking the logarithm often stabilizes the variance. β3.X3 and X6 = X0 . Design matrix expanded for fixed effects model Sequence + Period + Treatment Intercept X0 1 1 1 1 1 1 1 1 1 1 1 1 Sequence AB [X1] BA [X2] 0 0 0 0 0 0 1 [X3] 1 1 1 1 1 1 Period 2 [X4] 0 0 0 0 0 0 1 1 1 1 1 1 Treatment A [X5] B [X6] 0 0 0 0 0 0 337 . and hence set β4 = 0 and β6 = 0. Using that model. since log(x) = ln(x)/ln(10). Crossover example continued The fixed effects are given in equation (5). Hence set β2 = 0. the design matrix X would be expanded into that presented in Table 14-8. The vector of fixed effects corresponding to that design matrix would be β = (β0. If the norm of the vector after GS divided by the norm of the original vector is less than δ. only scaled by ln 10. X4 = X0 . then the vector is called singular. resulting in better fit. or log transformation (log10). β5. If the standard deviation is proportional to the mean. Notice that X2 = X0 . Similarly. Transformation of the response The transformation pull-down menu provides three options: No transformation. This implicitly assumes that the error structure is multiplicative with lognormal distribution. Note that all estimates using log will be the same estimates found using ln. β2. β6).X1.X5.Linear Mixed Effects Modeling Fixed effects 14 Singularity tolerance If intercept is in the model. β1. Table 14-8. then center the data. β4. Perform a Gram-Schmidt orthogonalization process. ln transformation (loge).

338 . The Random effects specify Z and the corresponding elements of G = Var(γ). Only 1 repeated effect may be specified.14 User’s Guide Table 14-8. The user may have 0. G. Design matrix expanded for fixed effects model Sequence + Period + Treatment (continued) Intercept X0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 Sequence AB [X1] BA [X2] 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 1 [X3] 0 0 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 Period 2 [X4] 1 1 1 1 1 1 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 Treatment A [X5] B [X6] 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 WinNonlin Variance structure The LinMix Variance Structure page specifies the Z. The Repeated effect specifies the R = Var(ε). 1. and R matrices defined in “The linear mixed effects model” on page 326. or more random effects specified.

T2. All variables used on this page of the LinMix wizard must be classification variables. and that serial correlations among the measurements are likely. Suppose further that this correlation is affected by the treatment itself. An example will make this easier to understand. The effect is to create additional parameters of the same variance structure. ng. Suppose 5 subjects are randomly assigned to each of 3 treatment groups. Group will accept any model term. Suppose the variance blocking model term has b levels. and ΙN is the N × N identity matrix. Several variance structures are possible. Put the model term in the Repeated Specification box. and T3. then R = σ2 ΙN is used. heterogeneous compound symmetry. where N denotes the number of observations.Linear Mixed Effects Modeling Variance structure 14 Repeated effect The repeated effect is used to model a correlation structure on the residuals. Suppose further that each subject is measured in periods t = 0. The variance Σ is specified using the Type pull-down menu.g. Then R would have the form R = Ιb ⊗ Σ where Σ is the variance structure specified. The autoregressive is a first-order autoregressive model. If no repeated effect is specified. …. one must have a classification model term that uniquely identifies each individual observation. 2. T2.. To model heterogeneity of variances. The variance blocking model term creates a block diagonal R matrix. and further suppose that the data are sorted according to the variance blocking variable. one may assume that the data are sorted by treatment group. or a nested term (e. including unstructured. Specifying the repeated effect is optional. then the variance for observations within group g will be Σg. and no-diagonal factor analytic. Without loss of generality. 1. autoregressive. 2. so the correlation structure will differ among T1. Time(Subject)). 3.. use the group variable.g. If a group variable has levels g = 1. 339 . Time*Subject). then subject within treatment group. and T3. To specify a particular repeated effect. See the Help file for the details of the variance structures. Note that a model term can be a variable name. call the treatment groups T1. an interaction term (e.

340 . I5 is used because there are five subjects within each treatment group. Within each subject. Often compound symmetry will effectively mimic autoregressive in cases where autoregressive models fail to converge. Other variance types are also possible. the variance structured specified is 1 ρg ρg ρg 2 2 3 σg 2 ρg 1 3 2 ρg ρg 2 ρg ρg 1 ρg ρg ρg ρg 1 This structure is the autoregressive variance type. It produces a variance structure that looks like WinNonlin R1 0 0 R = 0 R2 0 0 0 R3 where each element of R is a 15×15 block.14 User’s Guide First consider the effect of the group. The output will consist of six parameters: σ2 and ρ for each of the three treatment groups. the form of each Rg is I5 ⊗ Σg. Each Rg has the same form. Because the variance blocking variable is specified.

341 . It is possible to put more than one model term in the Random Effects Model. if no entries are made in the random effect. it expands the Z matrix by taking the Kronecker product of the Variance Blocking Variables with the columns produced by the intercept check box and the Random Effects Model terms. It is not possible to delete all random effects. To delete a previously specified random effect. it is possible to specify up to 10 random effects. The variance matrix appropriate for that type is sized according to the number of columns produced in that random effect. The random intercept check box puts an intercept column (i.Linear Mixed Effects Modeling Variance structure 14 Random effects Unlike the (single) repeated effect. The Variance Blocking Variables has the effect of inducing independence among the levels of the Variance Blocking Variables. constructed in the same way that columns are produced in the X matrix.e. but each random page will correspond to a single variance type. then it will be ignored. all 1’s) into the Z matrix. however. click the Delete Random button. The Group model term is used to model heterogeneity of the variances among the levels of the Group model term.. Additional random effects can be added by clicking the Add Random button. Mathematically. similarly to Repeated Effects. Model terms entered into the Random Effects Model produce columns in the Z matrix.

build the resulting Z and G matrices. This example models the variance using the compound symmetry variance type. Suppose further that R1 has three levels. while R2 has two levels. For R1+R2 on Random 1 page. To illustrate this. A. γ3. multiple effects on one random effect Suppose one has two random effect variables. The hypothetical data are shown as follows. put the columns of each variable in the Z matrix to get the following. Y and Z. B. R1 and R2. γ5). R1 A B C A B C R2 Y Y Y Z Z Z WinNonlin Specifying a random effect of R1+R2 will produce different results than specifying R1 on Random 1 and R2 on Random 2 pages.14 User’s Guide Multiple random effects vs. R1 and R2 share the same variance matrix G where: 342 . and C. R1 A 1 0 0 1 0 0 R2 C 0 0 1 0 0 1 B 0 1 0 0 1 0 Y 1 1 1 0 0 0 Z 0 0 0 1 1 1 The random effects can be placed in a vector γ = (γ1. γ4. γ2.

and R2 is placed on Random 2 page. the columns in the Z matrix and the corresponding G is: R2 1 1 1 0 0 0 Y Z 0 0 0 1 1 1 In this case: ⎡G 0 ⎤ G=⎢ 1 ⎥ ⎣ 0 G2 ⎦ 343 .Linear Mixed Effects Modeling Variance structure 14 Now. For R1. the Z matrix columns are: R1 B 0 1 0 0 A 1 0 0 1 C 0 0 1 0 A 1 0 0 1 0 0 R1 B 0 1 0 0 1 0 C 0 0 1 0 0 1 For R2. consider the case where R1 is placed on Random 1 page.

Thus. the average of the x1j is 5. βj is the effect of the jth level of factor B. sample means cannot be used to compare levels of A because they contain different functions of β1 and β2. The preceding examples constructed linear combinations of parameters. and εij is a random error with zero expected value. αi is the intercept for the ith treatment. that represent the expected values of sample means in balanced data. Suppose there are two treatments. and εijk is a random error with zero expected value. one compares the linear combinations μ + (β1 + β2)/2 + α1 and μ + (β1 + β2)/2 + α2.6β1+0. one can estimate α1 + 10β and α2 + 10β where the overall mean of the covariable is used in each linear combination. and the average of the x2j is 15. in the presence of unbalanced data. The respective expected values of the averages of all values on level 1 of A and the averages of all values on level 2 of A are μ+(0. Suppose there are six observations for the combinations where i = j and four observations for the combinations where i ≠ j. If there are covariables with unequal means for the different levels. The model is: yijk = μ + αi + βj + εijk where yijk is the observed response on the kth individual on the ith level of factor A and the jth level of factor B. or if there are unbalanced data. This is the idea behind least squares means. Now consider a 2 × 2 factorial design. The model is: yij = αi + xij β + εij where yij is the observed response on the jth individual on the ith treatment. though the process can be repeated for different defining terms for the same model. β is the slope with respect to x. These are not comparable because of the different coefficients of β. μ is the over-all mean. This section describes least squares means. The respective expected values of the sample means are α1 + 5β and α2 + 15β.6β2)+α2. The defining 344 . the subsample means are not estimates that can be validly compared. αi is the effect of the ith level of factor A. Consider a completely randomized design with a covariable. Instead.4β1+0. Least squares means are given in the context of a defining term. xij is the value of the covariable for the jth individual in the ith treatment.14 User’s Guide WinNonlin Least squares means Sometimes one would like to see the mean response for each level of a classification variable or for each combination of levels for two or more classification variables. statistics that make proper adjustments for unequal covariable means and unbalanced data in the linear mixed effects model. Instead.4β2) +α1 and μ+(0.

average values of the variables are the coefficients. three sets of coefficients are created. Since Drug has three levels. DrugA. The next twelve elements are: [ 1 0 0 ] ⊗ [0. A set of coefficients are created for each of all combinations of levels of the classification variables in the defining term.25 0.25 ] = [0. the average of a four factor indicator variable with balanced data.25 0. the average of Age.Linear Mixed Effects Modeling Least squares means 14 term must contain only classification variables and it must be one of the terms in the model. 0. To get means for each level of Drug. For all variables in the model. When the user requests least squares means. and factor A is the defining term in the second example. and 0. Build the coefficients associated the first level of Drug. No new principles would be illustrated by finding the coefficients for the Form least squares means. the products are formed using the same rules used for constructing rows of the X matrix as described in the section on Fixed Effects. The value 1/k would be the actual average if the data were balanced. The next coefficient is 32.25 0. The first coefficient is 1 for the implied intercept. The results for DrugB and DrugC are also shown in Table 14-9. It is possible that some least squares means will not be estimable. LinMix automatically generates the coefficients lj of expression (5) and processes them almost as it would process the coefficients specified in an estimate statement. Form is not in the defining term. so average values are used. The coefficients for the Drug*Form least squares means would be like representative rows of X except that Age would be replaced by average of Age. the indicator variables associated with DrugA. The values of all variables in the model have now been defined. suppose the fixed portion of the model is: Drug + Form + Age + Drug * Form. This chapter describes generation of linear combinations of elements of β that represent least squares means.25 0. The next three coefficients are 1. assume the average of each indicator variable is 1/k.25 0. above. The average value of a numeric variable (covariable) is the average for all cases used in the model fitting. the defining term is Drug. If some terms in the model are products.25. Treatment is the defining term in the first example. For example. For a classification variable with k levels. but not in the defining term.25 0 0 0 0 0 0 0 0] The final result is shown in the DrugA column of Table 14-9.17. The next four coefficients are all 0.25 0. 345 .

25 0.17 0.25 0.25 32.25 0 0 0 0 0 0 0 0 DrugB 1 0 1 0 0.25 0.17 0 0 0 0 0. Coefficients of least squares means lj Column of X Intercept Drug DrugA DrugB DrugC Form capsule drops syrup tablet Age Drug*Form DrugA*capsule DrugA*drops DrugA*syrup DrugA*tablet DrugB*capsule DrugB*drops DrugB*syrup DrugB*tablet DrugC*capsule DrugC*drops DrugC*syrup DrugC*tablet βj β0 β1 β2 β3 β4 β5 β6 β7 β8 β9 β10 β11 β12 β13 β14 β15 β16 β17 β18 β19 β20 DrugA 1 1 0 0 0.14 User’s Guide WinNonlin Table 14-9.25 0.25 0.25 0 0 0 0 DrugC 1 0 0 1 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 0.25 32.25 0.25 32.25 Contrasts A contrast is a linear combination of the parameters associated with a term where the coefficients sum to 0. Contrasts facilitate the testing of linear com346 .17 0 0 0 0 0 0 0 0 0.25 0.25 0.25 0.25 0.

Using the first approach. in a single test with a predetermined level of the test.e. To see the difference. and High Dose. For example.Linear Mixed Effects Modeling Contrasts 14 binations of parameters. This requirement is enforced in the wizard. Contrasts are tested using the Wald (1943) test statistic: –1 T –1 Tˆ ˆ T ˆ ( Lβ ) ( L( X VX ) L ) ( Lβ ) ˆ ˆ where β and V are estimators of β and V. i. compare Placebo to Low Dose and Placebo to Medium Dose. Low Dose. One could write this hypothesis as: H0: μPlacebo . This is true generally: contrasts are invariant under changes in scaling. This approach will produce two independent tests. both of which have one column. This tests both hypotheses jointly. consider a completely randomized design with four treatment groups: Placebo.(μLow + μMedium + μHigh) = 0..(μLow + μMedium + μHigh)/3 = 0 or equivalently: H0: 3μPlacebo . Note that this number must be no more than one less than the number of distinct levels for the model term of interest. Contrast #1 Joint tests 347 Title . Both of these are contrasts since the sum of the coefficients is 0. The user may wish to test the hypothesis that the average of the dosed groups is the same as the average of the placebo group. The Wald statistic is compared to an F distribution with rank(L) numerator degrees of freedom and denominator degrees of freedom estimated by the program or supplied by the user. Medium Dose. Note that both contrasts make the same statement. A second approach would be to put the same term in the neighboring contrast. each of which is at the specified level. Joint versus single contrasts Joint contrasts are constructed when the number of columns for contrast is changed from the default value of 1 to a number larger than 1. enter the coefficients as follows. L must be a matrix such that each row is in the row space of X.

at the possible expense of some power. Placebo Treatment Treatment Placebo Low Dose Medium Dose High Dose -1 0 1 0 This produces two tests. See “Substitution of estimable functions for nonestimable functions” on page 334 for the details. H0: – 1 1 0 0 β = –1 0 1 0 0 0 Now compare this with putting the second contrast in its own contrast. H01: [-1 1 0 0] β = [0] H02: [-1 0 1 0] β = [0] Note that this method inflates the overall Type I error rate to approximately 2α. whereas the joint test maintains the overall Type I error rate to α. Placebo Treatment Treatment Placebo Low Dose Medium Dose High Dose -1 1 0 0 Contrast #2 Medium vs. Contrast #1 Title Effect Contrast #1 #2 #3 #4 Low vs. testing the following hypothesis. then an estimable function will be substituted for the nonestimable function. 348 .14 User’s Guide Effect Contrast #1 #2 #3 #4 Treatment Treatment Placebo Low Dose Medium Dose High Dose WinNonlin -1 1 0 0 -1 0 1 0 This will produce one test. Nonestimability of contrasts If a contrast is not estimable.

To override the default degrees of freedom.95 will yield a very narrow confidence interval with coverage just less than 1%. The degrees of freedom must greater than 0. The numerator degrees of freedom is always calculated as rank(L). If the check box is not checked. The confidence level must be between 0 and 100. estimates do not support joint estimates.. it will enter the estimable function used instead of the nonestimable function. The Show Coefficients check box will produce extra output. enter an appropriate tolerance in the Estimability Tolerance text box (see “Substitution of estimable functions for non-estimable functions” on page 334). then the default denominator degrees of freedom will be used. Unlike contrasts. i. As such. Other options To control the estimability tolerance.e. including the coefficients used to construct the tolerance. then it will repeat the contrasts entered. Note that it is entered as a percent: Entering 0. The confidence intervals are marginal. Satterthwaite and residual degrees of freedom yield the same number. Estimates The estimates page produces estimates output instead of contrasts. Check the box for residual degrees of freedom. the coefficients need not sum to zero. Note: For purely fixed effects model. All other options act similarly to the corresponding option in the Contrasts. multiple model terms may be included in a single estimate. check the box. 349 . If the contrast entered is estimable. the Satterthwaite degrees of freedom. If the contrasts are not estimable. and enter an appropriate number.Linear Mixed Effects Modeling Estimates 14 Degrees of freedom The degrees of freedom check box allows the user to control the denominator degrees of freedom in the F approximation. For details of the Satterthwaite approximation. see the section “Satterthwaite’s Approximation for Degrees of Freedom”. Additionally.

click the Previous Model button. Note: To reuse the model last run in LinMix (during the same WinNonlin session).14 User’s Guide WinNonlin LinMix Wizard The general steps to create a linear mixed effects model are outlined below.. 3. See the Examples Guide for further linear mixed effects modeling and bioequivalence examples. save the model to a LinMix Command file (*. Refer to “An example” on page 323 for an in-depth discussion of linear mixed effects modeling in the context of an example study. depending which calculations are included. by clicking the Save Model button. Choose Tools>Linear Mixed Effects Wizard from the WinNonlin menus or click the Linear Mixed Effects Wizard tool bar button. 4. 6.ANV) can also be loaded into LinMix so long as bioequivalence is not included. ANOVA command files (*. Set up any other optional calculations and options as described under: – “Variance structure” on page 353 – “Contrasts” on page 356 – “Estimates” on page 358 – “Least squares means” on page 359 – “LinMix general options” on page 360 5. Open the worksheet to be analyzed. Click Calculate to start the calculations and close the LinMix wizard. To set up a linear mixed effects model: 1.. 350 . button and open the LinMix Command file (*.LML) for future re-use. as described under “Fixed effects” below. Some steps are optional. To reload a saved model. click the Load.LML). LinMix limits and constraints Cell data may not include question marks (?) or either single (') or double (") quotation marks. 2. (Optional) At any point in the LinMix wizard. Define the model in the Fixed Effects dialog.

valid characters include numbers and underscore (my_file_2). Estimates.000 256 65.g. variable names may not have single or double quotes in them.).*. Table 14-10. Titles (in Contrasts. /. question marks (?) or semicolons (.000 1. They also cannot include parentheses (). After the first character. They cannot include spaces or the following operational symbols: +.000 10.. . The following are maximum counts for LinMix variables and other objects. output title or chart title) Maximum 30 10 16 128 255 256 10 256 100 100 10. =. LinMix limits Item model terms factors in a model term sort keys dependent variables covariate/regressor variables variables in the data set random and repeated statements variables contrast statements estimate statements combined length of all variance parameter names (total characters) combined length of all level names (total characters) levels per variable path and file name (total characters) number of records passed from WinNonlin to other programs characters per data cell column name length (characters) titles (e.535 128 128 5 lines 351 . Finally. or ASCII) can have single or double quotation marks—but not both.Linear Mixed Effects Modeling LinMix Wizard 14 Variable names must begin with a letter.

enter the fixed effects model: a. See also “Weighting” on page 241. type the model. In the box labeled Model Specification. temperature or body weight. the analysis will be performed for the missing level and MISSING will be printed as the current value of the sort variable. e. For discussion of the model. Sort variables define subsets of data to be processed separately.. formulation. see “The linear mixed effects model” on page 326. treatment. the data are averages and weights are sample sizes associated with the averages.. The Weight Variable cannot be a classification variable. Hence: Seq+Subject(Seq)+Period+Form 352 • . a separate analysis will be done for each level of the sort variables. 2.. drag variable names and click on the operators to be used in the model. each row of data is multiplied by the square root of the corresponding weight. and gender.14 User’s Guide Fixed effects To define the model: WinNonlin 1. that is. The weights are used to compensate for observations having different variances. In the most common situation. drug concentration. A weight variable can be included if weights for each record are included in a separate data column. It can also be used in the model. Regressor variables or covariates are continuous independent variables. Drag and drop variables to categorize those needed in the LinMix model: • • • • Dependent variables contain the values to which the model will be fit. using an example to illustrate. Classification variables or factors are categorical independent variables.g. If the sort variable has missing values. Weight variable values should be proportional to the reciprocals of the variances.g. e. or b. When a weight variable is specified. It must have been declared as a regressor/covariate before it can be used as a weight variable. Note: Parentheses in the model definition represent nesting of model terms. e.g.

6.) It provides a variety of covariance structures. Select the transformation from the pull-down list for Dependent Variables Transformation. The choices for variance structures are shown in Table 14-11. 3. CAUTION: If the dependent variable data in the workbook are already log transformed. Mixed linear models contain both fixed effects and random effect parameters. Accept the default setting for Fixed Effects Confidence Level or enter the confidence interval for calculations of the fixed effects model.Linear Mixed Effects Modeling LinMix Wizard 14 is a valid use of parentheses and indicates that Subject is nested within Seq. 5. Note that this will change the parameterization of the classification variables. See “Construction of the X matrix” on page 328 for discussion.LML) for future re-use. Select the No Intercept option to eliminate the intercept in the model. 4. (n = 1 if the group option is not used) t = number of time points in the repeated context b = dimension parameter: for some variance structures. Variance structure The Variance Structure dialog follows the Fixed effects dialog in the LinMix wizard. by clicking the Save Model button. 7. Click Next to move to the Variance structure or Calculate to run the analysis. The variances of the random-effects parameters become the covariance parameters for this model. (See “LinMix Wizard” on page 350. as this will transform the data a second time. The table uses the following notation: n = number of groups. See “Variance structure” on page 338 for discussion of variance in linear mixed effects models. Drug + Disease + (Drug*Disease) is not a valid use of parentheses in the model definition. Check the Dependent Variables Transformation check box if the dependent variable values should be log-transformed before analysis. the num353 . do not select a log transformation from the pull-down list. At any time in the LinMix wizard. it is possible to save the model to a LinMix Command file (*.

Variance structure types Description Variance Components Number of Parameters ith. same as Unstructured. otherwise. j. same as No-Diagonal Factor Analytic Autoregressive (1) Heterogeneous Autoregressive (1) n×2 n × (t + 1) σ ρ 2 i–j i–j σi σj ρ See “Random coefficients” below to specify traditional variance components and/or to specify random coefficients. for others. 0 otherwise Table 14-11. otherwise.14 User’s Guide ber of bands. n×t n × b if b < t . n ) ∑ k=1 λ ik λ jk Banded No-Diagonal Factor Analytic (b) n × b ⁄ 2 ( 2t – b + 1 ) if b < t . the number of factors 1(expression) = 1 if expression is true. 354 . jth element WinNonlin n σ k 1 ( i = j ) and i corresponds to kth effect 2 Unstructured Banded Unstructured (b) if Toeplitz Banded Toeplitz Compound Symmetry Heterogeneous Compound Symmetry No-Diagonal Factor Analytic n × t × (t + 1) ⁄ 2 n × b ⁄ 2 ( 2t – b + 1 ) b < t . same as Toeplitz n×2 n × (t + 1) n × t(t + 1) ⁄ 2 σi–j 2 σ ij σ ij ( i – j < n ) σi–j + 11( 2 +1 i – j < n) σ1 + σ 1 ( i = j ) σ i σ j [ ρ1 ( i ≠ j ) + 1 ( i = j ) ] min ( i. or “Repeated measurements” on page 356 to set covariance structures for repeated measurements on subjects. otherwise.

” on page 354. 4. This variable must be a classification model term. All observations having the same level of the group effect have the same covariance parameters. Variance Blocking Variables (Subject): (Optional) Drag the appropriate variable to this field. Random Effects Model: Enter the model by typing it or by dragging the variable names and operators to be used in the model from the appropriate boxes to the Random Effects Model box using the mouse. 355 . or nested variables) should not appear in both. products.Linear Mixed Effects Modeling LinMix Wizard 14 Random coefficients The Random tab of the Variance structure dialog provides a means to incorporate random effects in a model. The model can be built from classification variables and/or regressors and the operators at the top of the dialog. The variable must be a classification model term. The random effects can be either classification or continuous. It can be used to specify traditional variance components and/or to specify random coefficients. The same term (single variables. Use parentheses to indicate nesting. 3. The options are explained in Table 14-11. 2. Note: The same variable cannot be used in both the fixed–effects specification and the random effects specification unless it is used differently—as part of a product. for instance. The default is off-no random intercept. Group: (Optional) This variable defines an effect specifying heterogeneity in the covariance structure. “Variance structure types. To specify random effects: 1. This is most commonly employed when a subject is specified in the Variance Blocking field. Other random models (tabs) can be added by clicking the Add Random button. Complete independence is assumed among subjects. Thus. The Repeated tab is used to specify covariance structures for repeated measurements on subjects. Each new level of the group effect produces a new set of covariance parameters with the same structure as the original group. Type: Select the type of covariance structure. Random Intercept: Check this box to indicate that the intercept is random. This Subject field identifies the subjects in a data set. 5. Subject produces a block diagonal structure with identical blocks.

(See “Fixed effects” on page 352 to select classification variables. 2. Click Next to proceed to the Contrasts or Calculate to run the analysis. Each new level of the group effect produces a new set of covariance parameters with the same structure as the original group. the same is true for nested terms. The repeated effect must contain only classification variables. To specify the repeated variance structure: WinNonlin 1. This Subject field identifies the subjects in a data set. 5. “Variance structure types.14 User’s Guide Repeated measurements The Repeated tab of the Variance structure dialog provides the means to specify the R matrix in the mixed model. 356 . All observations having the same level of the group effect have the same covariance parameters. It defines an effect specifying heterogeneity in the covariance structure. 4.” on page 354. Complete independence is assumed among subjects. Note: Interactions can be used as contrasts only if the interaction is a model term. 3. Variance Blocking Variables (Subject): (Optional) This must be a classification variable. Repeated Specification: enter the model by typing or by dragging the classification variable names and operators from the appropriate boxes to this field. Type: Select the type of covariance structure from the options explained in Table 14-11. Contrasts provide a mechanism for creating custom hypothesis tests. R 2 is assumed to be equal to σ I .) Further discussion of contrasts is provided in “Contrasts” on page 346. Contrasts The Contrasts dialog is next in the LinMix wizard after the Variance structure. Syntax rules for the repeated specification are the same as those for the fixed– effects model. Subject produces a block diagonal structure with identical blocks. Group: (Optional) This must be a classification variable with no regressors or operators. Contrasts can be computed only for classification variables. If no Repeated statement is specified.

If the contrast is not estimable. a nearby estimable function will be used instead. Note: If all values of the dependent variable are missing or excluded for a given level of the effect variable. Enter a descriptive label for the contrast in the Title cell. Once contrast #1 has been specified. The Show Coefficients of Contrasts check box ensures that the contrast is estimable. If the algorithm doesn’t seem to be using appropriate choices for the degrees of freedom. Once Contrast #1 is entered. Each contrast is tested independently of other contrasts. These columns are tested jointly. b. check User-specified [denominator] degrees of freedom and enter an integer for df greater than 1. a. then this level of the effect variable will not appear in the Contrast vector. Specify the Settings for the contrast entered. 357 . Effect variables may be either numeric or alphabetic. the grid automatically expands to allow for Contrast #2. d. 3. Number of columns for contrast: check this box to test multiple linearly independent combinations simultaneously. 4. but they may not be both. the settings for the contrast can be specified. Enter the appropriate contrast coefficients in the empty cells associated with Contrasts. up to a maximum of 100. 2. Note: The coefficients for single contrasts must sum to zero. the column cannot be used as an effect variable when building contrasts. If a column contains both numeric and alphabetic data. Drag the appropriate Model term to the cell labeled Effect. These settings are specific to the active contrast. This shows the actual coefficients used in the output. This grid will allow additional contrasts.Linear Mixed Effects Modeling LinMix Wizard 14 To specify contrasts: 1. They are specific to that contrast. The Estimability tolerance indicates the closeness of the zero vector before invoking the algorithm to make it estimable. c.

Click Next to proceed to the Estimates. 3. then the term Intercept is included with the model terms. Repeat as necessary for other terms within any one estimate. 4. Enter the appropriate coefficients in the empty cells associated with Effects. 7. drag another model term below the existing estimate. Note: Coefficients for estimates are not bound by the same limits as are those for Contrasts. Interaction terms and nested terms can be used if they appear in the model. Set the Confidence Level for confidence interval calculations. To enter estimates: WinNonlin 1. Drag the appropriate Model term to the cell labeled Effect. 8. The grid expands to display every unique level of that term. Enter the appropriate coefficients in the cells associated with the new term. it works like a regressor. Specify the Settings for the Estimate: a. The grid automatically expands up to the maximum number of estimates. Repeat as necessary for other estimates. Enter a descriptive label for the estimate in the cell labeled Title. only one number will be entered in the grid. Estimates The LinMix Estimates dialog is similar to the Contrasts dialog—except that multiple effects can be used in each estimate statement. 358 . 6. See also “Estimates” on page 349 and “The linear mixed effects model” on page 326. (Optional) To create compound estimates. or Calculate to run the analysis. If the fixed–effects model includes an intercept (the default). If the Intercept term is used as an effect for the estimate. The grid expands to include the new term and its levels.14 User’s Guide 5. 2. 5. Note: The marginal confidence interval is generated for each estimate.

no change is made. If the algorithm doesn’t seem to be using appropriate choices for the degrees of freedom. The Estimability tolerance indicates the closeness of the zero vector before invoking the algorithm to make it estimable. Checking Show Coefficients of Estimates displays in the output the actual coefficients used and. Click Next to proceed to the Least squares means. 359 . check User-specified [denominator] degrees of freedom and enter an integer for df greater than 1. b. estimable function. Least Squares Means can be computed for any classification model term. Checking Show Coefficients of LSM’s ensures that the contrast is estimable. c. (See “LinMix Wizard” on page 350. See “Fixed effects” on page 336 to set classification variables. a. If the algorithm doesn’t seem to be using appropriate choices for the degrees of freedom. They are estimates of what the mean values would have been had the data been balanced. the least squares means will be identical to the raw (observed) means. If the hypothesis is estimable. d. 3. That is. or Calculate to run the LinMix model. The Least Squares Means dialog follows the Estimates dialog in the LinMix wizard. See also “Least squares means” on page 344. these are the means predicted by the ANOVA model. this option will apply a nearby. If the test is not estimable.Linear Mixed Effects Modeling LinMix Wizard 14 b. c. Drag each model term for which the least square means are to be calculated up to the Least Squares Means box. applies a nearby. Least squares means Least squares means are generalized least-squares means of the fixed effects. To specify Least Squares Means: 1. estimable function. The linear mixed effects model terms are provided in the Model Classifiable Terms field.) 2. select the User-specified [denominator] degrees of freedom check box and enter an integer for df greater than 1. if the estimate is not estimable. If a data set is balanced. Set the Confidence Level for confidence interval calculations. The settings selected are applied to all terms. (Optional) Select the Settings for the Least Squares Means. 9.

When entering a multi-line title. 7. See “LinMix output” on page 361. 8. Check ASCII output to include text output in addition to workbook output. To save the model to a LinMix command file (*. reduced data matrix. If these are not specified. discussed under “Convergence criterion” on page 369. Title: Enter a title for the ASCII output (optional). (Optional) Select Intermediate Calculations to include the design matrix. Singularity Tolerance: (Optional) Enter the level. The title can include up to 5 lines. e.LML) for later re-use. This is the number of iterations in the Newton fitting algorithm. 6. 4. the application uses the method of moments estimates. 360 . (Optional) If the model has random effects. Convergence Criterion: (Optional) Specify the desired option. Satterthwaite computes the df base on χ approximation to distribution of variance. and final variance matrix in the ASCII output. Hessian. The Estimability tolerance indicates the closeness of the zero vector before invoking the algorithm to make it estimable. The columns in X and Z are eliminated if their norm ≤ this number. asymptotic covariance matrix of variance parameters. 2. Click OK. Degrees of freedom calculation form: (Optional) Residual is the same as in a 2 purely fixed effects model. The Compute Pairwise differences of LSM’s check box provides additional output with the differences of confidence intervals and LSM’s. use the Save Model button. 4. Click Next to proceed to the LinMix general options or Calculate to run the LinMix analysis. press ENTER to move to the next line.14 User’s Guide d. 9. 3. This function tests η 1 = η 2 . maximum iterations: (Optional) Enter the number of maximum iterations. select the Initial Variance Parameters button to edit a grid of the parameters. LinMix general options To specify LinMix general options: WinNonlin 1. 5.

t-statistic. See also “Computational details” on page 366. or Back to return to the Least squares means and other previous settings. Table 14-12. Click Calculate to run the LinMix analysis. F-statistic. but each term is tested last Estimates. Linear mixed effects worksheets Item Diagnostics Sequential Tests Partial Tests Final Fixed Parameters Final Variance Parameters Parameter Key Contrasts Estimates Contents Model-fitting information Tests of fixed effect for significant improvement in fit.Linear Mixed Effects Modeling LinMix output 14 10. in sequence. and confidence interval 361 . pvalue. including all output as well as analysis settings and any errors that occur. LinMix output LinMix produces the following text and/or workbook output. Depending which model options have been set up. standard error. and confidence intervals Estimates and units of the final variance parameters Information about the variance structure Hypothesis. Tests of fixed effects. Linear mixed effects text This is the optional ASCII output selected in the LinMix general options. p-value. Linear mixed effects workbook The Linear Mixed Effects Workbook contains summary tables of the output. Specific output depends on the options set in the LinMix Wizard (see “LinMix Wizard” on page 350). workbook may contain the items listed in Table 14-12. standard error. It contains a complete summary of the analysis. Details for specific output follow below. t-statistic. see “LinMix output” on page 361. df. For details on ASCII and workbook output. Depends upon the order that model terms are written. and p-value Estimate.

Then it is tested whether the third model term should enter. standard error. and confidence interval Dependent variable(s). The tests are performed by finding appropriate L matrices and testing the hypotheses H0: Lβ = 0 for each model term. This is computed by modifying the basis created by the QR factorization to yield a basis that more closely resembles that found in balanced data. partial tests are invariant under the order in which model terms are listed in the Fixed Effects dialog. p-value. given that the first term is in the model. This is computed using a QR factorization of the XY matrix.14 User’s Guide Table 14-12. Partial tests Partial tests test each model term given every other model term. The QR factorization is segmented to match the number of columns that each model term contributes to the X matrix. Linear mixed effects worksheets (continued) Item Least Squares Means Residuals User Settings Initial Variance Parameters Iteration History Contents Least squares means. Sequential tests The sequential tests test each model term sequentially. Then it is tested whether the second model term should enter the model. 362 . Unlike sequential tests. It does this by factoring out of each model term the contribution attributable to the remaining model terms.. t-statistic. etc. First it is tested whether the first model term should enter the model. standard error and residuals Information about model settings Initial values for variance parameters Data on the variable estimation in each iteration WinNonlin Tests of hypotheses The tests of hypotheses are the mixed model analog to the analysis of variance in the fixed model case. units. given that the first two terms. until all model terms are exhausted. observed and predicted values. df.

. The mixed effects tests have similar properties. s = rank(X) + dim(θ) ).e. The partial ANOVA is designed to eliminate the contamination problem.Linear Mixed Effects Modeling LinMix output 14 Discussion For fixed effects models.e. For the sequential ANOVA. 363 . r is the rank of the fixed effects design matrix X. some terms in ANOVA may be contaminated with undesired effects. and s is the rank of the fixed effects design matrix X plus the number of parameters in θ (i.. However. the sums of squares are statistically independent. the sum of the individual sums of squares is equal to the model sum of squares. Schwarz’s Bayesian Criterion (SBC) The Linear Mixed Effects module uses the smaller-is-better form of Schwarz’s Bayesian Criterion: SBC = – 2L R + s log ( N – r ) where LR is the restricted log-likelihood function evaluated at the final estiˆ ˆ mates β and θ . the ANOVA represents a partitioning of the model sum of squares. certain properties can be stated for the two types of ANOVA.e. Akaike’s Information Criterion (AIC) The Linear Mixed Effects module uses the smaller-is-better form of Akaike’s Information Criterion: AIC = –2LR + 2s where LR is the restricted log-likelihood function evaluated at final fixed ˆ ˆ parameter estimates β and the final variance parameter estimates θ . but the sums of squares are correlated and do not add up to the model sums of squares. and s is the rank of the fixed effects design matrix X plus the number of parameters in θ (i. i. Also. s = rank(X) + dim(θ)).. N is the number of observations used.

The variance of the prediction is as follows. This is controlled on the Fixed Effects page of the wizard. then the predicted values are also transformed. T-type confidence intervals for the predicted values are also computed.14 User’s Guide Hessian eigenvalues The eigenvalues are formed from the Hessian of the restricted log likelihood. the residuals are the transformed values minus the predicted values. the user should supply the desired points in the input data set with missing dependent variables. Predicted values and residuals In the Linear Mixed Effects module. Positive eigenvalues indicate that the final parameter estimates are found at a maximum. and therefore the predicted values ˆ ˆ ˆ for the input data are equal to y = Xβ . Predicted values are estimated by using the rows of the fixed effects design matrix X as the L matrices for estimation. Note that if the user requested a log-transform. The Lower_CI and Upper_CI are the confidence bounds on the predicted response. Fixed effects parameter estimates In the Linear Mixed Effects module. or in the case of transforms. predicted values for the dependent variables are computed at all of the input data points. The confidence level is given in Conf_Level. Tˆ T ˆ Var ( y ) = X ( X V X ) X –1 WinNonlin The standard errors are the square root of the diagonal of the matrix above. where β are the final fixed parameters estimates. The residuals are then the observed values minus the predicted values. To obtain predicted values for data points other than the observed values. There is one eigenvalue per variance parameter. the final fixed effects parameter estiˆ mates β are computed by solving ˆ T00 β = y0 364 .

This provides the opportunity to inspect what combination is actually used to see if it represents the intentions. Otherwise. two indices are appended to the variance parameter names in order to clarify which random or repeated model. The second index is the group index indicating that the parameters with the same group index are for the same group level. or group. The first index indicates the random model or repeated model to which the parameter applies. and β are the final fixed parameters estimates. the variance parameter addresses. The highest index indicates the repeated model if there was one. a convention is followed such that if any column of T00 is a linear combination of preceding columns. then zero iterations will be displayed.Linear Mixed Effects Modeling LinMix output 14 where T00 and y0 are elements of the reduced data matrix as defined in the ˆ section on Restricted Maximum Likelihood Estimation. When there are multiple solutions. then the estimable replacement will be displayed. Source Random if the parameter is from a random model Repeated if the parameter is from a repeated specification Assumed if the variance parameter is the residual Variance type specified by the user. or Identity. etc. Coefficients If Display Coefficients is checked on any of the page of the LinMix wizard. An index of 1 indicates the model on the Random 1 tab in the user interface. the coefficients of the parameters used to construct the linear combination will be displayed. then indices are not used in the parameter names. If the initial estimates are the REML estimates. then the coefficients will be the same as those entered. If the linear combination is estimable. If the Group option is not used in the model or if there is only one random or repeated model. if the residual variance 365 Type . Iteration history This worksheet holds the fitting algorithm results. Parameter key If the model included random or repeated effects. If the linear combination is not estimable. the Parameter column of this grid contains the variance parameter names assigned by LinMix. then the correˆ sponding element of β is set to zero.

REML would be the same as ML. and not a function of the fixed effects parameters. if there was a subject Group term for this variance structure. In contrast. if there was a group WinNonlin Computational details Restricted maximum likelihood For linear mixed effects models that contain random effects.14 User’s Guide Bands Subject_Term Group_Term Group_Level Number of bands or factors if appropriate for the variance type Subject term for this variance structure. Let θ be a vector consisting of the variance parameters in G and R. restricted maximum likelihood estimation (REML) maximizes a likelihood that is only a function of the variance parameters θ and the observations y. Hence for models that do not contain any fixed effects. The linear mixed effects model described under “The linear mixed effects model” on page 326 is: y = Xβ + Zγ + ε. V = Variance(y) = ZGZT + R. To obtain the restricted likelihood function in a form that is computationally efficient to solve. the first step in model analysis is determining the maximum likelihood estimates of the variance parameters associated with the random effects. This is accomplished in LinMix using Restricted Maximum Likelihood (REML) estimation. if there was a group Level of the group that this parameter goes with. the data matrix [X | Z | y] is reduced by a QR factorization to an upper triangular matrix. The full maximum likelihood procedure (ML) would simultaneously estimate both the fixed effects parameters β and the variance parameters θ by maximizing the likelihood of the observations y with respect to these parameters. Let [R0 | R1 | R2] be an orthogonal matrix such that the multiplication with [X | Z | y] results in an upper triangular matrix: R0 T T 00 T 01 y 0 0 T 11 y r 0 0 ye T R1 [ X Z y ] = R2 366 T .

For some balanced problems. ye)) = –ne/2 log(θe) – ½ (yeTye / θe) Vr = Variance(yr) θe = residual variance ne = residual degrees of freedom. and since yr and ye are independent. and ignores y0. see Corbeil and Searle (1976). [ y e1 y e2 … y en ] . the negative of the log of the restricted likelihood function is the sum of the separate negative log-likelihood functions: log(L(θ.y e ) ) = -.∑ y ei V y ei 2 2 i=1 n where V is the dispersion matrix for the multivariate normal distribution. and N = number of observations used. For more information on REML. Since ye has a distribution that depends only on the residual variance.Linear Mixed Effects Modeling Computational details 14 REML estimation maximizes the likelihood of θ based on yr and ye of the reduced data matrix. in order to increase the speed of the program and then the log-likelihood can be computed by the equivalent form: n 1 T –1 –1 log ( L ( θ . ye)) – (N–rank(X))/2 log(2π) where: log(L(θ. 367 . ye is treated as a multivariate residual. yr)) + log(L(θ. ye)) = log(L(θ. yr.log ( V ) – -. yr)) = –½ log (|Vr |) – ½ yr T Vr –1 yr log(L(θ.

or when it is unable to continue. Unstructured. a line search is done as described in Fletcher (1980) to find an approximate minimum along the direction vector.).. where H–1 is the inverse of the Hessian of the objective function. and g is the gradient of the objective function. The algorithm stops when a convergence criterion is met. a modification of Newton’s algorithm is used to minimize the negative restricted log-likelihood function with respect to the variance parameters θ. The algorithm needs a starting point. the first and second derivatives of the variance can WinNonlin 368 . Another modification to Newton’s algorithm is that. etc. the objective function used in the algorithm is the negative of the restricted log-likelihood without the constant term: objective function = – log(L(θ. Note from the discussion on REML estimation that the restricted log-likelihood is a function of the variance matrix Vr and the residual variance. ensuring a direction of descent: θi+1 = θi – ( Hi + Di )–1 gi The appropriate Di matrix is found by a modified Cholesky factorization as described in Gill. The gradient and Hessian of the objective function with respect to θ are therefore functions of first and second derivatives of the variances with respect to θ. it finds the next point using: θi+1 = θi – H–1(θi) g(θi). hence yielding the variance parameters that maximize the restricted likelihood. Variance Components. Murray. In the Linear Mixed Effects module. Since the variances can be determined through the types specified for G and R (e. ye)) Newton’s algorithm is modified by adding a diagonal matrix to the Hessian when determining the direction vector for Newton’s. once the direction vector d = – ( Hi + Di )–1 gi has been determined. so that the diagonal matrix plus the Hessian is positive definite.14 User’s Guide Newton’s Algorithm for maximization of restricted likelihood As a general review. and then at each iteration. Newton’s algorithm is an iterative algorithm used for finding the minimum of an objective function.g. Since a constant term does not affect the minimization. and Wright. yr)) – log(L(θ. which are in turn functions of the parameters θ.

LinMix will determine initial values. 0.01 for correlations. Sometimes the method of moments will yield estimates of θ that are REML estimates. indicating successful convergence at a local—and hopefully global—minimum. The algorithm converged.g. indicating that the model may be over-specified. Convergence criterion The convergence criterion used in the Linear Mixed Effects module is that the algorithm has converged if: gT (θ) H–1(θ) g(θ) < ι. It may be in a trough. via the Initial Variance Parameters… button. or on a flat surface. • 369 . where ι is the convergence criterion specified on the General Options page of the wizard. For variance structures that are linear in θ (Variance Components. In this case the program will not need to iterate. 1. Newton's algorithm converged with a modified Hessian. The output is suspect. LinMix will use values that should yield a reasonable variance matrix. Compound Symmetry. and hence the gradient and Hessian are evaluated in closed-form in the optimization algorithm. then LinMix will calculate a “sample” variance by solving for random effects and taking their sums of squares and cross-products. If the variance structure is nonlinear. General Options page. The default value is 1×10–10. Starting values Newton’s algorithm requires initial values for the variance parameter θ. and the system of equations will be solved to obtain initial parameter estimates. The possible convergence outcomes from Newton’s algorithm are: • Newton's algorithm converged with a final Hessian that is positive definite.0 for variance components and standard deviations. More often. e. The sample variance will then be equated with the parametric variance. at a saddle point. If either method fails. The user can supply these in the LinMix Wizard.Linear Mixed Effects Modeling Computational details 14 be evaluated in closed-form. and Toeplitz). the initial values are determined by the method of moments.. but not at a local minimum. Unstructured.

Intermediate variance matrix is not positive definite. This indicates an invalid starting point for the algorithm. The Wald statistic is asymptotically distributed under the null hypothesis as a chi-squared random variable with q = rank(L) degrees of freedom. WinNonlin Satterthwaite’s approximation for degrees of freedom The linear model used in linear mixed effects modeling is: Y ~ N(Xb. Wald (1943) developed a method for testing hypotheses in a rather general class of models. V) (10) where X is a matrix of fixed known constants. The algorithm cannot continue. Their methodology applies to variance component models with unbalanced data. and V is the variance matrix dependent on other unknown parameters. LinMix uses a method due to Allen (2001) for finding an F distribution to approximate the distribution of the Wald statistic. Using the chi-squared approximation results in a test procedure that is too liberal. Allen derived an extension of their technique to the general mixed model. Giesbrecht and Burns (1985) suggested a procedure to determine denominator degrees of freedom when there is one numerator degree of freedom. b is an unknown vector of constants. To test the hypothesis H0: Lb = 0 in (10). the Wald statistic has an exact F-distribution. ˆ 370 .14 User’s Guide • • • Initial variance matrix is not positive definite. In common variance component models with balanced data. Failed to converge in allocated number of iterations. L must be a matrix such that each row is in the row space of X. the Wald statistic is: T ˆ –1 –1 T –1 ˆ T ˆ ( Lβ ) [ L( X V X ) L ] ( Lβ ) (11) ˆ where β and V are estimators of β and V.

Assuming that the u-th term is distributed F(1. then: ˆ L(X V X) L 2 -------------------------------------.= -Var –1 T –1 ν L(X V X) L The approach is to use (13) to solve for ν.= -------------------------------------. the Wald statistic can be re-expressed as: T ˆ –1 –1 T ˆ 2 ˆ 2 L(X V X) L (Lβ) (Lβ) -------------------------------------. The distribution of the numerator is approximated by a chi-squared random variable with 1 df.∼ χ .Linear Mixed Effects Modeling Computational details 14 A one degree of freedom (df) test is synonymous with L having one row in (11). If: ˆ L(X V X) L ν -------------------------------------.-------–1 T –1 ν L( X V X) L 2 T –1 –1 T is true. When L has a single row. and λu is the u-th diagonal of Λ. Each term in (14) is similar in form to (12). νu). the expected value of W is: 371 .⁄ -------------------------------------T –1 –1 T T –1 –1 T ˆ –1 –1 T L(X V X) L L(X V X) L L(X V X) L (12) Consider the right side of (12). The objective is to find ν such that the denominator is approximately distributed as a chi-squared variable with ν df. Find an orthogonal matrix R and a diagonal matrix Λ such that: T ˆ –1 –1 T T L ( X V X ) L = RΛR T –1 –1 T (13) The Wald statistic (11) can be re-expressed as: q ˆ ˆ W = ( R Lβ ) Λ (R Lβ ) = T –1 T T ∑ u=1 ˆ ( ru L β ) ------------------ 2 (14) λu where ru is the u-th column of R.

by default. Prints “Not estimable” Not estimable parameters 372 . If a variable is not estimable. and between LinMix and PROC MIXED in SAS.’ LinMix provides the option to write out 0 instead. LinMix and ANOVA Table 14-13. LinMIx compared to WinNonlin ANOVA Feature Degrees of freedom Partial tests of model effects LinMix Uses Satterthwaite approximation for all degrees of freedom. User can also select Residual. Equating expected values and solving for ν. its expected value is qν/(ν-2). one obtains: ν = 2 E{W} / (E{W} – q) Comparing LinMix to ANOVA and SAS’s PROC MIXED This section outlines differences between LinMix and the previous WinNonlin ANOVA module. The partial tests in LinMix are not equivalent to As SAS’s PROC GLM Type III the Type III method in SAS though they coincide in most situations. ANOVA Uses only Residual.14 User’s Guide q WinNonlin E( W) = u–1 ------------∑ νu – 2 νu Assuming W is distributed F(q. See “Partial tests” on page 362. LinMix outputs ‘Not estimable. ν).

Convergence cri. and repeated statements. 373 . and hence will produce different final parameter estimates. In particular. Multiple factors on the same random command REML By default. This is how it is done in LinMix. where g is the gradient and H is the Hessian of the negative log-likelihood. put them on the same random tab. and the likelihood calculation method in LinMix also differs from SAS. It will determine when the variance is over-specified. This often occurs when the variance is over-specified (i. by default. Modified likelihood LinMix always tries to keep the final Hessian (and hence the variance matrix) positive definite. The LinMix criterion is equivalent to SAS with the ‘convh’ and ‘absolute’ options set. A userspecified variance structure takes priority over that supplied by LinMix. LinMIx compared to SAS PROC MIXED Feature Degrees of freedom LinMix Uses Satterthwaite approximation for all degrees of freedom. SAS allows the final Hessian be non positive definite. too many parameters in the model). To be consistent with LInMix. random. set the SAS ‘ddfm=satterth’ option. SAS modifies the REML by ‘profiling out’ residual variance. In addition. SAS uses the bigger-is-better form. In LinMix. AIC and SBC are functions of the likelihood. The default convergence criterion options in SAS are ‘convh’ and ‘relative’.Linear Mixed Effects Modeling Comparing LinMix to ANOVA and SAS’s PROC MIXED 14 LinMix and PROC MIXED Table 14-14. If independent. SAS sets the residual to some other number. the residual is not estimable. on separate tabs.e. effects associated with the matrix columns from a single random statement are assumed to be correlated per the variance specified. LinMix –log(likelihood) may not equal twice SAS ‘–2 Res Log Like’ if noprofile is not set. so LinMix and SAS will differ for models with multiple terms on a random statement using the VC structure.. does maximum likelihood estimation on the residual vector. PROC MIXED PROC MIXED chooses different defaults based on the model. and set some parameters to zero. set the ‘noprofile’ option on the PROC MIXED statement.The convergence criterion is gT H–1g < tolteria erance with H positive definite. Akaike and LinMix uses the smaller-is-better form of Schwarz’s criteria AIC and SBC in order to match WinNonlin and WinNonMix. as defined in Corbeil and Searle (1976). To make LinMix and SAS® consistent. Regardless of variance structure: If random effects are correlated. The SAS residual needs to be added to the other SAS variance parameter estimates in order to match the LinMix output. except that it does not follow this rule in the case of Variance Components. SAS behaves similarly. REML (Restricted Maximum Likelihood) estimation. User can also select Residual. Saturated variance structure If the random statement saturates the variance. LinMix sets the residual variance to zero.

CSH variance structures If a variable is not estimable. and Dii the sum of squares of the i-th column of the QR factorization. The Type III results can change depending on how one specifies the model. a time variable model.For VC structure. LinMix outputs SAS writes the value of 0. LinMix uses the standard deviations for the SAS prints the variances in the output for final parameters. Group as classifi.LinMix requires variables used for the subcation variable ject and group options to be declared as classification variables. Singularity tolerance Define dii as the i-th diagonal element of the QR factorization of X. force individual variances > 0. The i-th column of X is deemed linearly dependent on the proceeding columns if dii2 < (singularity tolerance)×Dii. ‘Not estimable. If intercept is present. Classification variables are treated similarly. For details. If the absolute diagonal element |dii| < (singularity tolerance)×Dii. consult the SAS manual or contact support@pharsight. 374 . but requires data to be sorted and considers each new regressor value to indicate a new subject or group. and the parameter estimates are flagged as biased. Repeated withLinMix requires an explicit repeated model. but will not nent.14 User’s Guide Table 14-14. See “Partial tests” on page 362. LinMix keeps the final If SAS estimates a negative variance compotive variance matrix positive definite.com.’ LinMix provides the option to write out 0 instead. and prints the standard devia. Not estimable parameters ARH. where Dii is the original diagonal of XTX. the first row of Dii is omitted.To run some SAS examples. Define dii as the diagonal element of the XTX matrix during the sweeping process. then the parameter associated with the column is set to 0.parameters. and assumes data are sorted appropriately. It instead creates an implicit time field able would need to be added. it outputs 0. Partial tests of model effects The partial tests in LinMix are not equivalent to the Type III method in SAS though they coincide in most situations. SAS doesn’t require an explicit repeated out any time vari. SAS allows regressors. terms of standard deviations. though the model is expressed in tions in the output. LinMIx compared to SAS PROC MIXED (continued) Feature LinMix PROC MIXED WinNonlin Keeping VC posi.

bioequivalence comparisons rely on a criterion. • • • • • • “Introduction to bioequivalence” on page 375 “The model and data” on page 378 “Average bioequivalence” on page 379 “Individual and population bioequivalence” on page 387 “Bioequivalence Wizard” on page 395 “Bioequivalence output” on page 403 Introduction to bioequivalence Basic terms Bioequivalence is said to exist when a test formulation has a bioavailability that is similar to that of the reference. Defined as relative bioavailability. Further details about WinNonlin bioequivalence are provided under the following headings. where the test and reference products can vary.Chapter 15 Bioequivalence Calculating average. and population bioequivalence WinNonlin includes a Bioequivalence Wizard that calculates average or population/individual bioequivalence. 375 . and calculation of a confidence interval for that criterion. There are three types of bioequivalence: Average. and Population. Individual. Although bioavailability and bioequivalence are closely related. depending upon the comparison to be performed. individual. a predetermined bioequivalence limit. bioequivalence involves comparison between test and reference drug products.

The objective of a bioequivalence study is to determine whether bioequivalence exists between a test formulation and reference formulation and to hence identify whether the two formulations are pharmaceutical alternatives that can be used interchangeably to achieve the same effect. population and individual bioequivalence. with random assignment to the two possible sequences of drug product administration. In addition to specifying this general approach. usually 80125% for the ratio of the product averages. Further. This approach is termed average bioequivalence and involves the calculation of a 90% confidence interval for the ratio of the averages of the test and reference products. the 1992 guidance also provided specific recommendations for (1) logarithmic transformations of pharmacokinetic data. To establish bioequivalence. the calculated confidence interval should fall within a bioequivalence limit. The US FDA Guidance for Industry (January 2001) Statistical Approaches to Establishing Bioequivalence recommends that average bioequivalence be supplemented by two new approaches. the 1992 guidance recommended that statistical analysis for pharmacokinetic parameters. such as area under the curve (AUC) and peak concentration (Cmax) be based on a test procedure termed the two one-sided tests procedure to determine whether the average values for pharmacokinetic parameters were comparable for test and reference formulations. Bioequivalence Bioequivalence between two formulations of a drug product. sometimes referred to as relative bioavailability. For example. and (3) methods to evaluate outlier data. indicates that the two formulations are therapeutically equivalent and will provide the same therapeutic effect. Population and individual bioequivalence include comparisons of both the averages and variances of the study measure. (2) methods to evaluate sequence effects. Individual bioequivalence assesses within-subject variability for the test and reference products as well as the subject-by-formulation interaction. the Food and Drug Administration (FDA) usually does not require a new drug 376 WinNonlin . to get approval to market a generic drug product.15 User’s Guide The July 1992 FDA guidance on Statistical Procedures for Bioequivalence Studies Using a Standard Two-Treatment Crossover Design recommended that a standard in vivo bioequivalence study design be based on the administration of either single or multiple doses of the test and reference products to healthy subjects on separate occasions. Population bioequivalence assesses the total variability of the measure in the population. Bioequivalence studies are important because establishing bioequivalence with an already approved drug product is a cost-efficient way of obtaining drug approval.

All types of bioequivalence comparisons are based on the calculation of a criterion (or point estimate). These added approaches are needed because average bioequivalence uses only a comparison of averages. The individual bioequivalence approach assesses within-subject variability for the test and reference products as well as the subject-by-formulation interaction. The concepts of population and individual bioequivalence are related to two types of drug interchangeability—prescribability and switchability. the confidence interval. Bioequivalence studies are also useful for testing new formulations of a drug product. Three different types of bioequivalence have been established by the FDA. and individual bioequivalence.Bioequivalence Introduction to bioequivalence 15 application submission if the generic drug company can provide evidence of bioequivalence between the generic drug product and the innovator product.” The FDA recommended administration of either single or multiple doses of the test and reference formulations to subjects. To establish bioequivalence. Both the confidence interval approach and the two one-sided t-tests are described further below. The FDA also recommended an equivalent approach using two. population and individual bioequivalence. A procedure for establishing average bioequivalence was recommended by the FDA in the 1992 guidance on “Statistical Procedures for Bioequivalence Studies Using a Standard Two-Treatment Crossover Design. These are average bioequivalence. of the bioequivalence measures. and on a predetermined acceptability limit for the confidence interval. and to determine the confidence interval for the difference. on the calculation of a confidence interval for the criterion. new routes of administration of a drug product. In contrast. and not the variances. The guidance then recommended that a statistical analysis of pharmacokinetic parameters such as AUC and Cmax be done on a log-transformed scale to determine the difference in the average values of these parameters between the test and reference data. The January 2001 guidance “Statistical Approaches to Establishing Bioequivalence” recommends that average bioequivalence be supplemented by two new approaches. one-sided t-tests. usually calculated at the 90% confidence level. and for a drug product that has changed after approval. should fall within a predetermined acceptability limit. population bioequivalence. Drug pre377 . The population bioequivalence approach assesses the total variability of the study measure in the population. population and individual bioequivalence approaches include comparisons of both the averages and variances of the study measure. with random assignment to the possible sequences of drug administration. usually within 20% of the reference average.

the Bioequivalence Wizard is based on a mixed effects model. 378 . If the data have many missing observations. such as when switching from an innovator drug product to a generic substitute within the same patient. See “The linear mixed effects model” on page 326. Missing data For population and individual bioequivalence. see the US FDA Guidance for Industry (August 1999). Variance structures The variance structures available in the Bioequivalence Wizard are the same as those available in the Linear Mixed Effects Modeling wizard.).*. Variable name limits for use in WinNonlin bioequivalence analyses. . the program assumes complete data for each subject. /. that subject is not included in the analysis. Finally. variable names may not have single or double quotes in them. See “Variance structure” on page 353. =. They cannot include spaces or the following operational symbols: +. They also cannot include parentheses (). For more details on the models required for different kinds of bioequivalence. Drug switchability is usually assessed by individual bioequivalence.15 User’s Guide scribability refers to when a physician prescribes a drug product to a patient for the first time. variable (column) names must begin with a letter. After the first character. Drug prescribability is usually assessed by population bioequivalence. consider imputing estimated values to produce complete records per subject. question marks (?) or semicolons (. and must choose from a number of bioequivalent drug products. This program does not impute missing values. valid characters include numbers and underscore (my_file_2). Drug switchability refers to when a physician switches a patient from one drug product to a bioequivalent product. WinNonlin The model and data Linear model Like the Linear Mixed Effects Modeling wizard. If a subject has a missing observation.

for the two formulations T and R. Crossover designs are further broken down into replicated and nonreplicated designs. For example. Replicated crossover designs should be used for individual bioequivalence studies. The FDA recommends that a bioequivalence study should use a crossover design unless a parallel or other design can be demonstrated to be more appropriate for valid scientific reasons. and parallel. Period 1 T R Period 2 R T Sequence 1 Sequence 2 Recommended models for average bioequivalence The recommended model to be used in average bioequivalence studies depends on the type of study design that was used – parallel. then the study is considered a replicated design. if all formulations have at least one subject given more than one dose of that formulation. Replicated designs involve multiple doses. nonreplicated crossover. subjects receive only one dose of the test formulation and only one dose of the reference formulation. 379 . In the Bioequivalence Wizard. each subject receives only one formulation in randomized fashion. In nonreplicated designs. In a parallel design. a 2×3 crossover design as described in the table below is a replicated crossover design. and can be used for average or population bioequivalence analysis.Bioequivalence Average bioequivalence 15 Average bioequivalence Study designs The most common designs for bioequivalence studies are replicated crossover. or replicated crossover. nonreplicated crossover. whereas in a crossover design each subject receives different formulations in different time periods. Period 1 T R Period 2 R T Period 3 T R Sequence 1 Sequence 2 An example of a nonreplicated crossover design is the standard 2x2 crossover design described in the 1992 FDA guidance and in the table below.

Fixed effects model is: • DependentVariable = Intercept + Sequence + Formulation + Period Random effects model is the nested term: • Subject(Sequence) There is no repeated specification. Using Subject as a random effect this way. 380 . the default model is as follows. the time variable is made explicit as Period. the correct standard errors will be computed for sequence means and tests of sequence effects. Nonreplicated crossover designs For nonreplicated crossover designs. but using maximum likelihood instead of method of moments to estimate inter-subject variance. so the default error model ε ~ N(0. This is equivalent to the classical analysis method.15 User’s Guide Replicated crossover designs For replicated crossover designs. σ2 I) will be used. the default model used in the Bioequivalence Wizard is the example model given for replicated designs in Appendix E of the FDA guidance: Fixed effects model is: • DependentVariable = Intercept + Sequence + Formulation + Period WinNonlin Random effects model is: • • • Treatment Variance Blocking Variable set to Subject Type set to Banded No-Diagonal Factor Analytic with 2 factors Repeated specification is: • • • • Period Variance Blocking Variables set to Subject Group set to Treatment Type set to Variance Components Rather than using the implicit repeated model as in Appendix E.

one must construct pseudo-F tests by hand to accomplish the same task. the first step is the computation of the least squares means (LSM) and standard errors of the test and reference formulations and the standard error of the difference of the test and reference least squares means. If.1 ) Intrasubject CV = sqrt( exp(Residual_Variance) . Fixed effects model is: • Formulation There is no random model and no repeated specification. These tables are also included in the text output. the user may supplement or modify the model to suit the analysis needs of the data set in consideration. Parallel designs For parallel designs. Note: In each case. F-statistic and p-value. Mean Squares (MS).Bioequivalence Average bioequivalence 15 Using a fixed effect model. and Residual_Variance is the intrasubject (within subject) variance. which contain the degrees of freedom (DF). for this default model. These 381 . Least squares means In determining the bioequivalence of a test formulation and a reference formulation. Var(Subj(Seq)) is the intersubject (between subject) variance. Note that the F-statistic and p-value for the Sequence term are using the correct error term since Subject(Sequence) is a random effect.1 ) Note that. Sum of Squares (SS). in addition to using the default model. for each of the model terms. the data is also ln-transformed. so the residual error term is included in the model. When this default model is used for a standard 2x2 crossover design. then the Final Variance Parameters tab will contain two additional parameters: Intersubject CV = sqrt( exp(Var(Subj(Seq)) . the default model is as follows. WinNonlin creates two additional tabs in the output called Sequential SS and Partial SS.

DiffSE = standard error of the difference in LSM (computed by LinMix). See also “Least squares means” on page 344. the ratio is obtained on arithmetic scale by exponentiating. Ratio(%Ref) = 100 (TestLSM / RefLSM) For ln-transform or data already ln-transformed. For ln-transform or data already ln-transformed. Ratio(%Ref) = 100exp(Difference) Similarly for log10-transform or data already log10-transformed. To simplify the notation for this section. fractionToDetect = (user-specified percent of reference to detect) / 100. Difference = TestLSM – RefLSM The ratio calculation depends on data transformation. For non-transformed.15 User’s Guide quantities are computed by the same process that is used for the Linear Mixed Effects module. TestLSM = test least squares mean (computed by LinMix). RefGeoLSM = 10RefLSM TestGeoLSM = 10TestLSM The difference is the test LSM minus the reference LSM. the ratio is Ratio(%Ref) = 100 · 10Difference 382 . WinNonlin The geometric LSM are computed for transformed data. RatioSE = standard error of the ratio of the least squares means. let: • • • • • RefLSM = reference least squares mean (computed by LinMix). RefGeoLSM = exp(RefLSM) TestGeoLSM = exp(TestLSM) For log10-transform or data already log10-transformed.

in which case the Bioequivalence module will print “Not Estimable. 95. and for the confidence level that the user gave on the bioequivalence tab if that value is different than 80. for example. 90. CI_Lower = 100 · 10lower CI_Upper = 100 · 10upper For no transform. df ⎠ ( TestLSM – RefLSM ) DiffSE RefLSM – t ( 1 – α ⁄ 2 ) . lower = Difference – t (1-alpha/2). CI_Lower = 100 · exp(lower) CI_Upper = 100 · exp(upper) For log10-transform or data already log10-transformed. Computing these intervals is an iterative procedure for which the NelderMead simplex algorithm is used. CI_Lower = 100 (1+lower/RefLSM) = 100 ⎛ 1 + ----------------------------------------------------. 90. and translated to percentages. or 95.” 383 .df · DiffSE upper = Difference + t (1-alpha/2). where alpha = (100 – confidenceLevel) / 100. Chow and Liu (2000). Similarly. To compute the classical confidence intervals.Bioequivalence Average bioequivalence 15 Classical and Westlake confidence intervals Output from the Bioequivalence module includes the classical confidence intervals and Westlake confidence intervals for confidence levels equal to 80. first the following values are computed using the students-t distribution. CI_Upper = 100 (1 + upper / RefLSM) The procedure for computing the Westlake confidence intervals is given in many statistical references.⋅ RatioSE⎞ ⎝ RefLSM – t ( 1 – α ⁄ 2 ) . page 86. It is possible that there will not be convergence. df RefLSM⎠ = where the approximation RatioSE = DiffSE / RefLSM is used.⋅ --------------------⎞ ⎝ TestLSM 100 ⎛ ---------------------------------------------------.df · DiffSE These values are then transformed if necessary to be on the arithmetic scale. See. For ln-transform or data already ln-transformed.

the CI_Lower and CI_Upper for the user-specified value of the confidence level are compared to the following lower and upper bounds. the first t-test is a left-tail test of the hypotheses: H0: Difference < ln(1 – fractionToDetect) (bioinequivalence for left test) H1: Difference ≥ ln(1 – fractionToDetect) (bioequivalence for left test) The test statistic for performing this test is: t1 = ( (TestLSM – RefLSM) – ln(1 – fractionToDetect) ) / DiffSE The p-value is determined using the t-distribution for this t-value and the degrees of freedom. If the p-value is <0. CI_Upper) is contained within LowerBound and UpperBound. Note that the upper bound for log10 or ln-transforms or for data already transformed is adjusted so that the bounds are symmetric on a logarithmic scale. Two one-sided T-tests For ln-transform or data already ln-transformed. For log10-transform or data already log10-transformed. To conclude whether bioequivalence has been achieved. average bioinequivalence has been shown. If the interval (CI_Lower. then the user can reject H0 at the 5% level. UpperBound). WinNonlin 384 .15 User’s Guide The bioequivalence conclusion that appears in the text output depends on the user-specified values for the confidence level and for the percent of reference to detect. these options are entered on the bioequivalence options tab of the Bioequivalence wizard. CI_Upper) is completely outside the interval (LowerBound. i. LowerBound = 100 – (percent of reference to detect) UpperBound (ln-transforms) = 100 exp( –ln(1 – fractionToDetect) ) = 100 (1 / (1 – fractionToDetect)) UpperBound (log10-transforms)=100 · 10^( –log10(1. the module has failed to show bioequivalence or bioinequivalence. the first test is done similarly using log10 instead of ln.05.e.0–fractionToDetect)) = 100 (1 / (1 – fractionToDetect)) UpperBound (no transform) = 100 + (percent of reference to detect) If the interval (CI_Lower. average bioequivalence has been shown. Otherwise. less than a 5% chance of rejecting H0 when it was actually true.

since ln(0. For data with no transformation. the first test is (where Ratio here refers to Ratio(%Ref) / 100 ): H0: Ratio < 1 – fractionToDetect H1: Ratio Š 1 – fractionToDetect The test statistic for performing this test is: t1 = [ (TestLSM / RefLSM) – (1 – fractionToDetect) ] / RatioSE where the approximation RatioSE = DiffSE / RefLSM is used. The two one-sided t-tests procedure is 385 . but for ln-transformed data. then the lefttail test is Pr(<80%). if the percent of reference to detect is 20%. For ln-transform or data already ln-transformed. the maximum of these p-values.25). the p-value for the second test above. the test will be symmetric on a logarithmic scale. the second test is: H0: Ratio > 1 + fractionToDetect H1: Ratio £ 1 + fractionToDetect The test statistic for performing this test is: t2 = ( (TestLSM / RefLSM) – (1 + fractionToDetect) ) / RatioSE where the approximation RatioSE = DiffSE / RefLSM is used.Bioequivalence Average bioequivalence 15 For data with no transformation. The second t-test is a right-tail test that is a symmetric test to the first.8) = – ln(1. and total of these p-values. the second test is done similarly using log10 instead of ln. The output for the two one-sided t-tests includes the p-value for the first test described above. the right-tail test is Prob(>125%). However for log10 or ln-transforms. the second test is a right-tail test of the hypotheses: H0: Difference > – ln(1 – fractionToDetect) (bioinequivalence for right test) H1: Difference £ – ln(1 – fractionToDetect) (bioequivalence for right test) The test statistic for performing this test is: t2 = ( (TestLSM – RefLSM) + ln(1 – fractionToDetect) ) / DiffSE For log10-transform or data already log10-transformed. For example.

these limits are entered on the bioequivalence options tab. In general. Power Power is the post-hoc probability of detecting a difference greater than or equal to a specified percentage of the reference least squares mean. Briefly.μR > θU vs. In the bioequivalence calculations.μR < θL vs. Power = 1 – (probability of a type II error) = probability of rejecting H0 when H1 is true. the Anderson-Hauck test is based on the hypotheses: H01: μT . this changes to: 386 . the hypotheses being tested are: H0: RefLSM = TestLSM H1: RefLSM ≠ TestLSM For the no-transform case. or page 99 of Chow and Liu (2000). That is. Anderson-Hauck test See Anderson and Hauck (1983).15 User’s Guide operationally equivalent to the classical confidence interval approach. then both the H0’s given above for the two tests are also rejected at the alpha level by the two one-sided t-tests.μR < θU where θL and θU are the lower and upper Anderson-Hauck limits. The Anderson-Hauck test statistic is tAH given by WinNonlin YT – YR – ( θL + θU ) ⁄ 2 t AH = -----------------------------------------------------Diff SE where Diff SE is the standard error of the difference in means. HA1: μT . HA2: μT . and data already ln-transformed. Rejection of both null hypotheses implies bioequivalence. if the classical (1-2×alpha) ×100% confidence interval for difference or ratio is within LowerBound and UpperBound. the power is the probability of rejecting H0 given: |TestLSM – RefLSM | ≥ fractionToDetect × RefLSM For ln-transform. Under the null hypothesis.μR > θL H02: μT . this test statistic has a noncentral t-distribution.

df – 0. Also the maximum probability of rejecting H0 occurs in the case of equality.2×RefLSM) = Pr (Difference/DiffSE > t (1-alpha/2). and no transform was done on the data.Bioequivalence Individual and population bioequivalence 15 |TestLSM – RefLSM | ≥ – ln (1 – fractionToDetect).2×RefLSM / DiffSE t2 = – t (1-alpha/2). RefLSM>0.2×RefLSM) Note that tstat is T-distributed.2. Because individual bioequivalence relies on estimates of within-subject. and similarly for log10-transform and data already log10-transformed.2×RefLSM) + Pr (Difference/DiffSE < –t (1-alpha/2).2×RefLSM) + Pr(tstat < t2 given |Difference| = 0.df given |Difference| = 0. For the sake of illustration.2×RefLSM) / DiffSE Then: Power = Pr(tstat > t1 given |Difference| = 0.2×RefLSM.2×RefLSM) Let: t1 = t (1-alpha/2).df – 0. assume fractionToDetect = 0. df given |Difference| = 0. Then: Power = 1 – (p1 – p2) Individual and population bioequivalence Introduction Individual and population bioequivalence are used to assess switchability and prescribability. and let p2 be the p-value associated with t2 (the area in the tail.2×RefLSM) = Pr (|Difference/DiffSE| > t (1-alpha/2).2×RefLSM / DiffSE tstat = (Difference – 0. Let p1 be the p-value associated with t1 (the area in the tail). So: Power = Pr (rejecting H0 at the alpha level given |Difference| = 0. note that p2 may be negligible). replicated crossover designs are 387 . |TestLSM – RefLSM| = 0. within-formulation variation.df given |Difference| = 0.

σP is called the Total SD standard.2 and θP = (ln(1 .20. Designs that can be appropriately analyzed include. This algorithm was developed by Francis Hsuan.. σR2 is computed by the program and is the total variance of the reference formulation. The default value for PercentReference is 0.15 User’s Guide required.and between-subject variance.< θ P if σ R ≤ σ P 1 2 -. i.PercentReference) + εP) /σP2. The criteria take the linearized form 2 2 2 2 η P1 = E ( μ T – μ R ) + σ T – ( 1 + θ P ) σ R < 0 388 2 2 2 if σ R > σ P .σ P 2 where σP defaults to 0. In the wizard. Sequence Period Design WinNonlin 2 4 4 4 2 3 2 6 2 6 4 4 2 3 5 3 3 3 4 4 TRTR/RTRT TRTR/RTRT/TRRT/RTTR TT/RR/TR/RT TRT/RTR/TRR/RTT TRRTT/RTTRR TRR/RTR/RRT RTR/TRT TRR/RTT/TRT/RTR/TTR/RRT TRRR/RTTT TTRR/RRTT/TRRT/RTTR/TRRR/RTTT The algorithm works for balanced and unbalanced data with an equal number of periods in each sequence and one measurement per subject in each period.< θ P if σ R > σ P 1 2 -. but are not limited to.e. Bioequivalence criterion The FDA population bioequivalence (PBE) criteria are E ( Y jT – Y j'R ) – E ( Y jR – Y j'R ) -----------------------------------------------------------------------. the sum of within.E ( Y jR – Y j'R ) 2 E ( Y jT – Y j'R ) – E ( Y jR – Y j'R ) -----------------------------------------------------------------------.

For constant scaling. The IBE criteria take the linearized form η I1 = E ( μ T – μ R ) + σ D + σ WT – ( 1 + θ I ) σ WR < 0 and 2 * 2 2 if σ WR > σ I η I2 = E ( μ T – μ R ) + σ D + σ WT – σ WR – θ I σ I < 0 where σ D will be defined below. use one of the following.< θ I 1 2 -.< θ I 1 2 -. σΙ is called the within-subject SD standard. Population Individual If σR > σP If σR ≤ σP If σWR > σI If σWR ≤ σI use ηP1 use ηP2 use ηI1 use ηI2 If the upper confidence bound on the appropriate η is less than 0. use ηP1 or ηI1. In the wizard. 389 .Bioequivalence Individual and population bioequivalence 15 η P2 = E ( μ T – μ R ) + σ T – σ R – θ P σ P < 0 The FDA individual bioequivalence (IBE) criteria are 2 2 2 2 if σ R ≤ σ P E ( Y jT – Y jR ) – E ( Y jR – Y j'R ) ----------------------------------------------------------------------. The default value for Percent Reference is 0.05.σ I 2 2 2 2 2 if σ WR > σ I if σ WR ≤ σ I where σI defaults to 0. For mixed scaling. use ηP2 or ηI2.20. * 2 * 2 2 2 if σ WR ≤ σ I For reference scaling. The method of calculating that upper confidence bound follows. σWR2 is computed by the program and is the within-subject variance of the reference formulation.2 and θI = (ln(1 – PercentReference) + εI) /σI2. then the product is bioequivalent in the chosen sense.E ( Y jR – Y j'R ) 2 E ( Y jT – Y jR ) – E ( Y jR – Y j'R ) ----------------------------------------------------------------------. and the default value for εI is 0. The confidence interval is set on the Bioequivalence Options page.

μR are the population mean responses of Y with treatments T and R respectively.σ TR} ˜ are defined as follows: 2 2 390 . Assume the mean responses of Y ij .15 User’s Guide Details of computations Let Yijkt be the response of subject j in sequence i at time t with treatment k. The components ˜ of Y are arranged in the ascending order of time.σ TT . ˜ ˜ u iI : = ( u iT – u iR ) . μ i follow a linear model ˜ WinNonlin ˜ μ i = Z iT μ T + Z iR μ R ˜ ˜ ˜ (1a) where μT. Let piT:= sum( Z iT ) and piR:= sum( Z iR ) be the number of occasions T and R ˜ respectively be˜assigned to the subjects in sequence i.˜ni > 1. be the number of subjects in sequence i. u iT : = p iT Z iT . and Y ij' be a vector of responses for subject j in sequence i. ˜ ˜ u iR : = p iR Z iR . σ WR . ˜ ˜ ˜ Assume that the covariances (Y ijkt.σ RR . ˜ ˜ ˜ and d i : = s T u iT – s R u iR .Y ijk't') follow the model –1 –1 –1 –1 ˜ ˜ Σ i = σ WT diag ( Z iT ) + σ WR diag ( Z iR ) + σ TT Z iT Z' iT ˜ ˜ + σ RR Z iR Z' iR + σ TR { Z iT Z' iR + Z iR Z' iT } ˜ ˜ ˜ ˜ ˜ ˜ ˜ ˜ 2 2 (1b) where the parameters σ = {σ WT . and Xi are design/model matrices. Let Z iT and Z iR be the ˜ design vector for treatment T and R respectively in sequence i.

Bioequivalence Individual and population bioequivalence 15 σ T = σ TT + σ WT and σ R = σ RR + σ WR are var ( Y ijTt ) and var ( Y ijRt ) . 2 ρ RR = σ RR ⁄ σ R = intra-subject correlation coeff. 2 ρ TR = σ TR ⁄ ( σ T σ R ) = intra-subject coeff.Y R) .Y R') . to the “subject-by-formulation” variance component in the mixed model discussed in the FDA guidance Section III. The quantity is similar. all the quantities above are non-negative when they exist. 1 > ρ TT > 0 .Y T') . 2 2 2 2 ρ TT = σ TT ⁄ σ T = intra-subject correlation coeff. corr (Y T. σ RR = ρ RR σ R and σ TR = ρ TR σ T σ R are intra-subject covariances. 2 2 σ WT = σ T – σ TT is the intra-subject variance ( Y T – Y T' ) ⁄ 2 . but not identical. it is useful to define additional parameters: 2 2 2 2 σ D = σ TT + σ RR – 2σ TR σ iT σ iR σ iI *2 *2 *2 * (2a) 2 = ( σ T – ( 1 – p iT )σ WT ) = ( σ R – ( 1 – p iR )σ WR ) * –1 2 –1 2 2 –1 2 2 –1 (2b) (2c) (2d) = ( σ D + p iT σ WT + p iR σ WR ) Except for (2a). 1 > ρ RR > 0 . 1 > ρ TR > – 1 . It satisfies the equation 391 . corr (Y R. For PBE and IBE investigations. σ TT = ρ TT σ T . σ WR = σ R – σ RR is the intra-subject variance ( Y R – Y R' ) ⁄ 2 . corr (Y T.

or TxRR/xRTT/RTxx/TRxx/xTRR/RxTT (Table 5. and Vi is the withinsequence sample covariance matrix.15 User’s Guide WinNonlin var ( Y ijT – Y ijR ) = σ D + σ WT + σ WR * 2 * 2 2 In general. The FDA mixed model is a special case of the multivariate model above. S iI = u' iI V i u iI ˜ ˜ ˜ ˜ ˜ ˜ S iWT = ( 1 – p iT ) ( tr { diag ( u iT )V i } – u' iT V i u iT ) ˜ ˜ ˜ S iWR = ( 1 – p iR ) ( tr { diag ( u iR )V i } – u' iR V i u iR ) ˜ ˜ ˜ where Y i• is the sample mean of the ith sequence.∑ n i d' i ∑ d i) ˜ ˜ i i (4a) S iT ∼ σ iT ( n i – 1 ) χ S iR ∼ σ iR ( n i – 1 ) χ S iI ∼ σ iI ( n i – 1 ) χ 2 –1 *2 *2 *2 –1 2 ni – 1 (4b) (4c) (4d) (4e) –1 2 ni – 1 –1 2 ni – 1 ( p iT – 1 ) ( n i – 1 ) S iWT ∼ σ WT ( p iT – 1 ) ( n i – 1 ) χ –1 2 392 . S iR = u' iR V i u iR . let ˆ Δ = ∑ d'i Yi• ˜ i S iT = u' iT V i u iT . while FDA’s σ D is always non-negative. It yields identical results to those described in the FDA guidance Appendices F & H when the design is TRTR/RTRT. this σ D may be negative. This method is applicable to a variety of higher-order two-treatment crossover designs including TR/RT/TT/RR (the Balaam Design). TRT/RTR. page 205).7 of Jones and Kenward. It can be shown that –1 –1 –1 –1 ˆ –1 Δ ∼ N (μ T – μ R. Given the ith sequence. This method for PBE/IBE is based on the multivariate model. * 2 and the quantity σ D is identical to σ D whenever the former is nonnegative.

SiWR (for IBE) are statistically independent. Then define the estimators of the components of the linearized criterion by Using the above notation. and that the four statistics Δ. one may define unbiased moment estimators for the PBE criteria: 393 . SiWT. SiI. it can be shown that {SiT. chosen to yield the method of moments estimates of the η’s.Bioequivalence Individual and population bioequivalence 15 S iWR ∼ σ WR ( p iR – 1 ) ( n i – 1 ) χ 2 –1 –1 2 ( p iT – 1 ) ( n i – 1 ) (4f) Furthermore. SiWR}. SiR} (for PBE) are statistically independent from {Δ} and {SiWT. Let αi’s be sets of normalized weights.

H and U statistics: 2 ˆ ˆ ˆ ( Δ ) ) 1 ⁄ 2⎞ ⎛Δ +t H0 = . 394 * 2 * . P2. P4. * 1⁄2 ˆ Hη = η + ( ∑ Uq ) where U q = ( 1 + θ P ) U q for q = P3. 2 Uq = ( Hq – Eq ) for all q where the degrees of freedom nq are computed using Satterthwaite’s approximation.n 0 ( var ⎝ ⎠ H q = n q E q ⁄ χ . Then. H q = n q E q ⁄ χ .95.95. This can be generalized to compute the following sets of nq. and U q = Uq for all other q.05. Hsuan and Holder (2000). I1 and I2.15 User’s Guide WinNonlin ˆ2 ˆ η P1 = Δ + E P1 + E P2 – ( 1 + θ P )E P3 – ( 1 + θ P )E P4 ˆ2 2 ˆ η P2 = Δ + E P1 + E P2 – E P3 – E P4 – θ P σ P and for the IBE criteria: ˆ2 η I1 = Δ + E I1 + E I2 – E I3a ˆ2 2 η I2 = Δ + E I1 + E I2 – E I3b – θ I σ I The FDA Guidance specifies a procedure to construct a 95% upper bound for η based on the TRTR/RTRT design using Howe’s approximation I and a modification proposed by Hyslop. the 95% upper bound for each η is . that indicates bioequivalence. nq 2 2 for q = P3. I3a and I3b. Hq > 0 fails to show bioequivalence. If Hq < 0. P4.nq for q = P1.

For a discussion of the analysis types. • • 4. To create a new analysis. as follows. see “Average bioequivalence” on page 379 and “Individual and population bioequivalence” on page 387.LML) file containing bioequivalence settings. 3. click the Previous Model button. Type of Bioequivalence: Select individual/population or average bioequivalence using the radio buttons at the upper right. Click Calculate to run the model. The Bioequivalence Wizard provides options to use prior model settings (optional).Bioequivalence Bioequivalence Wizard 15 Bioequivalence Wizard To set up and run a bioequivalence analysis: 1. 395 . 2. click the Load button and open a LinMix/bioequivalence command (*. For average bioequivalence. 5. Start the Bioequivalence Wizard by choosing Tools>Bioequivalence Wizard or by clicking on the Bioequivalence Wizard tool bar button. proceed to step 4 below. To reload the most recently run model. select the study type at the top left. Open the data set to be analyzed. • To load a previously-saved bioequivalence model.

Period. select the variables that identify the Subject. WinNonlin will try to match the variable names from the active data set with the Subject. including variable names.15 User’s Guide Based on the analysis and study types. At any point in the wizard. Click Next to proceed with the analysis: • • “Using average bioequivalence” on page 396 “Using population/individual bioequivalence” on page 401 Using average bioequivalence See the Examples Guide for an example using average bioequivalence. and Formulation. If WinNonlin has not already selected the appropriate variables. and Formulation fields. regressor(s)/covariate(s). 9. Period. Sequence. Sequence. to a command (*. 6. For Average bioequivalence other classification variables can be added by dragging them from the Variable Collection to the Classification 396 . only Formulation and its reference value need to be specified. Note: For Average bioequivalence with a parallel study type. WinNonlin 8. Select the reference formulation. and sort variable(s) must be specified. from the pull down list labeled Reference Value. 7. Average bioequivalence analysis settings fall under the following headings: • • • • • “Fixed effects for average bioequivalence” on page 396 “Random effects” on page 398 “Repeated variance” on page 399 “Bioequivalence options” on page 400 “Bioequivalence general options” on page 401 Fixed effects for average bioequivalence The dependent variable(s). the Save button can be used to save the analysis settings.LML) file for later re-use. against which to test for bioequivalence.

drug concentration. treatment. In the most common situation. It must have been declared as a regressor/covariate before it can be used as a weight variable. Classification variables or factors are categorical independent variables. temperature or body weight. see “The model and data” on page 378. Sort variables define subsets of data to be processed separately. When a weight variable is specified. See “Fixed effects” on page 352. the data are averages and weights are sample sizes associated with the averages.LML). The weights are used to compensate for observations having different variances. each row of data is multiplied by the square root of the corresponding weight. The Weight Variable cannot be a classification variable. To define the fixed effects model: 1. Select the weight variable if weights for each record are included in a separate column. i.. a separate analysis will be done for each level of the sort variables.g. e. It can also be used in the model.. See also “Weighting” on page 241. Regressor variables or covariates are continuous independent variables. and gender. If the sort variable has missing values. To reuse the most recently run model.Bioequivalence Bioequivalence Wizard 15 Variable field. click the Previous Model button. the analysis will be performed for the missing level and MISSING will be printed as the sort variable value. e.g. Edit the model as needed. For Average bioequivalence the Model Specification field automatically displays an appropriate fixed effects model for study type.. e. The Bioequivalence Wizard Fixed Effects dialog works very much like the LinMix Fixed Effects dialog. Drag and drop variables to categorize those needed in the LinMix model: • • • • Dependent variables contain the values to which the model will be fit.g. Weight variable values should be proportional to the reciprocals of the variances. click the Load Model button. formulation. For discussion of the model.e. Note: To reuse a model in a command file (*. • 397 ..

15 User’s Guide WinNonlin 2. If the input data are already transformed.) 1. 2. Fixed Effects Confidence Level: Accept the default setting [95] or enter the confidence level for the parameter estimates. To transform the dependent variable values before analysis. 3. Variance Blocking Variables (Subject): (Optional) This variable must be a classification model term. Classification variables and regressors: Drag and drop column names (or type them) to as needed for the random effects model(s). Random effects Use this tab of the Bioequivalence Wizard to set traditional variance components and/or to random coefficients. or Next to specify Random effects or Repeated variance. Note: The Bioequivalence Wizard Variance Structure dialog operates exactly as it does in the LinMix wizard. (See also “Variance structure” on page 353. else see “Repeated variance” on page 399. Complete 398 . Click Calculate to run the model with default settings. The confidence level for the bioequivalence test is specified later. select a log transformation from the pull down list for Dependent Variables Transformation. select Already Ln-transformed or Already Log10-transformed. It identifies the subjects in a data set. 4.

Subject produces a block diagonal structure with identical blocks. Group variable: (Optional) This variable must be a classification model term. R is assumed to be 2 equal to σ I . All observations having the same level of the group effect have the same covariance parameters.Bioequivalence Bioequivalence Wizard 15 independence is assumed among subjects. The repeated effect must contain only classification variables. 7. 4. It defines an effect specifying heterogeneity in the covariance structure. If no Repeated statement is specified. Thus. Click Calculate to run the analysis or Next to set Bioequivalence options. Type: If a random effects model is specified. select the type of covariance structure here. 3. Check Random Intercept to indicate that the intercept is random. 5. 399 . See “Variance structure” on page 353 for details on each type. Click the Add Random button to create multiple variance models. if needed. 6. This is most commonly employed when a subject is specified in the Variance Blocking field. Each new level of the group effect produces a new set of covariance parameters with the same structure as the original group. Repeated variance Select the Repeated tab of the Bioequivalence Wizard to specify the R matrix in the mixed model. See also “Repeated measurements” on page 356. The default is off—no random intercept.

See “Average bioequivalence” on page 379. 2. Click Calculate to run the analysis or Next to set Bioequivalence options.25 for log-transformed data. It defines an effect specifying heterogeneity in the covariance structure. “Variance structure” on page 353 for types. Complete independence is assumed across subjects. Group variable: (Optional) This must be a classification variable with no regressors or operators.2 for not transformed) 400 . The default model follows the FDA guidance. 3. 5. • • • • Confidence Level (90%) Percent of Reference to Detect (20%) Anderson-Hauck Lower Limit (0. Variance Blocking Variables (Subject): (Optional) This variable must be a classification model term. select the type of covariance structure. 4. Note: The default Repeated model depends upon whether the crossover study is replicated or not.15 User’s Guide To specify the repeated variance settings: WinNonlin 1. Thus. Repeated Specification: Enter the model by typing it or by dragging the classification variable names and clicking on the operators. Syntax rules for the repeated specification are the same as those for the fixed–effects model terms. To set the bioequivalence options: 1. All observations having the same level of the group effect have the same covariance parameters. The following options are pre-set to meet FDA requirements. Bioequivalence options This dialog of the Bioequivalence Wizard sets the range for bioequivalence. This Subject field identifies subjects in the data set. Subject produces a block diagonal structure with identical blocks.8) Anderson-Hauck Upper Limit (1. Each new level of the group effect produces a new set of covariance parameters with the same structure as the original group. 1. Type: If a Repeated Specification has been included.

How output that is not estimable is to be represented may be set. an optional ASCII text window). 401 . The average bioequivalence calculations begin and the output appears in a new workbook (and. The kind of degrees of freedom can be stipulated. up to a maximum of five lines. Click Calculate to run the analysis or Next for Bioequivalence general options. and Intermediate Calculations. See “Individual and population bioequivalence” on page 387 for a discussion of theory. When all the appropriate settings have been made. Press Ctrl+Enter to move to the next line. A title can be assigned as a header for the ASCII output. The maximum number of iterations can be set. Bioequivalence general options This dialog of the Bioequivalence Wizard sets the following operations: • • • • • • ASCII output can be selected as an additional output form.Bioequivalence Bioequivalence Wizard 15 2. Note: Population/Individual bioequivalence can only use replicated crossover study data. Each sequence must contain the same number of periods. For each period each subject must have one measurement. Convergence Criterion. Launch the Bioequivalence Wizard and set up the variables for the analysis as described under “Bioequivalence Wizard” on page 395. To specify population/individual bioequivalence information: 1. The advanced user options include: – Initial estimates for the Variance Parameters – Change defaults for Singularity Tolerance. click the Calculate button to run the model. Using population/individual bioequivalence Following is an overview of the basic steps to perform population/individual bioequivalence in the Bioequivalence Wizard. See the Examples Guide for an example.

Select Ln(x) or Log(x) to transform the values before analysis. 3.. See “Fixed effects” on page 352. WinNonlin Fixed effects for population/individual bioequivalence The Fixed effects model is prebuilt following FDA guidelines. button to reload the last-run model.02) . To complete the model: 1. Appropriate terms for study type are specified automatically as classification variables. If the input data are already transformed. regressors. Click Calculate to run the model with the default settings.. select Already Lntransformed or Already Log10-transformed.15 User’s Guide 2.2) and Epsilon (0. Drag the appropriate term(s). Bioequivalence options for population/individual bioequivalence To set the bioequivalence options: 1. e. and fixed effects model are disabled and preset to FDA guidelines. The Bioequivalence Wizard Fixed Effects dialog for population/individual bioequivalence works very much like that for LinMix. For Population bioequivalence. The following options are pre-set to meet FDA standards. See “Individual and population bioequivalence” on page 387. (Optional) Select a log transformation from the pull down list for Dependent Variables Transformation. from the Variable Collection to the Dependent Variables field. or Next to proceed to the Fixed effects for population/individual bioequivalence. or Next to proceed to the Bioequivalence options for population/individual bioequivalence.g. • • • 402 Confidence Level (95%) Percent of Reference to Detect (20%) Population Limits: Total SD Standard (0. concentration. Note: Use the Load Model button to load a previously-saved model. classification variables.. 2. Click Calculate to run the model. Use the Previous Model.

AIC. degrees of freedom. Units. parameter estimate. Dependent Variable(s).05) 2. click Calculate.Bioequivalence Bioequivalence output 15 • Individual Limits: Within Subject SD Standard (0. Variance parameter names assigned by WinNonlin 403 Sequential Tests Sequential SS Partial Tests Partial SS Final Fixed Parameters Final Variance Parameters Final estimates of the variance parameters Parameter Key . and p-value Only for 2x2 crossover designs using the default model: ANOVA sums of squares. and Hessian eigenvalues Sort Variable(s). residual degrees of freedom. F statistic. units. Average bioequivalence can produce text output as an option. number of observations used.2) and Epsilon (0. etc. and p-value Only for 2x2 crossover designs using the default model: ANOVA sums of squares. Hypothesis. dependent variable(s). F-tests and p-values for sequential test Sort Variable(s). Denominator degrees of freedom. Dependent Variable(s). F-tests and p-values for partial test Sort variables. mean squares. Hypothesis. To start calculating Population/Individual bioequivalence. effect level. degrees of freedom. Table 15-1. Population/individual bioequivalence always generates text output as well. residual variance. Denominator degrees of freedom. t-statistic. Bioequivalence output The Bioequivalence Wizard generates a new bioequivalence workbook. Average bioequivalence Average bioequivalence generates a workbook containing following output. mean squares. Units. confidence intervals. residual sums of squares. Hypothesis degrees of freedom. p-value. Hypothesis degrees of freedom. F statistic. restricted log likelihood. Average bioequivalence output worksheets Worksheet Average Bioequivalence Ratio Tests Diagnostics Contents Output from the bioequivalence analysis Ratios of reference to test mean computed for each subject Number of observations. SBC.

Population/individual output worksheets Worksheet Population/Individual Bioequivalence Ratio Tests User Settings Contents Output from the bioequivalence analysis Ratios of reference to test mean User-specified settings Depending which settings were specified in the wizard. the output also contains the ratios table. the first step is computation of the least squares means of the test and reference formulations and their standard errors. These are described in detail in “Introduction to bioequivalence” on page 375 and “Average bioequivalence” on page 379. Obj_function. and information on residual effects vs. and computation of the standard error of the difference between test and reference least squares means. and degrees of freedom for the difference. the output may not include all of these worksheets. Average bioequivalence output worksheets (continued) Worksheet Least Squares Means LSM Differences Residuals User Settings Initial Variance Parameters Iteration History Contents Sort Variable(s). For crossover designs. and the least squares means for the test formulation Difference in least squares means Sort Variable(s). 404 . described under “Ratio tables” on page 406. computes these quantities. Dependent Variable(s). Units. Therefore the output for average bioequivalence includes all the output obtained by running LinMix. Dependent Variable(s). and column for each variance parameter WinNonlin To determine average bioequivalence of a test and reference formulation. LinMix.15 User’s Guide Table 15-1. as described in “LinMix output” on page 361. The Linear Mixed Effects module. Population/Individual bioequivalence Table 15-2. predicted and observed effects User-specified settings Parameter and estimates for variance parameters Iteration. The output also contains the Average Bioequivalence Statistics. Units.

to determine if bioequivalence has been shown. Table 15-3. and also specified a ln-transform. to determine whether mixed scaling for population bioequivalence will use the constant-scaling values or the reference-scaling values. the Within Subject SD Standard. to determine whether mixed scaling for individual bioequivalence will use the constant-scaling values or the reference-scaling values. For example. Note: The FDA guidance suggests that both the ratio test and the tests listed below need to be passed to indicate bioequivalence. SigmaWR . The population/individual bioequivalence statistics are described further in “Introduction to bioequivalence” on page 375. if the user specified a percent of reference to detect of 20%.Bioequivalence Bioequivalence output 15 Errors and warnings in the computations appear in the ASCII output as well as the application log or data history. then Ratio(%Ref) needs to be in the interval (80%. SigmaR . They include: Difference (Delta) = difference in sample means between the test and reference formulations. Population/Individual Bioequivalence Calculations Statistic Const_Pop_eta Ref_Pop_eta Mixed_Pop_eta Const_Indiv_eta Ref_Indiv_eta Mixed_Indiv_eta Indication Test statistic for population bioequivalence with constant scaling Test statistic for population bioequivalence with reference scaling Test statistic for population bioequivalence with mixed scaling Test statistic for individual bioequivalence with constant scaling Test statistic for individual bioequivalence with reference scaling Test statistic for individual bioequivalence with mixed scaling 405 . Each of the following is given. the Total SD Standard. Ratio(%Ref) = ratio of the test to reference means expressed as a percent.value that is compared with sigmaP. 125%) to show bioequivalence for the ratio test. An upper confidence bound < 0 indicates bioequivalence. This is compared with the percent of reference to detect that was specified by the user.value that is compared with sigmaI. with its upper confidence bound and bioequivalence conclusion.

or log10-transform. For each subject. Let ref be the corresponding value for the reference formulation. For data that was specified to be ‘already ln-transformed. for the case in which only one dependent variable measurement is made. ln-transform.15 User’s Guide For crossover designs. and similarly for ref. This is in response to the FDA guidance “Measures for each individual should be displayed in parallel for the formulations tested. Ratio(%Ref) = 100 × test / ref. and similarly for ref.’ these values are back-transformed to be in terms of the original data: Difference = exp(test) – exp(ref). the ratios table contains: Difference = test – ref. the output also contains the ratio tables. the output also includes for each test formulation a table of differences and ratios of the original data computed individually for each subject. for the cases of no transform. except for the cases of ‘ln-transform’ and ‘log10-transform’ where the geometric mean should be used: ‘ln-transform’: WinNonlin ⎛ ⎞ test = exp ⎜ ∑ ln ( X i ) ⁄ N⎟ ⎝ i ⎠ N –1 ‘log10-transform’: test = 10 ∑ log i 10 X i where Xi are the measurements after administration of the test formulation. Ratio tables For any bioequivalence study done on a crossover design. If multiple test or reference values exist for the same subject.” Let test be the value of the dependent variable measured after administration of the test formulation for one subject. for each BE measure the ratio of the individual geometric mean of the T product to the individual geometric mean of the R product should be tabulated side by side for each subject. then let test be the mean of the values of the dependent variable measured after administration of the test formulation. Ratio(%Ref) = 100×exp(test – ref) = 100×exp(test)/exp(ref) 406 . below. In particular.

if the mean is used for test or ref. 407 .Bioequivalence Bioequivalence output 15 Similarly. for data that was specified to be ‘already log10-transformed’: Difference = 10test – 10ref Ratio(%Ref) = 100×10test – ref = 100×10test/10ref Note: For ‘already transformed’ input data. and the antilog of test or ref is taken above. then this is equal to the geometric mean.

15 User’s Guide WinNonlin 408 .

Running a simulation of a model can quickly uncover errors in the model specification. parameter values for that model. must be checked. and dosing are set up as for model fitting. Simulating responses Simulation of output values requires a set of time or other X-variable values. Simulations can use any compiled or ASCII model. and dosing constants (for some models). See the Examples Guide for an example using simulation to evaluate competing sampling schedules. data variables. as detailed under “Modeling” on page 193. Simulations are also useful when writing custom models. to illustrate. and an example of simulation using a user-defined ASCII model. An example is provided below. This capability is especially useful. The model. and the Simulate check box. comparing drug concentrations for different dosing regimens or for different values of model parameters (such as absorption or elimination rates). except that no Y variable is specified. in simulating steady state responses from single dose data. including a user model. to explore the model shape and behavior. a set of time (X variable) values. Note that a compiled or ASCII model could be used. or determining optimal sampling times. a model. Simulation of a compiled library model This example uses a compiled WinNonlin library model to illustrate the use of simulation to generate multiple-dose concentrations based on parameters from fitting single-dose data.Chapter 16 Simulation Generating response values based on a model. and doses WinNonlin simulations can use library models or user defined models to calculate response values at specified time points. in the Data Variables dialog. 409 . for example.

This example uses a data set with times ranging from 0 to 36 hours. or double-click on the column header. and =A1+0. 6.18 hours. highlight cell A2 and position the cursor over the lower right hand corner of the cell until the cursor changes into a small black cross. 410 . an automatic record of the data history. Enter time values in the first column. and drag down to Row 201. 3.16 User’s Guide Creating the input data Simulation requires a data set with X values defining the time points over which to simulate Y variable values. In the resulting dialog. To enter all of the times quickly. To create the time value data: WinNonlin 1. A new workbook appears with two worksheets: the first. the second. Next. Select File>New from the WinNonlin menus. Model 11 from WinNonlin's compiled pharmacokinetic library.18 in each row. When simulating. in increments of 0.18 in cell A2. for data. to name Column A: Time. 4. The formula will be copied. Setting up the model The kinetics will be modeled by a two-compartment open model with firstorder input. Hold the left mouse button down. enter 0 in the first row. Press the Enter key. incrementing by 0. select the Workbook radio button and click OK. the process of selecting the model (a compiled model or an ASCII model) is exactly the same as it would be when modeling. Use Data>Column Properties in the menus. 5. 2.

In the final dialog. Click Next. Check the Simulate Data check box. 4. click Finish to load the model. To set model variables: 1. Click Next. Choose Model>Data Variables from the menus.Simulation Simulation of a compiled library model 16 To select the PK model: 1. 3. Choose Tools>PK/PD/NCA Analysis Wizard from the menus or click the PK/PD/NCA Analysis Wizard tool bar button. 2. Select Pharmacokinetic in the Model Types dialog. Select Model 11. Data variables The Simulate Data check box in the Data Variables dialog selects simulation rather than model fitting. The time variable must be identified. 2. 3. Drag Time to the X Variable box. and confirm that the Compiled check box is selected. 411 .

3. To enter parameter values: 1. K21 = 1. Model parameters The simulation requires parameter values. as constants.5. 412 . K12 = 1. to calculate output. Enter the following parameter values into the grid as shown below: V_F = 10. Select Model>Model Parameters from the menus. Click OK.2. K01 = 5. 2. Click Apply (leave the dialog box open). K10 = 0.16 User’s Guide WinNonlin 4.

Dosing Constant Number of doses Dose #1 Time of dose #1 Dose #2 Time of dose #2 Value 2 150 0 150 24 3. Enter mg in the Current Units field. Click OK. To enter the dosing information: 1. 413 . 4. Enter the values shown in the table below. Open the Dosing Regimen tab. 2.Simulation Simulation of a compiled library model 16 Dosing regimen This example will dose 150 mg every 24 hours.

Click OK when finished. Select the PK Settings tab. and in the PK Chart output. 2. The simulated response data appear in the Summary Table worksheet of the PK Workbook. 414 .16 User’s Guide Model options To specify model options: WinNonlin 1. Running the simulation To start the simulation: 1. Click the Start Modeling button or choose Model>Start Modeling from the menus. and uncheck the Transpose Final Parameter Table check box. Click the Model Options tool bar button or choose Model>Model Options from the menus.

For example.Simulation Comparing dosing schemes 16 The Variance Inflation Factor (VIF) is useful in determining optimal sampling schedule for parameter estimation. Only the Dosing Regimen need be edited. WinNonlin remembers the values for the Data Variables. After the previous simulation (“Simulation of a compiled library model” on page 409). Lower values indicate better estimates. Model Parameters. re-run the prior simulation. keeping the total daily dose of 150. Note: The secondary parameter estimates in simulation output are based on the first dose. and Dosing Regimen. but this time administering 75 mg every 12 hours. Comparing dosing schemes Simulations provide an easy means of modifying a scenario to see the impact on concentrations or effects. 415 .

use doses of 75 mg at times 0. Choose Model>Start Modeling to re-run the simulation. 3. 2. 416 . 12 and 24.16 User’s Guide Dosing regimen For this new simulation. Enter the new doses and click OK. To enter dosing information: WinNonlin 1. Select Model>Dosing Regimen from the menus.

com to learn about WinNonlin automation software and services. users can devise script files. 417 . Terminology Object-oriented programming terms used in WinNonlin scripting follow. This section includes: • • • • • • “Terminology” on page 417: object-based programming terminology “Writing and editing a script” on page 418 “Executing a script” on page 420 “Scripting object properties and methods” on page 421: a quick reference guide to properties and methods for each WinNonlin object and engine “Script file components” on page 426 “Processing multiple profiles in script files” on page 428 See “Scripting Commands” on page 579 for details on each scripting method and property. See the WinNonlin Examples Guide for example scripts. which are in essence computer programs. to customize WinNonlin and run a number of WinNonlin operations in batch mode. Note: E-mail sales@pharsight. Using the scripting language.Chapter 17 Scripting WinNonlin’s object-oriented scripting language The WinNonlin scripting language is an object-based programming language for automating WinNonlin tasks and analyses.

Model objects. the Method LOAD causes an associated data file (an Object) to be loaded.17 User’s Guide WinNonlin Table 17-1.WDO. Engines Method Property Script File a. PLOT. and WinNonlin itself. Older script files will run but may require minor modifications to file types. The WinNonlin engines are: PK. properties are attributes or characteristics of an object. Writing and editing a script This section discusses the steps in writing and editing a script. Every object has properties that define how the object should appear and interact. In WinNonlin. but cannot be saved back to that file type. In short. MERGE and TOOLBOX.PWO for a workbook rather than a . DESC. Only one Engine may be in use at any one time. ANOVA is available for backward compatibility with older versions of WinNonlin. For example. For example. Graph objects. LINMIX. In short. A third property would define whether or not the named data file contains data headers (column names) as the first row.BIOEQ. a property of an output data object could contain a file name to identify that object. Within a script. output must be saved to an appropriate file type—a .PSC. Text objects. each type of “child window” that appears in the main WinNonlin parent window is an object. ANOVAa. There are five types of objects in WinNonlin: Data objects. TRANSFORM. Another property could be the file type to be used when loading the object. Note: Scripts written with early versions of WinNonlin have the extension *. They have the default extension of .PSC file extension. Scripting terminology Objects Object-oriented programming languages deal with objects. TABLE.PSC) is an ASCII file that contains WinNonlin scripting code. 418 . A method is a set of computer code that performs functionality on a specific object or engine. For example. there could be many Objects of each type and several Engines. The Method RUN causes an associated PK Model (Engine) to run the model.WSC. New script files use the . A Pharsight script file (*. Script files are ASCII text files. it invokes an action. They can be opened in this version of WinNonlin. Properties also set values for objects or engines. Engines execute tasks in WinNonlin. LinMix or Bioequivalence Wizards provide newer and more powerful options than ANOVA.

PSC) in the List Files of Type box. Click OK. click the text line where changes are needed. 6. Select Script Files (*. Select the appropriate drive and directory if necessary. To edit a script: 1. The File Save dialog appears. In the text window. O). In the File name box.Scripting Writing and editing a script 17 To write a script: 1. or click the Open button on the tool bar. 419 . Case does not matter for any of the Properties or Methods. 7. Select the radio button for Text. 3. 8. The File Open dialog appears. click Save. A text window appears. From the File menu. 4. using the file extension . Comments may be added to the Script file by adding lines to the script that begin with REMA. 5. Select Script Files (*. 5. 2. Select Open from the File menu (Alt+F. select Save As from the File menu (Alt+F. When the script is complete. or REMARK. A). 4. select New (Alt+F. 6. Double click on the name of the script file to be edited. The New dialog appears. type the desired file name. 3. Select the appropriate drive and directory.PSC. 9. Click Save. then type in the changes. The script file will appear in a text window.PSC) from the pull-down menu in the Save as type box. When done. Type the content of the script. 2. All script files in the selected directory will be listed. N).

The Run Script dialog appears. or WINNONLN. The script file executes and the output appears in a WinNonlin output window. followed by the script file name. Exit WinNonlin if it is already running. 2. type WINNONLN. Script files may be executed either from within WinNonlin or from a Windows command line. Select Run. again including the full file path.17 User’s Guide WinNonlin Executing a script Once a script is created. From the command line To execute a script from the command line in Windows NT or Windows 98/2000: 1. Click OK. 4. 3. including the script. Close all WinNonlin windows. When executing from the Windows command line. the following steps are used to run it. R). Click the Windows Start button.EXE along with the appropriate path. 2. Select the appropriate drive and directory and double click on the script file name or click on the script file name and click Open. From the File menu. no WinNonlin windows may be open. 420 . For example: C:\WINNONLN\WINNONLN C:\WINNONLN\EXAMPLES\EXMPL_1. A Windows dialog will appear. In the Open box. When running a script from within WinNonlin. WinNonlin must be closed. select Run Script (Alt+F.PSC 5. Within WinNonlin To execute a script while in WinNonlin: 1. 3.

Detail on each command is provided under “Scripting Commands” on page 579. See the WinNonlin Examples Guide for examples of their usage. Data. If no errors occur during a script file execution. WinNonlin object properties and methods Arrange Cascade CloseAll DeleteFiles ExportAll FileMask Load MsgBox PrintAll PrintData PrintGraph PrintModel Save TileHorizontal TileVertical Unload WorkingDir WindowState 421 . From file manager or explorer To execute scripts from Windows File Manager or from Explorer: Script files can be run simply by double clicking on their names in Windows Explorer (Windows 98/2000 or Windows NT). The Pharsight application log will contain appropriate entries for actions taken during the script file execution (exclusions. a log file is created.).PSC must be associated with WINNONLN.EXE.Scripting Scripting object properties and methods 17 Note: When a script file cannot be successfully executed because of an error or errors in the file. Objects There are five WinNonlin objects: the WinNonlin. a log file will not be created. To do so. etc. A quick reference table of applicable properties and methods is given below for each Object. Refer to this Quick Reference Guide to see which properties and methods are used for each operation. A text window will appear showing the contents of that log file. the file extension . Text. Scripting object properties and methods This section lists all properties and methods that are applicable to each WinNonlin object and engine. Table 17-2. Model and Graph objects.

17 User’s Guide Table 17-2. Model object properties and methods DoseUnit Filename FileType 422 LinkFile Load ParmFile Print Save Unload . Text object properties and methods Filename FileType Load Print Save Unload WindowState Table 17-5. WinNonlin object properties and methods (continued) Halt PrintText WinNonlin Table 17-3. Data object properties and methods Append AppendFilename AppendFileType AppendSheet AutoFileOptions CaseSensitive Column ConseqDelim Copy Cut Delimiter EndCol EndRow Exclude Filename FileType Find Font FontSize Headers Include Load Missing MultiGrid Operation Paste Print Remove RemoveCol RemoveRow RenameCol Replace Save SearchArea Sort SortVar( ) SortVars StartCol StartRow Tab Tolerance Unload Unit WindowState With Table 17-4.

PK engine properties and methods OutData OutGraph OutText Run RunWordExport WordExportFilename Table 17-8. Model object properties and methods (continued) LamzFile PartFile WindowState Table 17-6. Graph object properties and methods Filename FileType FootnoteFont FootnoteFontSize LegendFont LegendFontSize Load LoadTemplate MoveFirst MoveLast MoveNext MovePrevious MultiGraph Print PrintMulti PrintMultiAcross PrintMultiDown Save SaveAll SaveAllPrefix SaveTemplate SortLevel TitleFont TitleFontSize Uniform Unload WindowState XLabelsFont XLabelsFontSize XTitleFont XTitleFontSize XUnits YLabelsFont YLabelsFontSize YTitleFont YTitleFontSize YUnits Engines The tables below provide a quick reference to the properties and methods applicable to each WinNonlin engine.Scripting Scripting object properties and methods 17 Table 17-5. Descriptive statistics engine properties and methods Confidence InData OutData Percentiles SortVars SummaryVar( ) 423 . See “Scripting Commands” on page 579 for details on individual properties and methods. Table 17-7.

Plot engine properties and methods AutoSort DnErr( ) ErrType GroupVar( . ANOVA engine properties and methodsa Filename FileType OutData OutText Run a. Scripts using the ANOVA engine will continue to work. ) JoinGroupVars JoinIDVar( . ) GroupVars( ) InData( ) Legend( ) NumData OutGraph Run SameErrCol( ) ShowUnits SortVar( . ) JoinIDVars OutData PageFooter PageHeader RegVar( ) RegVars Run Stats TableNum . Table 17-10. ) JoinCrossVars JoinGroupVar( .17 User’s Guide Table 17-8. Descriptive statistics engine properties and methods (continued) Interval Run SortVar( ) SummaryVars Weight WinNonlin Table 17-9. but it is best that new scripts use the LinMix engine. ) SortVars( ) Title UpErr( ) XTitle XUnits XVar( ) YTitle YUnits YVar Table 17-11. The former ANOVA functions are now handled by the Linear Mixed Effects Modeling engine. Table engine properties and methods AggregateVar CrossVar( ) CrossVars DefinitionFile Filename GroupVar( ) GroupVarBreak( ) GroupVars 424 IDVars InData( ) JoinCrossVar( .

) IncludeVars( ) a.Function=Crossover Applicable properties and methods InData OutData Run Response SortVar( ) SortVars ShowUnits Subject TreatmentA TreatmentB Stacked 425 . LinMix and Bioequivalence engine properties and methods Filename FileType OutData OutText Run Table 17-15. InData( ) NumDataa OutData Run SortVar( . Must be equal to 2. Toolbox engine properties and methodsa Toolbox. Transform engine properties and methods DestArea DestCol Extra Function InData Run SourceCol Table 17-13. Table engine properties and methods (continued) IDVar( ) NumData Table 17-12. ) SortVars Table 17-14.function assignments Toolbox. Merge engine properties and methods CarryData IncludeVar( .Scripting Scripting object properties and methods 17 Table 17-11.

Note: Deconvolution commands from scripts written prior to WinNonlin version 5. Script file components While some operations are specified in Script files completely through the use of the scripting language commands described in this chapter.17 User’s Guide Table 17-15. Valid properties are listed for each Function property assignment. more complex.1. Function can be assigned one of four values (above) to select the analysis tool. 426 .function assignments Toolbox.Function=Semicompartmental Applicable properties and methods ConcCol Ecol InData Keo OutData ConcCol DoseUnit InData InitialDose MaintenanceDose Npoints OutData OutGraph AlphaTerms() Aterms() Bolus ConcCol DoseUnit Delta InputLag OutGraph Run SortVar( ) SortVars TimeCol Run SortVar( ) SortVars Tau TerminalLower TerminalPhase TerminalUpper TimeCol TimeRepeat Npoints NumTerms Run Smoothing SortVar( ) SortVars TimeCol WinNonlin Toolbox. The set of properties available to the Toolbox engine depends on the value assigned to the Function property. other. Toolbox engine properties and methodsa (continued) Toolbox.1 will function properly in WinNonlin 5.1. but the deconvolution output may differ due to enhancements made in version 5. This section describes how each engine is implemented in the scripting language.2.Function=Superposition Toolbox. operations are specified via files that are prepared in advance.Function=Deconvolution a.

for each sort level.CMD) from prior versions of WinNonlin can also be loaded. including ANOVA. The user specifies which columns are summary variables and a new output data grid is generated. ANCOVA. The merging of data sets are specified in Scripts via scripting language commands that use the merge engine properties and methods.ANV). To use that data set for any other purpose within the script. Descriptive Statistics are specified in Scripts via scripting language commands that address the descriptive statistics engine..LML). To use that data set for another purpose. etc. Note that although a workbook will be loaded and used for the modeling as a result of either the Command file or Pharsight Model Object. Compartmental models are specified in scripts via Pharsight Model Objects (*. Command files (*. it will be necessary to load it separately. The Load and Save operations on this Data Object then rely on a PKS file pointed to by the Data Object. population. PKS Access to data in the Pharsight Knowledgebase Server (PKS) is possible using scripting in WinNonlin. Command files (*. Linear mixed effects modeling. 427 . and linear regression can be included in scripts using a linear mixed effects command file (*. see “Processing multiple profiles in script files” on page 428.LML). To use Command files with different model parameters. including average. that data file is not available for any other processing because it does not have an alias assigned to it. Note: ANOVA command files (*. etc. Charts are specified in scripts via scripting language commands.. Scripting the PKS is done by adding an additional FileType property to the WinNonlin Data Object. that data file will not be available for any other processing. see “Processing multiple profiles in script files” on page 428. can be specified in scripts using a bioequivalence command file (*. Note that although a workbook will be loaded and used for the modeling as a result of either the Command file or Pharsight Model Object. that data file will not be available for any other processing because it will not have an alias assigned to it. The PKS file is the same format as the files generated by the Study Definition and Study Mappings dialogs used in the creation of a study in the PKS. Note that although a data grid will be loaded and used for the modeling as a result of loading the ANOVA command file (*.PMO). for each sort level. Any combination of data from two data sets can be merged and saved as one data set. To use Comtal analysisa mand files with different Lambda Z ranges.Scripting Script file components 17 Table 17-16. To use that data set for any other purpose within the script. and individual bioequivalence models.PMO). it will be necessary to load it separately.CMD) from prior versions of WinNonlin can also be loaded. The file format is XML and is defined by an embedded DTD. it will be necessary to load it separately. Descriptive statistics LinMIx Merging data sets Noncompartmen Noncompartmental models are specified in scripts via Pharsight Model Objects (*.ANV) from previous versions of WinNonlin can also be loaded. Script file operations and methods of implementation Engine Bioequivalence Charts Compartmental models Method of implementation Bioequivalence analysis.

The Final Parameters worksheet will be transposed if NCATRANS is included in the NCA specification or if the user has set up “transposed output” as the WinNonlin default. see “Chart templates” on page 129 WinNonlin Transformations Data transformations are specified in Scripts via scripting language commands that access properties and methods of the transform engine. but it is not possible to select a subset. A table definition file contains all of the information needed to create the table except the input data set(s) to be used. Also.TDF). See “Transpose final parameter table” on page 206. Script file operations and methods of implementation (continued) Engine Tables Method of implementation Tables may be specified in Script files in two different ways: through the use of scripting language commands. Summary statistics can be included. default formatting is used. or. A script can call any existing table definition file.PMO) . with a separate variable for each of the Final Parameters in the transposed output. Note on saving files CAUTION: When saving files with the Scripting Language. or using a table definition file (*. this option simply determines which format will appear in the worksheet entitled Final Parameters. Processing multiple profiles in script files Two options are available to process compartmental models and noncompartmental analyses in the Scripting Language: • 428 Loading a Pharsight model object (*. Both transposed a. It is possible to overwrite existing files in this way. When writing scripts that will use the NCA Data output. The NCA Data output can appear with the variables (columns) Parameter and Estimate (in addition to the sort and carry-along variables). a. WinNonlin does not check to see if that file name already exists.17 User’s Guide Table 17-16. it is important to be aware of the form in which this outpwill appear. For more information on table definition files. Scripting language commands can specify: The Input data set(s) Table template number Variable mappings Summary Statistics (all or none) Page Break options Using only the Scripting Language commands loses some flexibility.

LAMZFILE for specifying lambda_z range information – .dat” Pk. – .LINKFILE for specifying PK/PD models that require two sets of parameter values – .Filetype= “ASCII” MYMODEL.cmd” MYMODEL. Set MYMODEL = Model MYMODEL. however.dat” MYMODEL.Scripting Processing multiple profiles in script files 17 • Loading a Command file (*. store only one set of values to be used for all sort levels.PARTFILE for specifying partial area information The following example script files shows how a command file could be loaded. different values may be used for dosing.Load MYMODEL.Dosefile= “Doseinfo. The next section discusses the ASCII file formats that are expected when using the following methods with Model Objects.PartFile= “Partinfo.PARMFILE for specifying parameter values – . to load a command file. partial areas or lambda_z range for each sort level automatically. 429 . however. then load ASCII files that contain unique information for each sort level.File Name= “Profiles. The information contained in the files specified by these properties is expected to be in a very specific format. and the methods used. parameter values. These ASCII files can be created and saved within WinNonlin or a text editor.DOSEFILE for specifying dosing information – . Command files.CMD) from an earlier version of WinNonlin When using a Pharsight model object.OutData= “MYOUTDATA” REMA Text and graphics output objects may also be named at this time Pk. It is possible.Run Note: File references within scripts can use absolute or relative file paths.

nType is between 1 to 5 as follows: 1 lamdbaz range information 2 dosing information 3 model parameter values 4 link parameter values 5 partial areas nProfiles is a value < 32767 that indicates how many profiles or subsets of values are included. General structure Order in file 1st 2nd Contents “CheckSum or Identifier” nType Comments Must be included on the first line of file. DOSEFILE for dosing information. Note: These files may be created using a text editor or from within WinNonlin. followed by annotated examples for several. 3rd 4th nProfiles nRows. WinNonlin The general structure for all of the ASCII files is given. nRows & nCols are integers that indicate how many rows and columns of data are in the file. PARTFILE for partial area information (start and end times for computation of partial areas under the curve). • • • • • • LAMZFILE names a file containing lambda_z range information (start and end times for computation of Lambda Z in noncompartmental analysis). UNITS for preferred parameter units in non-compartmental analysis. using the Save buttons on the appropriate tabs of the Model Properties or Model Options dialogs. PARMFILE for initial parameter values and bounds (optional).17 User’s Guide Special file formats This section discusses the required ASCII file formats when using the following methods with an ASCII user model or Model Object. LINKFILE for PK/PD models that require two sets of parameter values. nCols 430 .

of profiles). Char = N as the data type is numeric for the 1st column. 4 nType = 1. etc. Char Comments For each column. as the number of decimal places. When nType is 1. nProfiles = 3. as the number of profiles (Subjects) used is 3. 2. of constants) or NPAR (no. Set(s) of values for each nRow. if any. so nRows = 3. is 4. 3. Lambda Z range information (nType=1) Data Grid with Lambda Z Range Information for three profiles. Table 17-17. Data values of column 1. Data values of each profile Note: When ntype = 1 or 5. nRow = nProfiles (no. 431 Comments 5th 4. n = 4. as 1 represents lambda_z range information. of parameters). N . Subject 1 2 3 Start time 3 3 3 End time 18 18 18 Exclusions 4 3 5 ASCII File with Lambda_z Range information of the above three profiles. nRow = NCON (no. respectively. Model used: NCA Model 200.Scripting Processing multiple profiles in script files 17 Order in file 5th(nColns+4)t h record Remaining records Contents n. the number of rows equals nProfiles (the number of profiles). these values indicate the number of decimal places and data type (“A” for strings or “N” for numeric) of the data. so nCols = 4. When nType is 1. the number of columns equals 2 + the maximum number of exclusions. Lambda Z file structure Record 1st 2nd 3rd 4th Contents CheckSum or Required Identifier 1 3 3. When ntype = 2 or 3 or 4.

(See Note below). n = 4. Numeric data in the 2nd column (EndTime). Subject 1 1 Constant Dose Time of Last Dose Value 100 0 432 . is 4. Invalid exclusions in the 4th column. Char = N as the data type is numeric for the 3rd column. WinNonlin Note: Values that are in the excluded part of the data grid but are not valid exclusions have a value of 0. if any. 1st profile. Numeric data in the 4th column (Exclusions). 1st profile. Numeric data in the 3rd column (Exclusions). as the number of decimal places. 3rd profile. Numeric data in the 2nd column (EndTime).17 User’s Guide Table 17-17. This does place the restriction that Time=0 may not be an excluded data point. Char = N as the data type is numeric for the 2nd column. (See Note below). Dosing information (nType = 2) Data Grid with Dosing Information for three profiles. 1st profile. 1st profile. 3rd profile. Lambda Z file structure (continued) Record 6th 7th 8th 9th 10th 11th 12th 13th 14th 15th 16th 17th 18th 19th Contents 4. is 4. Numeric data in the 2nd column (EndTime). (See Note below). N 3 18 4 5 3 18 3 0 3 18 -1e150 0 Comments n = 4. Numeric data in the 1st column (Start Time). 3rd profile. Numeric data in the 1st column (Start Time). if any. 3rd profile. 2nd profile. 2nd profile. 2nd profile. Numeric data in the 3rd column (Exclusions). 2nd profile. Model used: NCA Model 200. Invalid exclusions in the 3rd column. N 4. Numeric data in the 1st column (Start Time). as the number of decimal places. Invalid exclusions in the 4th column.

5th 6th 7th 8th 9th 10th 11th 12th 13th 14th 4. (See Note below on nRows). so nRows = 3. Null data in the 3rd column. sdosenormalization. It is possible to save dosing units and dose normalization information to ASCII *. n = 4. To do so. Can be used to save Identifier dosing units. Null data in the 3rd column. 1st profile. Null data in the 2nd column. as 2 represents dosing information. Table 17-18. nProfiles = 3. so nCols = 1. Numeric data in the 1st column (Value). 2nd profile. sdoseunit. When nType is 2. Null data in the 3rd column. 1st profile. 1st profile. 2 3 3. if any. 3rd profile. even without the units [hence it could be V32. Null data in the 2nd column. as the number of profiles (Subjects) used is 3. Char = N as the data type is numeric for the column. Dosing file structure Record 1st 2nd 3rd 4th Contents Comments CheckSum or Was not used before WinNonlin version 3. 3rd profile.2. the format for the first line becomes V32. Numeric data in the 1st column (Value). 433 . Numeric data in the 1st column (Value).DAT files. N 100 0 0 100 0 0 100 0 0 a. 2nd profile. 3rd profile. The commas must be included. the number of columns equals 1. When nType is 2. the number of rows equals NCON (the number of constants). 1 nType = 2.. is 4. as the number of decimal places. Null data in the 2nd column. See note belowa.Scripting Processing multiple profiles in script files 17 Subject 2 2 3 3 Constant Dose Time of Last Dose Dose Time of Last Dose Value 100 0 100 0 ASCII File with dosing information of the above three profiles. ]. 2nd profile.

the number of rows equals NPAR (the number of parameters). Model parameter values (nType = 3) Data grid with parameter values on two profiles. (See Note below on nCol). 3 Unused nType = 3. Char = N as the data type is numeric for the 1st column (Initial Value). Table 17-19. as the number of decimal places. N 4. these values should be set to 0. Model used: PK Model 3 Subject 1 1 1 2 2 2 Parameter Volume_F K01 K10 Volume_F K01 K10 Initial value 0. N 434 . Model parameter file structure Record 1st 2nd 3rd 4th Contents CheckSum or Identifier 3 2 3. When steady state is being calculated. if any.2 20 2 0.17 User’s Guide Note: When using the Model. the number of columns equals 3.DoseFile method with NCA models.15 33 1. When nType is 3. so nRows = 3. as 3 represents parameter values. Comments 5th 6th 4. is 4. is 4. nRows is equal to NCON+1 in order to accommodate the Time of Last Dose values. When nType is 3. as the number of decimal places. Char = N as the data type is numeric for the 2nd column (Lower Limit).5 Lower 0 0 0 0 0 0 Upper 0 0 0 0 0 0 WinNonlin ASCII File with parameter values of the above two profiles. as the number of profiles (Subjects) used is 2. so nCols = 3. nRows is equal to NCON+2 to accommodate the Tau values. n = 4. If steady state is not being calculated. if any. nProfiles = 2. n = 4.

Char = N as the data type is numeric for the 3rd column (Upper Limit).Scripting Processing multiple profiles in script files 17 Table 17-19. Numeric data of the Volume/F parameter in the 1st column (Initial Value). 1st profile.2 0 0 20 0 0 2 0 0 . 1st profile. Null data of the Volume/F parameter in the 2nd column (Lower Limit). Null data of the K10 parameter in the 3rd column. Numeric data of the K10 parameter in the 1st column (Initial Value). 2nd profile.5 0 0 Comments n = 4. 2nd profile. if any. Numeric data of the K01 parameter in the 1st column (Initial Value). Null data of the K10 parameter in the 2nd column. 2nd profile. Null data of the Volume/F parameter in the 3rd column (Upper Limit). Null data of the K10 parameter in the 3rd column (Upper Limit). 435 . Null data of the K01 parameter in the 3rd column (Upper Limit). 1st profile.15 0 0 33 0 0 1. 1st profile. Numeric data of the K10 parameter in the 1st column. Null data of the K01 parameter in the 3rd column. 2nd profile. Null data of the Volume/F parameter in the 3rd column (Upper Limit). as the number of decimal places. Numeric data of the Volume/F parameter in the 1st column (Initial Value). 1st profile. Model parameter file structure (continued) Record 7th 8th 9th 10th 11th 12th 13th 14th 15th 16th 17th 18th 19th 20th 21st 22nd 23rd 24th 25th Contents 4. 1st profile. is 4. 1st profile. 2nd profile. Null data of the K10 parameter in the 2nd column (Lower Limit). Null data of the K01 parameter in the 2nd column. 2nd profile. 2nd profile. N 0. 1st profile. Null data of the K01 parameter in the 2nd column (Lower Limit). Null data of the Volume/F parameter in the 2nd column (Lower Limit). 1st profile. Numeric data of the K01 parameter in the 1st column. 2nd profile. 2nd profile.

When nType is 4. so nCols = 3.15 33 1. if any. Models used: PK Model 3/PD Model 105 The following data grid is applied to the PK Model: Subject 1 1 1 2 2 2 Parameter Volume_F K01 K10 Volume_F K01 K10 Initial Value 0. are not always used. n = 4.2 20 2 0. Link PK parameter values Record 1st 2nd 3rd 4th Contents Comments CheckSum or Was not used before WinNonlin version 3.5 Lower 0 0 0 0 0 0 Upper 0 0 0 0 0 0 WinNonlin ASCII File with parameter values of the above two profiles. as 4 represents Link parameter values. Table 17-20. the number of columns equals 3. Data grids with link parameter values on two profiles. nProfiles = 2. so nRows = 3. 4 2 3. the Linkfile is included. Char = N as the data type is numeric for the 1st column (Initial Value). as the number of decimal places. nCol3 = Upper Limit. When nType is 4. (See Note below on nCol). as the number of profiles (Subjects) used is 2. nCol2 = Lower Limit.2. the number of rows equals NPAR (the number of parameters). Link parameter values (nType = 4) Since a PK/PD model actually requires two sets of parameter values. Can be used to Identifier save PK concentration units. is 4. but must still be part of the file.17 User’s Guide Note: The three columns: nCol1 = Initial Value. 3 nType = 4. 5th 4. N 436 . See note belowa.

N 0. if any. PD parameter values Record 1st Contents CheckSum or Identifier Unused Comments 437 . 1st profile. as the number of decimal places.). even without the units (hence it could be V32. Null data of the Volume/F parameter in the 2nd column (Lower Limit). Table 17-21.Scripting Processing multiple profiles in script files 17 Table 17-20. Numeric data of the Volume/F parameter in the 1st column (Initial Value). is 4.PKconcentrationunit. The following data grid is applied to the PD Model: Subject 1 1 1 2 2 2 Parameter Emax Ece50 Gamma Emax Ece50 Gamma Initial value 5 6 7 1 2 3 Lower 0 0 0 0 0 0 Upper 0 0 0 0 0 0 ASCII File with parameter values of the above two profiles. Char = N as the data type is numeric for the 2nd column (Lower Limit).2 0 0 Comments n = 4. as the number of decimal places. Char = N as the data type is numeric for the 3rd column (Upper Limit). the format for the first line becomes V32. The comma must be included. To do so. 1st profile.DAT files. 1st profile. The remainder of the file structure is the same as the file shown in Example 3. is 4. It is possible to save PK concentration units for the PK tab of PK/PD models to ASCII . if any. a. Link PK parameter values (continued) Record 6th 7th 8th 9th 10th Contents 4. n = 4. Null data of the Volume/F parameter in the 3rd column (Upper Limit). N 4.

Numeric data. so nCols = 3. Emax parameter in the 2nd column (Lower Limit). PD parameter values (continued) Record 2nd 3rd 4th Contents 4 2 3. Null data. 2nd profile. n = 4. Ece50 parameter in the 1st column. Emax parameter in the 3rd column (Upper Limit). Emax parameter in the 2nd column (Lower Limit). When nType is 4. Emax parameter in the 1st column (Initial Value). WinNonlin 5th 6th 7th 8th 9th 10th 11th 12th 13th 14th 15th 16th 17th 18th 19th 20th 21st 4. Null data. 2nd profile. 1st profile. as 4 represents Link parameter values. 1st profile.) n = 4. Gamma parameter in the 1st column. 1st profile. 2nd profile. so nRows = 3. Null data. as the number of decimal places. When nType is 4. is 4. nProfiles = 2. 1st profile. as the number of decimal places. if any. 1st profile. as the number of profiles (Subjects) used is 2. 1st profile. (See Note below on nCol. Ece50 parameter in the 3rd column. Ece50 parameter in the 2nd column. 1st profile. 1st profile. Ece50 parameter in the 2nd column. Numeric data. Null data. Char = N as the data type is numeric for the 1st column (Initial Value). Null data. Char = N as the data type is numeric for the 2nd column (Lower Limit). is 4. 3 Comments nType = 4. 2nd profile. Numeric data. Numeric data. the number of columns equals 3. 2nd profile. Gamma parameter in the 2nd column. N 5 0 0 6 0 0 7 0 0 1 0 0 2 0 438 . Ece50 parameter in the 1st column. Null data. Emax parameter in the 1st column (Initial Value). is 4. Null data. the number of rows equals NPAR (the number of parameters). Null data. Emax parameter in the 3rd column(Upper Limit). Gamma parameter in the 3rd column. N 4. n = 4. Null data. as the number of decimal places. if any. Char = N as the data type is numeric for the 3rd column (Upper Limit). if any. Numeric data. N 4.17 User’s Guide Table 17-21. 1st profile.

Note: The three columns: nCol1 = Initial Value. Ece50 parameter in the 3rd column. if any. are not always used. Table 17-22. Null data. Partial areas (nType = 5) Data grid with partial area information for three profiles. is 4. as 5 represents Partial Areas. Partial areas file format Record 1st 2nd 3rd 4th Contents CheckSum or Unused Identifier 5 3 3. so nCols = 2. 2nd profile. When nType is 5. Model used: NCA Model 200 Subject 1 2 3 Min Time . n = 4. Char = N as the data type is numeric for the 1st column (AUC1 Lower). When nType is 5. nCol3 = Upper Limit. 439 Comments 5th 4. nCol2 = Lower Limit. Null data. PD parameter values (continued) Record 22nd 23rd 24th 25th Contents 0 3 0 0 Comments Null data. nProfiles = 3. 2nd profile. as the number of decimal places. Gamma parameter in the 3rd column.25 Max Time 48 48 48 AUC1 Lower 5 8 2 AUC1 Upper 6 12 3 ASCII File with partial areas information of the above three profiles. Numeric data. Gamma parameter in the 2nd column. the number of rows equals nProfiles. Gamma parameter in the 1st column. N .25 .Scripting Processing multiple profiles in script files 17 Table 17-21.25 . as the number of profiles (Subjects) used is 3. but must still be part of the file. 2 nType = 5. 2nd profile. 2nd profile. the number of columns equals the maximum number of partial areas multiplied by 2. so nRows = 3.

Numeric data of the column AUC1 Lower of the 3rd profile. Load Note: 440 Rules follow the user interface. . if any. Numeric data of the column AUC1 Upper of the 3rd profile. can be accessed by scripting from WinNonlin. See “PKS file format” on page 441 for details. as the number of decimal places. The workflow rules associated with the manipulation of PKS data via Scripting can be encapsulated based on the method being invoked (i. should be given lower and upper values of 0. is 4. and the files stored as studies and scenarios in it. The Load and Save operations on this Data Object will rely on a PKS file pointed to by the Data Object. but are not valid partial areas. Load or Save). N 5 6 8 12 2 3 Comments n = 4.e. Numeric data of the column AUC1 Lower of the 1st profile. Numeric data of the column AUC1 Lower of the 2nd profile. The file format is XML and is defined by an embedded DTD. Partial areas file format (continued) Record 6th 7th 8th 9th 10th 11th 12th Contents 4.17 User’s Guide Table 17-22. Scripting the Pharsight Knowledgebase Server The Pharsight Knowledgebase Server (PKS). Scripting the PKS is done by adding an additional FileType property to the Data Object. Char = N as the data type is numeric for the 2nd column (AUC1 Upper). Numeric data of the column AUC1 Upper of the 2nd profile. WinNonlin Note: Values that are in the partial area portion of the data grid. This PKS file is the same format as the files generated by the Study Definition and Study Mappings dialogs used in the creation of a study in the PKS. See the WinNonlin Examples Guide for examples of scripts to run WinNonlin/ PKS operations. Numeric data of the column AUC1 Upper of the 1st profile.

Scenario? . workbooks) are present in WinNonlin. If a PKS session has been created via loading a study and scenario or a study only and other Objects (i. texts. texts. A new version will be created if and only if the scenario name being saved is the same as the scenario that was loaded. Save Note: A more involved set of rules based on the state of the application. 2. If the PKS file pointed to by the Data Object has only a study name. then the Save Method will update the Study (if it changed) and Save the scenario to a new version or an entirely new scenario. no connection has been made to the PKS). If no PKS session has been created (i. then the Save Method will update the Study only. then only a study will be loaded. 1.e.e. workbooks) are present in WinNonlin. 3. then both the study and latest version of the scenario will be loaded. Scripting will not load previous versions of a scenario.Mappings?)> <!ATTLIST PWRImport UserID CDATA #IMPLIED Password CDATA #IMPLIED Server CDATA #IMPLIED 441 . charts. then the Save Method will create a study in the PKS based on the information in the PKS file pointed to by the File Name Property of the Data Object. If a PKS session has been created via loading a study only and no other Objects (i.Scripting Scripting the Pharsight Knowledgebase Server 17 1.e. charts. It is saved as an XML file with the following format. 2. <?xml version="1.0"?> <!DOCTYPE PWRImport [ <!ELEMENT PWRImport (StudyInformation?. PKS file format The PKS file contains XML that describes load and/or save operations to and/ or from the PKS. If the PKS file pointed to by the Data Object has a study name and a scenario name.

17 User’s Guide Port CDATA #IMPLIED Root CDATA #IMPLIED > <!ELEMENT StudyInformation EMPTY> <!ATTLIST StudyInformation BlindingType CDATA #IMPLIED Compound CDATA #IMPLIED Description CDATA #IMPLIED Design CDATA #IMPLIED EndTime CDATA #IMPLIED Indication CDATA #IMPLIED Name CDATA #IMPLIED Portfolio CDATA #IMPLIED Project CDATA #IMPLIED StartTime CDATA #IMPLIED State CDATA #IMPLIED Type CDATA #IMPLIED Reason CDATA #IMPLIED > <!ELEMENT Scenario EMPTY> <!ATTLIST Scenario Name CDATA #IMPLIED Reason CDATA #IMPLIED GetLatest (TRUE|FALSE) #IMPLIED > <!ELEMENT Mappings (Sources?. Mapping*)> <!ATTLIST Mappings Type (SAMPLE|DOSE|BOTH) #IMPLIED > <!ELEMENT Sources EMPTY> <!ATTLIST Sources Observation CDATA #IMPLIED Dose CDATA #IMPLIED Demographics CDATA #IMPLIED > 442 WinNonlin .

Scripting Scripting the Pharsight Knowledgebase Server 17 <!ELEMENT Mapping (FromFields*. FieldOption*)> <!ATTLIST Mapping FromField CDATA #IMPLIED ToField CDATA #IMPLIED FieldType (Subject_ID|Subject_Data|Observation|DCP_Description|Actual_Clock_Time|Relative_Nomina l_Time|Sample_Number|Relative_Actual_Time|Relative_Ac tual_End_Time|Treatment|Period|Phase|Undefined|Dose) #IMPLIED TreatAsDCP (TRUE|FALSE) #IMPLIED > <!ELEMENT FromFields EMPTY> <!ATTLIST FromFields Name CDATA #IMPLIED > <!ELEMENT FieldOption EMPTY> <!ATTLIST FieldOption OptionName (Unit|Sample_Description|Sample_Matrix|Analytical_Metho d|Status|LLOQ|ULOQ) #IMPLIED IsColumn (TRUE|FALSE) #IMPLIED Value CDATA #IMPLIED > ] > 443 .

17 User’s Guide WinNonlin 444 .

scripts and results among the members of a drug development team. This chapter covers the use of WinNonlin as a PKS client: loading PKS data to WinNonlin and saving data from WinNonlin as PKS studies and scenarios. models. The combination also supports compliance with the U. • • • • • • • • • • “The PKS information structure” on page 446 “Dosing” on page 448 “Accessing the PKS from WinNonlin” on page 449 “Creating a study” on page 452 “Loading a study or scenario” on page 463 “Creating scenarios” on page 465 “Optimizing PKS/WNL performance” on page 467 “Change tracking and audit information” on page 470 “Dosing information” on page 472 “Appending or updating a study” on page 474 445 .Chapter 18 Using the Pharsight Knowledgebase Server Data access in a secure environment The Pharsight Knowledgebase Server (PKS) is a database system for secure storage of study data and analyses. which provides additional operations.S. The PKS provides a secure data management system with complete auditing capabilities. This chapter covers WinNonlin/PKS interactions under the following headings. The PKS can also be accessed by way of a web interface. and an XML API. The combination of the PKS and WinNonlin Enterprise provides an effective means to share source data. Food and Drug Administrations’s 21 CFR Part 11 regulation. These functions and operations are covered in the Pharsight Knowledgebase Server’s documentation.

First. measurements of blood pressure. concurrent measurements for any subject (e. A study starts with a collection of raw data that must include longitudinal observations.g. Non-WinNonlin documents. baseline covariates.. first name. A study may include time-dependent observation and dosing data. reports and graphics that are related to the study. model.e. each data collection point can have multiple observations. chart. the analyses.18 User’s Guide • • • “PKS libraries and object shares” on page 475 “PKS Information” on page 479 “WinNonlin and PKS differences” on page 480 WinNonlin See also “Scripting the Pharsight Knowledgebase Server” on page 440. As analyses are performed on these data. a study must include at least one column containing subject identifiers. PKS saves observations and related data as a PKS study. etc. Analysis settings and output are added to the study as scenarios. A collection of fields taken together can be used for subject identification (e. Those objects fit into specific niches in the PKS data structure. can be loaded into the PKS and associated with a scenario using WinNonlin’s “Other documents” feature. last name). Studies The basic unit of information in the PKS is a study. If the observation data include multiple. In this case. subject-specific information such as body weight or gender. i. data) based application. heart rate. The PKS information structure WinNonlin is an object (text. and may include dosing and subject demographics. A PKS study requires specific fields. observation times and subject identifiers. As the PKS manuals explain. such as analysis code files.g. can be added as scenario(s) within the study. A study may include (but is not required to have) multiple columns of subject data—time-invariant. including model settings and results... multiple assays performed on a given sample or multiple sam- 446 . a sampling variable is used to differentiate the samples. Observation data may include records of concentration.

etc. 3 for the 1st. (Relative time is time elapsed since the start of the study.Using the Pharsight Knowledgebase Server The PKS information structure 18 ples during a given visit). The metadata are saved with the study. Scenario When WinNonlin analysis is performed on a PKS study. etc. values 1. See also “Creating scenarios” on page 465. Compound. and/or statistical tests to be applied to a particular data set for fitting or analysis. Portfolio. Note: To create a study that does include longitudinal observation data. and the analytical output. Study Design. a duplicate column will be created for the actual times. a study must include relative nominal time. the combination of data. if included) were scheduled in the protocol. Indication.). the relative nominal end time must be identified as well. See “Creating a study” on page 452 and “Loading a study or scenario” on page 463. a model with model parameters. representing the times that observations or doses really happened. Scenarios are naturally associated with (i. include a dummy time column. initial parameters. in one place and under one name. and (optional) results data can then be saved in the PKS as a scenario of the original study. are children of) a particular data set (study). This is in contrast to relative actual time. period or phase. although it is possible to create further scenarios from the study (more children) or from an existing scenario (grandchildren.e. A scenario can include a structural model. the data set must include a variable that differentiates the measurements (e.g. Study metadata When a study is created.) If the data uses infusion dosing. 447 . indicating the times at which observations (and doses. Time fields Because a study generally includes time-dependent observation data. 2nd and 3rd samples). This metadata provides ways to search and filter the data within the database. other model settings. Scenarios are roughly comparable to the WinNonlin concept of a workspace in that it is possible to save. related work— the data source. as well as any derived output from those tests or analysis. This dialog includes information about the study: Study Name... model. with values for each row (as placeholders). in the study. Project. 2. information about the study can be entered in the Study Definition dialog (see below). dosing. If only the Relative Nominal Time is defined in the data set. Study Type.

Users can select which dose columns to use for a given scenario. from a separate worksheet (not the worksheet containing observation data) Append dosing after study creation. such as study reports. See “Dosing information” on page 472 for details. at minimum. Dosing worksheet It is possible to include dosing data when creating a study in the PKS.18 User’s Guide Other documents The PKS can store documents that are not related to WinNonlin. That worksheet is selected as the dose source during study creation. subject identifiers. The dosing information must appear in a different worksheet from that containing the study observation data. The PKS will use those columns to match observation data to dose data for given subject(s). from a dosing worksheet Dosing dialog: enter dosing as part of a modeling scenario. as detailed under “Dosing information” on page 472. SAS or NONMEM code files. using the WinNonlin Dosing Regimen dialog. and its variables are assigned as part of mapping the study variables (see “Creating a study” on page 452). or a graphic created by saving a WinNonlin chart to a JPEG file. and relative nominal dose times. The worksheet must contain. The columns representing subject identifiers and nominal time must be named exactly as the corresponding columns are in the workbook containing the observation data. Dosing is generally saved as part of a PKS study. Dose data can be added to a study by one of three methods. If multiple scenarios are included in a study. It must contain the sort variables required to distinguish profiles. The dosing worksheet can contain columns for more than one analysis scenario. 448 . dose amounts. • • • Dosing worksheet: load dosing during study creation. See “Adding non-WinNonlin documents to a scenario” on page 466 for instructions. WinNonlin Dosing Dosing information is a special case of study data. each may use different dose data. Sets of such documents can be associated with scenarios. though it is not required.

Thereafter. at minimum. the dosing is transferred to a Dosing worksheet inside the study workbook. model parameters and dosing information. The worksheet must contain. The PKS will use the subject and time columns to match dose data to records for specific subjects at specific times. and creates a new scenario Opens the mapping wizard to set up a new PKS study from the current workbook 449 . close all WinNonlin windows. and loads a PKS study or scenario Saves changes to the study and creates a new version of the current scenario. PKS menu Menu item Load Save Save As Create Study Used when Always A PKS study is loaded A PKS study and scenario are loaded No PKS session is open and a workbook is loaded Function Closes the current PKS session. and a model is selected. if any. The menu items include: Table 18-1. any changes to the dosing then must be done on the Dosing worksheet rather than through the WinNonlin Dosing Regimen. then dosing data can be entered via the WinNonlin Dosing Regimen or Dosing Data Variable dialog. and relative nominal dose times. resides with the scenario information.Using the Pharsight Knowledgebase Server Accessing the PKS from WinNonlin 18 Append dosing To append dosing data to a PKS study. Dosing entered for a simulation. and subsequently saved to a scenario. if any. The WinNonlin dosing dialog can still be used for simulations and for user models. Dosing dialog When a PKS study or scenario with no dosing data is loaded in WinNonlin. subject identifiers. Accessing the PKS from WinNonlin Access to the Pharsight Knowledgebase Server occurs via the PKS menu in WinNonlin Enterprise. then open a workbook containing the dose information in a single worksheet. dose amounts. and does not become the study dosing data. Use the PKS Append/Update feature to map the new data columns to PKS study fields. if any Saves changes to the study. When the scenario is saved to the PKS with model.

18 User’s Guide Table 18-1. and tracks all changes. WinNonlin knows whether data came from the PKS. PKS menu (continued) Menu item Append/Update Study Dosing Add columns Used when No PKS session is open and a workbook is loaded Function Opens the mapping wizard. 450 . so that data columns from the current workbook can be added to an existing PKS study WinNonlin A PKS study or scenario is Sets dosing column((s) to be used in the loaded current model or analysis (scenario) A PKS study is loaded Loads additional study data columns from PKS into WinNonlin Opens the WinNonlin Document Manager. If data are then saved back to PKS (whether to an existing study or scenario or to a new one). While the link between WinNonlin and the PKS is not constant. any changes are documented for audit trail purposes. if one is loaded Other documents Always Share Always Libraries Always Status A library or share item is loaded A PKS study is open Information PKS sessions The Enterprise edition of WinNonlin has a concept of a PKS session. A PKS session begins when WinNonlin retrieves and uses data from the PKS. parameter names. storing a binary file containing code. and other items to be used as standards across many studies and analyses Checks whether any currently-loaded library or share items are up to date with the most recent version in PKS Displays the PKS study identifier and scenario identifier. share link can be saved with a scenario Loads or creates a PKS system object. to save or load non-WinNonlin documents to or from a PKS scenario Loads a non-WinNonlin document (“other document”) from any PKS scenario for use in the current analysis.

Click the Configure button to edit the PKS setup.Using the Pharsight Knowledgebase Server Accessing the PKS from WinNonlin 18 Connecting to the PKS The PKS configuration must be set up before WinNonlin can contact the PKS. 451 . This information need be entered once only. Note: The system administrator who set up the PKS database should provide the values for the server. port and root. 5. Select any available item in the PKS menu to open the PKS Logon dialog. The configuration settings are accessed from the PKS Logon dialog. enter the port number under Port.. as more than one PKS database may be available on the local network. To configure the PKS connection: 1. Enter a name describing the PKS database under PKS Alias. many sites have one production database and one example database. 4. Click OK. Enter the PKS server machine name under Server. choose PKS>Load. WinNonlin will save it. 2. 3. used to run examples from PKS Examples Guide. and root directory for the PKS database under Root. For example. This will be used to select a specific PKS database instance when logging in. Similarly. For example..

If dosing and/or demographics data are to be loaded. A PKS study stores raw data to be modeled and analyzed in one or more scenarios. open those as well. Data requirements: Observation data must contain. Note: See “Dosing” on page 448 for methods to add dosing information to a study. who must set up an account for each user to access the PKS (see the PKS User’s Guide for details).e. at minimum.18 User’s Guide Logging in User name and password Each time data are loaded to or from the PKS. phase. and indicators of period. Dosing must be in a separate worksheet from observations. and must contain the same data collection point identifiers. time fields. Demographic data must be time-invariant and subject-specific. WinNonlin Creating a study Studies are the fundamental unit of PKS organization. To create a study: 1. 452 . If logging in for the first time. etc. See “Connecting to the PKS” on page 451. subject identifiers. See “Studies” on page 446 for further discussion. sample values and nominal (scheduled) times. i. and can be loaded from the same worksheet as the observations. columns for subject identifiers. The log in user name and password are provided by your PKS system administrator.. Choose PKS>Create Study from the WinNonlin menus. Open sample data in a WinNonlin workbook. the user must log in and provide a password. set up the PKS configuration. 2.

Enter the user name and password supplied by your PKS administrator.) These attributes are called study metadata. You must have WRITE system access in order to create studies. 2002 or mm/dd/yyyy. Study metadata fields Attribute Study namea Description Used to identify and locate study in PKS. year. indication.g. Study definition Creating a study requires specification of the following information. (Required fields are marked with an asterisk in the Study Definition dialog. must be unique within the PKS database Data entry Typed Format 50 character max Study start time Dates of first and last days in and end time the study Typed Month day. WinNonlin will contact the PKS and open the Study definition dialog. See your database administrator for local rules for metadata values.g. 4. These are not necessarily the same as those used to log onto you Windows network. They are useful in database searches for studies on a given compound. e. Select the desired PKS database under PKS Alias and click OK. etc.: May 5.: 05/05/2002 453 .Using the Pharsight Knowledgebase Server Creating a study 18 3. Table 18-2. e.

g. e. Study metadata fields (continued) Attribute Description Portfolio Project Indication Blinding typea Study typea Study designa Compound Status Description Text describing the study. Click OK to proceed to the Study Creation Wizard.18 User’s Guide Table 18-2. Study Creation Wizard 454 .. 1. parallel or crossover Typed or selected 100 character from list of prior values max Typed or selected 50 character max from list of prior values Typed or selected 30 character max from list of prior values Typed or selected 30 character max from list of prior values Typed or selected 30 character max from list of prior values Main drug compound(s) under Typed or selected 50 character max development from list of prior values Whether the study will be locked against changes or not Selected Normal or locked a. use Ctrl+Enter for line breaks Data entry Typed Format 2000 character max WinNonlin Development portfolio to which Typed or selected 50 character max the study belongs from list of prior values Development project to which the study belongs Primary indication(s) for the study blinding level for the study type of study Experimental structure. Required field.

2.PKS file extension. if needed.Using the Pharsight Knowledgebase Server Creating a study 18 1. as an XML file with a *. Note: If dosing and/or demographics worksheets are used. Select dosing and demographics worksheets (optional). To view them. only columns that appear in all worksheets. Subject identifiers 1. Select the worksheet for the observation data under Observation Worksheet. 3. or to assign units to a column: 455 . Click Next to map Subject data. The Load button reloads the settings from that file. Subject ID attributes If units already exist for a given column in the WinNonlin worksheet. including file paths/names and variable mappings. Note: The Save button saves all study creation settings. as detailed under “Subject data attributes” below. 3. 2. Click Next to proceed to the Subject identifiers. those units are loaded to the PKS. will appear in the Subject Identifiers list. named identically. Drag the variable(s) that identify individual subjects to the PK Fields box. Select a variable under PKS Fields and click the Attributes button to set units for the subject ID variable(s).

subject-specific columns to the PKS Fields box. 2. To load units for each record from a column in the data set.18 User’s Guide 1. Drag columns that contain subject demographics (baseline covariates) and other time-invariant. CAUTION: Improper variable mappings can slow PKS operations substantially. 456 . Subject data WinNonlin 1. as detailed under “Subject data attributes” below. to be used in the PKS study. Other units can be any string. Select the variable under PKS Fields and click the Attributes button to set units. 3. and must follow WinNonlin column name limits. Click Next to specify variables related to Time. unit conversions will not be available for it. check Static and enter or select the unit under Unit. rather than one record per subject. To enter a new name for a column. Observations include one record for every data collection point. 4. drag that column to the Unit field. select that column under PKS Fields and click the Rename button. mapping subject-specific data as observations slows performance. controlled list. but if a unit is not defined in the PKS. In partic- ular. Time units must come from a fixed. 3. The name may include up to 50 characters. See also “Optimizing PKS/WNL performance” on page 467 and “Performance Tips” in the PKS System Administrator’s Guide 2. Click the Hide button to return to the Subject identifiers. To set one unit for an entire column.

2. unit conversions will not be available for it. 3. To set one unit for an entire column. expressed relative to the start of the study. Note that infusion dosing requires Relative Actual End Time. drag that column to the Unit field. check Static and enter or select the unit under Unit. See your system administrator for a list of allowed units. Time unit: Either choose a column containing the units for each record. Relative Nominal Time: Drag the variable containing scheduled times. to this field.Using the Pharsight Knowledgebase Server Creating a study 18 Subject data attributes If units already exist for a given column in the WinNonlin worksheet. or to assign units to a column: 1. Time 1. Set other time variables (optional). 3. if included. Time units must come from a fixed. controlled list. but if a unit is not defined in the PKS. All time variables must exist with identical names in both the observations worksheet and the dosing worksheet. 457 . Click Next to map Sample data. those units are loaded to the PKS. To view them. See “Time fields” on page 447 for a discussion of the PKS time variables. Click the Hide button to exit the Attributes dialog and return to the Subject data dialog. To load units for each record from a column in the data set. or check Static and enter the unit under Time Unit. 2. 4. Other units can be any string.

2. Select a variable under PKS Fields and click the Attributes button to edit the Sample attributes. Drag each column containing sample data to the PKS Fields box. Sample attributes All attributes are optional. drag the variable that differentiates measurements to Sample Number. 458 .18 User’s Guide Sample data WinNonlin 1. If a column includes multiple. 3. 4. concurrent measurements for any given subject. Click Next to set Dosing.

Click the Hide button to exit the Attributes dialog and return to the Sample data. – Analytical Method: assay or technique used to obtain the sample data. e. the Map Dosing Data screen appears. or a unique value for each observation record (from a column in the data set). Dosing 1.) 459 . To set one attribute value for an entire column. 3. If a dosing worksheet is used. – LLOQ: lower limit of quantification for the assay. 1. – Sample Description: further information about the sample. Valid data are “finite. To load attribute values for each record from a column in the data set. drag that column to the appropriate field. 2.g. or missing due to dropout or non-adherence. as needed.” The list of statuses is defined by your PKS administrator. A number value. – ULOQ: upper limit of quantification for the assay. (See “Dosing attributes” below. 2. A number value. 3. Drag the column(s) containing dose amounts to the PKS Fields box. Select a dose variable under PKS Fields and click the Attributes button to set units and other optional attributes. – Unit: units of measurement.Using the Pharsight Knowledgebase Server Creating a study 18 Each attribute can take one value for an entire column. below a limit of quantification (BQL).. – Status: indicates data status. check Static and enter or select the value.

– Unit: units of measurement. To load attribute values for each record from a column in the data set.” The list of statuses is defined by your PKS administrator. To set one attribute value for an entire column. e.g. drag that column to the appropriate field. – State: further text information about the sample. Each can take one value for an entire column. Click the Hide button to exit the Attributes dialog and return to the Dosing. or a unique value for each observation record (from a column in the data set). Valid data are “finite. check Static and enter or select the value. Dosing attributes WinNonlin All attributes are optional.. – Status: indicates dose status. Click Next to continue to the Data Collection Point (DCP) screen. – Dose Description: further text information about the sample. 1. 3. – Analytical Method: assay or technique used to obtain the sample data.18 User’s Guide 4. missing due to dropout or nonadherence. 2. 460 .

Drag variables representing the study period (required for crossover designs only). 2. an indicator variable noting whether a given record contains a measurement of plasma concentration or pain score. for example. Treatment 461 . 3. phase (optional) and other study descriptors to the appropriate boxes. Click Next to set Treatment variables. DCP Description: (Data Collection Point description) use this field for variables that distinguish different measurements stored a single observation column.Using the Pharsight Knowledgebase Server Creating a study 18 Data Collection Point 1.

Study creation This dialog is the final step in study creation. When all necessary data has been entered. uncheck Reload Study.) for each dose. Treatment description: additional descriptive field. d. IM. 2. 3. a confirmation dialog will appear. If the study does not need to remain open in WinNonlin for modeling or other further work. Click Finish when the variable mappings are complete. 2. Route: variable identifying route (oral. Formulation: variable identifying drug formulation for each dose. f. 1. g. c. (See “Creating a study” on page 452. Treatment Code: variable identifying the treatment. 3. e. 462 . Use the Preview button to review the variable mappings.18 User’s Guide 1. b.) WinNonlin Note: The study name cannot be changed in this dialog. The Study creation dialog appears. Once all data are successfully loaded to the PKS. Use Save to save mappings to a re-loadable file. click Next to save the PKS study. Fasted/fed: variable identifying whether subject ate before dosing. IV. Regimen: variable identifying dose regimen. causing study data to be saved to PKS. and the Back button to make any adjustments. Drag variables to the following (optional) to add treatment descriptors per your company’s standards: a. State: additional descriptive field. etc. Click OK.

5. Use the Preferences button to organize the study tree by study name. 6. 2. 7. 4. indication. the preferences select whether all scenarios will be displayed (“All”). 3. In addition. See “Scenario search”. A green scenario icon means that the study data have changed since the scenario last loaded the study data.. Use the Search button to search by study attributes.. Log into PKS as detailed under “Connecting to the PKS” on page 451. that is. Note: A red scenario icon indicates that at least one derived output object is out of sync with its source. compound. Launch WinNonlin.Using the Pharsight Knowledgebase Server Loading a study or scenario 18 Loading a study or scenario To load a PKS study into WinNonlin: 1. from the WinNonlin menus. Click OK. Choose PKS>Load. 463 . Loading the study as Read Only improves performance significantly. Do not leave any windows open. if needed. PKS will close them before loading any study. the final leaf on each scenario branch in the study tree. or only the most-recent (“Latest”). WinNonlin will contact the PKS database and load a tree view of available studies and scenarios. Un-check the Read Only check box if any alterations or additions are to be made to the existing study or scenario (as opposed to creating a new scenario). assigned by your PKS administrator. and/or portfolio. The list is determined by your user access rights.

for studies. including portfolio.18 User’s Guide 8. Select dose data Select the columns to display in the dosing worksheet and click OK to load the data into WinNonlin. and click OK. compound and indication. Scenario search The Search button available while Loading a study or scenario allows a search of available studies by study name. to select data columns to appear in observation and dose worksheets. 464 . Select the study or scenario and click OK to load the scenario or. Select observations WinNonlin Drag and drop columns to include in the observation worksheet when the study is loaded into WinNonlin. scenario name and/or value of study metadata.

Load the PKS study to analyze in WinNonlin as detailed under “Loading a study or scenario” on page 463. in WinNonlin. creating a new version of the scenario. choose PKS>Save from the menus. 3. If analysis results will be saved in the scenario. Each scenario can be updated. Load the model or analysis. it can be added with the scenario. 465 . 2. 4. run the model or analysis. Once the analysis is complete.Using the Pharsight Knowledgebase Server Creating scenarios 18 Creating scenarios Each scenario contains analysis settings and results for one analysis. including any related settings. To save analysis results to PKS: 1. If the PKS study does not include dose data. A scenario can be derived from an existing study or from another scenario.

choose PKS>Save from the WinNonlin menus. WinNonlin Drag study data columns from the left side to the Selected Fields box to add those data columns to the Observations worksheet for the scenario. a dialog will appear requesting names.. Log in to PKS if required. See also “PKS share” on page 475 and “Libraries” on page 476. Such documents may include graphics files.18 User’s Guide 5. or graphics created from WinNonlin charts. The PKS Client for Word and the PKS Client for Excel can view XL and Word documents. 7. To save the updated scenario. Both are required. Adding non-WinNonlin documents to a scenario Non-WinNonlin documents can be saved to (or loaded from) a PKS with a scenario. Click OK. The Scenario Field Selection dialog appears. Other documents can be attached. 6. If any output windows were not saved and named. Click OK. and viewed in an appropriate software program. but may not be viewable until extracted from the PKS. Adding data columns to a scenario After a scenario is created. saved to disk.. The new scenario will be saved with the original study. PKS can display JPG files in the Web Administrator interface. Enter a scenario name and reason for the study (scenario) update. A final “Save complete” dialog will confirm scenario creation. among others. from the WinNonlin menus. you can add columns from the study to the scenario by loading the scenario and choosing PKS>Add Columns. Enter a scenario description (optional). data sets from other analysis programs. Any file can be attached to a scenario. 466 . Microsoft Word reports created with the PKS Reporter.

3. If the study data are loaded as read-only–allowing the user to analyze them in WinNonlin but not change data values in the source data–the loading can be done much more quickly. WinNonlin and the PKS must do a significant amount of work to track any changes to the study data. To commit modifications of the document manager list to the scenario in the PKS. Choose PKS>Other Documents from the WinNonlin menus to open the Document Manager. Select a document and click the Delete button to remove it from the scenario. as part of the current scenario. Click OK when finished.Using the Pharsight Knowledgebase Server Optimizing PKS/WNL performance 18 To save or load a non-WinNonlin document as part of a PKS scenario: 1. 6. 2. 4. 467 . choose PKS>Save in the menus. Read-only access When loading data. Optimizing PKS/WNL performance A number of issues impact the performance of operations between WinNonlin and the PKS. 7. 5. See “Loading a study or scenario” on page 463. Use the Add button to locate and save a file to the PKS. Load the scenario. Select a document in the list and click the Extract button to save it to a hard drive or network location.

Appropriate field mappings Study data fields must be mapped to an appropriate data type when initially loaded from the WinNonlin client into the PKS. Thus. If baseline covariates are mapped as observations rather than subject data. Some data types (such as subject identifiers and subject data) are constant across time.18 User’s Guide Note: The read-only check box on the PKS Load dialog applies to the study data and not the scenario. Actual time differs from nominal time when subject adherence to the protocol schedule is imperfect. This data may also be called baseline covariates or demographics. while others (observation data) vary. read-only. Nominal times are the dose and observation times designated in the study protocol. the more they must be checked when loaded and the slower the loading is. or (4) Observation. and combinations). nominal. Subject identifiers (up to five fields) uniquely identify a subject. WinNonlin 468 . Make sure data that varies over time are mapped as observation data while demographic data are mapped as subject data. the loading/reloading process becomes much slower because those data cells must be checked for changes. WinNonlin and the PKS track any changes to the data. the observation data are checked for any changes. While they can be changed (corrected). the more that data are mapped as observation data. (3) Time (actual. When studies are loaded. In creating a study or scenario and loading/reloading. by their very nature. Only new versions of scenarios can be created. Subject data (unlimited number of fields) are data related to a subject that does not change over time. (2) Subject data. Time data identify the nominal and actual time points for the observed data. The basic data types are (1) Subject identifier. Changes are captured in an audit. Users must have system level rights and right to the original study in order to create a new scenario. they are not normally modified in the course of a study and its analysis. Observation data (unlimited) include sample or dose data that change over time for a given subject. All scenarios are.

Highlight the Text object in the list box and press F2. Therefore. Enter the password and click OK. See “Loading a study or scenario” on page 463. 7. 4. 8. Base Scenario). Once the study is loaded. This allows WinNonlin to load the study data in a more condensed format if read-only study data access is selected. Give the object a name (i. create a text object describing the base scenario (File>New then Text). Enter a scenario name for the base scenario (i.. It is not possible to change the data types in a study once it has been created.e. 5. The logon dialog appears. Creation of a base scenario for a study Loading a scenario is faster than loading a study. File>Close All (it may take some time to close all since a bit of memory is being freed up). 3.e.e. this will be much faster than loading the source study... Scenario Description). Selecting the correct data types can ensure performance is optimized. Load the study from the PKS into WinNonlin. Save the base scenario via PKS>Save. An Update Reason dialog is displayed. create a base scenario that contains all the columns that will be used for a set of scenarios on a study. Modeling and other analysis can be done and saved as new scenarios using PKS>Save As. to save time on subsequent operations. 469 . Base Scenario). All subsequent scenarios should be derived from this base scenario. Name the Text object by right clicking on it. Now all work can be done by loading the base scenario with read-only access to the study. and selecting Object Name. Enter a reason (i. Example: 1.Using the Pharsight Knowledgebase Server Optimizing PKS/WNL performance 18 Note: Mapping fields in the creation of a study is particularly important. Click OK. 6. 2. Click Save.

the system will display the following dialogs to document all changes. 470 . the Audit Information dialog presents only one Reason field. See your system administrator to inquire about standard reason text or click the Lookup button for a list of company-standard values. including date. Note: If 100 records or fewer have changed. A reason for change is required for every alteration to study data. If more than 100 records have changed. Enter information for all changes. This reason will be recorded for every change that was made.18 User’s Guide Note: Make sure that all dosing data are loaded into the study. WinNonlin Change tracking and audit information If any changes are made to a PKS study or scenario. are stored and made available via the PKS web client’s audit trail. Dosing data are considered part of the study and cannot be added after studies are loaded in readonly mode. and for each scenario update. Use the Copy and Paste All buttons to speed the process. and the new data are saved back to the PKS. fill in a reason for each item. All change records. the Audit Information dialog appears. PKS user name and change reason. Audit information If the study data have changed. A change reason must be entered for every changed value in study data. change made.

Worksheets with fewer cells will save individual data values to PKS. Use the Lookup button to select a reason for change from a standard list. Worksheets may be saved as binary files. Use Ctrl+Enter to insert line breaks. Check Native format for large worksheets to save all worksheets larger than a set limit as binary objects. Enter the scenario name (for new scenarios).Using the Pharsight Knowledgebase Server Change tracking and audit information 18 Update reason When saving a new scenario or scenario version. but make data values unavailable to query and other tools within the PKS Web Client. Check items to save in binary format. Charts may be saved in JPG or Windows metafile (WMF) format. rather than saving individual data values to the databse. created by your system administrators. Enter the minimum worksheet size to be saved in binary format under # cells and click Apply. The Format button opens the PKS Save Native Format dialog. update reason and (optional) scenario description. you will see the Update Reason dialog. 471 . rather than saving individual data values to the databse. instead of as Pharsight chart objects. This will speed loading of those data objects to and from PKS. which provides a means to save worksheets and charts in binary formats.

18 User’s Guide WinNonlin Once the Audit Information dialog is complete. Name objects WinNonlin will prompt you to provide names for any unnamed objects in the scenario. then click on it again to edit the name. PKS asks whether to create a base scenario with the updated data and any other changes that have been made in WinNonlin. Dosing information Dose data are added to PKS studies either through a separate dose worksheet at study creation. Clicking Yes adds a new scenario with the current WinNonlin settings and objects. click OK to enter the update reason(s). If no scenario was loaded. or as part of a modeling scenario (via the WinNonlin Dosing 472 . Select an object name.

and one that records the data history (History). See “Dosing worksheet” on page 448 for data requirements. After loading a study or scenario that contains dose information. and applied to different scenarios by including different data columns [Dose_A. after loading the model). Thus different dosing schedules can saved in a study or scenario. Since the IV and Oral formulations require different NCA models. one containing dosing data (Dosing). See “Dosing” on page 448 for discussion. WinNonlin contains a workbook with three worksheets: observation data (Observations). Changing the dose regimen When the dosing data exists on the Dosing worksheet. Dose times and amounts for the two formulations appear in four columns of the Dosing worksheet: • • • • IV_Time IV_Amount Oral_Time Oral_Amount. and WinNonlin can load only one model at a time. Dose_B. it is possible to indicate which fields apply to the current scenario. two scenarios are required: • • Oral scenario: uses NCA model 200 and dose columns Oral_Time and Oral_Amount IV scenario: uses NCA model 201 and dose columns IV_Time and IV_Amount The selection of dose columns for each scenario is made as follows. Non-compartmental analysis will be performed for each. For example. etc.Using the Pharsight Knowledgebase Server Dosing information 18 Regimen dialog. suppose that a PKS study contains time-concentration data for a crossover trial comparing IV bolus and oral drug delivery.] as follows. The Dosing worksheet can contain dose information for one or more analysis scenarios. To choose dosing columns to apply in the current scenario: 1. 473 . Load a study or scenario containing multiple dose regimens as described above.

3. 8. 3. 2. Close any PKS study or scenario that is open in WinNonlin. select the study to which to append data. Note: Append/Update is a convenient way to add dosing information to the study or scenario. additional data can be added to the study. In the Pharsight Knowledgebase Server Load dialog. The Study Creation Wizard appears.. 4. Choose PKS>Append/Update Study. from the menus.. 2. Log in. (See “Logging in” on page 452 for details. from a workbook. or existing data can be updated. The Dosing Field Selection dialog appears. add them beneath the existing study variables. such as extra subject data or other observations. or in which to update data.) 5. The column(s) containing the subject identifiers must be identified. Variables listed in the Non-Dosing Columns box will not be used in the current scenario. 474 . Select PKS>Dosing. Click OK.18 User’s Guide 1. Click OK. Click OK. 7. Drag the dosing fields for the current analysis to the Dosing Columns box. Open the data to be added in a WinNonlin workbook window. 6. to map the variables in the data set to the selected study. Drag variables from the workbook and drop them on or under the corresponding study variables. If there are new data variables. This method can also be used to add subject data to the study. To append/update a study: 1. WinNonlin Appending or updating a study After a study is created in the PKS.

the PKS library feature supports creation of system-level objects that can be referenced by any scenario. 3. First. choose the study and scenario containing the file(s) to be shared. it is possible to specify or convert units in the PKS without loading the data in WinNonlin. 475 . When the scenario is loaded in WinNonlin. • • “PKS share” on page 475 “Libraries” on page 476 PKS share A new analysis in WinNonlin can share non-Pharsight analysis and data files contained in any PKS scenario. Whenever there are significant differences between the actions in WinNonlin and those in PKS. PKS libraries and object shares PKS provides two features for sharing models. Details appear under the following headings. When an analysis that uses shared files is saved to PKS. the documentation systems point them out.Using the Pharsight Knowledgebase Server PKS libraries and object shares 18 Note: The PKS documentation set covers a number of other operations that affect studies and scenarios in the PKS. Log in to the PKS. For example. any shared files may be updated to reflect changes in their source files as described under “Library and share status” on page 478. as described under “Logging in” on page 452. code and other data objects across studies and scenarios. the share associations are saved with the new scenario. any PKS scenario can share data objects that are part of another scenario. In the Pharsight Knowledgebase Server Load dialog. It is also possible to load data into WinNonlin then specify units or convert units—then save them back to the PKS. Choose PKS>Share from the WinNonlin menus. To load files from the Pharsight Knowledgebase Server: 1. Second. data. and maintain a link to that object so that updates are shared. 2.

Libraries Note: Each application has it’s own folder/container for storing the binary files. Select items to load.18 User’s Guide WinNonlin 4. applications cannot see other applications’ library systems. 6. Consequently. Hold down the Ctrl key to select multiple items. Click OK. The Select Share Item(s) dialog will appear. PKS Client may add a library item and it will not be visible by the WinNonlin application. For example. presenting a list of non-Pharsight files that can be referenced in the current analysis. 5. 476 . Click OK to load the items.

The library functions include the following: • • “Loading a library from the PKS” “Adding or updating a library in the PKS” on page 477 Loading a library from the PKS Use WinNonlin to load standardized parameter names. and other files stored as Libraries in the PKS. Log in to the PKS. and other files as Libraries in the PKS. Adding or updating a library in the PKS Use this feature to save standardized parameter names. parameter names. 1. Select the library(ies). Hold down the Ctrl key to make multiple selections. and other items that will be used as standards across many studies and analyses. NONMEM or SAS code. NONMEM or SAS code.Using the Pharsight Knowledgebase Server PKS libraries and object shares 18 WinNonlin can save any type of binary file as a PKS system level object. as described under “Connecting to the PKS” on page 451. The library function is intended to store a small number of files containing code. 4. The Select Library Item(s) dialog will appear. Library items are not associated with any PKS study or scenario. 3. Choose PKS>Libraries>Load from the WinNonlin menus. 477 . Click OK. and load it back again. using the Library function. 2.

To synchronize shared files and/or libraries with the latest versions in the PKS: 1. 2. as described under “Connecting to the PKS” on page 451. choose PKS>Status. 6. Items that are out-of-date can be updated as follows. Click OK. Make any desired changes. following any standards set within your company. Choose PKS>Libraries>Add/Update from the WinNonlin menus. 478 . Library and share status WinNonlin tracks whether PKS share items and Libraries are current with the latest version in the PKS. The Select Library Item(s) dialog appears. 7.18 User’s Guide To save a file to the PKS as a library or update an existing library: WinNonlin 1. Log in to the PKS. 3.. 4. You may save the file to a local disk to save interim changes before uploading them to the PKS. 8. A dialog will appear. from the menus. Click OK after the save operation is complete. enter the Library Description and the Reason for Update. Open the file in the WinNonlin using File>Open. 5. Hold down the Ctrl key to select multiple items. Click Yes. Select the item(s) to be loaded to the PKS and click OK. or load an existing PKS library as described under “Loading a library from the PKS” on page 477. In the PKS SAS Save dialog.. With PKS share and/or library files loaded in WinNonlin. to confirm that you wish to add PKS data. Items that have newer versions available in the PKS appear in red. The Share and Library Status dialog appears.

It shows the following information. – Description: description of the most-recent version of the study. will use this time zone. PKS Information Information about a study and/or scenario loaded into WinNonlin is summarized in the PKS Information dialog. The PKS Information dialog appears. the study is loaded as read-only. To view summary information for the currently-loaded study: 1. – User selected read only: if true. With a PKS study or scenario loaded in WinNonlin. – Type: study type. – Name: study name in PKS. All dates and times. no changes may be saved to PKS. Study information: – Locked: if the study is locked. – Internal ID: study identifier used in PKS. choose PKS>Information from the menus. its data cannot be updated in PKS. Check the items to be synchronized with the latest PKS version and click OK. • • Server Information: time zone of the PKS server.Using the Pharsight Knowledgebase Server PKS Information 18 2. including time stamps for study updates in PKS. WinNonlin will load the latest version from the PKS. entered by the user when the study was last saved or updated. 479 .

However. (The units would be stored in an adjacent column. – BlindingType: study blinding scheme.) In WinNonlin. • • 480 . but they do not affect calculations. – State: state of the study in the PKS. Thus there are situations when valid data from a PKS data set cannot load correctly in WinNonlin until conversions are performed to make units uniform within each column. etc. because each is also designed to work with other tools as well. New units can be added and defined in the PKS. WinNonlin performs calculations and conversions using only its internal set of units.18 User’s Guide – Design: study design. Relative Nominal Time. there are some differences that are important to note. for dosing and observations) use the same time unit. • • • Scenario Name in PKS Scenario Description: entered when the scenario was last saved in PKS. This level determines whether the user can save changes to this study in the PKS. custom units can be entered and carried through operations. they are never required. each record in a given column can use different units. The PKS provides the means perform this conversion. with conversion factors to define relationships to existing units. Units • • Units are required for some fields in the PKS.. Units can be handled at the cellular level in the PKS. Conversions are not available for custom units in WinNonlin. Current user rights: PKS access rights of the PKS user account used to load the study into WinNonlin. WinNonlin WinNonlin and PKS differences WinNonlin is closely integrated with the Pharsight Knowledgebase Server. By default. That is. units are associated with an entire column. While units are useful in WinNonlin. PKS includes all of WinNonlin’s default units. See the PKS user documentation for a description of various levels of user access. PKS requires that all time-dependent data (Relative Actual Time. This is not required in WinNonlin.

WinNonlin analyzes only one drug at a time. If a data set does not include time or subject data. dummy columns must be created.” Time and subject data must be present to save and import data from WinNonlin to the PKS. except that workspaces require all child windows to have filenames prior to being saved. • • 481 .Using the Pharsight Knowledgebase Server WinNonlin and PKS differences 18 Studies and scenarios • Saving scenarios to PKS is analogous to saving a workspace in WinNonlin. The PKS can handle multiple drugs within one study. PKS requires the child windows to have “object names.

18 User’s Guide WinNonlin 482 .

P is the number of parameters. a placeholder. accumulation index actual clock time AIC 483 . N is the number of observations with positive weight. Appropriate only for comparing models that use the same weighting scheme. Italics indicate terms that are defined elsewhere in the glossary. actual time The time that a study dose or observation actually happened. this time may differ from the nominal time.Appendix A Glossary WinNonlin and PKS terminology The following terms and definitions pertain specifically to objects and operations in WinNonlin and its operations with Pharsight Knowledgebase Server. A measure of goodness of fit based on maximum likelihood. When comparing several models for a given data set. available for entering clock times from research data. WRSS is the weighted residual sum of squares. Not used in any WinNonlin calculations. N. In Deconvolution A and alpha refer to the parameters of a polyexponential unit impulse response function of the form: UIR(t) = sum[A(j) * exp(-alpha(j) * t)]. A A In Modeling. j = 1 …. 1 -------------------------------–( λz × τ ) [1 – e ] In PKS. the model associated with the smallest AIC is regarded as giving the best fit. the zero time intercept associated with the Alpha phase. Akaike Information Criterion (AIC). AIC = N log (WRSS) + 2P for modeling in WinNonlin.

This is derived from a noncentral t distribution.anv) Legacy ANOVA command file (*. Sometimes denoted ALPHA_HL. These versions can open and run and AVNOVA command files and call them in scripts. Replaced with the LinMix command file (*.A WinNonlin User’s Guide ˆ ˆ In LinMix. then: b a b AUC AUC = ∫ c ( t ) dt a AUCall 484 AUC computed from time zero to the time of the last Y value. nE and nS are the group sample sizes. for use in scripting. j = 1 …. time curve. .LML) in WinNonlin versions 4. but cannot save to that format.0 and later. the macro rate constant associated with the distribution phase. ANOVA command file (*.ANV) containing all information required to recall and reproduce a given ANOVA model. The p-value of the Anderson-Hauck test statistic. If C(t) denotes the concentration of analyte at time t. A and alpha refer to the parameters of a polyexponential unit impulse response function of the form: UIR(t) = sum[A(j) * exp(-alpha(j) * t)]. and S is the estimate of the residual standard deviation.( A + B ) 2 t = ---------------------------------------------S 1 ⁄ nE + 1 ⁄ nS where the X's are the respective sample means. Alpha half-life The half life associated with the macro constant Alpha. Alpha In modeling. Area under a concentration of analyte vs. and variance components for linear mixed-effects modeling calculations. AndersonHauck pval AndersonTest statistic applicable for testing equality of means in a bioequivalence Hauck statistic study. N. A and B are the lower and upper limits. The test statistic t is: 1 X E – X S – -. In Deconvolution. AIC is defined as: AIC = – 2 L R + 2S where L R is the restricted likelihood evaluated at converged estimates. and s is the rank of the fixed-effects design matrix plus the number of variance parameters.

If a subject drops out of one sequence. AUMCINF = AUMC ∞ 0 AUMClast Area under the moment curve computed to the last observation. AUMClast = AUMC t last 0 B B balanced design The zero time intercept associated with the beta phase. An experimental design in which the number of observations is the same for every combination of the experimental effects. “Macro” rate constant associated with the elimination phase. Area under the moment curve: b AUMC = a b ∫ c ( t )t dt a AUMCINF Area under the moment curve when the time concentration curve is extrapolated to infinity. For example. There are three types of bioequivalence: average. Average bioequivalence states that aver485 Beta Beta half-life bioequivalenc e . AUC computed from time zero to the time of the last positive Y value. Also called BETA_HL. The half life associated with the macro constant Beta. the design is no longer balanced. individual. and population. each with two levels.Glossary B AUCINF AUClast AUMC A AUC from time zero extrapolated to infinity. is balanced if each of the four combinations of formulation by period has the same number of observations. a crossover design with effects for formulation and period. Bioequivalence is said to exist when a test formulation has a bioavailability that is similar to that of the reference.

it is recommended that Lower and Upper bounds be used routinely. For models with parameters that may be either positive or negative.A WinNonlin User’s Guide age bioavailability for test formulation is the same as for the reference formulation. C C The zero–time intercept associated with the absorption phase for an extravascular model or the zero time intercept associated with the gamma phase for a 3 compartment IV bolus model. bounds Upper and lower constraints on the initial parameter estimates for compartmental modeling. carry along variables Cavg 486 . and is similar to population bioequivalence. The WinNonlin default bounds are 0 for one bound and 10 times the initial estimate for the other bound. Individual bioequivalence is meant to assess switchability of products. but will be copied and included in the output data set. The whiskers end at the Max and Min points unless there are outliers beyond 1. If they are used. This can be very valuable if the parameter values become unrealistic or the model will not converge. Although bounds are not always required. BQL Below the lower limit of quantification (LLOQ) for a given assay. box and whiskers plot The box and whiskers plot is designed to show the distribution of values. For steady-state data: computed as AUC(0-τ) divided by τ. Variables that are not required for the current analysis. user-defined bounds are preferred. and the median between them. Profiles with fewer than five data points plot the data points directly with no box. The lower and upper values limit the range of values the parameters can take on during model fitting. every parameter must be given lower and upper bounds. Population bioequivalence is meant to assess equivalence in prescribability and takes into account both differences in mean bioavailability and in variability of bioavailability. The box marks the 25th and 75th percentiles.5 times the H-spread [the distance between the hinges] from the hinges.

g. rather than continuous values.Glossary C CL classification variables A Total body clearance. but not save *. Replaced by Pharsight Model Objects (*.CMD) from an earlier version of WinNonlin. and dosing times. For steady-state data: an estimate of the total body clearance. A command file can contain all information required to reproduce a compartmental or noncompartmental model. Legacy command file (*. Clss/F Cmax Cmin The peak or maximum concentration. CL= Dose/AUC In a LinMix model.CMD files. magnitude of each dose. For steady-state data: the minimum concentration between 0 and τ (corresponding to Tmin). For extravascular models the fraction of dose absorbed cannot be estimated. It may include one or more formulations. command file compound condition number (COND) constant 487 . WinNonlin can open and run. Dosing information such as number of doses. COND is the square root of the ratio of the largest to the smallest eigenvalue of the matrix of partial derivatives. smoking status). Each library model requires specific constants. The concept is used in PKS to group study objects that pertain to the same molecule. See also LinMix command file and table definition file. Condition number of the matrix of partial derivatives.PMO). therefore Clearance for these models is actually Clearance/F where F is the fraction of dose absorbed: Dose = --------------τ AUC 0 clock time CLR Clss. Time expressed as a date and/or time. rather than relative to the first dose or the start of a study. gender. A drug molecule under study. classification variables or covariates are independent variables that employ values from a discrete set (e. Renal Clearance. See also regressors.

etc. 10mg. A variable such as “drug concentration” is usually a dependent variable because it depends on factors like “formulation. suppose that effect Trt has three levels: Placebo.” “period.5=0. See also nested factors. or body weight.5. Independent variables in a LinMix model. linear combinations of levels of an effect. possibly unknown. 10mg=-0. Concentration that achieves 50% of predicted maximum effect in an Emax model. and 20mg.. depending on the particular model's requirements. It may also include the dose cycle. For example. This compares the average of the 10mg and 20mg groups versus placebo.” See also independent variable. to the independent variables. To compare the two dose groups against placebo.” or “time of sample. dosing regimens E E EC50 Denotes Effect. This normally includes the designated dose and the time of the dose. It is not required that data be available at every intersection of the factors. 488 . The convergence criterion determines when WinNonlin can stop with a successful fit. Note that 1-0. Convergence is achieved when the relative change in the residual objective function is less than the convergence criteria. the scheduled times for observations. blood pressure. Generally. Covariates are continuous measurements. convergence criterion covariates crossed factors curve stripping A graphical method for estimating parameters associated with models that can D dependent variable Dependent variables have some functional relationship. such that the sum of the coefficients is zero. construct the contrast: Placebo=1. The schedule of dosing.A WinNonlin User’s Guide contrasts In LinMix. In a LinMix crossed-classification model every level of every factor could be used in combination with every level of every other factor. This method can be used to determine initial estimates required as a prelude to nonlinear least-squares regression. In this way. 20mg=-0. some subject characteristic of interest. be written as a sum of exponentials. the factors “cross” each other. such as age.5.5-0.

Used to denote rate constants when the transfer rates are proportional to the amount in the source compartment. Fraction dissolved over time in-vitro. exponential A graphical method for estimating parameters associated with models that can curve stripping be written as a sum of exponentials. WinNonlin engines are: PK. as in standard linear regression. ANOVA. Scripting language engines execute tasks in WinNonlin using methods. The eigenvalues included in WinNonlin's modeling output are associated with the variance . as in a standard linear model. WinNonlin prints a condition number that is equal to the square root of the ratio of the largest to the smallest eigenvalue. For compartmental modeling. The point of dose input into a structural (PK/PD) model. Generally associated with explanatory variables. Maximum effect. such that Ax = λx for some vector x. LINMIX.Glossary F eigenvalue A An eigenvalue of matrix A is a number λ. The variables can be either qualitative. or quantitative. Eigenvalues and their associated eigenvectors can be thought of as building blocks for matrices. MERGE. F Fabs Fdiss first-order Fraction absorbed over time. for fitting multiple functions to one data set. TRANSFORM. A column in the data set that indicates which observations belong with which model functions. BIOEQUIVALENCE. For steady-state data: % fluctuation is computed as 100*(Cmax-Cmin)/Cavg. Very large values of the condition number may be indicative of instability in the model fitting process. E0 Emax engine Effect at time 0. PLOT. as in traditional ANOVA. DESC. TABLE. estimated from in-vivo time-concentration data. where x is the eigenvector. Required for simultaneous link models. where Cmin and Cmax were obtained between 0 and Tau. This method can be used to determine ini- tial estimates required as a prelude to nonlinear least-squares regression. 489 fixed effects % fluctuation formulation function variable .covariance matrix.

or as an information line above the column titles. group variables are much like ID variables. For the Table Wizard.covariance matrix of the parameters. 1 HM = ⎛ -⎝n ∑ ---⎞ Y⎠ 1 –1 490 . first-order model.A WinNonlin User’s Guide G Gamma The terminal elimination rate associated with a three compartment. group variables allow for heterogeneity of variances while retaining the same mean model. Gamma halflife GaussNewton algorithm geometric mean group variables H harmonic mean The harmonic mean is the reciprocal of the arithmetic mean of the reciprocals of the Ys (Y > 0). A model fitting algorithm. They are printed only when their value changes and at page breaks. with the exception that only unique values of group variables are printed. Multiple group variables may be used. The half-life associated with the macro constant Gamma. This requires that each Y > 0. Modifications of the Gauss-Newton algorithm include the Hartley and Levenberg modifications. Sometimes denoted GAMMA_HL. Note that group variables may appear as a column in the body of the table. Hartley Modification: This method is not as robust as others but is very fast. In LinMix. This method requires the estimated variance . a group variable breaks the data into separate plot lines. This method is the default. as selected using the Tools>Options menu item. 1 2 3 n In the Chart Wizard. The geometric mean is the nth root of the product of the n Y's: GM = n Y Y Y …Y . including ill-conditioned data sets. however the Levenberg modification suppresses the magnitude of the change in parameter values from iteration to iteration to a reasonable amount. Levenberg Modification: A model fitting algorithm that performs well on a wide class of problems.

of the response may be inhibited or stimulated. For LinMix ANOVA and regression models. and its related uses in formulation development. Interaction is the degree to which the effect of one factor differs with varying values of another factor. For example: dF F(P(1) + Δ) – F(P(1)) -------------. if a treatment has a different effect in one period than it does in another period. The default is 0. and that this may be a convenient way to include units. In-vitro in-vivo correlation. The program estimates these derivatives using a forward difference. R.g.001. the value of the response when all effects are zero. kout. formulation) or do not depend on other factors. Indirect Pharmacodynamic Response Model. Factors controlling the input or production. of a response may be either inhibited or stimulated. a model describing the relationship between dissolution in-vitro and absorption in-vivo. For example. or determinants of loss. When summary statistics are selected these variables are not considered. Nonlinear algorithms require derivatives of the models with respect to the parameters. Note that multiple ID variables may be used. The measured response. interaction intercept IPR iterative reweighting IVIVC 491 . See also dependent variable.≅ -------------------------------------------------------dP ( 1 ) Δ where Δ = Increment times the parameter value.Glossary I A I ID variables Used by the Table Wizard to identify data in adjacent “Data” variable columns. increment for partial derivatives independent variable indirect response models The independent variables are controlled by the experimenter (e. kin. to a drug may be produced by an indirect mechanism. then Treatment and Period are said to interact.. Interaction terms would be included in a LinMix or Bioequivalence model using the “*” symbol: Treatment*Period. A weighting scheme where weights associated with the individual observations are functions of the predicted values.

A

WinNonlin User’s Guide

J

join variables

In the Tables Wizard, when merging data from two workbooks into one table, joins identify corresponding variables to be used in merging profiles from the two data sets, e.g., “Subj” and “Subject.”

K

K half-life K01

The half-life associated with the rate constant K. Sometimes denoted K_HL. The rate at which the drug enters the central compartment from outside the system. It is sometimes denoted as Ka. The rate at which the drug leaves the system from the central compartment. The elimination rate. The half-life associated with the rate constant K10. Often denoted K10_HL. For first-order rate constants; the rate from Compartment i to Compartment j. Zero-order input or infusion rate constant. Michaelis constant for models involving nonlinear kinetics.

K10

K10 half-life Kij K0 KM

L

lag time (LT)

Lag time is a time delay between drug administration and the onset of absorption. An example of such a process is the stomach-emptying time, following oral administration of a drug that is not absorbed from the stomach. When a drug is administered by bolus intravenous injection there is no lag phase. phase. This is estimated via linear regression of time vs. log concentration.

Lambda Z (λz) First order rate constant associated with the terminal (log-linear) elimination least squares fitting least squares means

Minimizes the sum of the squares of the deviations between the observed and predicted values of the dependent variable. Least squares means are estimates of what the mean values would have been had the data been balanced (see balanced design). That is, least squares means are the means predicted by the fixed-effects model. If a data set is balanced, the least square means will be identical to the raw (observed) means.

492

Glossary M linear kinetics linear regression

A

A model in which the transfer rates are first order. A model relating an independent variable to dependent variables such that the unknown parameters enter into the model linearly. In the simplest case, a straight line is fit to the data. The intercept and slope are determined by least squares. The slope and intercept are called model coefficients, or parameters. File (*.LML) storing all information needed to reproduce a LinMix analysis, often used in scripting. See “Linear Mixed Effects Modeling” on page 323. Lower limit of quantification, the concentration whose lower bound (of variability in the assay) includes zero. Analyte may be detectable below this concentration even though it is not quantifiable.

LinMix command file LLOQ

M

macro rate constants

Compartmental models involving first order kinetics can be written as a sum of exponentials. The associated parameters are called macro rate constants and are functions of the underlying micro rate constants. Maximum value. The arithmetic average. Arithmetic average of the logarithms of the Ys. The variance is normally a function of weighted residual SS/df, or the mean square. For certain types of problems, when computing the variance the mean square needs to be set to a fixed value. It is possible to use a specified value other than the residual mean square in the estimates of the variances and standard deviations. This value replaces the residual sum of squares in the estimates of variances and standard deviations.

max mean mean of the logs mean square

method

In the scripting language, a method is a set of computer code that performs functionality on an object or engine. For example, the method 'LOAD' causes an associated data file (an object) to be loaded. The method 'RUN' causes an associated PK Model (engine) to run the model. A type of model often employed for models involving nonlinear kinetics. The V = V max × C ⁄ ( K m + C ) Michaelis-Menten equation is:

493

MichaelisMenten

A

WinNonlin User’s Guide

micro rate constants min MRT

Transfer rates associated with a compartmental model. See also macro rate constants. Minimum value. Mean Residence Time is the average amount of time a particle remains in a compartment or system, equal to AUMC / AUC. Mean Residence Time when the drug concentration profile is extrapolated to infinity. This can be estimated from single dose data or steady state data. Mean Residence Time when the drug concentration profile is not extrapolated to infinity, but rather is based on values up to and including the last measured concentration: MRTlast = AUMClast / AUClast

MRTINF

MRTlast

N

N Nmiss Nelder-Mead simplex algorithm

In modeling output, the number of observations with non-missing data. In modeling output, the number of observations with missing data. A model fitting algorithm that does not require the estimated variance—covariance matrix as part of its algorithm. It often performs very well on ill-conditioned data sets (data sets that do not contain enough information to precisely estimate all of the parameters in the model). When used to fit a system of differential equations, it can be very slow unless the model has been compiled. nested factor are unrelated to one another and are nested within a level of another factor. For example, in a crossover study, Subjects are assigned to Treatment Sequences. This nested factor would be included in a LinMix or bioequivalence model using parentheses: Subject(Sequence). Subjects are not crossed with Sequences, because the subjects within one sequence are unrelated to the subjects within another sequence. See also crossed factors.

nested factors In a nested classification, unlike a crossed classification, the levels of the

Nobs nominal time

The number of observations. The time specified in the protocol for a dose or observation—which may differ from the actual time that a dose or observation was taken. Noncompartmental, or model independent analysis describes the concentration-time curve by estimating noncompartmental parameters such as: area

noncompartm ental analysis 494

Glossary O

A

under the curve (AUC), Tmax, Cmax, and Lambda Z (the rate constant associated with the terminal elimination phase), among others. Noncompartmental parameters are estimated using the Linear or Linear/Log trapezoidal rule. If Lambda Z can be estimated, then the noncompartmental parameters will be extrapolated to infinity.

nonlinear kinetics

Kinetics that can be modeled by nonlinear differential equations, for example, V max C dc V ---- = ---------------dt Km + C

nonlinear model

In nonlinear models, at least one of the parameters appears as other than a coefficient of an independent variable. One such model is the sum of two or more exponentials, such as Y = A1*exp(-B1*X) + A2*exp(-B2*X). A nonlinear model is one for which one or more of the partial derivatives with respect to the model parameters involve model parameters. Note that nonlinear models may arise from compartmental models having either linear or nonlinear kinetics. The process of fitting nonlinear models to data.

nonlinear regression

O

ODBC

Open Database Connectivity. A standard for data exchange, in common use on Microsoft Windows platforms. It is an application programming interface for database access that uses Structured Query Language (SQL) as its database access language. In scripting, objects are described by one or more property/ies, and acted upon by methods. In WinNonlin, there are five main objects: data (workbook) objects, text objects, model objects, graph objects, and WinNonlin itself.

object

P

partial tests

LinMix tests of whether each model term contributes to the fit after all other model terms have been included. Partial tests are invariant to the model specification order.

495

A

WinNonlin User’s Guide

percentiles

The pth percentile divides the distribution at a point where p percent of the distribution are below this point. Models of drug effect as a function of a drug concentration.

pharmacodyn amic (PD) models

pharmacokine Models of drug concentration as a function of time. tic (PK) models PK/PD link

A PK/PD Link model uses drug concentration data to estimate concentrations at an effect site, which are then used to model the pharmacodynamics. A confidence interval obtained from the tangent planes to the joint confidence ellipsoid of all the parameter estimates. In the Bioequivalence Wizard, the probability of detecting a difference greater than or equal to a specified percentage of the reference mean. A set of observed data to be fit at one time, producing one set of individual model parameters. A profile is identified as data records with a given, unique set of values for all sort variables. Attribute or characteristic of a object in scripting, defining how the object should appear and interact. Properties also set values for an object or engine.

planar confidence interval power

profile

property

Q R

random effect

Also known as covariance parameters, random effects distinguish a linear mixed effects models from standard linear models. In general, random effects are additional unknown random parameters assumed to impact the variability of the data. The variances of the random effects, commonly known as variance components, become the covariance parameters for this structure. Rank of the matrix of partial derivatives. If the matrix is of full rank, Rank= # of parameters. If the Rank is less than the number of parameters, then there is not enough information in the data to estimate all parameters in the model.

rank

496

Glossary S regressors

A

In a LinMix model, regressor variables or covariates are independent variables that employ values from a continuous set (e.g. temperature, body weight), rather than categories. See also classification variables. In PKS, the time that observations or doses really happened, as opposed to relative nominal time. See also relative actual end time. In PKS, the times when infusion doses actually ended, as opposed to relative nominal end time. See also relative actual time. In PKS, scheduled dose and observation times, as set in the study protocol. Compare with relative actual time. See also relative nominal end time. In PKS, the times when infusion doses were scheduled to end, as set in the study protocol. Compare with relative actual end time. See also relative nominal end time. Elapsed time since the first dose or the beginning of the study, period or phase. Observed minus predicted values, where the predicted values come from fitting the specified model.

relative actual time relative actual end time relative nominal time relative nominal end time relative time

residuals

S

S

Standard error of the weighted residuals: S = ( WRSS ) ⁄ ( DF ) where WRSS is the weighted residual sums of squares, and DF is the number of observations minus the number of zero weights and the rank of the variance/ covariance matrix. The actual times at which the concentration or effects are observed. Schwarz Bayesian Criterion. A measure of goodness of fit based on maximum likelihood. When comparing several models for a given data set, the model with the smallest SBC value is regarded as giving the best fit. Appropriate only for comparing models that use the same weighting scheme. SBC = NOBS × log ( WRSS ) + log ( NOBS ) × NPARM ˆ For LinMix, the SBC calculation is: SBC = – 2 L R + s log ( N – r ) where ˆ ˆ is the restricted log-likelihood function evaluated at final estimates β L

R

sampling times SBC

497

A

WinNonlin User’s Guide

**ˆ and θ ; r is dim(x); s is the number of parameters; and N is the number of observations.
**

scenario

In PKS, a set that includes one structural model, initial parameters, and/or statistical tests to be applied to a data set for fitting or analysis. See also study. Standard deviation. Standard deviation of the logs of the Ys. Standard error. Parameters that are functions of the primary model parameters. For example, AUC and half lives are secondary parameters. Note that, in the case of multiple-dose data, secondary parameters dependent on dose use the first dose. Standard error of Predicted Y. Denotes the “S” shaped nature of certain models. A separate analysis or plot is done for each level of the sort variable(s). If gender and smoking are sort variables, a separate set of analysis results or plots will be created for each of female non-smokers, male non-smokers, female smokers and male smokers. In these models, which can be written as a sum of exponentials, dose does not appear explicitly in the model. Rather, the macro constants are a function of dose. Thus, when fitting macro constant models with user-specified initial estimates, the user must specify the associated dose. For single dose macro constant models, when WinNonlin determines initial parameter estimates by curve stripping, the stripping dose is identical to the administered dose.

SD SD of the Logs SE secondary parameters

SE - Yhat sigmoid sort variables

stripping dose Dose associated with initial parameter estimates for macro constant models.

structural model study

The set of functions defining the shape of a model, i.e., a structural PK, PD, or PK/PD model. It does not include parameter distributions. In PKS, a study comprises a collection of data that can include observations, dosing, and protocols. A study can also contain one or more scenarios. The variable(s) for which descriptive statistics will be calculated.

summary variables

498

Glossary T

A

T

table definition A file (*.TDF) that stores a user-defined table format and settings. See also file “Table definition files” on page 148. Tau (τ) TI Tmax Tmin

The (assumed equal) dosing interval for steady-state data. Infusion length. The time of peak concentration. For steady-state data: time of minimum concentration based on samples collected during a dosing interval. A combination of one or more drugs given at specific doses and time(s).

treatment

U

UIR

Unit impulse response function, describing the PK profile following an instantaneous dose of one unit of drug into the central compartment. A confidence interval based on standard error. The univariate confidence interval does not take into account correlations with other parameters. Libraries of user-defined ASCII models. See “User Models” on page 217.

univariate confidence interval user libraries

V

V variance

**Volume of distribution of the central compartment. This is also denoted VC.
**

n

The variance, v, in descriptive statistics:

∑ ( Yi – Y )

i=1

2

⁄ (n – 1)

Note:

VIF

SD =

v

Variance Inflation Factor, the multiplier of the residual mean square, used in deriving the asymptotic variance of the parameter estimates. Useful in simulations, to find optimal experimental design. Smaller values are better. Theoretical maximum rate of process in Michaelis-Menten kinetics.

499

Vmax

A

Vss

WinNonlin User’s Guide

An estimate of the volume of distribution at steady state. Vss = MRTINF*Cl, where MRT = Mean Residence Time and CL= total body clearance. For steady-state date, Vss = MRTINF*Clss. Not computed for model 200, as calculating MRT from oral data requires estimating Mean Input Time, which requires compartmental methods. Volume of distribution based on the terminal phase. For extravascular models, the fraction of dose absorbed cannot be estimated. Therefore, Volume for these models is actually Volume/F, where F is the fraction of dose absorbed.

VZ, VZ/F

W

weight variable

The weight variable is a numeric data column used in LinMix fitting to compensate for systematic differences in variance across observations, e.g., variance that increases with observed value. Each observation is multiplied by the square root of its weight. Weight values should generally be proportional to reciprocals of variances. This variable should not be included with classification variables, but can be among regressors or used in the model. Square root of the weight times Y Predicted. Predicted Y times the square root of the weight. Residual Y times the square root of the weight.

weighted computed Y weighted predicted Y weighted residual Y

weighted - SS Weighted Sum of Squared Residuals. See also WRSS. Westlake confidence interval workspace

Confidence intervals that are constrained to be symmetric about 100%.

A workspace saves all open windows and settings in WinNonlin, including model and/or analysis settings. Weighted residual sums of squares, an estimate of the variance of the residuals. This is the quantity WNL minimizes during modeling.

WRSS

500

Glossary X

A

X

X variable

The independent variable to be used in PK modeling or Noncompartmental analysis, usually, time relative to the first dose.

Y

Y variable

The dependent variable to be used PK modeling or Noncompartmental analysis, generally Concentration.

501

A

WinNonlin User’s Guide

502

Appendix B

**WinNonlin Model Libraries
**

Parameterization and equations for WinNonlin compiled and ASCII models

WinNonlin is distributed with an extensive library of models, including most of the commonly used one, two and three compartment pharmacokinetic (PK) models, linear, pharmacodynamic (PD), link, indirect responseA and Michaelis-Menten models, as well as seven “models” for noncompartmental analysis of plasma, urine or drug effect data. The following are included in the library of compiled models and detailed in the sections below, following the syntax noted under “Notation and conventions” on page 504.

Table B-1. Model libraries Model “Noncompartmental analysis” on page 505 “Pharmacokinetic models” on page 519 “Pharmacodynamic models” on page 533 “PK/PD link models” on page 536 “Indirect response models” on page 539 “Michaelis-Menten models” on page 542 “Simultaneous PK/PD link models” on page 545 “Linear models” on page 546 “Sigmoidal/dissolution models” on page 547 with IVIVC licence only Compiled Yes Yes Yes Yes Yes No No Yes Yes ASCII No Yes Yes No No Yes Yes No No

Some of the models are provided in ASCII (text) model libraries; others are compiled into machine language. Each PK and PD model is included in both

A. Dayneka NL, Garg V, Jusko W (1993); Jusko W (1990); Nagashima R, O’Reilly RA, Levy G (1969). 503

B

WinNonlin User’s Guide

ASCII and compiled forms. Compiled models run faster. The ASCII versions provide the user with examples and starting templates for custom models. Note: When using the ASCII models, be sure to examine the ASCII code to ascertain what dosing constants need to be created. Although WinNonlin can be used to estimate the parameters in any system of nonlinear equations, these libraries contain classical PK and PD models. Note that it is possible to create custom model libraries, as described under “User Models” on page 217.

**Notation and conventions
**

WinNonlin Model Libraries number the central compartment 1, and the exterior to the compartments, zero. First-order rate constants are expressed as Kij, meaning the rate from Compartment i to Compartment j. Thus, K01 is the rate at which drug enters the central compartment from outside the system, and K10 is the rate at which drug leaves the system from the central compartment. Some models compute secondary parameters, such as half-lives and areas, based the estimates of the primary parameters. WinNonlin also estimates standard errors for these secondary parameters. For the two and three-compartment models some pharmacokineticists prefer to estimate the micro rate constants (Kij) as primary parameters; others prefer to estimate the macro rate constants (alpha and beta). The WinNonlin libraries contain both forms. Thus, Model 11, for example, is the two-compartment model with first-order input and alpha and beta as secondary parameters; Model 13 is the same model with alpha and beta as primary parameters and the rate constants K01, K12, and K21 as secondary parameters. Since these models are normally used to fit concentration data as a function of time, t represents the independent variable, and C(t) is the mathematical equation being fit to the observations. All of the models require the dose(s) and time(s) of dosing to be input as a constants. In addition, for those models defined in terms of the macro constants, the dose associated with the initial parameter estimates (this dose is hereafter referred to as the stripping dose) must also be input. The reason for this is as follows. Such models can be written as: C(t) =

504

∑ Ai × exp ( –αi × t )

WinNonlin Model Libraries Noncompartmental analysis

B

Note that the dose, D, does not appear directly in the expression for C(t). Instead, the intercepts, {Ai}, are functions of D (as well as the micro constants). To support multiple-dose data, WinNonlin defines the macro constants model as: C(t) =

∑ Dj × ----- × exp ( αi × t ) D

j

Ai

where Dj denotes the jth administered dose. Writing the model in this way enables WinNonlin to fit multiple dose data for the macro constant models. In order to do so, WinNonlin requires the stripping dose (D). With the exception of the Noncompartmental Analysis “models,” WinNonlin models assume that the Time variable has been coded such that the time of the first dose is 0. This applies for both WinNonlin modeling and simulation. The factor that converts dose amount to concentration is denoted as V, volume. In absorption models there is no way to estimate both volume and fraction of dose absorbed, so both of these are included in V (thus V/F is estimated where F is the fraction of the dose that was absorbed). The presentation of the mathematical equations of the models follows FORTRAN notation. For example, exp(-K01 · t) is the constant e with an exponent of the negative product of time, t and the rate constant K01. The models with first-order input are presented with and without a lag time. Since this amounts only to a shift on the time axis, the lag time is not explicitly shown in the expression for C(t), and the estimated Tmax should have the lag time added to it for those models that include a lag time. Note: For discussion of one and two-compartment models with first-order exchange see Gibaldi and Perrier (1982).

Noncompartmental analysis

The WinNonlin Model Libraries include seven compiled “models” for noncompartmental analysis of plasma, urine or drug effect data.

505

dose interval (Tau)a Plasma or blood Extravascular Plasma or blood Bolus IV Plasma or blood Constant infusion Dose. time of last dose. Models 200-202 can be used for either single-dose or steady-state data. see Gouyette (1983) and Chan and Gilbaldi (1982). time of last dose. Tau is required for steady-state data only. Noncompartmental analysis models Model Type of data 200 201 202 210 211 212 220 Dose input Constants required Dose. time of last dose Dose. baseline effect. dose interval (Tau)a Dose. 506 . NCA output parameters and their computation formulas NCA output parameters and their computation are covered under: • • • • “Plasma or serum data” on page 507 “Urine data” on page 512 “Sparse sampling (pre-clinical) data” on page 514 “Drug effect data (model 220)” on page 515 See also “Computational rules for WinNonlin’s noncompartmental analysis engine” on page 181 and “Noncompartmental Analysis” on page 159. time of last dose Dose time. threshold (optional) Constant infusion Dose.B WinNonlin User’s Guide Table B-2. Models 210-212 (urine concentrations) assume single-dose data. and that the data are from a “final” dose given at steady-state. time of last dose a. For steady-state data. dose interval (Tau)a Urine Urine Urine Drug effect Extravascular Bolus IV Any Dose. time of last dose. For additional reading. length of infusion. the computation assumes equal dosing intervals (Tau) for each profile.

When using steady-state data. if the concentration at the dosing time is missing. if the input data do not contain an observation at time zero. or the elapsed time since the time of the first dose. Assumed to be zero unless otherwise specie fied. “Parameters that do not require λz. Output Models 200-202 estimate the parameters in the following tables. Given only for bolus IV models. then that concentration is set equal to the minimum concentration observed in the dosing interval. the entire curve is considered.” on page 507 Table B-4.” on page 511 Table B-3. Maximum observed concentration divided by dose. it is estimated by back-extrapolating from the first two concentration values according to the rules given under “Insertion of initial time points” on page 182. Time of maximum observed concentration. Tlag Tmax Extravascular input (model 200) only. Tmax corresponds to points collected during a dosing interval. Parameters that do not require λz C0 Initial concentration. Tlag is the time prior to the first measurable (non-zero) concentration.WinNonlin Model Libraries Noncompartmental analysis B Plasma or serum data Data structure NCA for blood concentration data requires the following input data: • • Time of each sample Plasma or serum concentrations For extravascular and constant infusion models (models 200 and 202). For steady-state data. 507 Cmax Cmax_D . Used mainly with steady-state data. which may code time as the time elapsed since the first dose. “Calculations for steady-state data. Dosing_tim Time of last administered dose.” on page 508 Table B-5. “Parameters available only if λz is estimable. Maximum observed concentration. WinNonlin will generate one. For non-steady-state data. • • • Table B-3. It is equal to the first observed concentration value if that value occurs at the dose time. occurring at Tmax. Otherwise.

For non-infusion models: MRTlast = AUMClast/ AUClast. If λz cannot be estimated. but can be obtained by checking Output intermediate calculations in the Model Options. zero. Lambda_z AUC_%Bac Computed for Bolus IV (model 201) only: the percentage of AUCINF that was k_ Ext due to back extrapolation to estimate C(0). Area under the moment curve from the time of dosing (Dosing_time) to the last measurable concentration. and parameters called “*_pred” used the predicted value for Clast. otherwise. Area under the curve from the time of dosing (Dosing_time) to the last measurable concentration.Lambda_z * Tlast). the C(0) is estimated by log-linear regression of the first two time points. Concentration corresponding to Tlast.B WinNonlin User’s Guide Table B-3.TI/2. If C2<C1. _upper Note: Parameters that are extrapolated to infinity. For infusion models: MRTlast = (AUMClast/AUClast) . in Table B-4 below. which is defined as: Clast_pred = exp(intercept . Area under the curve from the time of dosing (Dosing_time) to the time of the last observation. Parameters that do not require λz (continued) Tlast Clast AUClast AUCall Time of last measurable (positive) concentration. The value for intercept is not given in the normal output. The values intercept and lambdaZ are those values found during the regression for Lambda_z. adjusted for the number of points used in the estimation of λz. Otherwise. Goodness of fit statistic for the terminal elimination phase. AUMClast MRTlast No_points_ Number of points used in computing λz. If the last concentration is non-zero AUClast=AUCall. 508 . AUCall will be greater than AUClast as it includes the additional area from the last measurable concentration down to zero. are calculated two ways: Parameters called “*_obs” use the last observed value for Clast. Table B-4. AUClower (Optional) user-requested area(s) under the curve from time lower to upper. Parameters available only if λz is estimable Rsq Rsq_adjusted Goodness of fit statistic for the terminal elimination phase. C(0) is set to C1. Mean residence time from the time of dosing (Dosing_time) to the time of the last measurable concentration.

r Lambda_z_uppe Upper limit on Time for values to be included in the calculation of λz. based on the last observed concentration (obs). i. Estimated by linear regression of time vs.WinNonlin Model Libraries Noncompartmental analysis Table B-4. Lambda_z_lowe Lower limit on Time for values to be included in the calculation of λz.⋅ 100 AUCINF_obs Dose -----------------------------------------λ z ⋅ AUCINF_obs Volume of distribution based on the terminal phase. = Dose/AUCINF_obs. based on the last predicted concentration. log concentration. First order rate constant associated with the terminal (log-linear) portion of the curve. Vz_F_obsa AUCINF_obs – AUClast ----------------------------------------------------------. r HL_Lambda_z AUCINF_obs Terminal half-life = ln ( 2 ) ⁄ λ z AUC from Dosing_time extrapolated to infinity. AUC from Dosing_time extrapolated to infinity. = Vz_obs.. below. = Cl_obs. obs = AUClast + -------------------- Clast λz AUCINF_D_obs AUCINF_obs divided by dose. See also AUCINF_pred. AUC_%Extrap _obs Percentage of AUCINF_obs due to extrapolation from Tlast to infinity. Cl_F_obsa AUCINF_pred Total body clearance for extravascular administration. Percentage of AUCINF_pred due to extrapolation from Tlast to infinity.⋅ 100 AUCINF_pred 509 . Parameters available only if λz is estimable (continued) Corr_XY Lambda_z B Correlation between time (X) and log concentration (Y) for the points used in the estimation of λz.e. = AUCINF_pred – AUClast -------------------------------------------------------------. concentration at the final observation time estimated using the linear regression performed to estimate Lambda Z. pred = AUClast + ---------------------- Clast λz AUCINF_D_pre d AUC_%Extrap _pred AUCINF_pred divided by dose.

Note that for oral model 200. = Tlast ⋅ Clast pred Clast pred AUMClast + --------------------------------------.TI/2. MRTINF includes Mean Input Time as well as time in systemic circulation. MRTINF includes Mean Input Time as well as time in systemic circulation.+ ---------------------2 λz λz AUMC_%Extrap Percent of AUMCINF_pred that is extrapolated.TI/2. Parameters available only if λz is estimable (continued) Vz_pred.⋅ 100 AUMCINF_obs AUMCINF_pred Area under the first moment curve (AUMC) extrapolated to infinity.B WinNonlin User’s Guide Table B-4. MRTINF_pred 510 . Mean residence time (MRT) extrapolated to infinity. For non-infusion models = AUMCINF_obs/AUCINF_obs. _obs AUMCINF_obs – AUMClast = ---------------------------------------------------------------------. = Tlast ⋅ Clast obs Clast obs AUMClast + -----------------------------------. Vz_F_preda Volume of distribution based on the terminal phase. = Dose/AUCINF_pred Area under the first moment curve (AUMC) extrapolated to infinity. where TI represents infusion duration.. Note that for oral model 200. _pred AUMCINF_pred – AUMClast = -----------------------------------------------------------------------.⋅ 100 AUMCINF_pred MRTINF_obs Mean residence time (MRT) extrapolated to infinity. based on the last observed concentration. based on the last predicted concentration. Cl_F_preda AUMCINF_obs Total body clearance for extravascular administration. For infusion models = (AUMCINF_obs/AUCINF_obs) . concentration at the final observation time estimated using the linear regression for Lambda Z. where TI represents infusion duration. i.+ -------------------2 λz λz AUMC_%Extrap Percent of AUMCINF_obs that is extrapolated.e. For non-infusion models = AUMCINF_pred/AUCINF_pred. For infusion models = (AUMCINF_pred/AUCINF_pred) . = -------------------------------------------- Dose λ z ⋅ AUCINF_pred Cl_pred.

the fraction of dose absorbed cannot be estimated. Mean residence time (MRT) extrapolated to infinity. Minimum concentration between dose time and dose time + Tau (at Tmin). for Cmin and Cmax between dose time and Tau. based on the last observed (obs) or last predicted (pred) concentration. AUC from dose time to Tau divided by (dose time + Tau). Parameters available only if λz is estimable (continued) VSS_obs VSS_pred B An estimate of the volume of distribution at steady state = MRTINF*Clss. therefore Volume and Clearance for these models are actually Volume/F or Clearance/F where F is the fraction of dose absorbed.WinNonlin Model Libraries Noncompartmental analysis Table B-4. a.– ---τ 2 AUC 0 τ τ τ τ Infusion: where TI represents infusion duration. Not computed for model 200. 511 . The following additions and changes are made for steady-state data. Non-infusion: AUMC 0 + τ ( AUCINF_obs – AUC 0 ) -------------------------------------------------------------------------------------------τ AUC 0 AUMC 0 + τ ( AUCINF_obs – AUC 0 ) TI -------------------------------------------------------------------------------------------. Clss_Fa The (assumed equal) dosing interval for steady-state data. Time of minimum concentration sampled during a dosing interval. Calculations for steady-state data Tau Tmin Cmin Cavg Accumulation Index Clss. For extravascular models (model 200). Table B-5. = 1 ---------------------–λz τ 1–e Dose ---------------τ AUC 0 An estimate of the total body clearance = Cmax MRTINF_obs Maximum concentration between dose time and dose time + Tau (at Tmax). %_Fluctuation = 100*(Cmax-Cmin)/Cavg. as MRTINF for oral models includes Mean Input Time as well as time in systemic circulation and therefore is not appropriate to use in calculating Vss.

For extravascular models (model 200).Starting time) 512 . Calculations for steady-state data (continued) MRTINF_pred Mean residence time (MRT) extrapolated to infinity. a. Vz_Fa = Dose ------------------------τ λ z ⋅ AUC 0 VSS_obs VSS_pred An estimate of the volume of distribution at steady state = MRTINF*Clss.B WinNonlin User’s Guide Table B-5. based on the last observed (obs) or last predicted (pred) concentration. as MRTINF for oral models includes Mean Input Time as well as time in systemic circulation and therefore is not appropriate to use in calculating Vss. Non-infusion: τ τ AUMC 0 + τ ( AUCINF_pred – AUC 0 ) ---------------------------------------------------------------------------------------------τ AUC 0 AUMC 0 + τ ( AUCINF_pred – AUC 0 ) TI ---------------------------------------------------------------------------------------------. models 210-212 compute the following for the analysis: • • Midpoint of each collection interval Excretion rate for each interval (amount eliminated per unit of time) = (Concentration * Volume) / (Ending time . the fraction of dose absorbed cannot be estimated. Not computed for model 200. Vz. Urine data Data structure NCA for urine data requires the following input data: • • • Starting and ending time of each urine collection interval Urine concentrations Urine volumes From this data. therefore Volume and Clearance for these models are actually Volume/F or Clearance/F where F is the fraction of dose absorbed.– ---τ 2 AUC 0 τ τ Infusion: where TI represents infusion duration.

Parameters that are estimated only when λz is estimable Rsq Rsq_adjusted Corr_XY Lambda_z Goodness of fit statistic for the terminal elimination phase. Midpoint of collection interval associated with the maximum observed excretion rate. First order rate constant associated with the terminal (log-linear) portion of the curve. Maximum observed excretion rate. Correlation between midpoints and log excretion rates for the points used in the estimation of λz. log excretion rates. Midpoint of collection interval associated with Rate_last. 513 Lambda_z_lower . AUClower_upper (Optional) partial area(s) under the curve from time lower to upper. Area under the urinary excretion rate curve from time 0 to the last measurable rate. Goodness of fit statistic for the terminal elimination phase. adjusted for the number of points used in the estimation of λz. = Σ(Concentration * Volume) Sum of Urine Volumes 100*Amount_Recovered/Dose Area under the urinary excretion rate curve from time 0 to the last rate. No_points_lambda_z Number of points used in the computation of λz. This equals AURC_last if the last rate is measurable. Last measurable (positive) rate. Table B-6. Lower limit on midpoint for values to be included in λz estimation. Table B-7. Midpoint prior to the first measurable (non-zero) rate. Cumulative amount eliminated. Parameters that do not depend on λz Dosing_time Tlag Tmax_Rate Max_Rate Mid_Pt_last Rate_last AURC_last Amount_ Recovered Vol_UR Percent_ Recovered AURC_all Time of the last administered dose (assumed to be zero unless otherwise specified).WinNonlin Model Libraries Noncompartmental analysis B Output Models 210-212 estimate the following parameters. Lambda Z. This is estimated via linear regression of midpoints vs.

Area under the urinary excretion rate curve extrapolated to infinity. SE_Cmax SE_AUClast Standard error of data at Tmax (time of maximum mean concentration).B WinNonlin User’s Guide Table B-7. Standard error of the area under the mean concentration curve from dose time to Tlast. Standard error of the area under the mean concentration curve from dose time to the final observation time. Note that AURC_INF is theoretically equal to the amount recovered but will differ due to experimental error. it estimates the same parameters normally calculated for plasma or urine data. AURC_INF_pred AURC_%Extrap_pred Percent of AURC_INF_pred that is extrapolated. the data are treated as a special case of plasma or urine concentration data. plus those listed below.e. The NCA engine computes the mean concentration or rate at each unique time value or interval. based on the last observed excretion rate. Using the mean concentration curve across subjects. based on the last predicted rate. Parameters that are estimated only when λz is estimable Lambda_z_upper HL_Lambda_z AURC_INF_obs Upper limit on midpoint for values to be included in λz estimation.. Sparse sampling (pre-clinical) data When an NCA model is loaded with the Sparse Sampling option (see “NCA model selection” on page 160). Plasma or serum concentration parameters Sparse sampling methods for plasma data (models 200-202) compute the following parameters in addition to those listed in “Plasma or serum data” on page 507. AURC_%Extrap _obs Percent of AURC_INF_obs that is extrapolated. excretion rate at the final midpoint estimated using the linear regression for λz. Terminal half-life = ln(2) / λz Area under the urinary excretion rate curve extrapolated to infinity. See “Sparse sampling computations” on page 188 for additional details of NCA computations with sparse data. where Tlast is the time of last measurable (positive) mean concentration. SE_AUCall 514 . i.

SE_AURC_all Individuals by time This sheet includes the individual subject data. SE_Max_Rate SE_AURC_last Standard error of the data at the time of maximum mean rate. i.WinNonlin Model Libraries Noncompartmental analysis B Urine concentration parameters Sparse sampling methods for urine data (models 210-212) compute the following parameters in addition to those listed under “Urine data” on page 512. entered in the Dosing regimen Baseline response value. Standard error of the area under the mean urinary excretion rate curve from dose time through the final interval. response or effect values (continuous scale) Independent variable.e. entered in the Therapeutic response tab of the Model Properties 515 . the mean and standard error for each unique time value (plasma data) or unique midpoint value (urine data). For nontime-based data.. entered in the Therapeutic response tab of the Model Properties or pulled from a Dosing worksheet as described below Threshold response value (optional). Standard error of the area under the mean urinary excretion rate curve from dose time through the last interval that has a measurable (positive) mean rate. such as concentration-effect data. along with N (number of nonmissing observations). and the following constants: • • • Dose time (relative to observation times). generally the time of each observation. take care to interpret output parameters such as “Tmin” (independent variable value corresponding to minimum dependent variable value) accordingly. Drug effect data (model 220) Data structure NCA for drug effect data requires the following input data variables: • • Dependent variable.

WinNonlin will use the response value at dose time as baseline. the dosing time is assumed to be zero and WinNonlin will insert the point (0. these slopes are reported as their actual value rather than their negative. the PD model can compute partial areas under the curve. “Above” means towards increasing Y values. • • • B = baseline effect value (discussed above). Instead of Lambda Z calculations. Like other NCA models. the NCA PD model can calculate the slope of the time-effect curve for specific data ranges in each profile. See “Lambda Z or slope estimation settings” on page 162 for more information. In that case. Partial areas are set up as described under “Partial areas” on page 184. even for inhibitory effects. Unlike Lambda Z. baseline). the user must supply a baseline value (see “Therapeutic response” on page 168). T = user-supplied threshold effect value. the user must enter a baseline response value unless working from a PKS study that includes a Dosing worksheet. as computed in NCA PK models. Estimated parameters Output parameter names use the following conventions.B WinNonlin User’s Guide By default. If the dosing time is not specified (“Dosing regimen” on page 171). if no baseline is provided by the user. If the data set does not include a measurement at dose time. 516 .

effect value at dose time. = 100 * Time_Below_B/(Tfinal . Minimum observed response value. Additional parameters for model 220 with threshold Threshold Threshold value used. Correlation between time (X) and effect (or log effect. Total time that Response >= Baseline. = AUC_Above_B . 517 .WinNonlin Model Libraries Noncompartmental analysis B Table B-8. Goodness of fit statistic for slope 1 or 2. for log regression) (Y) for the points used in the slope estimation. Slope of the first (or second) segment of the curve. Upper limit on Time for values to be included in the slope calculation. See “Lambda Z or slope estimation settings” on page 162.Tdose) where Tfinal is the final observation time and Tdose is dosing time. This is likely to be a negative value for inhibitory effects. Total time that Response < Baseline. Maximum observed response value.AUC_Below_B. or. adjusted for the number of points used in the estimation. Area under the response curve that is above the baseline (dark gray areas in the above diagram). When a threshold value is provided. Area that is below the baseline and above the response curve (combined blue and pink areas in the above diagram). Rsq_adj_Slope1 (or 2) Corr_XY_Slope1 (or 2) No_points_Slope1 The number of data points included in calculation of slope 1 or 2. (or 2) Slope1_lower or Slope2_lower Slope1_upper or Slope2_upper Rmax Rmin Time_Above_B Time_ Below_B Time_%Below_B Lower limit on Time for values to be included in the slope calculation. Rsq_Slope1 (or 2) Goodness of fit statistic for slope 1 or 2. Table B-9. for PKS studies with no user-supplied baseline value. Estimated parameters for model 220 Baseline AUC_Above_B AUC_Below_B AUC_Net_B Slope1 (or 2) Baseline response (Y) value supplied by the user. model 220 also computes the following.

Time that the response first crosses the threshold coming from the direction of the baseline value. Total time that Response >= Threshold. = 100 * Time_Below_T/(Tlast .Tdose) where Tlast is the final observation time and Tdose is dosing time. 518 . Use caution in interpreting Tonset and Toffset for noisy data if Baseline and Threshold are close together. Tmax Tmin AUClower_upper Time of maximum observed response value (Rmax). Area that is below the threshold and above the response curve (pink area in the above diagram).B WinNonlin User’s Guide Table B-9.a Toffset Diff_Toffset_Tonset = Toffset . Total time that Response < Threshold. a. (See “NCA settings” on page 177. For this reason. (Optional) user-requested area(s) under the curve from time lower to time upper. If the log trapezoidal rule is appropriate for the area under a curve. The time will be interpolated using the calculation method selected in the model options. = AUC_Above_T .) Use caution with log trapezoidal since areas are calculated both under the curve and above the curve. Note: For PD data. Tonset = Tdose if the first response value is across the threshold. and similarly for threshold and AUC_Below_T. relative to baseline. as shown in the above diagram.AUC_Below_T. Time of minimum observed response value (Rmin).)a Time greater than Tonset at which the curve first crosses back to the baseline side of threshold. as shown in the diagram above. Additional parameters for model 220 with threshold (continued) AUC_Above_T AUC_Below_T AUC_Net_T Time_Above_T Time_Below_T Time_%Below_T Tonset Area under the response curve that is above the threshold value (combined light and dark gray areas in the above diagram). the program will use linear trapezoidal for area where the curve is below baseline when computing AUC_Below_B. the default and recommended method for AUC calculation is the Linear Trapezoidal with Linear Interpolation (set as described under “NCA settings” on page 177. then it would underestimate the area over the curve.Tonset Time_Between_BT Total time spent between baseline and threshold (sum of length of green arrows in diagram).

519 .LIB. Each is available in both compiled and ASCII text forms. Model Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10 Model 11 Model 12 Model 13 Model 14 Model 15 Model 16 Model 17 Model 18 Model 19 Input IV-bolus IV-infusion 1st order 1st order 1st order 1st order IV-bolus IV-bolus IV-infusion IV-infusion 1st order 1st order 1st order 1st order IV-bolus plus IV-infusion IV-bolus plus IV-infusion IV-bolus plus IV-infusion IV-bolus IV-infusion Number of compartments 1 1 1 1 1 1 2 2 2 2 2 2 2 2 1 2 2 3 3 Parameterization Lag time no no no yes K10 = K01 no K10 = K01 yes micro no macro no micro no macro no micro no micro yes macro no macro yes micro no micro macro macro macro no no no no Elimination rate 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order 1st order Note: All models except models 15-17 accept multiple dose data. The models shown in this section give equations for compartment 1 (central). ASCII models and files WinNonlin PK models are available as compiled models.LIB .WinNonlin Model Libraries Pharmacokinetic models B Pharmacokinetic models The WinNonlin Model Libraries include 19 pharmacokinetic (PK) models. as follows.PK10. and also in the ASCII library files PK1.

LIB PK2.LIB PK9.LIB PK5. 18 19 File name PK1. 4 5.LIB Model 1 One compartment with bolus input and first-order output C (T ) = D/V C (T ) = (D/V) exp(-K10 · T ) Required constants Estimated parameters V = Volume K10 = elimination rate Secondary parameters AUC = D/V/K10 K10 half-life Cmax = D/V CL AUMC MRT Vss Clearance estimated V CL Clearance secondary AUC K10 half-life Cmax K10 AUMC MRT VSS N doses dose N time of dose N (Repeat for each dose) Model 2 One-compartment with constant IV input.LIB PK3. 2 3.B WinNonlin User’s Guide Model # 1.LIB PK4. 14 15.LIB PK6.LIB PK8. 8 9.LIB PK7. 6 7. 10 11. first-order absorption 520 . 12 13.LIB PK10. 16 17.

⎞ ----------. Required constants Estimated parameters V = Volume K10 = elimination rate Secondary parameters K10 half-life Cmax = C(TI) CL AUMC MRT Vss Clearance estimated CL Clearance secondary AUC K10 half-life Cmax K10 AUMC MRT VSS N doses dose N start time N end time N (Repeat for each dose) AUC = D/V/K10 V Model 3 One-compartment with first-order input and output. TSTAR = T-TI for T>TI. no lag time DK 01 C ( T ) = ------------------------------.[ exp ( – K 10 TSTAR ) – exp ( – K 10 T ) ] ⎝ TI ⎠ VK 10 where TI = infusion length.WinNonlin Model Libraries Pharmacokinetic models B 1 D C ( T ) = ⎛------.[ exp ( – K 10 T ) – exp ( – K 01 T ) ] V ( K 01 – K 10 ) Required constants Estimated parameters V_F Secondary parameters AUC = D/V/K10 Clearance Clearance estimated secondary V_F K01 CL_F AUC K01 half-life K10 half-life K10 Tmax N doses dose N (Repeat for each dose) K01=absorption rate K01 half-life CL_F Tmax = time of Cmax time of dose N K10=elimination rate K10 half-life ln ( K01 ⁄ K10 ) = --------------------------------( K01 – K10 ) Cmax = max concentration Cmax 521 . TSTAR = 0 for T≤TI.

equal first-order input and output with lag time Identical to Model 5. with an additional estimated parameter: Tlag = lag time. with an additional estimated parameter: Tlag = lag time. no lag time C (T ) = (D/V) K · T · exp(-K · T ) Required constants Estimated parameters V_F elimination rate Secondary parameters AUC = D/V/K CL_F Tmax = time of Cmax = 1/K Cmax = max concentration Clearance estimated V_F CL_F Clearance secondary AUC K half-life K Tmax Cmax N doses dose N time of dose K=absorption and K half-life N (Repeat for each dose) Model 6 One-compartment.B WinNonlin User’s Guide Model 4 One-compartment with first-order input and output with lag time Identical to Model 3. Model 5 One-compartment. equal first-order input and output. 522 .

micro-constants as primary parameters C (T ) = A exp(-ALPHA · T )+B exp(-BETA · T ) where: A = (D/V) (ALPHA .BETA) B = (D/V) (BETA .K21)/(ALPHA . 1 to 2 ALPHA K21 = transfer rate.WinNonlin Model Libraries Pharmacokinetic models B Model 7 Two-compartment with bolus input and first-order output.K21)/(ALPHA .BETA) where -ALPHA and -BETA (ALPHA > BETA) are the roots of the quadratic equation: (r · r + (K12 + K21 + K10) · r + K21 · K10 = 0). Required constants Estimated parameters V1 = Volume1 K10=elimination rate Secondary parameters AUC = D/V/K10 K10 half-life Clearance estimated V1 CL V2 CLD2 Clearance secondary AUC K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B Cmax K10 AUMC MRT Vss K12 K21 N doses dose N (Repeat for each dose) time of dose N K12 = transfer rate. 2 to 1 BETA ALPHA half-life BETA half-life A B Cmax = D/V CL AUMC MRT Vss V2 CLD2 523 .

exp(-ALPHA · TSTAR)] + B1[exp(-BETA · T ) .-----------------------------------------------------------------TI ( V ) ( ALPHA – BETA )ALPHA 524 . macro-constants as primary parameters As Model 7 with macroconstants as the primary (estimated) parameters. microconstants as primary parameters C (T ) = A1[exp(-ALPHA · T ) . Required constants stripping dose Estimated parameters Secondary parameters A B ALPHA BETA AUC = A/ALPHA + B/BETA V1 K10 half-life ALPHA half-life BETA half-life K10 K12 K21 Cmax CL AUMC MRT Vss V2 CLD2 N doses dose N time of dose N (Repeat for each dose) Clearance parameters are not available for Model 8.exp(-BETA · TSTAR)] where TI = infusion length. TSTAR = 0 for T≤TI. Model 9 Two-compartment with constant IV input and first-order output. TSTAR = T-TI for T>TI. D ( K21 – ALPHA ) A 1 = ------------.B WinNonlin User’s Guide Model 8 Two-compartment with bolus input and first-order output.

Model 10 Two-compartment with constant IV input and first-order output. 1 to 2 K21 = transfer rate. macroconstants as primary parameters As Model 9 with macroconstants as the primary (estimated) parameters. 525 .WinNonlin Model Libraries Pharmacokinetic models B –D ( K21 – BETA ) B 1 = ------------.------------------------------------------------------------TI ( V ) ( ALPHA – BETA )BETA and -ALPHA and -BETA (ALPHA > BETA) are the roots of the quadratic equation: r · r + (K12 + K21 + K10) · r + K21 · K10 = 0. 2 to 1 Secondary parameters K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B Cmax = C(TI) CL AUMC MRT Vss V2 CLD2 Clearance estimated CL V2 CLD2 Clearance secondary AUC K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B Cmax K10 AUMC MRT Vss K12 K21 N doses dose N start time N end time N (Repeat for each dose) AUC = D/V/K10 V1 A and B are the zero time intercepts following an IV injection. Required constants Estimated parameters V1 = Volume1 K10 = elimination rate K12 = transfer rate.

Model 11 Two-compartment with first-order input. Clearance parameters are not available in Model 10. no lag time and micro-constants as primary parameters C (T ) = A exp(-ALPHA · T ) + B exp(-BETA · T ) + C exp(-K01 · T ) where: ⎛ D⎞ K01 ( K21 – ALPHA ) --⎝ V⎠ A = --------------------------------------------------------------------------------------( ALPHA – BETA ) ( ALPHA – K01 ) ⎛ – D⎞ K01 ( K21 – BETA ) --⎝ V⎠ B = ---------------------------------------------------------------------------------( ALPHA – BETA ) ( BETA – K01 ) 526 .B WinNonlin User’s Guide Required constants Estimated parameters V1 K21 ALPHA BETA Secondary parameters K10 K12 K10 half-life AUC ALPHA half-life BETA half-life A B Cmax CL AUMC MRT Vss V2 CLD2 N doses dose N start time N end time N (Repeat for each dose) A and B are the zero time intercepts following an IV injection. first-order output.

2 to 1 BETA ALPHA half-life BETA half-life A B CL_F V2_F CLD2_F Tmaxa Cmaxa a.WinNonlin Model Libraries Pharmacokinetic models B ⎛ D⎞ K01 ( K21 – K01 ) --⎝ V⎠ C = ----------------------------------------------------------------------------( BETA – K01 ) ( ALPHA – K01 ) and -ALPHA and -BETA (ALPHA > BETA) are the roots of the quadratic equation: r · r + (K12 + K21 + K10) · r + K21 · K10 = 0. 527 . Required constants Estimated parameters V1_F K01 = absorption rate K10 = elimination rate Secondary parameters K01 half-life K10 half-life Clearance Clearance estimated secondary AUC K01 half-life K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B K10 K12 K21 Tmax Cmax K01 CL_F V2_F CLD2_F N doses dose N time of dose N (Repeat for each dose) AUC = D/V/K10 V1_F K12 = transfer rate. with an additional estimated parameter: Tlag = lag time. lag time and micro-constants as primary parameters As Model 11. Model 12 Two-compartment with first-order input. first-order output. Estimated for the compiled model only. 1 to 2 ALPHA K21 = transfer rate.

no lag time and macro-constants as primary parameters As Model 11.B WinNonlin User’s Guide Model 13 Two-compartment with first-order input. Model 15 One compartment with simultaneous bolus IV and constant IV infusion CB (T ) = (Db/V) exp(-K10 · T ) D IV 1 C IV ( T ) = ⎛ -------. lag time and macro-constants as primary parameters As Model 13. with macroconstants as the primary (estimated) parameters.[ exp ( – K10 ⋅ TSTAR ) – exp ( – K10 ⋅ T ) ] ⎝ TI ⎠ V ( K10 ) 528 .⎞ -----------------. Estimated for the compiled model only. first-order output. Required constants Stripping dose Estimated parameters A B K01 = absorption rate ALPHA BETA Note: C = -(A + B) Secondary parameters K10 K12 K21 AUC = D/V/K10 K01 half-life K10 half-life ALPHA half-life a. with an additional estimated parameter: TLAG = lag time. Model 14 Two-compartment with first-order input. first-order output. BETA half-life V1_F CL_F V2_F CLD2_F Tmaxa Cmaxa N doses dose N time of dose N (Repeat for each dose) Clearance parameters are not available for Model 13.

------------------------------------------------------------TI ( V ) ( ALPHA – BETA )BETA 529 .exp(-BETA · TSTAR)] where: D IV ( K21 – ALPHA ) A 2 = ------------.--------------------------------------------V ( ALPHA – BETA ) – D B ( BETA – K21 ) B 1 = --------.--------------------------------------------V ( ALPHA – BETA ) and: CIV (T ) = A2[exp(-ALPHA · T ) . micro constants as primary parameters CB (T ) = A1 exp(-ALPHA · T ) + B1 exp(-BETA · T ) where: D B ( ALPHA – K21 ) A 1 = -----. TSTAR = 0 for T≤TI Required constants bolus dose IV dose length of infusion = TI Estimated Secondary parameters parameters V = Volume K10 CL K10 half-life Clearance Clearance estimated secondary V CL K10 K10 half-life Model 16 Two compartment with simultaneous bolus IV and constant infusion input. TSTAR = T-TI for T>TI.WinNonlin Model Libraries Pharmacokinetic models B C (T ) = CB (T ) + CIV (T ) where TI = infusion length.exp(-ALPHA · TSTAR)] + B2[exp(-BETA · T ) .-----------------------------------------------------------------TI ( V ) ( ALPHA – BETA )ALPHA – D IV ( K21 – BETA ) B 2 = ------------.

530 .B WinNonlin User’s Guide where TI = infusion length. Required constants stripping dose bolus dose IV dose length of infusion (= TI) Estimated parameters A B ALPHA BETA K10 K12 K21 K10 half-life ALPHA half-life Secondary parameters BETA half-life V1 CL V2 CLD2 Clearance parameters are not available in Model 17. TSTAR = 0 for T≤TI C (T ) = CB (T ) + CIV (T ) and -ALPHA and -BETA (ALPHA > BETA) are the roots of the quadratic equation: r · r + (K12 + K21 + K10) · r + K21 · K10 = 0. TSTAR = T-TI for T>TI. Required constants bolus dose IV dose length of infusion (= TI) Estimated parameters V1 K10 K12 K21 Secondary parameters K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B CL V2 CLD2 Clearance estimated V1 CL V2 CLD2 Clearance secondary K10 half-life ALPHA BETA ALPHA half-life BETA half-life A B K10 K12 K21 Model 17 Two compartment with simultaneous bolus IV and constant infusion input and macro constants as primary parameters As Model 16 with macroconstants as the primary (estimated) parameters.

first-order output and macro constants as primary parameters C (T ) = A exp(-ALPHA · T ) + B exp(-BETA · T ) + C exp(-GAMMA · T ) Required constants stripping dose Estimated parameters A B C ALPHA BETA GAMMA Cmax V1 K21 K31 K10 K12 K13 K10 half-life ALPHA half-life BETA half-life Secondary parameters GAMMA half-life AUC CL AUMC MRT Vss V2 CLD2 V3 CLD3 N doses dose N time of dose N (Repeat for each dose) Clearance parameters are not available in Model 18.WinNonlin Model Libraries Pharmacokinetic models B Model 18 Three-compartment with bolus input. 531 .

macro constants as primary parameters C (T ) = A1 [exp(-ALPHA · TSTAR ) .⎞ ( K21 – ALPHA ) ( K31 – ALPHA ) ⎝ TI ⎠ A 1 = --------------------------------------------------------------------------------------------------------------------------------V ( ALPHA ) ( GAMMA – ALPHA ) ( BETA – ALPHA ) D ⎛ -------⎞ ( K21 – BETA ) ( K31 – BETA ) ⎝ TI ⎠ B 1 = ----------------------------------------------------------------------------------------------------------------------V ( BETA ) ( GAMMA – BETA ) ( ALPHA – BETA ) D ⎛ -------⎞ ( K21 – GAMMA ) ( K31 – GAMMA ) ⎝ TI ⎠ C 1 = --------------------------------------------------------------------------------------------------------------------------------------V ( GAMMA ) ( ALPHA – GAMMA ) ( BETA – GAMMA ) where TI = infusion length. TSTAR = T-TI for T>TI.exp(-BETA · T )] + C1 [exp(-GAMMA · TSTAR ) . TSTAR = 0 for T≤TI Required constants Estimated parameters V1 K21 K31 Secondary parameters Cmax K10 K12 K10 half-life ALPHA half-life BETA half-life MRT Vss V2 N doses dose N start time of dose N 532 .B WinNonlin User’s Guide Model 19 Three compartment model with constant IV infusion.exp(-GAMMA · T )] where: D ⎛ ------.exp(-ALPHA · T )] + B1 [exp(-BETA · TSTAR ) .

Model # 101.WinNonlin Model Libraries Pharmacodynamic models Required constants end time of dose N (Repeat for each dose) Estimated parameters ALPHA BETA GAMMA Secondary parameters K13 A B C GAMMA half-life AUC CL AUMC CLD2 V3 CLD3 B Clearance parameters are not available in Model 19. 106 File name PK51. Available PD models in the WinNonlin Model Libraries include the following. Model Model 101 Model 102 Model 103 Model 104 Model 105 Model 106 Model 107 Model 108 Description Simple Emax Model Simple Emax Model Inhibitory Effect Emax Model Inhibitory Effect Emax Model Sigmoid Emax Model Sigmoid Emax Model Effect at C=0 0 E0 Emax Emax 0 E0 Effect at C=infinity Emax Emax 0 E0 Emax Emax 0 E0 Inhibitory Effect Sigmoid Emax Model Emax Inhibitory Effect Sigmoid Emax Model Emax ASCII library These models are available as compiled models and also as ASCII library files. B and C are the zero time intercepts following an IV injection.LIB PK53. A. 104 105.LIB PK52. The ASCII library files are stored in the WINNONLN\LIB subdirectory.LIB 533 . 102 103. Pharmacodynamic models Load and set up pharmacodynamic (PD) models as described under “Loading the model” on page 193 and “Model properties” on page 196.

E0 534 . E0. EC50 Secondary parameter: Emax . 108 File name PK54. EC50 Model 102 Simple Emax model with a baseline effect parameter E = E0 + (Emax .E0 Gamma Definition Maximum effect Baseline effect Drug concentration required to produce 50% of the maximal effect (Emax.B WinNonlin User’s Guide Model # 107. Model notation Term Emax E0 EC50 Emax.LIB Table B-10.E0 ) [C / (C + EC50)] Estimated parameters: Emax.E0) Drug-induced maximal effect Shape parameter Model 101 Simple Emax model E = (Emax · C) / (C + EC50) Estimated parameters: Emax.

WinNonlin Model Libraries Pharmacodynamic models B Model 103 Inhibitory effect model E = Emax [1 . EC50.(Emax . Gamma 535 . EC50 . E0 Secondary parameter: Emax . EC50 Model 104 Inhibitory effect model with a baseline effect parameter E = Emax .E0 Model 105 Sigmoid Emax model E = (Emax · Cγ ) / (Cγ + ECγ 50)] Estimated parameters: Emax.E0 ) [C / (C + EC50)] Estimated parameters: Emax.(C / (C + EC50))] Estimated parameters: Emax.

E0. Gamma Secondary parameter: Emax . Gamma Model 108 Sigmoid inhibitory effect model with a baseline effect parameter E = Emax . EC50.Cγ / (Cγ + ECγ 50)] Estimated parameters: Emax.E0) [ Cγ / (Cγ + ECγ 50)] Estimated parameters: Emax.E0 Model 107 Sigmoid inhibitory effect model E = Emax [1 . a pharmacodynamic model is applied to characterize the relationship between drug concentrations and effect. EC50. E0. When the pharmacologic response takes time to develop and the observed response is not 536 .E0 PK/PD link models When pharmacological effects are seen immediately and are directly related to the drug concentration. Gamma Secondary parameter: Emax .(Emax . EC50.E0) [ Cγ/ (Cγ +ECγ 50)] Estimated parameters: Emax.B WinNonlin User’s Guide Model 106 Sigmoid Emax model with a baseline effect parameter E = E0 + (Emax .

Drug induced maximal effect Shape parameter Definition Linking PK and PD models WinNonlin can generate concentrations at effect sites using any PK model in the compiled library. and these concentrations are then used as input to the PD model. the PK/PD Link models treat the pharmacokinetic parameters as fixed. and using these concentrations in any PD model. Select PK/PD Link in the Model Types dialog. The PK/PD Link models can use any combination of WinNonlin compiled Pharmacokinetic models and Pharmacodynamic models. Thus. Effect at time 0 Rate of drug loss from the effect compartment. The PK model is used to predict concentrations. Model parameter information will be required for the PK model in order to simulate the concentration data. Notation Term Emax ECe50 E0 Ke0 Emax–E0 Gamma Maximum effect Concentration required to produce one-half of the maximal drug effect. To link any PK and PD models: 1. and generate concentrations at the effect site to be used by the PD model. Note: To fit PK and PD data simultaneously. the PK data are not modeled. 537 . Click the PK/PD/NCA Analysis Wizard button on the tool bar or choose Tools>PK/PD/NCA Analysis Wizard from the menus. see “Simultaneous PK/PD link models” on page 545.WinNonlin Model Libraries PK/PD link models B directly related to plasma concentrations of the drug a link model is usually applied to relate the pharmacokinetics of the drug to its pharmacodynamics. 2. The Ke0 T1/2 is the half life of the amount of time required to collapse the hysteresis curve.

Notice that the Model Parameters dialog now has a two tabs. Output parameters The output parameters for the PD models differ from the parameters for these models when they are not linked. and model options as for other compartmental models. Select the PD Model tab. To assign initial estimates and bounds for the PD model: 1. 5.B WinNonlin User’s Guide 3. 4. Click OK. 3. Click the Model Parameters button or choose Model>Model Parameters from the menus. Select the PK Model tab. The model parameters are set and the dialog closes. Select a PK model from those detailed under “Pharmacokinetic models” on page 519. 7. Click Next. Model Parameters. Enter the parameter values. 5. 538 . as described under “Model properties” on page 196. The initial parameter values must be entered. 2. Enter initial estimates and bounds. Enter the Data Variables. Select a PD model from those detailed under “Pharmacodynamic models” on page 533. Note: The units for PK concentration can be entered in the PK Concentration Unit field or set using the Units Builder button. one for the PK Model and one for the PD Model. 6. Click Next. 2. Click Apply or OK. Click Finish. 4. The Selection Summary dialog appears. Table B-11 lists the PD model parameters. To assign parameter values for the PK model: 1. 3. Dosing Regimen/ Constants.

E0. Gamma. and then use these concentrations as input to the indirect response model. a pharmacodynamic model is applied to characterize the relationship between drug concentrations and effect. ECe50 Emax. Four basic models have been developed for characterizing indirect pharmacodynamic responses after drug administration. KE0 Emax. Gamma. 539 . KE0 Emax. KE0 Emax. When the pharmacologic response takes time to develop and the observed response is not directly related to plasma concentrations of the drug a link model is usually applied to relate the pharmacokinetics of the drug to its pharmacodynamics. ECe50. ECe50. Link model output parameters PD model 101 Simple Emax 102 Simple Emax with baseline effect 103 Inhibitory effect 104 Inhibitory effect with a baseline effect 105 Sigmoid Emax 106 Sigmoid Emax with a baseline effect 107 Sigmoid inhibitory effect 108 Sigmoid inhibitory effect model with a baseline effect Estimated parameters Emax. ECe50. Garg. The indirect response models differ from other models in the WinNonlin Model Libraries in that they use a PK model to predict concentrations. ECe50. KE0 Emax. These models are based on drug effects (inhibition or stimulation) on the factors controlling either the input or the dissipation of drug response. E0. ECe50. E0. KE0 Emax-E0 Emax-E0 Emax-E0 Emax-E0 Secondary parameter Indirect response models When pharmacological effects are seen immediately and are directly related to the drug concentration. E0. ECe50. Gamma. and Jusko (1993). One kind of link model available in WinNonlin is the family of indirect pharmacodynamic response (IPR) models proposed by Jusko and co-workers Jusko (1990) and Dayneka. Gamma. KE0 Emax. KEO. ECe50.WinNonlin Model Libraries Indirect response models B Table B-11. KE0 Emax.

B WinNonlin User’s Guide Note: These models are available only as compiled models Models Notation Term R kin kout Cp Ce IC50 Emax EC50 Measured response to a drug The zero order constant for the production of response The first order rate constant for loss of response Plasma concentration of a drug Drug concentration at the effect site The drug concentration that produces 50% of maximum inhibition Maximum effect The drug concentration that produces 50% of maximum stimulation Definition Model 51 Inhibition of input Cp dR = k ⎛ 1 – ---------------------.= k in – k out ⎛ 1 – ---------------------.⎞ – k R ----in ⎝ out dt C p + IC 50⎠ Estimated parameters Kin Kout IC50 Model 52 Inhibition of output Cp dR ----.⎞ R ⎝ dt C p + IC 50⎠ Estimated parameters Kin Kout IC50 540 .

7. 2. (Otherwise. Select an Indirect Response model. The PK/PD Analysis Wizard appears.⎞ – k out R ⎝ dt C p + EC 50⎠ Estimated parameters Kin Kout EC50 Emax Model 54 Stimulation of output Emax ⋅ C p dR = k – k ⎛ 1 + -----------------------. 541 . The Selection Summary box appears. Select Indirect Response from the selections in the Compartmental Models group box. If everything described in this summary box is correct. displaying the Model Types dialog. click Finish.) The modeling window appears. The list of Indirect Response Models appears. The WinNonlin Compiled Models dialog appears. 5. 3.WinNonlin Model Libraries Indirect response models B Model 53 Stimulation of input Emax ⋅ C p dR ----.⎞ R ----in out ⎝ dt C p + EC 50⎠ Estimated parameters Kin Kout EC50 Emax Running an indirect response model To select an Indirect Response model: 1. 6. click Back and return to correct the mistake. Click the PK/PD/NCA Analysis Wizard button on the tool bar or select the PK/PD/NCA Analysis Wizard from the Tools menu. Click Next. 4. Select a PK model. Click Next.= k in ⎛ 1 + -----------------------.

Click Apply or OK to specify the settings and close the dialog. (Optional) Enter the PK units in the PK Concentration Unit field. Enter the Data Variables. and model options as for other compartmental models. Select the IPR Model tab in the Model Properties dialog. To assign parameter values for the PK model: 1. as described under “Model properties” on page 196. Model Parameters. 2. 6. Click Apply. Michaelis-Menten models The WinNonlin library contains four Michaelis-Menten models: 542 . Note: Initial parameter values for the PK model must be specified. 2. 5. To assign initial estimates and bounds for the IPR model: 1. Click the Model Parameters button or choose Model>Model Parameters from the menus. 3. Enter the parameter values for the PK model. Select the PK Model tab located under the parameters table. Notice that this dialog now has tabs for both the PK Model and the indirect response model. and Model Options precisely as for other models. Initial estimates and bounds can be specified for the IPR model. Dosing Regimen/ Constants. Dosing Regimen/Constants. 4. It is possible to choose to have WinNonlin generate the initial parameter values or to specify user-supplied initial values. 3.B WinNonlin User’s Guide Note: Enter the Data Variables. Enter initial estimates and bounds.

be sure to examine the ASCII code to determine which dosing constants need to be created. WinNonlin assumes that the time of the first dose is zero.LIB. For these models.WinNonlin Model Libraries Michaelis-Menten models B Model Description Model 301 One compartment with bolus input and Michaelis-Menten output Model 302 One compartment with constant IV input and Michaelis-Menten output Model 303 One compartment with first order input. times in the data set must correspond to the dosing times. These models parameterize Vmax in terms of concentration per unit time. Model 301 One-compartment with bolus input and Michaelis-Menten output C (T ) is the solution to the differential equation: dC/dt = (-VMAX · C)/( KM + C ). For discussion of difficulties inherent in fitting the Michaelis-Menten models see Tong and Metzler (1980) (1980) and Metzler and Tong (1981) (1981). with initial condition C(0) = D/V. Michaelis-Menten output and a lag time These models are stored as ASCII library files in \LIB\MM. As with all WinNonlin ASCII models. and weight these observations as 0. In this case. Required constants Estimated parameters V = Volume VM = max elimination rate KM = Michaelis constant Secondary parameters AUC = (D/V)(D/V/2 + KM)/ VMAX N doses Dose N Time of dose N 543 . even if these times contain no observations. Michaelis-Menten output and no time lag Model 304 One compartment with first order input. include a column for Weight on the data set.

Required constants Estimated parameters V = Volume VM = max elimination rate KM = Michaelis constant Secondary parameters (none) N doses Dose N Start time for dose N End time for dose N Model 303 One-compartment with first-order input. dC/dt = .VMAX · C /(KM+C ) for T <=TI.VMAX · C /(KM+C ) for T > TI. Michaelis-Menten output and no lag time C (T ) is the solution to the differential equation: dC/dt = K01 · (D/V) exp(-K01 · T ) . with initial condition C(0) = 0.B WinNonlin User’s Guide Model 302 One-compartment with constant IV input and Michaelis-Menten output C (T ) is the solution to the differential equations: dC/dt = (D/TI/V) . with initial condition C(0) = 0. Required constants Estimated parameters V_F K01 = absorption rate VM = max elimination rate KM = Michaelis constant Secondary parameters K01 half-life N doses Dose N Time of dose N 544 .VM · C /(KM+C ).

Data The data for these models must contain the concentration and effect data in the same column. 1 148 145 123 . this library includes each PK models with PD Model 105.. 121 Function 1 1 1 .. Function 2 should include the time and effect data. with a function variable in a separate column to distinguish the two.. As with all of the WinNonlin ASCII models. with one additional estimated parameter: Tlag = lag time. Function 1 should include the time and concentration data...WinNonlin Model Libraries Simultaneous PK/PD link models B Model 304 One-compartment with first-order input... 36 Response 0 5 15 .5 1 ... Michaelis-Menten output and lag time As Model 303.. 1 2 2 2 . Time 0 0.. An example is shown below. 36 0 0. As it is difficult to rewrite these models for changes in the PK model. Simultaneous PK/PD link models WinNonlin has a library of link models that simultaneously fit PK and PD data. be sure to examine the ASCII code to see what dosing constants need to be created. 545 . but easy to rewrite them for a different PD model. These models are described below.5 1 . 2 ASCII library These models are stored as ASCII library files in the \LIB subdirectory..

constant IV input.bolus input + constant infusion.1st order input.LIB KEO4. See also “Linear Mixed Effects Modeling” on page 323 for more sophisticated linear models. defined in macroconstants 3 compartment . includes a lag time 1 compartment . defined in microconstants 2 compartment .LIB KEO4.LIB KEO1. defined in macroconstants 2 compartment . includes a lag time 2 compartment .LIB KEO8.1st order input.extravascular input.LIB Model 19 Linear models WinNonlin includes a selection of models that are linear in the parameters.extravascular input 1 compartment .LIB PK model Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10 Model 11 Model 12 Model 13 Model 14 Model 15 Model 16 Model 17 PK model description 1 compartment .LIB Model 18 KEO10.LIB KEO5. defined in macroconstants 2 compartment .bolus input + constant infusion 2 compartment .LIB KEO6.constant IV input.bolus input 1 compartment .LIB KEO5. defined in macroconstants 3 compartment . 546 .LIB KEO6. defined in macroconstants KEO10.extravascular input.bolus input. equal input and output rates 1 compartment . defined in macroconstants 2 compartment . equal input and output with a lag time 2 compartment .extravascular input.LIB KEO2. defined in microconstants 2 compartment .constant IV input 1 compartment .1st order input. defined in macroconstants.LIB KEO9.bolus input + constant infusion. defined in microconstants 2 compartment .LIB KEO3.LIB KEO7.bolus input.bolus input.LIB KEO8. defined in macroconstants.LIB KEO7.1st order input. defined in microconstants 2 compartment . with a lag time 1 compartment .constant infusion.LIB KEO2.LIB KEO3.B WinNonlin User’s Guide Model 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 File KEO1.

Weibull. The linear models require no dosing information (constants). See “Units” on page 208 to set preferred units for the output parameters. Note: The default computation method. Gauss-Newton (Hartley) (Model Options>PK Settings). the WinNonlin model library includes the following models for smoothing dissolution data as part of analysis of in vivo-in vitro correlations (IVIVC): Hill. and the model option “Do Not Use Bounds” (Model Options>Model Parameters) are recommended for all linear models. Sigmoidal/dissolution models If you have purchased and installed an IVIVC Toolkit for WinNonlin license. Linear models Plot Model # Name 501 Constant Model Estimated parameters (units) y(t) = constant CONSTANT (y-unit) 502 Linear y(t) = INT + SLOPE * t INT (y-unit) SLOPE (y-unit / t-unit) 503 Quadratic y(t) = A0 + A1*t + A2*t2 A0 (y-unit) A1 (y-unit / t-unit) A2 (y-unit / t-unit2) 504 Cubic y(t) = A0 + A1*t + A2*t2 + A3*t3 A0 (y-unit) A1 (y-unit / t-unit) A2 (y-unit / t-unit2) A3 (y-unit / t-unit3) Where y-unit is the preferred unit of the y (concentration) variable and t-unit is the preferred unit of the t (time) variable. 547 .WinNonlin Model Libraries Sigmoidal/dissolution models B Table B-12. Double Weibull and Makoid-Banakar.

This model includes an optional lag time in the form of a step function. Model parameters can be estimated or set to a fixed value. you may need to fix a parameter while estimating the others. See “Units” on page 208 to set preferred units for the output.int)*tb) / (MDTb + t b) where the estimated parameters are: Finf = amount released at time infinity. 548 . It may not be possible to estimate all parameters simultaneously.tlag)b) where U(x) = x if x >= 0. especially the slope factor b.: y(t) = int + ((Finf . See “Model parameters” on page 197 for details. The IVIVC models require no dosing (constants). i. Model 601 Hill y(t) = int + ((Finf . using the preferred units for y.B WinNonlin User’s Guide For these models. U(x) = 0 if x < 0.tlag)b) / (MDTb + U (t . using the preferred units for y MDT = mean dissolution time.int)*U(t . in the preferred units for x (time) b = slope factor (no units) int = y-intercept.e. the parameter Tlag may be strongly correlated with other parameters. zero (default) or estimated intercept.

WinNonlin Model Libraries Sigmoidal/dissolution models B Model 602 Weibull y(t) = int + (Finf . using the preferred units for y.int) * (1 . zero (default) or estimated intercept. U(x) = 0 if x < 0. Model 603 Double Weibull y(t) = int + f1 * (Finf . using the preferred units for y MDT = mean dissolution time. i.exp[-(t / MDT)b]) where the estimated parameters are: Finf = amount released at time infinity.: y(t) = int + (Finf . setting the fraction due to each Weibull (no units) Finf = amount released at time infinity.int) (1-exp[(-t / MDT2)b2]) where the estimated parameters are: f1 = weighting factor.tlag) / MDT)b]) where U(x) = x if x >= 0. This model includes an optional lag time in the form of a step function. MDT2 = mean dissolution time for each Weibull. (x units) 549 . in the preferred units of x (time) b = slope factor (no units) int = y-intercept.int) (1-exp[(-t / MDT1)b1]) + (1-f1) * (Finf .int) * (1 . using the preferred units for y MDT1.e.exp[-(U(t .

i. zero (default) or estimated intercept. for t > Tmax where the estimated parameters are: Fmax = maximum y value. 550 . Model 604 Makoid-Banakar y(t) = int + (Fmax .: y(t) = int + f1 * (Finf .tlag)/Tmax ) ] where U(x) = x if x >= 0.U(t .B WinNonlin User’s Guide b1. This model includes an optional lag time in the form of a step function.e. using the preferred units for y. zero (default) or estimated intercept.int) (1 .: y(t) = int + (Fmax .exp[-(U(t . U(x) = 0 if x < 0.tlag) / MDT2)b2]) where U(x) = x if x >= 0. for t <= Tmax y(t) = Fmax.int) * [ U(t .e.int) * (t/Tmax)b * exp[b * (1 . This model includes an optional lag time in the form of a step function.t/Tmax)]. i. b2 = slope factors (no units) int = y-intercept.tlag) / MDT1)b1]) + (1-f1) * (Finf . using the preferred units for y.exp[-(U(t . U(x) = 0 if x < 0. using the preferred units for x (time) b = slope factor (no units) int = y-intercept.tlag)/Tmax]b * exp[ b * (1 . using the preferred units for y Tmax = time of maximum y value.int) (1 .

Table C-1. .. Worksheet functions Function(arguments) ABS(number) ACOS(number) ACOSH(number) ADDRESS(row. . column. False if at least one argument is false.Serial number refers to the stored value for a date or time entered in a cell.Appendix C Worksheet Functions Syntax. . arguments. . Creates a cell address as text. B5. wherein: .Expression list is a comma-separated list of up to 30 expressions. and examples for functions in WinNonlin worksheets WinNonlin worksheets support the functions detailed in Table C-1. if possible.Number is a number or reference to a cell that contains a number.[ ] denotes optional arguments. In DBCS systems this function returns a copy of text in which the double-byte characters are converted to single-byte characters.Value list is a comma-separated list of up to 30 values or cell references.Precision is the number of decimal places allowed. Returns the arcsine of a number. Returns True if all arguments are true. . Returns the inverse hyperbolic cosine of a number.Number list is a comma-separated list of up to 30 numbers or cell references.a1] [. . ref_type[. Characters that cannot be converted are left unchanged.sheet] ) AND(logical_list) ASC(text) Description Absolute value of a number. 551 ASIN(number) .Text represents a text string. e.g. . Returns the arc cosine of a number.Significance is the multiple to which to round. .

Returns the number of values in the supplied list. month. Rounds a number up to the nearest multiple of a specified significance. item_list) CLEAN(text) CODE(text) COLUMN(reference) COLUMNS(range) CONCATENATE(text1. Returns the day of the month that corresponds to the given date. text2. factor] ) DOLLAR(number [. Returns the arithmetic mean of the supplied numbers. Worksheet functions (continued) Function(arguments) ASINH(number) ATAN(number) ATAN2(y. period [. Returns the hyperbolic cosine of a number. significance) CHAR(number) CHOOSE(index. Returns the number of cells within a range that meet the given criteria. Returns a character that corresponds to the supplied ASCII code. Example DATEVALUE (“3/6/94”) returns 34399. Returns the inverse hyperbolic tangent of a number. Example: DAY(34399) returns 6. Removes all nonprintable characters from the supplied text. Joins several text items into one item. x) ATANH(number) AVERAGE(number list) CEILING(number. life. precision] ) EVEN(number) Description Returns the inverse hyperbolic sine of a number. period [. Returns the column number of the supplied reference. salvage. Returns the arctangent of the specified coordinates. Returns a value from a list of numbers based on the index number supplied. Returns a numeric code representing the first character of the supplied text. Returns the real depreciation of an asset for a specific period of time using the fixed-declining balance method. Returns the serial number of a date supplied as a text string.months] ) DDB(cost.C WinNonlin User’s Guide Table C-1. DAY(6-21-94) returns 21. Rounds the specified number up to the nearest even integer. salvage. Returns the depreciation of an asset for a specific period of time using the double-declining balance method or a declining balance factor. 552 . Returns the number of columns in a range reference. Returns the number of non-blank values in the supplied list. criteria) DATE(year. Returns the arctangent of a number. Returns the cosine of an angle. Returns the specified number as text. Returns the serial number of the supplied date. life. day) DATEVALUE(text) DAY(text) DB(cost.…) COS(number) COSH(number) COUNT(value list) COUNTA(expression list) COUNTIF(range. using the local currency format and the supplied precision.

Returns the factorial of the specified number. Determines if the specified expression returns a logical value. Returns the hour component of the specified time in 24-hour format. Searches for a string of text within another text string and returns the character position at which the search string first occurs. Example: FIXED(2000.precision][. True is returned if the expressions are identical. HLOOKUP(search_item. column] [. based on regular payments and a fixed periodic interest rate. start_position] ) FINDB(search_text.500. row_index ) HOUR(serial_number) IF(condition. Returns e raised to the specified power. Returns the contents of a cell from a specified range. FV(interest. case-sensitive matches. search_range. guess] ) ISBLANK(reference) ISLOGICAL(expression) ISNONTEXT(expression) ISNUMBER(expression) ISREF(expression) ISTEXT(expression) Searches the top row of a table for a value and returns the contents of a cell in that table that corresponds to the location of the search value. Determines if the specified expression is text.Worksheet Functions C Table C-1. Returns the interest payment of an annuity for a given period. expression2) EXP(number) FACT(number) FIND(search_text. false_value) INDEX(reference [. text [. row] [. Rounds a number to the supplied precision. 553 .000. text [. False is returned if they are not. formats the number in decimal format. 1) returns 2010.5. Rounds a number down to the nearest multiple of specified significance. a1] ) INT(number) IPMT(interest. Rounds the supplied number down to the nearest integer. Returns the future value of an annuity based on regular payments and a fixed type] ) interest rate. [fv]. Worksheet functions (continued) Function(arguments) EXACT(expression1. [type] ) IRR(cash_flow [. payment [. range_number] ) INDIRECT(ref_text [. true_value. -1. Determines if the specified expression is a number.5. per. start_position] ) FIXED(number [. nper. significance) Description Compares two expressions for identical. 3) returns 2. pv. pv] [. nper.no_commas] ) FLOOR(number. Returns internal rate of return for a series of periodic cash flows. Determines if the specified expression is a range reference. Determines if the specified cell is blank. Determines if the specified expression is not text. Returns the contents of the cell referenced by the specified cell. and returns the result as text. Searches for a string of text within another text string and returns the byte position at which the search string first occurs. FIXED(2009. Tests the condition and returns the specified value.

beginning with the specified starting position. reinvest_rate) MOD(number. pf [. Returns the net present value of an investment based on a series of periodic payments and a discount rate. beginning with the specified starting position. Returns the current date and time as a serial number. omitting the base assumes a base of 10. LEN(“1-3”) returns 3. Returns the natural logarithm of a number. start_position. num_bytes) MIN(number list) MINUTE(number list) MIRR(cash_flows. Returns the smallest value in the specified list of numbers. Worksheet functions (continued) Function(arguments) LEFT(text [. 4. LEFT(“2nd Quarter”. start_position. fv] [. num_chars) MIDB(text. Returns the number of characters in the supplied text string. Returns the minute that corresponds to the supplied date. Examples: LEFT(“2nd Quarter”) returns 2. Returns the number of periods of an investment based on regular periodic payments and a fixed interest rate.base]) LOG10(number) LOWER(text) MAX(number list) MID(text. Examples: LEN(“3rd Quarter”) returns 11. Returns the left-most byte from the specified text string. (Empty parentheses are required). Returns the number of bytes in the supplied text string.num_bytes] ) LEN(text) LENB(text) LN(number) LOG(number. Returns the modified internal rate of return for a series of periodic cash flows. [. 554 . Returns the largest value in the specified list of numbers. Example: MID(“07/04/96”. Returns the month that corresponds to the supplied date. Tests the supplied value and returns the value if it is a number. 3) returns 2nd. Returns the remainder after dividing a number by a specified divisor. 2) returns 04. Returns the specified number of characters from a text string. finance_rate.C WinNonlin User’s Guide Table C-1. Rounds the specified number up to the nearest odd integer. divisor) MONTH(serial_number) N(value) NOW() NPER(interest. Changes alphabetic characters in the specified string to lower case. Returns the base 10 logarithm of a number. pmt. value_list) ODD(number) OR(logical_list) PI() Description Returns the left-most characters from the specified text string. Returns True if at least one of a series of logical arguments is true. type] ) NPV(discount_rate. Returns the logarithm of a number to the specified base. Returns the value of p. num_chars]) LEFTB(text [. Returns the specified number of bytes from a text string.

start_position. number. num_bytes. number) RIGHT(text [. pmt. SEARCHB(search_text. [type] ) PRODUCT(number list) Description Returns the periodic payment of an annuity. Multiplies a list of numbers and returns the result. pv. repl_text) REPT(text. Returns the second that corresponds to the supplied date. Repeats a text string the specified number of times. “7”) returns “For the year: 1997". based on regular payments and a fixed periodic interest rate. 1. fv] [. nper. Example: SEARCH(“/”. nper. type] Returns the present value of an annuity. SEARCH(search_text. type] ) PPMT(interest. pv [.Rounds a number down. 555 . Returns the sine of the supplied angle. given a series of constant guess] ) cash payments made over a regular payment period. Returns the right-most bytes from the given text string.Worksheet Functions C Table C-1. Returns the principle paid on an annuity for a given period. start_position] ) SECOND(serial_number) SIGN(number) SIN(number) Locates the position of the first byte of a specified string within another string. num_chars. type] [. num_chars] ) RIGHTB(text [. Returns the sign of the specified number. num_chars] ) ROUND(number. REPLACE(orig_text. Returns the interest rate per period of an annuity. fv] [. precision) Replaces part of a text string with another text string. numberOfDigits) ROWS(range) Rounds the given number up to the supplied number of decimal places. start_position. “07/04/96”) returns 3. RATE(nper. text [. nper. 18. OfDigits) ROUNDUP(number. pmt [. Example: REPLACE(“For the year: 1996". per. repl_text) REPLACEB(orig_text. considering a series of constant pay) ments made over a regular payment period. Returns the right-most characters from the given text string. Returns the number of rows in a range reference. Worksheet functions (continued) Function(arguments) PMT(interest. text [. PV(interest. Replaces part of a text string with another text string. RAND() Returns a number selected randomly from a uniform distribution greater than or equal to 0 and less than 1. pv [. [fv]. ROUNDDOWN(number. start Locates the position of the first character of a specified text string within position]) another text string. (Empty parentheses are required). fv] [. Rounds the given number to the supplied number of decimal places.

Returns the tangent of the specified angle. Replaces a specified part of a text string with another text string. Example: SUBSTITUTE(“07-04-97”.precision]) TYPE(expression) UPPER(text) VAR(number list) Description Returns the hyperbolic sine of the specified number. Multiplies the cells in the given ranges and returns the sum of those products. column_index) WEEKDAY(serial_number) YEAR(serial_number) 556 Searches the first column of a table for a value and returns the contents of a cell in that table that corresponds to the location of the search value. Returns the given number as text. Returns the hyperbolic tangent of a number. Returns the argument type of the given expression. [. factor] method of depreciation. Returns the depreciation of an asset for a specified period using a variable start_period. end_period [. Returns the sum of the specified numbers. new_text [.format) TIME(hour. Returns the standard deviation of a list of as many as 30 numbers. Squares each of the supplied numbers and returns the sum of the squares. life) SQRT(number) STDEV(number list) SUBSTITUTE(text. salvage. . life. Returns a serial number for the supplied text representation of time. Worksheet functions (continued) Function(arguments) SINH(number) SLN(cost. salvage. VDB(cost. Truncates the given number to an integer. Returns the day of the week that corresponds to the supplied date. Returns the depreciation of an asset for a specific period of time using the straight-line balance method. old_text. Removes all spaces from text except single spaces between words. minute. instance]) SUM(number list) SUMIF(range. search_range. sum_range) SUMPRODUCT(range_list) SUMSQ(number list) SYD(cost. using the specified formatting. Returns the sum of the specified cells based on the given criteria. period) TAN(number) TANH(number) TEXT(number. salvage. Returns the depreciation of an asset for a specified period using the sum-ofyears method. Returns the square root of the specified number. Returns the variance of a population based on a sample of up to 30 values. “/”) returns 07/04/97. Returns the current date as a serial number. criteria.C WinNonlin User’s Guide Table C-1. method] ) VLOOKUP(search_item. “-”. Returns the year that corresponds to the supplied date. life. second) TIMEVALUE(text) TODAY() TRIM(text) TRUNC(number [. Returns a serial number for the supplied time. Changes the characters in the specified string to uppercase characters.

i.e. #. BEGIN BTIME COMMAND Usage BTIME lower_val.. See “Drug effect data (model 220)” on page 515 for details. upper_val.Appendix D Commands. etc. Arrays and Functions Commands and syntax for ASCII text user models Notation Items enclosed in brackets. are optional.” E BASELINE COMMAND Usage Description Example BASE N Sets the (optional) baseline value for use analyzing effect data with NCA model 220. 557 . #. numexcl. The symbol “|” means “or. [ ]. BASE 0 BEGIN COMMAND Usage Description Example BEGIN Indicates that all commands have been specified and processing should start.

D

WinNonlin User’s Guide Description Selects points within the Lambda Z time range to be excluded from computations of Lambda Z. In the following example, Lambda Z is to be computed using times in the range of 12 to 24, excluding the concentrations at 16 and 20 hours. These “excluded” points will, however, still be used in the computation of the pharmacokinetic parameters. BTIME 12, 24, 2, 16, 20 If there are no times to be excluded, set numexcl to zero or leave it blank. Do not mark a point for exclusion by setting the case weight to 0 as that will also exclude this data point from the calculation of the PK parameters.

Example Notes

F

CARRYALON G

COMMAND

Usage Description Examples CARRY #, etc. Specifies variables from the input data grid to be copied into each of the output worksheets. These variables are not required for the analysis. CARRY 3 CARR 1 CARRYALONG 2 See the NCARRY command. How Carry Along Data is Migrated to Output Multigrids The data for carry along columns is moved to all output worksheets from PK (PK/PD) or NCA analyses, with the exception of the last worksheet, named Settings. For all output worksheets other than the Summary Table, the carry along data will appear in the right-most column(s) of the worksheet. The first cell of data for any given profile within a carry along column will be copied over the entire profile range. There is one exception: generation of the “Summary Table” worksheets. The carry along columns will be the first columns, preceded only by any sort key columns. Also, as there are almost the same number of observations in the raw data set as in the Summary Table, all the cells within a profile of a carry along column are copied over to the Summary Table—not just the first value. There is one special case here as well: an NCA without a first data point of t=0, c=0. When such a data point does not exist on the raw data set, depending on which NCA model is selected, the core program may generate a pseudo-point of t=0, c=0 to support all necessary calculations. In this event, the first cell within a carry along column is copied 'up' to the core-generated data value of 0,0.

Notes

558

Commands, Arrays and Functions F COMMANDS

D

MODEL BLOCK

Usage COMMANDS Statements END The COMMANDS block allows WinNonlin commands to be permanently inserted into a model. Following the COMMANDS statement, any WinNonlin command such as NPARAMETERS, NCONSTANTS, PNAME, etc., may be given. To end the COMMAND section use the END statement. COMMANDS NCON 2 NPARM 4 PNAME 'A', 'B', 'Alpha', 'Beta' END

Description

Example

CON

ARRAY

Usage Description Example Note CON(N) Contains the values of the constants, such as dosing information, used to define the model. The index N indicates the Nth constant. CON(3) The values specified by the CONSTANT command, described below, correspond to CON(1), CON(2), etc. respectively.

CONSTANTS

COMMAND

Usage Description Examples Note CONSTANTS [are | =] C1, C2, …, CN Specifies the values of the constants (such as dosing information) required by the model. The arguments C1, C2, …, CN are the values. CONSTANTS 20, 2.06, -1 CONS are 5, 2 The NCONSTANTS command must appear before CONSTANTS.

CONVERGEN COMMAND CE Usage CONVERGENCE [is | =] C Description Examples Specifies the convergence criterion, C. 0 < C < 0.1. The default value is 10-4. CONVERGENCE = 1.e-6 CONV .001 559

D

WinNonlin User’s Guide Note The test for convergence is

**SSnew – SSold ------------------------------------- < C , where SS is the residual sum of squares. SSold
**

When CONVERGENCE = 0, convergence checks and halvings are turned off (useful for maximum likelihood estimates).

G

DATA

COMMAND

Usage Description DATA [[are in,] 'path'] Begins data input and optionally specifies an external data file. If arguments are omitted, the data are expected to follow the DATA statement. If the arguments are used, path is the full path of the data set, up to 64 characters. DATA DATA 'C:\MYLIB\MYFILE.DAT'

Examples

DATE

COMMAND

Usage Description Example DATE Places the current date in the upper right corner of each page in ASCII output. DATE

DHEADERS

COMMAND

Usage Description Example Note DHEADERS Statement indicating that data included within the model file contain variable names as the first row. DHEADERS If a DNAMES command is also included, it will override the names in the data.

DIFFERENTI AL

MODEL BLOCK

Usage DIFFERENTIAL Statements END Statements in the DIFFERENTIAL - END block define the system of differential equations to fit. The values of the equations are assigned to the DZ array.

Description

560

Commands, Arrays and Functions G Example DIFFERENTIAL DZ(1) = -(K1+KE)*Z(1) + K2*Z(2) DZ(2) = K1*Z(1) - K2*Z(2) END

D

Note

NDERIVATIVES must be used to set the number of differential equations to fit. NFUNCTIONS specifies the number of differential equations and/or other functions with associated data.

DINCREMEN T

COMMAND

Usage Description Examples DINCREMENT [is | =] D Specifies the increment to use in the estimation of the partial derivatives. The argument D is the increment. 0 < D < 0.1. The default value is 0.0001 DINCREMENT is 0.001 DINC 0.005

DNAMES

COMMAND

Usage Description DNAMES [are | =] 'name1' , 'name2' , . . . , 'nameN' Assigns names to the columns on the data set. The arguments 'name1', 'name2', . . . ,'nameN' are comma- or space-separated variable names. Each may use up to 8 letters, numbers or underscores, and must begin with a letter. DNAMES 'Subject', 'Time', 'Conc' The names appear in the output and may also be used in the modeling code. If both DNAMES and DHEADERS are included, DNAMES takes precedence.

Example Note

DTA

ARRAY

Usage Description Example DTA(N) Returns the value of the Nth column for the current observation. T = DTA(1) W = DTA(3)

DTIME

COMMAND

Usage Description DTIME # Specifies the time of the last administered dose. This value will appear on the Final Parameters table as Dosing_time. Computation of the final parameters will be adjusted for this dosing time. 561

D

WinNonlin User’s Guide Example DTIME 48

DZ

ARRAY

Usage Description Example Note DZ(N) Contains the current values of the differential equations. The index N specifies the equation, in the order listed in the DIFFERENTIAL block. DZ(3) = -P(3)*Z(3) The maximum value of N is specified by the NDERIVATIVES command.

H

F

VARIABLE

Usage Description Example F Returns the value of the function for the current observation. F = A * exp(-B*X)

FINISH

COMMAND

Usage Description Example FINISH Included for compatibility with PCNonlin Version 4. FINISH should immediately follow the BEGIN command. BEGIN FINISH

FNUMBER

COMMAND

Usage Description FNUMBER [is | =] N Indicates the column in the data set that identifies data for the FUNCTION block when NFUNCTIONS > 1. The function variable values must correspond to the FUNCTION numbers, i.e., 1 for FUNC 1, 2 for FUNC 2, etc. FNUMBER is 3 FNUM 3 FNUMBER should be specified prior to DATA.

Examples Note

562

Commands, Arrays and Functions I FUNCTION

D

MODEL BLOCK

Usage FUNCTION N Statements END The FUNCTION - END block contains statements to define the function to be fit to the data. The argument N indicates the function number. FUNC 1 F=A*EXP(B) + C*EXP(D) END FUNC 1 F=A*EXP(B) + C*EXP(D) END FUNC 2 F=Z(1) END A function block must be used for every data set to be fit. NFUNCTIONS specifies the number of function blocks. The variable F must be assigned the function value. WT can be assigned a value to use as the weight for the current observation.

Description Examples

Note

I

INITIAL

COMMAND

Usage Description Examples INITIAL [ARE | =] I1, I2, I3, . . . , IN Specifies initial values for the estimated parameters. The arguments I1, I2, I3, . . . , IN are the initial values, separated by commas or spaces. INITIAL = 0.01, 5.e2, -2.0 INIT are 0.02, 10, 1.3e-5 INIT 0.1 .05 100 INITIAL must be preceded by NPARAMETERS.

Note ITERATIONS

COMMAND

Usage Description Examples ITERATIONS [are | =] N Specifies the maximum number of iterations to be executed, N. ITERATIONS are 30 ITER 20

563

D

WinNonlin User’s Guide Note If ITERATIONS = 0, the usual output is produced using the initial estimates as the final values of the parameters.

J

LOWER

COMMAND

Usage Description Example Note LOWER [are |=] L1, L2, L3, . . . , LN Specifies lower bounds for the estimated parameters. The arguments L1, L2, L3, . . . , LN are the lower bounds, separated by commas or spaces. LOWER 0 0 10 If LOWER is used, a lower bound must be provided for each parameter, and UPPER must also be used. LOWER must be preceded by NPARAMETERS.

K

MEANSQUAR COMMAND E Usage MEANSQUARE [is | =] M Description Allows a specified value other than the residual mean square to be used in the estimates of the variances and standard deviations. The argument M replaces the residual sum of squares in the estimates of variances and standard deviations. If given, MEANSQUARE must be greater than 0. MEANSQUARE = 100 MEAN=1.0

Examples

METHOD

COMMAND

Usage METHOD [is ⏐ =] N

564

Commands, Arrays and Functions K Description

D

In modeling, the METHOD command specifies the type of fitting algorithm WinNonlin is to use. The argument N takes the following values: 1. Nelder-Mead Simplex 2. Levenberg and Hartley modification of Gauss-Newton (default) 3. Hartley modification of Gauss-Newton In noncompartmental analysis, METHOD specifies the method to compute area under the curve. The argument N takes on the following values: 1. Lin/Log - Linear trapezoidal rule up to Tmax, and log trapezoidal rule after 2. Linear trapezoidal rule (Linear interpolation) (default) 3. Linear up/Log down rule – Uses the linear trapezoidal rule any time the concentration data is increasing and the logarithmic trapezoidal rule any time that the concentration data is decreasing. 4. Linear Trapezoidal (Linear/log Interpolation) METHOD is 2 METH 3

Example

MODEL

COMMAND

Usage Description MODEL [N[[,] 'path']] In a command file, MODEL specifies the compiled or source model to use. If “MODEL N” is specified, model N is used from the built-in library. If “MODEL N, PATH” is used, model N from the source file path is used. MODEL 5 MODEL 7, 'PK' MODEL 3 'D:\MY.LIB'

Examples

MODEL LINK

COMMAND

Usage Description MODEL LINK M1 [,] M2 MODEL with the option LINK specifies that compiled models are to be used to perform PK/PD or Indirect Response modeling. The arguments M1 and M2 are model numbers where: M1 is the PK model number, and M2 is the model number of the PD model or Indirect Response model. When this option is used a PKVAL command must also be used. MODEL LINK 4 101 PKVAL 10 1.5 1.0 1.0 MODE LINK 1, 53 PKVA 10, 2.5 See also PKVALUE.

Examples

Note

565

D

WinNonlin User’s Guide

L

NCARRY

COMMAND

Usage Description Examples Note NCARRY [are | =] N Specifies the number of carry-over variables specified. The argument N is the number of carry-over variables. NCAR are 3 NCAR 1 NCARRY must be specified before CARRYALONG.

NCATRANS

COMMAND

Usage Description NCATRANS When using NCA, NCATRANS specifies that the Final Parameters output will present each parameter value in a separate column, rather than row. See “Transpose final parameters” in the WinNonlin User’s Guide. NCATRANS NCAT

Examples

NCONSTANT S

COMMAND

Usage Description Examples NCONSTANTS [are | =] N The NCONSTANTS command specifies the number of constants, N, required by the model. If used, NCONSTANTS must precede CONSTANTS. NCONSTANTS are 3 NCON 1

NDERIVATIV ES

COMMAND

Usage Description Examples NDERIVATIVES [are | =] N Indicates the number, N, of differential equations in the system being fit. NDERIVATIVES are 2 NDER 5

NFUNCTION S

COMMAND

Usage NFUNCTIONS [are | =] N

566

Commands, Arrays and Functions L Description Examples Note

D

Indicates the number, N, of functions to be fit (default = 1). There must be data available for each function. If used, NFUNCTIONS must precede DATA. NFUNCTIONS are 2 NFUN 3 The functions can contain the same parameters so that different data sets can contribute to the parameter estimates. When fitting multiple functions one must delineate which observations belong to which functions. To do this, include a function variable in the data set or use NOBSERVATIONS. See FNUMBER or NOBSERVATIONS.

NOBOUNDS

COMMAND

Usage Description Examples Note NOBOUNDS Tells WinNonlin not to compute bounds during parameter estimation. NOBOUNDS NOBO Unless a NOBOUNDS command is supplied WinNonlin will either use bounds that the user has supplied, or compute bounds.

NOBSERVATI COMMAND ONS Usage NOBSERVATIONS [are | =] N1, N2, N3, . . . , NF Description Provides backward compatibility with PCNonlin command files. When more than one function is fit simultaneously, NOBS can indicate which observations apply to each function. The arguments N1, N2, . . . , NF are the number of observations to use with each FUNCTION block. The data must appear in the order used: FUNC 1 data first, then FUNC 2, etc. Note that this statement requires counting the number of observations for each function, and updating NOBS if the number of observations changes. NOBS must precede DATA. NOBS 10, 15, 12 The recommended approach is to include a function variable whose values indicate the FUNCTION block to which the data belong. (See FNUMBER.)

Example Note

NOPAGE

COMMAND

Usage Description Examples NOPAGE Tells WinNonlin to omit page breaks in the ASCII output file. NOPAGE NOPAGE BREAKS ON OUTPUT 567

D

WinNonlin User’s Guide

NOPLOTS

COMMAND

Usage Description Example Note NOPLOTS Specifies that no plots are to be produced. NOPLOTS This command turns off both the high resolution plots and ASCII output.

NOSCALE

COMMAND

Usage Description NOSCALE Turns off the automatic scaling of weights when WEIGHT is used. By default, the weights are scaled such that the sum of the weights equals the number of observations with non-zero weights. NOSCALE This command will have no effect unless WEIGHT is used.

Example Note NOSTRIP

COMMAND

Usage Description Example NOSTRIP Turns off curve stripping for all profiles. NOSTRIP

NPAUC

COMMAND

Usage Description Examples Note NPAUC # Specifies the number of partial area under the curve computations requested. NPAUC 3 See PAUC.

NPARAMETE RS

COMMAND

Usage Description Examples Note NPARAMETERS [are | =] N Sets the number of parameters, N, to be estimated. NPARAMETERS must precede INITIAL, LOWER or UPPER. (LOWER and UPPER are optional.) NPARAMETERS 5 NPAR 3 See PNAMES.

568

Commands, Arrays and Functions M NPOINTS

D

COMMAND

Usage Description Example Notes NPOINTS [is | =] N Determines the number of points in the Predicted Data file. Use this option to create a data set to plot a smooth curve. 10 <= N <= 20,000. NPOINTS 200 For compiled models without differential equations the default value is 1000. If NDERIVATIVES> 0 (i.e., differential equations are used) this command is ignored and the number of points is equal to the number in the data set. The default for user models is the number of data points. This value may be increased only if the model has only one independent variable.

NSECONDAR COMMAND Y Usage NSECONDARY [are | =] N Description Examples Note Specifies the number, N, of secondary parameters to be estimated. NSECONDARY 2 NSEC = 3 See SNAMES.

NVARIABLES

COMMAND

Usage Description Examples NVARIABLES [are | =] N Specifies the number of columns in the data grid. N is the number of columns (default = 2). If used, NVARIABLES must precede DATA. NVARIABLES 5 NVAR 3

M

P

ARRAY

Usage Description Examples P(N) Contains the values of the estimated parameters for the current iteration. The index N specifies the Nth estimated parameter. P(1) T=P(3) - A 569

D

WinNonlin User’s Guide Note The values of the INITIAL command correspond to P(1), P(2), etc., respectively. NPARAMETERS specifies the value of N. PNAMES can be used to assign aliases for P(1), P(2), etc.

PAUC

COMMAND

Usage Description PAUC #, # etc. Lower, upper values for the partial AUC computations. If a specified time interval does not match times on the data set, then WinNonlin will generate concentrations corresponding to those times. PAUC 0, 5, 17, 24 See NPAUC.

Examples Note PKVALUE

COMMAND

Usage Description Examples PKVALUE N1, N2,. . . ,NPK Used to assign parameter values to the PK model when using MODEL LINK. MODEL LINK 4 101 PKVAL 10 1.5 1.0 1.0 MODE LINK 1, 53 PKVA 10, 2.5

PNAMES

COMMAND

Usage Description PNAMES [are | =] 'name1', 'name2', . . . , 'nameN' Specifies names associated with the estimated parameters. The arguments 'name1', 'name2', . . . ,'nameN' are names of the parameters. Each may have up to eight letters, numbers, or underscores, and must begin with a letter. These names appear in the output and may be used in the modeling code. PNAMES are 'ALPHA', 'BETA', 'GAMMA' PNAM 'K01', 'K10', 'VOLUME' The number of names specified must equal that for NPARAMETERS.

Examples Note PUNITS

COMMAND

Usage Description PUNITS [are | =] 'unit1', 'unit2', . . . , 'unitN' Specifies the units associated with the estimated parameters. The arguments 'unit1', 'unit2', . . . ,'unitN' are units and may be up to eight letters, numbers, operators or underscores in length. These units are used in the output.

570

Commands, Arrays and Functions N Examples Note PUNITS are 'hr', 'ng/mL', 'L/hr' PUNI 'hr' The number of units specified must equal that for NPARAMETERS.

D

N

REMARK

COMMAND

Usage Description Examples Note REMARK [comment] Allows comments to be placed in a command file. REMARK Y transformed to log(y) REMA GAUSS-NEWTON If a command takes no arguments or a fixed number of arguments, comments may be placed on the same line as a command after any required arguments.

REWEIGHT

COMMAND

Usage Description REWEIGHT W In the PK modeling module, REWEIGHT specifies that iteratively reweighted least squares estimates of the parameters are to be computed. The weights have the form WT = FW where W is the argument of REWEIGHT and F is the current value of the estimated functions. If W < 0 and F < 0 then the weight is set to 0. REWEIGHT 2 REWE -.5 REWEIGHT is useful for computing maximum likelihood estimates of the estimated parameters. The following statement in a model is equivalent to a REWEIGHT command: WT = exp(W log(F)), or WT = F**W This command is not used with noncompartmental analysis.

Examples Note

O

S

ARRAY

Usage S(N)

571

if the command SNAMES 'K10_HL' and a PNAMES command with the alias K10 have been included: SECONDARY K10_HL=0. no data fitting is performed. All other output such as plots and secondary parameters are also produced. Model size sets default memory requirements and may range from 1 to 6.D WinNonlin User’s Guide Description Example Note Contains the estimated secondary parameters. S(2). The index N specifies the Nth secondary parameter. using SNAMES. The secondary parameters are defined in the SECONDARY section of the model. if one is included.693/K10 END Or. Assignments are made to the array S. SECONDARY S(1)=0.END block. Aliases may be assigned for S(1). SIMULATE SIMULATE may be placed in any set of WinNonlin commands. The default value is 2. It must precede the DATA statement. SIZE COMMAND Usage Description SIZE N N is the model size. When it is encountered. S(1) = -P(1)/P(2) The range of N is set by NSECONDARY. etc. only the estimated parameters and predicted values from the initial estimates are computed. SECONDARY MODEL BLOCK Usage SECONDARY Statements END The secondary parameters are defined in the SECONDARY . 572 .693/K10 END NSECONDARY must be used to set the number of secondary parameters. Description Examples Note SIMULATE COMMAND Usage Description Example Note SIMULATE Causes the model to be estimated using only the time values in the data set and the initial values.

'name2'. Arrays and Functions O Example Note SNAMES SIZE 3 Applicable only when using ASCII models. etc. 1 SORT 1 3 2 SORT 1.Commands. The default is to not use sparse methods. . 1. 2 SORT must be on one line. Causes WinNonlin to segment a large data file into smaller subsets. SORT 1. . it can not use the line continuation symbol '&. They appear in the output and may also be used in the programming statements. and must begin with a letter. STARTING MODEL BLOCK Usage STARTING Statements END 573 . 'nameN' are names of the secondary parameters. Each may be up to eight letters. 'Tmax' SNAM 'K10_HL' 'K01_HL' Examples SORT COMMAND Usage Description SORT 1 [. Arguments s1 and the options s2. numbers or underscores in length. S(2).] s1 [. .’ Examples Note SPARSE COMMAND Usage Description SPARSE Causes WinNonlin to use sparse data methods for noncompartmental analysis. 'name2'. i. It is included to provide compatibility with PCNonlin version 4. 'Cmax'. . See “Sparse sampling computations” on page 188. . aliases for S(1). .. . . are the column numbers of the sort keys. Each subset will be run through the modeling separately. The arguments 'name1'. etc.e. 'nameN' Specifies the names associated with the secondary parameters. SNAMES are 'AUC'. D COMMAND Usage Description SNAMES [are | =] 'name1'.] s2 . s3. . The 1 that follows the command SORT has no function.

. operators or underscores in length. Description 574 . Examples Note P TEMPORARY MODEL BLOCK Usage TEMPORARY Statements END Statements placed on the TEMPORARY . . STEADY STATE 12 STEA 12 SUNITS COMMAND Usage Description SUNITS [are | =] 'unit1'. . 'unit2'.END block specify the starting values of the differential equations. Parameters can be used as starting values.D WinNonlin User’s Guide Description Statements in the STARTING . # is Tau. . . . Values are placed in the Z array. START Z(1) = CON(1) Z(2) = 0 END Example STEADY STATE COMMAND Usage Description Examples STEADY STATE # Informs WinNonlin that data were obtained at steady state. . 'unit2'. 'unitN' Specifies the units associated with the secondary parameters. 'mL/hr' SUNI '1/hr' The number of units specified must equal NSECONDARY. These units are used in the output. The arguments 'unit1'. the (assumed equal) dosing interval. . so that different PK parameters are computed.'unitN' are units and may be up to eight letters. numbers. SUNITS are 'ml'.END block define global temporary variables that can be used in the MODEL statements.

TITLE Experiment EXP-4115 TITLE 1 Experiment EXP-4115 TITLE 2 Subject #1 Examples TRANSFORM MODEL BLOCK Usage TRANSFORM Statements END Statements placed in the TRANSFORM . 1 <= N <= 5. See “Drug effect data (model 220)” on page 515 for details. To specify any title other than the first. 575 Description . a value N must be included with TITLE.END block are executed for each observation before the estimation begins.Commands. Arrays and Functions P Example TEMP T=X DOSE1 = CON(1) TI=CON(2) K21=(A*BETA + B*ALPHA)/(A+B) K10=ALPHA*BETA/K21 K12=ALPHA+BETA-K21-K10 END D THRESHOLD COMMAND Usage Description Example THRE N Sets the (optional) threshold value for use analyzing effect data with NCA model 220. Up to five titles are allowed per run. TIME TITLE COMMAND Usage Description TITLE [N] Indicates that the following line contains a title to be printed on each page of the output. THRE 0 TIME COMMAND Usage Description Example TIME Places the current time in the upper right hand corner of each output page.

and the concentration in the urine. U3. U2. … . . URINE COMMAND Usage Description URINE LOWER [is | =] C1 UPPER [is | =] C2 VOLUME [is | =] C3 [is | =] CONCENTRATION [is | =] C4 Used to specify variables when doing noncompartmental analysis for urine data (Models 210-212). Variables must be specified for all four keywords. 1. UPPER 1. The user must supply: the beginning time of the urine collection interval. Examples Note R WEIGHT COMMAND Usage WEIGHT [W] 576 . .D WinNonlin User’s Guide Example TRANSFORM Y = exp(Y) END Q UPPER COMMAND Usage Description Example Note UPPER [are |=] U1. URINE LOWER=C1 UPPER=C2 VOLUME=C3 CONC=C4 URINE CONC C1 VOLU C2 LOWE C3 UPPE C4 The keywords may appear in any order. UN are the bounds. The arguments U1. the volume of urine collected. 200 If UPPER is used. . U3. .C4 give their column numbers. and LOWER must also be specified. separated by commas or spaces. the ending times of the urine collection interval. The four keywords LOWER UPPER VOLUME. U2. an upper bound must be provided for each parameter. UN Sets upper bounds for the estimated parameters. and only the first four letters of each keyword is required. C1 . UPPER must be preceded by NPARAMETERS. and CONCENTRATION identify the required data columns.

WTNUMBER 4 WTNU is 7 If WEIGHT is used and WTNUMBER is not. when used in a command file. See also WTNUMBER. a column in the data set provides weights— by default.5 Uses as the weight 1/Sqrt(Y) WEIGHT WTNUM 7 Uses the values in Column 7 as the weights The weights are scaled so that their sum = the number of observations with non-zero weights. WT contains the weight for the current observation. The argument W sets the weight value: YW. Another column may be set using WTNUMBER. WT. the default column for weights is 3. F = exp(-X) By default. REWEIGHT and NOSCALE. WTNUMBER COMMAND Usage Description Examples Note WTNUMBER [is | =] N Specifies which column of the data set contains weights. S X VARIABLE Usage Description Example Note X Contains the current value of the independent variable. See also “WEIGHT” on page 576. WT = 1/(F*F) If the variable WT is included in a model. the third column.Commands. If W is not set. then any weighting indicated via the Variables and Weighting dialog will be ignored. the values of X are taken from the first column of the data set unless the XNUMBER command has been used. Examples Note WT VARIABLE Usage Description Example Note WT In PK modeling. 577 . WEIGHT Uses the values in Column 3 as the weights WEIGHT -. Not used for NCA. Arrays and Functions S Description D Activates weighted least squares computations.

XNUMBER indicates the column number that the differential equations are integrated with respect to. XNUMBER is 3 XNUM 1 XNUMBER should precede DATA. If differential equations are being fit. Y is assumed to be in the 2nd column unless YNUMBER is used. YNUMBER is 4 YNUM 3 The YNUMBER command should precede the DATA command. Y = log(Y) By default. The index N specifies the Nth differential equation. B = A*Z(2) The maximum value of N is set by NDERIVATIVES. of the independent variable. U Z ARRAY Usage Description Example Note Z(N) Contains the current values of the solutions of the differential equations. When running Version 4 command files. YNUMBER COMMAND Usage Description Examples Note YNUMBER [is | =] N The YNUMBER command specifies the column number. of the dependent variable. N. 578 . Examples Note T Y VARIABLE Usage Description Example Note Y Current value of the dependent variable.D WinNonlin User’s Guide XNUMBER COMMAND Usage Description XNUMBER [is | =] N Sets the column number. N. the default value is 2. The default value is 1.

refer to “Scripting” on page 417.84 579 . Only one Aggregate Variable may be used. Table. A AggregateVar Property Applies To Description Syntax Remarks Example Table For tables 9 and 10. sets the variable from Data Set 1 that composes the body of the table.AlphaTerms(2)=3.AlphaTerms(1)=.AggregateVar = “Conc” AlphaTerms() Property Applies To Syntax Remarks See also Example Toolbox Toolbox. syntax and usage For instructions on writing and executing WinNonlin scripts. The Aggregate Variable is divided into separate columns based on values of the cross variable(s).Appendix E Scripting Commands WinNonlin scripting language commands.229 Toolbox.AggregateVar = VariableName Similar to a Variable in tables 4-8. See the Examples Guide for example scripts. Table.AlphaTerms(n)=number (n) corresponds to the number set by NumTerms NumTerms Toolbox.

E WinNonlin User’s Guide Append Method Applies To Description Syntax Remarks Data Add content of a file to another file. properties. • All load options are in effect.Append See also Example AppendFilename Property Applies To Description Syntax See also Data Specifies a file to be appended to an existing data object. Use Find/Replace to standardize the missing value codes. Delimiter.Append method. Missing MYDATA. just as the Data.PWO to another file. . WinNonlin will not update these automatically. AppendFileType AppendFileType Property Applies To Description Syntax Remarks See also Data Specifies the file format of the file to be appended to an existing data object.Load method would. objectname. AppendFilename. Only the values are appended.PWO file types are not supported with this property. AppendFilename 580 . Append.AppendFileType = EXCEL MYDATA.Append Data can be appended to a Pharsight workbook object (*. • The ReadOnly status is not recalled for the appended file.PWO). objectname. When using the Data.Headers. • Formulas from Excel files will be lost. When using Append: • If the two data sets use different missing value codes. The ReadOnly status of the original file will be maintained. Objectname.AppendFilename = “C:\MYDIR\newdata.xls” MYDATA. etc. if the file type is ASCII. but it is not possible to append a .Delimiter.AppendFileType = {ASCII | EXCEL} . Headers.AppendFilename = string Append. it will use the . AppendFileType.

Append Arrange Method Applies To Description Syntax WinNonlin Arranges the icons representing any minimized windows in WinNonlin.3 Toolbox.Aterms(2)=8.AppendFilename = “C:\MYDIR\newdata.xls” MYDATA.Scripting Commands A AppendFileType (continued) Property Example MYDATA.Aterms(n)=number The argument (n) corresponds to the number set by NumTerms. Used only with the Append method.AppendFilename=”Test2” MYDATA1. Toolbox. WinNonlin Arrange Aterms() Property Applies To Syntax Remarks Example Toolbox Toolbox. MYDATA1.Filename=”Test1” MYDATA1.Aterms(1)=12. MYDATA.1 581 .Load MYDATA1.Append E AppendSheet Property Applies To Description Syntax Remarks Example Data Selects the sheet of a multisheet workbook to be used for an append operation (similar to the Multigrid Property).AppendSheet=”Sheet 1” <-selected sheet to append MYDATA1.Multigrid=”Sheet 3” MYDATA1.AppendSheet=”Final Parameters” Used with the Append method of the Data property.AppendFileType = EXCEL MYDATA.

AutoSort = False B Bolus Property Applies To Description Syntax Example Toolbox Sets whether WNL will constrain the estimated input rate to zero at the initial time (lag time). the X variable column will be sorted before the plot is made. Toolbox. Plot.Bolus=True C CarryData Property Applies To Description Syntax Merge Sets whether or not “Carry along data for like sort levels” is used. The default is True (yes).CarryData = {True | False} Cascade Property Applies To 582 WinNonlin .E WinNonlin User’s Guide AutoFileOptions Property Applies To Description Syntax Data Turns on automatic file options when loading a data file. Merge. DataObject.AutoSort = {True | False} Plot.Bolus={True|False} Toolbox.AutoFileOptions = True/False AutoSort Property Applies To Description Syntax Example Plot If AutoSort = “True” (default).

Scripting Commands C Cascade (continued) Property Description Syntax See also Example Causes the open objects to be cascaded in the WinNonlin window. RemoveCol. Save and Unload Column Property Applies To Data 583 . Exclude.CaseSensitive=False MYDATA. RenameCol and Replace.Tolerance=”0” MYDATA.Find=”3” MYDATA. Remove. Objectname. WinNonlin.Cascade E CaseSensitive Property Applies To Description Syntax Remarks Example Data Specifies whether a search is case sensitive (True) or not (False). MYDATA.SearchArea=”CONC” MYDATA.With=”Missing” MYDATA. WinNonlin.Cascade TileHorizontal and TileVertical WinNonlin.Operation=”=” MYDATA.Replace CloseAll Method Applies To Description Syntax Remarks See also WinNonlin Closes all window objects within WinNonlin.CaseSensitive = {True | False} Use with Include.CloseAll Any information that had not been saved will be lost.

the system will use the value specified on the most-recent Load method. put data in the column first (e. Interval.ConseqDelim = {True | False} Used only when loading ASCII data. Toolbox.Column = string Use with RemoveCol and RenameCol.. In order to create and rename a new column. Percentiles Desc.Column=”a” MYDATA. Objectname. Desc.” Load treats consecutive delimiters as one. If a value is not set. using a Paste or Append operation) then rename it. MYDATA.ConcCol=column name Toolbox. 584 . Note that WinNonlin recognizes only columns that contain data.Confidence = True ConseqDelim Property Applies To Description Syntax Remarks Data When ConseqDelim = “True.RenameCol Syntax Remarks Example ConcCol Property Applies To Description Syntax Example Toolbox Sets the input column for Concentration data.With=”Subject” MYDATA.E WinNonlin User’s Guide Column (continued) Property Description Specifies a column to be renamed or removed.ConcCol=”Concentration” Confidence Property Applies To Description Syntax Remarks See also Example Desc Sets whether confidence intervals will be calculated with descriptive statistics (True) or not.g. objectname.Confidence = {True | False} The default is False. by column name (or letter).

EndCol=b MYDATA.Copy MYDATA. and EndRow. StartRow.StartCol=a MYDATA. the cross variable must be from Data Set 1. IDVar( ). EndCol. which define the columns when breaking a variable into separate columns within a table. Missing. To specify an entire column (or entire columns) use only StartCol and EndCol.CrossVar(2) = “Form” CrossVars Property Applies To Table 585 . 9.Copy = string Optionally. RegVar( ) Table. specify a selection of cells for copying using StartCol.Paste Example CrossVar( ) Property Applies To Description Syntax Remarks See also Example Table Defines cross variable(s). Headers E Copy Method Applies To Description Syntax Remarks Data Copies a specified range of data and stores it for use with Paste.EndRow=20 MYDATA. CrossVars.CrossVar(1) = “Subject” Table. 6.CrossVar(Variable no.) = Variable Name Applies to tables 4. MYDATA.StartCol=e MYDATA. Table. 7.Filename = <file name> MYDATA.StartRow=1 MYDATA. 5. For table 9.StartRow=1 MYDATA. 8. Objectname.Scripting Commands C ConseqDelim (continued)Property See also Delimiter. To specify an entire row (or entire rows) use only StartRow and EndRow. GroupVar( ). CrossVars sets the number of cross variables.

7. for rows. GroupVars.StartCol=A MYDATA. and EndRow.Cut MYDATA. IDVars.StartRow=5 MYDATA.StartCol=A MYDATA. MYDATA.Cut = <string> To select cells to cut.EndRow=10 MYDATA. Table.StartRow=1 MYDATA.Filename = <file name> MYDATA. 6. Objectname.E WinNonlin User’s Guide CrossVars (continued)Property Description Syntax Remarks See also Example Defines the number of cross variables for the Table engine.DefinitionFile = string 586 . except the data file(s) to use.CrossVars = integer Applies to tables 4. StartRow and EndRow. 9. EndCol. StartRow. 8.CrossVars = 2 Cut Method Applies To Description Syntax Remarks Example Data Cuts a specified range of data from a data set and stores it for use with the Paste function. use StartCol.EndCol=J MYDATA. Table. RegVars Table. 5. To select an entire column (or columns) use only StartCol and EndCol.Paste D DefinitionFile Property Applies To Description Syntax Table Loads a definition file containing all specifications for a table.

the system will use the value specified for the most-recent Load method.FileMask = “data. Headers MYDATA. ConseqDelim.DefintionFile=“ ” to nullify the definition file before using the commands.* are not allowed in file specifications. 587 . WinNonlin.Delimiter = “ ” Delta Property Applies To Description Toolbox User supplied smoothing parameter.Delimiter = character Used only while loading ASCII data files.Scripting Commands D DefinitionFile (continued)Property Remarks E Any input data file that includes the exact variable names used in the table definition may be used with a definition file. then create another table using Scripting Language commands.” MYDATA. If this value is not specified. Use the command: TABLE.DefinitionFile=“c:\winnonln\mytable.tdf” Example DeleteFiles Method Applies To Description Syntax Remarks See also Example WinNonlin Deletes a file selected using the FileMask property. objectname. It must be created and saved by the table Generator. Table.DeleteFiles Delimiter Property Applies To Description Syntax Remarks See also Example Data Specifies the field delimiter to use with the Load method. Missing. The definition file is machine readable only.xls” WinNonlin.DeleteFiles Directory names and *. FileMask WinNonlin. It is necessary to turn a table definition file “off” if a script will use a table definition file.Delimiter = “.

DestCol = string Spaces are not allowed in the new column name. UpErr( ). is required even if there is only one data source for the plot.DnErr(Data Set no. The column can be either inserted adjacent to the SourceCol or appended to the end of the data set.DnErr(1) = “Minimum” 588 .Delta=2 DestArea Property Applies To Description Syntax Remarks See also Example Trans If DestCol is defined as a new column name.DestArea = {Adjacent | Append} If DestCol uses an existing column.E WinNonlin User’s Guide Delta (continued)Property Syntax Example Toolbox. Trans.DestArea = Append DestCol Property Applies To Description Syntax Remarks See also Example Trans This property defines the destination column for the transform function. DestArea Trans.) = string Data Set no.Delta=number Toolbox. DestCol Trans.DnErr(1) = “SE” Plot. then DestArea determines its location. then DestArea is ignored.DestCol = “LnConc” DnErr( ) Property Applies To Description Syntax Remarks See also Example Plot Defines the column containing down error data for a plot with error bars. Trans. ErrType Plot. Plot.

See “Special file formats” on page 430 for details on this file. Toolbox. Its value is a string (column name) or number (position).DoseUnit={unit} Toolbox.DoseUnit=”ng/ml” E Ecol Property Applies To Description Syntax Remarks Example Toolbox Specifies the input column for effect data. but the column number (position from the left) must be used to designate a new column.Ecol=”Effect” EndCol Property Applies To Description Data Sets the last column in a range. PartFile Model.DoseFile = “c:\dosedata. Model. Toolbox.dat” DoseUnit Property Applies To Description Syntax Example Toolbox Specifies unit for dosing. 589 .Ecol=column name Not used in calculations. Note that the column name can be used to reference an existing column. LamzFile.Scripting Commands E E DoseFile Method Applies To Description Syntax Remarks See also Example Model Loads an ASCII data file containing model dosing information. ParmFile.DoseFile=string See “Special file formats” on page 430. Toolbox. LinkFile.

EndCol = string | number Use with RemoveRow. 590 . setting EndRow = –1 selects all rows below and including StartRow.StartRow=5 MYDATA.EndCol = MyData.EndRow = MyData. MYDATA. Cut.EndCol = MyData.StartRow MyData.Copy MyData. The following example removes all rows after row 4.E WinNonlin User’s Guide EndCol (continued) Property Syntax Remarks See also Examples Objectname. EndRow MyData. StartRow <= EndRow.EndRow = MyData. Set StartCol <= EndCol. Cut and Copy methods. Objectname. StartCol. is to be included or excluded. Objectname.Copy = 1 = “Time” 12 “Time” = 1 = 1 12 3 EndRow Property Applies To Description Syntax Remarks Data Sets the last row in a range.RemoveRow Example EntireRow Property Applies To Description Syntax Remarks Data Specifies that the entire row.StartCol MyData.StartCol MyData.EndRow=-1 MYDATA. StartRow. Copy or Font and FontSize.StartRow MyData. If StartRow is set and is not –1. as opposed to individual cell(s).EntireRow = {True | False} Use with Include and Exclude.EndRow = number Use with the RemoveRow.

objectname.EntireRow=True MYDATA.ErrType = {Absolute | Relative} The default value is Relative.CaseSensitive=False MYDATA. Tolerance. Use the “Print” properties to set the object type(s) to export.Scripting Commands E EntireRow (continued)Property Example MYDATA. CaseSensitive MYDATA. See “Data exclusion and inclusion” on page 60.Exclude E ErrType Property Applies To Description Syntax Remarks See also Example Plot Determines handling of error data by the Plot engine. UpErr( ).SearchArea=”entire data set” MYDATA.Exclude ExportAll Method Applies To Description WinNonlin Exports all open objects to Microsoft Word. The example below exports all charts.Tolerance=”0” MYDATA.Find=”censor” MYDATA. See “Error bars” on page 120. SearchArea. using the same options as PrintAll.SearchArea=”entire data set” MYDATA. EndRow.Operation=”<” MYDATA.EntireRow=True MYDATA. Operation. DnErr( ) Plot.Exclude Find.Find=”0” MYDATA. Plot. EntireRow. 591 .ErrType = Absolute Exclude Method Applies To Description Syntax See also Example Data Excludes the selected data. StartRow.

DestArea TRANS_CHANGEBASE: Trans. SourceCol. Extra.defines Column Y String .Extra = {string | number} Function used TRANS_LN TRANS_LOG10 TRANS_SQRT TRANS_ABS TRANS_CHANGEBASE TRANS_PCTCHANGE TRANS_RATIOBASE TRANS_MULTCOL TRANS_METABOLITE TRANS_ADDCOL TRANS_SUBTRACTCOL TRANS_E TRANS_INV TRANS_POWER TRANS_MULT TRANS_DIV TRANS_ADD TRANS_SUBTRACT Use Not used Not used Not used Not used Variable Name .specifies Time column String .E WinNonlin User’s Guide ExportAll (continued)Method Syntax See also Example WinNonlin.Extra = 2 592 .PrintText=False WinNonlin.Extra = “Hour” TRANS_POWER: Trans.defines Column Y Not used Not used number number number number number See also Example InData. as detailed under “Remarks” below.ExportAll Extra Property Applies To Description Syntax Remarks Trans Contains additional information required for some transformations. Column Y String . DestCol.PrintData=False WinNonlin.defines Column Y String .ExportAll PrintAll WinNonlin. Each transformation requires different information.defines the denominator.PrintModel=False WinNonlin. Trans.PrintGraph=True WinNonlin.specifies Time column Variable Name .specifies Time column Variable Name .

Graph. Text Sets the file format for Load and Save.wsp” WinNonlin.\data\profiles.DeleteFiles Filename Property Applies To Description Syntax Remarks Example WinNonlin.Filename = “. specifies a workspace to load. Data. WinNonlin can load WDO. When the FileType is set to anything other than PCO.WSP). the FileType must be set to PCO.Scripting Commands F E F FileMask Property Applies To Description Syntax Remarks Example WinNonlin Specifies a file in the current working folder.FileMask=”bguide2. WMO. WTO. Data. For the WinNonlin object.filename = “c:\winnonln\examples\profiles.* to select files. WinNonlin.pwo” MyData. WGO.FileType = {ASCII | ANV | EXCEL | EXCEL95 | EXCEL97 | PWO | BMP | WMF | PCO | PMO | RTF | PTO | LML | SAS | ODBC | PKS}a When loading graph objects.dat” WinNonlin. WinNonlin.Filename = <absolute or relative path to data file> Use with FileType except when using with the WinNonlin object (assumes file type . Text Sets a file name for Load and Save. Save 593 See also . and EXCEL file types but cannot save back to them.FileType = “PWO” MyData.FileMask =string Use with DeleteFiles.Load FileType Property Applies To Description Syntax Remarks ANOVA. Filename. WinNonlin. WinNonlin. Model. Graph.. objectname. Load. Save is equivalent to export. Do not use directory names or *. Model.load or MyData.filename = <absolute or relative path to workspace file> or objectname.

ObjectName. Remove and Replace.EndRow=1 MyData.Replace Font Property Applies To Description Syntax See also Example Data Sets the font for the current selection.EndCol=10 MyData.Operation=”=” MYDATA. MYDATA.Find = string Use with Exclude. StartCol.StartCol=1 MyData. Include.Font = string FontSize.Find=”0” MYDATA.001” MYDATA. Find Property Applies To Description Syntax Remarks Example Data Specifies data to search for. Use Excel97 to load Excel2000 files.E WinNonlin User’s Guide FileType (continued)Property Example MyData.SearchArea=”CONC” MYDATA. EndRow. EndCol MyData.With=”0” MYDATA.FileType = EXCEL a. StartRow. ObjectName.StartRow=1 MyData.Tolerance=”0.FontSize = number .Font=”Arial” FontSize Property Applies To Description Syntax 594 Data Sets the font size for the current selection. objectname.

StartRow=1 MyData.FootnoteFont = “Arial” FootnoteFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the chart footnote.EndRow=1 MyData. ObjectName. ObjectName.StartCol=1 MyData. EndCol MyData. EndRow.FontSize=12 E FootnoteFont Property Applies To Description Syntax See also Example Graph Sets the font for the chart footnote. 595 .FootnoteFontSize=number FootnoteFont MyGraph. StartRow.EndCol=-1 MyData.FootnoteFont=string FootnoteFontSize MyGraph. StartCol.Scripting Commands F FontSize (continued)Property See also Example Font.FootnoteFontSize = 12 Function Property Applies To Description Transform Defines the function to be used by the transform engine.

SourceCol. GroupVars. IDVar( ). DestArea Trans. which sets the number of group variables. Table.GroupVar(1) = “Subject” 596 . ) when using the Plot engine.E WinNonlin User’s Guide Function (continued)Property Syntax Trans.Function = {TRANS_LN | TRANS_LOG10 | TRANS_SQRT | TRANS_ABS | TRANS_CHANGEBASE | TRANS_PCTCHANGE TRANS_RATIOBASE | TRANS_MULTCOL | TRANS_METABOLITE | TRANS_ADDCOL| TRANS_SUBTRACTCOL| TRANS_E | TRANS_INV | TRANS_POWER | TRANS_MULT | TRANS_DIV | TRANS_ADD | TRANS_SUBTRACT} LN(x) Log10(x) Square Root(x) Absolute Value(x) Change from Baseline %Change from Baseline Ratio from Baseline x*y x/y x+y x-y e^x 1/x x^n x*n x/n x+n x-n Remarks See also Example The string must match exactly (not case-sensitive) one of the strings above.) = string This property must be specified after GroupVars. CrossVar( ). GroupVarBreak( ). GroupVar applies to all tables except 9 and 10 (see JoinGroupVar( .Function = TRANS_LN G GroupVar( ) Property Applies To Description Syntax Remarks See also Example Table Defines the group variables for the Table engine.GroupVar(Variable no. RegVar( ) See GroupVar( . Extra. Table. DestCol. )). InData.

. GroupVar( ).) = {True|False} 597 . RegVars Table. The example given below sets the first GroupVar for Data Set 1 to Subject.GroupVars(Data set no. CrossVar( ). RegVar( ) Plot.GroupVars = integer Applies to all tables except tables 9 and 10. IDVars.GroupVars = 1 GroupVars( ) Property Applies To Description Syntax Example Plot Sets the number of group variables from a given data set. CrossVars. IDVar( ).GroupVar(2.Scripting Commands G E GroupVar( .GroupVar(Data set no.1) = “Subject” Plot.1) = “Patno” GroupVars Property Applies To Description Syntax Remarks See also Example Table Sets the number of group variables for the Table engine.GroupVars(1) = 2 GroupVarBreak( ) Property Applies To Description Syntax Table Specifies whether text output has a page break when the group variable values change.) = integer Plot. Plot. Table.GroupVarBreak (Variable no.) = string This property must be specified after GroupVars( ). Plot. see JoinGroupVars. When using table 9 or 10. Variable no. which sets the number of group variables. ) Property Applies To Description Syntax Remarks See also Example Plot Defines the group variables for each data set for the Plot engine. for the Plot engine.GroupVar(1. Table. GroupVar( ).

objectname.E WinNonlin User’s Guide GroupVarBreak( ) (continued) Property See also Example GroupVar( ).Headers = {True | False} Not used when loading PWO files. 2. WinNonlin.” the first row of a data file contains column names rather than data. When using table 9.) = string Applies to table no. RegVar( ) . The number of ID variables is set by IDVars. ).GroupVarBreak(1) = True H Halt Method Applies To Description Syntax Remarks Example See also WinNonlin During runtime. leaving WNL and all windows open.Halt= string Useful in debugging scripts. GroupVars Table. Delimiter. see JoinIDVar( . Missing. 6.Halt=”See the output” MsgBox Headers Property Applies To Description Syntax Remarks See also Data When Headers = “True.IDVar(Variable no. GroupVar( ). WinNonlin. Halt stops the script and presents a dialog asking “Do You Want to Continue?” with Yes/No buttons. CrossVar( ). Table. 7. The default value is that used in the last Load operation. 10. Similar to MsgBox. ConseqDelim I IDVar( ) Property Applies To Description Syntax Remarks See also 598 Table Defines the ID variables for a table. 5. 8. Selecting No stops the run. 3.

IncludeVar(2. Find.SearchArea="conc" MYDATA. When using table 9.2) Merge. Tolerance.IDVar(2) = “Time” E IDVars Property Applies To Description Syntax Remarks See also Example Table Sets the number of ID variables for the Table engine. 7. 10. CaseSensitive MYDATA.Variable no.IDVars = integer Applies to table no. 6.. see JoinIDVars. SearchArea.2) = = = = “AUC_PO” “Cmax_PO” “AUC_IV” “Cmax_IV” 599 .Include IncludeVar( .Include Exclude.3" MYDATA. CrossVars.Find="4. Table.1) Merge.IncludeVar(1.IDVars = 2 Include Method Applies To Description Syntax See also Example Data Includes a range of data.IncludeVar(2.) = string IncludeVars( ) Merge. IDVar( ) defines each variable.1) Merge.IDVar(1) = “Subject” Table.Scripting Commands I IDVar( ) (continued) Property Example Table. Merge. 8. 2. EntireRow. GroupVars. objectname. RegVars Table. Operation.IncludeVar(Data Set no. ) Property Applies To Description Syntax See also Example Merge Tells the Merge engine which columns to include from each data set. 5. Set IncludeVars( ) first. 3.IncludeVar(1.

IncludeVars(Data Set no. NumData.E WinNonlin User’s Guide IncludeVars( ) Property Applies To Description Syntax See also Example Merge Defines the number of variables to be included from each data set in a merge operation. Plot Defines the data source for the above engines.IncludeVars(1) = 3 Merge. Trans Defines the data source for the above engines. InData( ) Trans.InData(1) = MyOral Merge.InData(2) = MyIV where MyOral and MyIV are previously defined data objects InitialDose Property Applies To 600 Toolbox . engine.InData = string The value supplied must be a valid Data object in the script file.InData = MYDATA InData( ) Property Applies To Description Syntax Remarks See also Example Table. NumData. InData Merge. ) Merge. Merge.InData( ) = string The value supplied must be a valid Data object in the script file. Must be set after NumData. engine.) = integer IncludeVar( . Merge.IncludeVars(2) = 2 InData Property Applies To Description Syntax Remarks See also Example Desc.

JoinCrossVar(Data Set no.InputLag={True|False} Toolbox. Percentiles Desc. ) 601 .Interval = number Confidence. Desc. Variable no. JoinCrossVars.InitialDose=number Toolbox. ). JoinGroupVar( . Table. ) Property Applies To Description Syntax Remarks See also Table Defines a common cross variable when merging two data sets using table 9. Toolbox. JoinCrossVars sets the number of join cross variables.Scripting Commands J InitialDose (continued) Property Description Syntax Example Sets a value for initial dose. JoinIDVar( .InitialDose=11 E InputLag Property Applies To Description Syntax Example Toolbox Selects whether WinNonlin will constrain the derivative of the estimated input rate to zero at the initial time (lag time).InputLag=False Interval Property Applies To Description Syntax See also Example Desc Specifies the value of the confidence interval for descriptive statistics. Toolbox.) = Variable Name Both arguments are required even if there is only one data source and/or join cross variable.. CrossVar( ).Interval = 95 J JoinCrossVar( . The default is 95.

). JoinIDVars.1) Table.JoinCrossVar(1. JoinCrossVar( . ) Table. JoinGroupVars sets the number of join group variables.1) Table.E WinNonlin User’s Guide JoinCrossVar( . Table. Variable no.1) Table.JoinGroupVar(1.JoinCrossVar(2.JoinGroupVar(Data Set no.JoinCrossVar(2. ).JoinGroupVar(1. JoinIDVar( .1) = “Trt” JoinGroupVars Property Applies To Description Syntax Remarks Table Defines the number of join group variables when merging two data sets with table 9.JoinGroupVars = integer The two data sets must have the same number of group variables.JoinCrossVars = integer The two data sets must have the same number of cross variables. CrossVars. JoinGroupVars Table. 602 ..) = Variable Name Both arguments are required even if there is only one data source and/or join group variable. JoinCrossVar( . ) Property Applies To Description Syntax Remarks See also Example Table Defines a common group variable when merging two data sets with table 9. GroupVar( ). JoinGroupVars. Table.JoinCrossVar(1. Table. ) (continued)Property Example Table.1) = = = = “Form” “Subject” “Form” “Subject” JoinCrossVars Property Applies To Description Syntax Remarks See also Example Table Defines the number of join cross variables for table 9.JoinCrossVars = 2 JoinGroupVar( .1) = “Form” Table.

IDVar( ). Variable no. IDVars. Table. JoinGroupVar( . Must be from 0 to 1. JoinGroupVars.JoinIDVar(1.Keo=number Toolbox. ). ). Table. JoinCrossVar( .JoinIDVar(Data Set no. ) Table. JoinGroupVar( .1) = “Subject” Table.JoinIDVars = integer The two data sets must have the same number of ID variables.2) = “PatNo” JoinIDVars Property Applies To Description Syntax Remarks See also Example Table Defines the number of join ID variables from each data set for table 9.. ). JoinCrossVars Table.Scripting Commands K JoinGroupVars (continued) Property See also Example GroupVars. JoinIDVars sets the number of join ID variables.JoinGroupVars = 1 E JoinIDVar( . Toolbox.JoinIDVar(1. JoinIDVars Table. JoinIDVar( .JoinIDVars = 1 K Keo Property Applies To Description Syntax Example Toolbox Sets the value for Keo.5 603 .Keo=. JoinCrossVars. ) Property Applies To Description Syntax Remarks See also Example Table Defines a common ID variable when merging two data sets with table 9. JoinIDVars.) = Variable Name Both arguments are required even if there is only one data source and/or one join ID variable.

Legend(2) = “Urine” Plot. ObjectName. Plot. . PartFile Model. Plot.Legend(1) = “Plasma” Plot. Model.LamzFile = “c:\lamzdata.Legend(1) = “” LegendFont Property Applies To Description Syntax See also Example Graph Sets the font for the chart legend.LegendFont = “Arial” LegendFontSize Property Applies To Description 604 Graph Sets the font size for the chart legend.Legend(Data Set no. DoseUnit.dat” Legend( ) Property Applies To Description Syntax Remarks Example Plot Defines a string to appear in the legend of an overlay plot (instead of the data set name).LegendFont=string LegendFontSize MyGraph.E WinNonlin User’s Guide L LamzFile Method Applies To Description Syntax Remarks See also Example Model Loads an ASCII data file containing Lambda_z information for NCA models.LamzFile = string See “Special file formats” on page 430 for details on this file.) = string Applies only when creating overlaid plots.

WinNonlin Loads a file into the system using the current file name. FileType.LinkFile = “c:\linkdata. Graph. When used with the WinNonlin object. Save MyGraph. When using the pre-compiled PK/PD or PK/IPR models.Load When using graph objects. Model. objectname. Filename. and possibly lower/upper bounds. ParmFile reads in the model parameters and possibly lower/upper bounds. See “Special file formats” on page 430 for details of these ASCII files. ASCII PK.LinkFile = string When using compiled.LegendFontSize = 12 E LinkFile Method Applies To Description Syntax Remarks Model Load an ASCII data file containing model parameters information for PK/PD and IPR models. This method should not be used for NCA models. The scenario name is captured from the PKS file pointed to by the Data Object Filename property. ParmFile Model. or User-ASCII models. loading the Data Object as file type PKS with the file name pointing to a PKS file will load the study with or without a scenario. this method loads workspaces.LegendFontSize=number LegendFont MyGraph. When used with the Data Object associated with a PKS study. the Load method works only for file type PCO. LinkFile reads in the PK model parameters and ParmFile reads the PD or IPR parameters.dat” See also Example Load Method Applies To Description Syntax Remarks Data. Text.Load See also Example LoadTemplate Method Applies To Graph 605 . DoseUnit.Scripting Commands L LegendFontSize (continued) Property Syntax See also Example ObjectName. Model.

MoveFirst objectname. Toolbox.E WinNonlin User’s Guide LoadTemplate (continued)Method Description Syntax See also Example Loads a template on the current graph. objectname.LoadTemplate “c:\winnonln\temps\xy. MoveFirst MovePrevious MoveNext MoveLast Methods Applies To Description Syntax Graph In a group of graphs created with a sort variable. objectname.gtp” M MaintenanceDose Property Applies To Description Syntax Example Toolbox Sets a value for maintenance dose.MoveFirst See also Example 606 .MovePrevious objectname.MoveNext objectname.Missing = . selects which graph is active or displayed.MaintenanceDose=11 Missing Property Applies To Description Syntax See also Example Data Sets the alphanumeric string that represents missing data. objectname.MoveLast SortLevel MyGraph.Missing = string ConseqDelim. Headers MYDATA. Delimiter.LoadTemplate = string SaveTemplate MyGraph.MaintenanceDose=number Toolbox.

Similar to Halt.MsgBox = string The script will pause until the user clicks OK to clear the message box from the screen.MultiGraph = “X vs. Final Parameters When sorting data. If Multigrid is used on a data set that is currently showing the history.MultiGraph = Graph Name where available Graph Names are: • Modeling: X vs. Observed Y and Predicted Y” Example MultiGrid Method Applies To Description Syntax Remarks Data Selects which sub-grid (worksheet) in a multigrid (workbook) is active. N Npoints Property Applies To Toolbox 607 . Observed Y vs.MsgBox = “Script complete” MultiGraph Method Applies To Description Syntax Graph Changes the visible graph type in multigraphs. and Sort commands. the data will be toggled into view for that object. use this command to select the appropriate worksheet before specifying the SortVars. WinNonlin.g. WinNonlin. Weighted Predicted Y vs. Weighted Predicted Y. e.. Weighted Residual Y. Partial Derivatives • NCA: X vs. objectname. or displayed. SortVar( ). Observed Y Predicted Y MyGraph. Weighted Residual Y. X vs.Scripting Commands N E MsgBox Method Applies To Description Syntax Remarks Example WinNonlin Presents a message box during script runtime.MultiGrid = Grid Name where Grid Name is the worksheet name. objectname. Observed Y and Predicted Y.

NumData = integer When used with Merge. InData( ) Table. engine.Npoints=100 NumData Property Applies To Description Syntax Remarks See also Example Table. Merge Indicates the number of data sets to be used with an engine. Objectname.E WinNonlin User’s Guide Npoints (continued) Property Description Syntax Example Number of output data points. and Replace. with table or Plot. Plot. 608 . NumData must be 2.NumTerms=number Range is 1 to 9 Toolbox.Operation = {<> | < | > | <= | >= | = } Use with Exclude. Include.NumData = 1 NumTerms Property Applies To Description Syntax Remarks Example Toolbox Number of exponential terms in the unit impulse response. Remove.NumTerms=4 O Operation Property Applies To Description Syntax Remarks Data Specifies search operator. Toolbox. NumData must be 1 or 2.Npoints=number Toolbox. Value must be 2 or greater. Toolbox.

OutData = “MYDATA” OutGraph Property Applies To Description Syntax Remarks Example Plot.Replace E OutData Property Applies To Description Syntax Remarks Example ANOVA.OutData = “” specifies a Null alias. table.Find="3" MYDATA. LinMix.OutText= “” specifies a Null alias. created when Run is invoked. PK. Desc Stats (Box and Whiskers plot) Sets the name of the text object. no output will be generated. PK.CaseSensitive=False MYDATA. no output will be generated. Desc. Plot. engine.SearchArea="CONC" MYDATA. PK Creates a graph object during the Run of an engine. Merge Sets the name of the output data object generated by the Run method of any engine.OutText = “MYTEXT” 609 . engine. if any.Tolerance="0" MYDATA. Bioeq. PK.OutGraph = “MyGraph” OutText Property Applies To Description Syntax Remarks Example ANOVA.With="Missing" MYDATA.OutText = string engine.Scripting Commands O Operation (continued)Property Example MYDATA. PK. engine.OutGraph = string engine.Operation="=" MYDATA. no output will be generated.OutGraph = “” specifies a Null alias.OutData = string engine.

See “Special file formats” on page 430 for details on these files.PageFooter = string A footer specified using this command will override any footers specified in a table definition file. Table. PageFooter. Accepts a quote delimited string up to 255 characters. DoseUnit. LinkFile reads in the PK model parameters and ParmFile reads the PD or IPR parameters. However. Model. optionally.ParmFile = “c:\parmdata.dat” See also Example 610 . Accepts a quote delimited string up to 255 characters.E WinNonlin User’s Guide P PageFooter Property Applies To Description Syntax Remarks See also Example Table Sets a footer for a scripted table.ParmFile = string When using compiled or ASCII PK models or User-ASCII models.PageFooter=“This is my Footnote” PageHeader Property Applies To Description Syntax Remarks See also Example Table Sets a title for a scripted table. if both are used in a single run. ParmFile reads in the initial values for model parameters. and. if both are used in a single run. This method should not be used for NCA models. LinkFile Model. DefinitionFile Table. lower/upper bounds. when using SCI PK/PD or PK/IPR models. Table. PageHeader.PageHeader=“This is my Title” ParmFile Method Applies To Description Syntax Remarks Model Loads an ASCII data file containing the model parameter and boundaries information.PageHeader = string A title specified using this command will override any titles specified in a table definition file. and optional lower/upper bounds. DefinitionFile Table.

PartFile = string See “Special file formats” on page 430 for the required format of these ASCII files.StartRow=1 MYDATA.EndRow=10 MYDATA.Paste Use with the Cut or Copy method.StartCol=A MYDATA.Percentiles = {True | False} The default is False. Desc. Objectname.PartFile = “c:\partdata.EndCol=J MYDATA.Scripting Commands P E PartFile Method Applies To Description Syntax Remarks See also Example Model Loads an ASCII file containing time ranges for the computation of partial areas in NCA.” percentile calculations will be included with descriptive statistics. Confidence.Paste StartCol. Model. EndCol. DoseUnit.Cut MYDATA.StartRow=5 MYDATA.Filename = <file name> MYDATA. MYDATA. Interval 611 . StartRow. EndRow Applicable Properties Percentiles Property Applies To Description Syntax Remarks See also Desc When Percentiles = “True.dat” Paste Method Applies To Description Syntax Remarks Example Data Pastes a range of data selected with the Cut or Copy property to a data set. LamzFile Model.StartCol=A MYDATA.

PrintData = {True | False} PrintAll. only the current data grid will be printed.Print If used on a multigrid data object. Text Prints the specified object.E WinNonlin User’s Guide Percentiles (continued) Property Example Desc. Model. PrintModel. Graph.PrintAll Print. PrintText WinNonlin.Percentiles = True Print Method Applies To Description Syntax Remarks See also Example Data. WinNonlin.PrintData=False PrintGraph Property Applies To 612 WinNonlin . PrintText WinNonlin. PrintModel. PrintGraph.PrintAll PrintData Property Applies To Description Syntax See also Example WinNonlin PrintData = False excludes Data windows from printing with PrintAll. PrintData.Print PrintAll Method Applies To Description Syntax See also Example WinNonlin Prints all open windows or data grids. objectname. the history will be printed rather than the data. PrintAll MyData. WinNonlin. including all worksheets of open workbooks. The default is True. PrintGraph. If the history of a data object is visible.

PrintModel. Chartobject. PrintMultiDown.PrintModel=False PrintMulti Property Applies To Description Syntax See also Chart If True.PrintGraph = {True | False} PrintAll. PrintText. The default is True.Scripting Commands P PrintGraph (continued)Property Description Syntax See also Example E PrintGraph = False excludes Graph windows from printing with PrintAll. PrintAll PrintMultiAcrossProperty Applies To Description Syntax See also Chart Sets the number of charts printed across the page.PrintGraph=False PrintModel Property Applies To Description Syntax See also Example WinNonlin PrintModel = False excludes the Model window from PrintAll. multiple charts are printed per page. PrintData.PrintModel = {True | False} PrintAll. PrintText. WinNonlin.PrintMulti = {True | False} PrintMultiAcross.PrintMultiAcross = number PrintMulti. PrintData WinNonlin. PrintMultiDown PrintMultiDown Property Applies To Chart 613 . Chartobject. WinNonlin. PrintGraph WinNonlin. The default is True.

the regular variables are the variables from Data Set 2. PrintMultiAcross PrintText Property Applies To Description Syntax See also Example WinNonlin PrintText = False excludes Text windows from printing with PrintAll. CrossVar( ).RegVar(1) = “Tmax” Table. PrintModel.PrintText=False Q R RegVar( ) Property Applies To Description Syntax Remarks See also Example Table Defines a ”regular” variable that composes the body of a table.RegVar(3) = “AUClast” RegVars Property Applies To 614 Table .RegVar(Variable no.PrintMultiDown = number PrintMulti. The default is True. RegVars sets the number of regular variables. For table 8 only one regular variable may be specified.PrintText = {True | False} PrintAll. PrintGraph WinNonlin. WinNonlin. Chartobject. IDVar( ). GroupVar( ) Table.) = string Applies to all table templates.RegVar(2) = “Cmax” Table. For tables 9 and 10. Table. PrintData.E WinNonlin User’s Guide PrintMultiDown (continued)Property Description Syntax See also Sets the number of charts printed down the page.

Table. Operation. For tables 9 and 10. 615 . the regular variables are the variables from Data Set 2. SearchArea.Find="3" MYDATA. Identify the first and last row in the range by row number. GroupVars Table. Identify the column by column name or letter. IDVars. Tolerance. Objectname.SearchArea="TIME" MYDATA. CrossVars.RegVars = 3 E Remove Method Applies To Description Syntax See also Example Data Removes rows fitting search criteria from a workbook. CaseSensitive MYDATA.RegVars = integer Applies to all table templates. CaseSensitive MYDATA.Remove RemoveCol Method Applies To Description Syntax See also Example Data Removes a specified column from a workbook.Scripting Commands R RegVars (continued)Property Description Syntax Remarks See also Example Defines the number of regular variables for a table.Column="a" MYDATA. objectname. For table 8 only one regular variable may be specified.RemoveCol Column.RemoveCol RemoveRow Method Applies To Description Data Removes a range of rows from a workbook. RegVar( ).Remove Find.Operation="=" MYDATA.

CaseSensitive=False MYDATA.RemoveRow MYDATA.Operation="=" MYDATA. Operation.RenameCol Replace Method Applies To Description Syntax See also Example Data Replaces data specified by the Find property with data specified by the With property. EndRow Applicable Properties RenameCol Method Applies To Description Syntax See also Example Data Renames a column. With.RenameCol Column.SearchArea="CONC" MYDATA. objectname.Replace 616 . Tolerance.Find="3" MYDATA.EndRow=18 MYDATA. SearchArea.With=”Subject” MYDATA. CaseSensitive MYDATA. objectname. Identify the column to be renamed by column name or letter. CaseSensitive MYDATA.E WinNonlin User’s Guide RemoveRow (continued)Method Syntax Example objectname.RemoveRow StartRow.With="Missing" MYDATA.Column=”a” MYDATA. With.StartRow=1 MYDATA.Replace Find.Tolerance="0" MYDATA.

creates a new document.RunWordExport 617 . Trans.WordExportFilename = “bg1. use WordExportFilename to have Word automatically save the output file. RunWordExport requires Word to be installed on the same machine as WinNonlin. LinMix.Run Be sure to name all possible output objects prior to using the Run method.doc” PK.Response=”Response” Run Method Applies To Description Syntax Remarks Example ANOVA. Bioeq. and populates it with the PK output. Plot. Run.RunWordExport Optionally. engine. table. PK. Desc. Merge Invokes the specified engine with its currently defined properties.pmo” MyModel.Filename = “bg1.Run RunWordExport Method Applies To Description Syntax Remarks See also Example PK Runs the PK analysis and launches Word (if not already running). Desc.Scripting Commands R E Response Property Applies To Description Syntax Example Toolbox Specifies column for response data.Load PK. WordExportFilename Set MyModel = Model MyModel.FileType = “PMO” MyModel. PK. Toolbox.Response=column name Toolbox.

Saving PKS: Saving a PKS workspace requires that the save operation be performed on the study workbook data object as a file of type PKS.SaveAll 618 . WinNonlin Saves a file to disk using the current file name. objectname. The PKS file contains all information necessary to save the workspace. Graph. DnErr( ) Plot.E WinNonlin User’s Guide S SameErrCol( ) Property Applies To Description Syntax Remarks See also Example Plot Determines whether the same column is to be used for both up and down error bars. Filename. Model. objectname. Saving Workspaces: Saving a workspace requires that all WinNonlin objects be saved first.Save See also Example SaveAll Method Applies To Description Syntax Example Graph Saves each graph in a multipart graph to a file. The default is False. if the FileType is anything other than PCO. Save is equivalent to export.SaveAll MYOUTGRAPH. UpErr( ).filetype="BMP" MYOUTGRAPH. Text. All multigrids must be saved as PWO's.SaveAllPrefix="graf" MYoutgraph.SameErrCol(1) = True Save Method Applies To Description Syntax Remarks Data. if one exists. Load MyGraph. FileType. then the column defined by UpErr( ) is used for the error data. Plot.Save When working with graph objects.SameErrCol(Data Set no. The Model. must be saved as a PMO.) = {True | False} The argument is required even if there is only one data source. File names use SaveAllPrefix plus a number. If True.

Scripting Commands S E SaveAllPrefix Property Applies To Description Syntax Remarks Example Graph Specifies file name prefix for saving multipart graphs.SaveAllPrefix = string This prefix must be between 1 to 4 characters in length.SaveTemplate = string LoadTemplate MyGraph. objectname.Operation="=" MYDATA.Replace 619 . MYDATA.SearchArea="CONC" MYDATA.WMF files MYOUTGRAPH.Tolerance="0" MYDATA.gtp” SearchArea Property Applies To Description Syntax Remarks Example Data Specifies where the Exclude.CaseSensitive=False MYDATA. Remove or Replace should focus.filetype="wmf" MYOUTGRAPH.With="Missing" MYDATA. rema Save all plots as separate . Use with SaveAll. Include.SaveAllPrefix="plot" MYOUTGRAPH.SearchArea = {column name | “entire data set”} It is not possible to search a “Current Selection” using the Scripting Language. as a graph template.Find="3" MYDATA. objectname. objectname.SaveTemplate “c:\winnonln\temps\xy.SaveAll SaveTemplate Method Applies To Description Syntax See also Example Graph Saves the current graph settings to a file.

Sequence = column name Toolbox. SortVars SortLevel Method Applies To 620 Graph . Plot.Smoothing=”Automatic” Sort Method Applies To Description Syntax See also Data Sorts a data set using the specified sortkeys. objectname.Sequence=”Sequence” ShowUnits Property Applies To Description Syntax See also Example Plot When ShowUnits = True. The default value is True.E WinNonlin User’s Guide Sequence Property Applies To Description Syntax Example Toolbox Specifies input column for sequence data.Sort SortVar( ). YUnits Plot. any units defined by XUnits or YUnits will appear after the respective axis title.ShowUnits={True|False} XUnits.ShowUnits=False Smoothing Property Applies To Syntax Example Toolbox Toolbox.Smoothing={Automatic|None|User} Toolbox. Toolbox.

Scripting Commands S SortLevel (continued)Method Description Syntax Remarks See also Example E Allows navigation among plots containing sort keys by going directly to a specified sort level. Desc Defines the sort keys to be used by the specified engine. For the Desc engine. MovePrevious.) = string For the Plot engine. SortLevel must contain a value for each key in order.SortLevel = string If there is more than one sort key. Must be set afterSortVars or SortVars( ). ). ) Property Applies To Description Syntax Remarks See also Plot. Merge Defines the number of sort keys to be used by the specified object or engine.SortLevel = “SubjA Cap” SortVar( ) Property Applies To Description Syntax See also Example Data.SortVar(2) = “Time” SortVar( . no. GroupVar( ) Data. Merge Defines the sort keys to be used.SortVar(Variable no. Desc. MoveLast MyGraph.SortVar(1) = “Form” Data. MoveFirst. GroupVar( ) SortVars Property Applies To Description Syntax Data. MoveNext. Must be set after SortVars. separated by a single space.SortVars = integer or objectname. Engine. SortVars( ).SortVars = integer 621 . objectname. Data. engine..SortVar(Data Set no. there must be a value even if there is only one sort key. Sort Var. both arguments are required even if there is only one data source and/or sort key. SortVars( ).) = string SortVar( .

SourceCol = string DestCol Trans.stacked=True 622 . SortVar( ). Therefore SortVars is used. False=separate Toolbox. even though there are two data sources. the number of SortVars must be the same for each.SortVars = 1 SortVars( ) Property Applies To Description Syntax Remarks See also Example Plot Defines the number of sort keys to be used by the Plot engine. SortVars( ). Trans. GroupVars( ) Desc.) = integer There must be a value even if there is only one data source.stacked={True|False} True=stacked. SortVars. GroupVars( ) Plot. SortVar( ). Plot.SortVars(1) = 2 SourceCol Property Applies To Description Syntax See also Example Trans Defines the source column for transformations.SortVars(Data Set no.SourceCol = “Conc” Stacked Property Applies To Description Syntax Remarks Example Toolbox Sets whether treatment data are in separate columns or stacked in one column in the input. and not SortVars( ).E WinNonlin User’s Guide SortVars (continued)Property Remarks See also Examples When working with the Merge engine. Toolbox.

If StartRow is –1 and EndRow is not set.Scripting Commands S E StartCol Property Applies To Description Syntax Remarks Data Specifies the first column in a range. Cut and Copy or Font and FontSize.Copy See also Examples = 1 = “Time” 15 “Time” = 1 = 1 15 3 StartRow Property Applies To Description Syntax Remarks Example Data Specifies the first row in a range for RemoveRow.Copy MyData.RemoveRow Stats Property Applies To Table 623 .StartCol MyData. EndCol MyData. unless StartRow = –1.StartRow MyData. Objectname.StartRow MyData. The latter is required to designate a new column. all rows will be selected.EndRow = MyData. Both are required. StartRow.StartCol = {string | number} Use with RemoveRow.EndCol = MyData. Set StartCol <= EndCol.EndCol = MyData. Cut and Copy. MYDATA.StartCol MyData.EndRow=18 MYDATA.StartRow = number Set StartRow <= EndRow. EndRow. StartCol can be a string identifying the column name (for columns already containing data) or a number identifying the column position.StartRow=1 MYDATA. objectname.EndRow = MyData.

Must follow SummaryVars. Table. 3.SummaryVars = 1 624 . 6.SummaryVar(Data Set no. Desc. 10.SummaryVars = integer SummaryVar( ) Desc. 9. 4. Table.E WinNonlin User’s Guide Stats (continued)Property Description Syntax Remarks Example Sets whether to generate descriptive statistics in the Table engine.SummaryVar(1) = “Conc” SummaryVars Property Applies To Description Syntax See also Example Desc Sets the number of summary variables for descriptive statistics. Toolbox.Subject = column name Toolbox. 7. 1. 8. Desc. Causes all of the available statistics to be printed.) = string SummaryVars Desc.Subject=”Subject” SummaryVar( ) Property Applies To Description Syntax See also Example Desc Specifies a variable to for which to calculate descriptive statistics.Stats = {True|False} Applies to table no.Stats = True Subject Property Applies To Description Syntax Example Toolbox Specifies input data set column for subject identifiers.

the . Toolbox.Tab = {Data | History} If MultiGrid is used on a data set that is currently showing the history.TableNum = integer Tables may be defined in a script file either by using scripting language commands.Tau=1 TerminalLower Property Applies To Description Syntax Example Toolbox Lower range for terminal phase.Scripting Commands T E T Tab Property Applies To Description Syntax Remarks Data Activates the Data or History tab of a data object. or using a table definition file.TAB value will automatically be set to Data–that is. the data will be toggled into view for that object.TableNum = 2 Tau Property Applies To Description Syntax Example Toolbox Specifies value for Tau. Toolbox. DefinitionFile Table. including TableNum.TerminalLower=1 625 . TableNum Property Applies To Description Syntax Remarks See also Example Table Identifies a table template number. Table.TerminalLower = number Toolbox.Tau = number Toolbox. objectname.

TileVertical TileHorizontal and Cascade WinNonlin.TerminalPhase=1 TerminalUpper Property Applies To Description Syntax Example Toolbox Upper range for terminal phase.TerminalUpper=5 TileHorizontal Property Applies To Description Syntax See also Example WinNonlin Arranges open objects to fill the WinNonlin window without overlapping.tilehorizontal TileVertical Property Applies To Description Syntax See also Example WinNonlin Arranges open objects to fill the WinNonlin window without overlapping. WinNonlin. Toolbox.TileHorizontal TileVertical and Cascade WinNonlin. WinNonlin.E WinNonlin User’s Guide TerminalPhase Property Applies To Description Syntax Remarks Example Toolbox Sets whether to include a user-defined terminal phase with Nonparametric Superposition.TerminalPhase = { 1 | 0 } 1=True. optimized for width.tilevertical 626 . Toolbox. optimized for height. If True. use with TerminalLower and TerminalUpper.TerminalUpper = number Toolbox. Toolbox. 0=False.

Toolbox.Title = “My Plot Title” TitleFont Property Applies To Description Syntax See also Example Graph Sets the font for the chart title.Title = string XTitle. Toolbox.TimeCol = column name Toolbox.TimeRepeat = number Toolbox. Plot. YTitle.TitleFont = string TitleFontSize MyGraph.TimeRepeat=5 Title ‘Property Applies To Description Syntax See also Example Plot Defines the string to be displayed in the title of the plot. ObjectName. Legend( ) Plot.Scripting Commands T E TimeCol Property Applies To Description Syntax Example Toolbox Specifies input data column for Time.TitleFont = “Arial” 627 .TimeCol=”Time” TimeRepeat Property Applies To Description Syntax Example Toolbox Sets repeat dosing interval to every number time units.

Toolbox. objectname.TreatmentA = ”Treatment1” TreatmentB Property Applies To Description Syntax 628 Toolbox Specifies input data set column to be used for second treatment.TreatmentA = column name Toolbox. Remove and Replace.With="Missing" MYDATA.CaseSensitive=False MYDATA. MYDATA.Tolerance = number Use with Exclude.Operation="=" MYDATA. ObjectName.SearchArea="CONC" MYDATA.TreatmentB = column name . Include.E WinNonlin User’s Guide TitleFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the chart title.Replace TreatmentA Property Applies To Description Syntax Example Toolbox Specifies input data set column to be used for first treatment. Toolbox.TitleFontSize=12 Tolerance Property Applies To Description Syntax Remarks Example Data Specifies tolerance range for numeric data searches.Tolerance="0" MYDATA.Find="3" MYDATA.TitleFontSize = number TitleFont MyGraph.

Used in conjunction with Column.DoseFile="DOSES.Scripting Commands U TreatmentB (continued)Property Example Toolbox. Objectname.Uniform = {True | False} MYOUTGRAPH. See also “Special file formats” on page 430.TreatmentB=”Treatment2” E U Uniform Property Applies To Description Syntax Example Graph Specifies that plots created with a sort variable should have uniform axes across sort levels.Unit = ”ng/ml” Units Property Applies To Description PK engine Assigns a units file to provide preferred parameter units for non-compartmental analysis.Filename="MODEL.Run Example 629 .Load MYMODEL.OutData=OUTDATA1 Pk.txt" Pk.Units="units.FileType="ASCII" MYMODEL.OutText=OUTTEXT1 Pk.Column = ”Concentration” MyData. MyData.LIB" MYMODEL.LamzFile="LAMBDAZ.DAT" Pk.DAT" MYMODEL.UNIFORM = True Unit Property Applies To Description Example Data Object Assigns a unit to a column. Set MYMODEL=Model MYMODEL. The units file can be created in a text editor or using the Save button on the Units tab of the Model Options dialog.

” then UpErr( ) is used for both up and down error.Weight = string If a weight variable is not specified.Weight = “Weight” 630 .) = string The argument is required even if there is only one data source. SameErrCol( ) Plot. uniform weighting will be used. DnErr( ).UpErr(Data Set no.Unload UpErr( ) Property Applies To Description Syntax Remarks See also Examples Plot Defines the column containing up error data for a plot with error bars. Desc. Plot.Uniform = {True | False} Unload should be applied to any object that will not be referenced any longer in a script.E WinNonlin User’s Guide Unload Method Applies To Description Syntax Remarks See also Example Plot Specifies that plots created with a sort variable should have uniform axes across levels. Load MyModel. Objectname.UpErr(1) = “SE” V W Weight Property Applies To Description Syntax Remarks Example Desc Sets the weight column for weighted descriptive statistics. If SameErrCol( ) = “True. Using Unload on the WinNonlin object causes the application to close. Desc.UpErr(1) = “Maximum” Plot.

With = string MYDATA.FileType = “PMO” MyModel. Word must be installed on the same machine as WNL. Model.WindowState = Minimized With Property Applies To Description Syntax Example Data Sets a replacement string for use with RenameCol and Replace. but not save.Column=”a” MYDATA.With=”Subject” MYDATA. If WordExportFilename is not set. Text. Objectname.Load PK.WindowState = {Minimized | Maximized | Normal} WinNonlin.WordExportFilename = string Use only with RunWordExport. WinNonlin Sets the window state of the specified object.pmo” MyModel.doc” PK.RenameCol WordExportFilename Property Applies To Description Syntax Remarks Example PK Sets the file name for the Word export method. RunWordExport will create. Set MyModel = Model MyModel.Scripting Commands W E WindowState Method Applies To Description Syntax Example Data.RunWordExport WorkingDir Property Applies To WinNonlin 631 . Using this property on the WinNonlin object allows a script to run with the application minimized.WordExportFilename = “bg1. the output document. objectname.Filename = “bg1. Graph. PK.

ObjectName.” as it removes the need to specify absolute paths for data files in the script—i.e. WorkingDir = string This is particularly useful for making script files “portable.XLabelsFont = string XLabelsFontSize MyGraph. file name alone will suffice.XTitle = string Title. Plot. if the files are located in the working directory. ObjectName.XLabelsFont = “Arial” XLabelsFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the X axis labels.XLabelsFontSize = 12 XTitle Property Applies To Description Syntax See also Plot Defines the string to be displayed in the x-axis title. WorkingDir = “C:\Newdir” Example X XLabelsFont Property Applies To Description Syntax See also Example Graph Sets the font for the X axis labels.E WinNonlin User’s Guide WorkingDir (continued)Property Description Syntax Remarks Sets the directory in which WinNonlin will load and save files..XLabelsFontSize = number XLabelsFont MyGraph. YTitle. Legend( ) 632 .

ObjectName.XTitleFont = “Arial” XTitleFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the X axis title.XTitleFontSize = 12 XUnits Property Applies To Description Syntax See also Example Graph Defines the units to be displayed in the x-axis title (in parentheses after the title).Xunits = string YUnits MyGraph. ObjectName.XTitle = “Time” E XTitleFont Property Applies To Description Syntax See also Example Graph Sets the font for the X axis title.XTitleFontSize = number XTitleFont MyGraph.XUnits = ‘hr” XUnits Property Applies To Description Plot Defines the units to be displayed in the x-axis title (in parentheses after the title).XTitleFont = string XTitleFontSize MyGraph.Scripting Commands X XTitle (continued) Property Example Plot. 633 . ObjectName.

) = string The argument is required even if there is only one data source. ObjectName. Plot. ObjectName.XVar(1) = “Time” Y YLabelsFont Property Applies To Description Syntax See also Example Graph Sets the font for the Y axis labels. Plot. YUnits Plot.YLabelsFontSize = number YLabelsFont MyGraph.E WinNonlin User’s Guide XUnits (continued) Property Syntax See also Example Plot.YLabelsFont = “Arial” YLabelsFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the Y axis labels.YLabelsFontSize = 12 634 .YLabelsFont = string YLabelsFontSize MyGraph.XVar(Data Set no.XUnits=string ShowUnits.XUnits = “hr” XVar( ) Property Applies To Description Syntax Remarks Example Plot Defines the X variable column for a plot.

YTitleFontSize = 12 YUnits Property Applies To Description Syntax See also Example Graph Defines the units to be displayed in the y-axis title of the plot (in parenthesis after the title). ObjectName. ObjectName. ObjectName.YUnits = string XUnits MyGraph. Legend( ) Plot.YTitleFontSize = number YTitleFont MyGraph.Scripting Commands Y E YTitle Property Applies To Description Syntax See also Example Plot Defines the string to be displayed in the y-axis title. XTitle.YTitleFont = “Arial” YTitleFontSize Property Applies To Description Syntax See also Example Graph Sets the font size for the Y axis title.YTitleFont = string YTitleFontSize MyGraph.YTitle = “Concentration” YTitleFont Property Applies To Description Syntax See also Example Graph Sets the font for the Y axis title.YTitle = string Title. Plot.YUnits = “mg/ml” 635 .

Plot.YVar(1) = “Concentration” 636 . Plot. Plot.YUnits = “mg/ml” YVar Property Applies To Description Syntax Remarks Example Plot Defines the Y variable column for a plot.YUnits = string ShowUnits.E WinNonlin User’s Guide YUnits Property Applies To Description Syntax See also Example Plot Defines the units to be displayed in the y-axis title of the plot (in parenthesis after the title).YVar(Data Set no. XUnits Plot.) = string The argument is required even if there is only one data source.

C = A+ *B).Appendix F Warnings and Error Messages Messages by category There are six sources of error messages in WinNonlin. Compartmental modeling See also “Troubleshooting modeling problems” on page 209. Program terminated. The preceding statement contains an extra left parenthesis. as follows: Source Microsoft Visual Basic & OCX'sa Compartmental modeling LinMix and Bioequivalence wizards Table Wizard Descriptive statistics Noncompartmental analysis IVIVC Scripting language Message numbers 1-9999 10000-10999 11000-11999 12000-12999 13000-13999 14000-14999 15000-15999 19000-19999 a. 637 . See Microsoft’s documentation for information on error messages from Visual Basic and the Windows OCX's.g. Program terminated. Each has a unique set of error message numbers. The preceding statement contains a syntax error (e. 10002 Unmatched parenthesis. 10001 Invalid statement.

STARTING. 10005 Invalid number. Program terminated. The model was too complex for the default memory settings. The maximum number of variables and operators has been exceeded (2000). If it occurs. Program terminated. Rerun with larger model size (SIZE command). Function numbers must be integers. DIFFERENTIAL. 638 .e.g. This is an unusual error. &) 10010 Model too complex. COMMANDS. /. or too many arguments were encountered in a subscribed variable or function (e. please bring it to the attention of Pharsight customer support. -.B)). The previous equation contains an invalid character. 0-9. 10012 Invalid function number. There was an unmatched or missing right parenthesis. Rerun with larger model size (SIZE command). greater or equal to 1. Workspace is not large enough to accommodate the model. SECONDARY. Program terminated. in the Model Options dialog box). 10004 Missing END statement. **. not A-Z.1) or ABS (A. (i.. assign expressions to temporary variables (in the TEMP section) before using them to define the model. +. TEMPORARY. 10006 Model too complex. 10011 Unmatched parenthesis. or the Model size option on the PK Settings tab. An invalid FUNCTION number was assigned. FNUMBER. The previous equation contains an illegal number (constant). Program terminated. *. An “END” statement was missing for one or more of the following labeled groups of commands: TRANSFORM.F WinNonlin User’s Guide 10003 Improper statement found when evaluating model. Program terminated. 10007 Invalid variable name or character. Program terminated. Program terminated. DZ(1. Try the model again after increasing the model size (via the SIZE command. Program terminated. (). To help alleviate this problem. and less than or equal to the number of functions in the NFUNCTIONS command.

with a GOTO) which was never defined. 10020 Undefined (GOTO) label. 10023 No matching IF.B)) 10015 Model section does not exist. Split it up into two or more parts using temporary variables. Program terminated. An ENDIF statement was encountered without a matching IF statement. but the secondary parameters were never defined. for an ELSE statement.. or if the FUNCTION block(s) is missing. THEN. 10022 Model too complex. 10021 Invalid (GOTO) label name. For example. an NSEC command was specified. 639 .g. Program terminated. 10017 Equation too complicated.g. Program terminated. 10019 Missing ENDIF statement. 10014 Missing argument in function or array.g. 10018 No IF corresponding to ENDIF. STRT in place of START). An argument was missing for a function (e. Change the label to a valid name. Program terminated.Warnings and Error Messages Compartmental modeling 10013 F Invalid group identifier. Check to see if NSEC > 0 but no secondary parameters are defined. or is missing the ending colon “:”. A label was referenced (e.. Program terminated. At least one of the equations was too complicated to evaluate. A labeled section does not exist. Program terminated. Program terminated. Rerun with larger model size (SIZE command). The label name is invalid. An ENDIF statement is missing. Program terminated. The maximum number of temporary variables and labels (100) has been exceeded. Program terminated. An invalid label was assigned to a group of commands (e. MAX(A) instead of MAX(A.. Program terminated. Either the label is missing entirely.

DTA (11)).g. Program terminated.F WinNonlin User’s Guide An ELSE statement was encountered without corresponding IF . etc. 10027 Invalid array subscript. Change the label to a new name. 10028 Invalid statement encountered when evaluating model. An invalid comparison occurred during execution. or 0 or a negative number. The argument for log10(x) was invalid. Program terminated. The argument for loge(x) was invalid. 10025 Cannot branch out of program section. Attempt to divide by zero. An attempt was made to branch from one block of statements (TEMP. Invalid argument for LOGE(X). Program terminated. 10024 IF statements nested too deeply. IF statements can only be nested as deep as 10 levels.THEN statements. 10026 Duplicate label name. Program terminated. 10034 Invalid argument for LOG10(X). FUNC. Program terminated. The model has gone outside allowable limits when accessing some variable which is an array (e. Check to see that the model is specified correctly. Perhaps X was undefined. Program terminated. Program terminated. Check to see if subscript exceeded array bound. 640 . Program terminated. 10030 10031 10033 Operation attempted using A missing VALUE. Program terminated. An invalid branching occurred during execution. or 0 or a negative number.) to a labeled section in another block of statements using a GOTO. 10029 Invalid statement encountered when evaluating model. A duplicate label name was encountered. Perhaps X was undefined. An invalid array access was attempted. Check to see that the model is specified correctly. This is not supported. Program terminated.

Program terminated.Y). Program terminated.Warnings and Error Messages Compartmental modeling 10035 F Invalid argument for SQRT(X).com. NEXT statement missing corresponding DO. Perhaps X was undefined or a negative number. Check the arguments and syntax of any DO statements. Perhaps either X or Y was undefined or a negative number. Program terminated. 10050 DO statement missing a NEXT statement. 10037 Problem exceeds available resources. or perhaps only one argument was given. Program terminated. No input record may exceed 256 characters in length. Maximum number of DO statements is 999. Notify Pharsight technical support for assistance at support@pharsight. 10051 Too many nested DO statements. Program terminated. 10054 10055 10056 Maximum length of DO statement is 788 characters. 10058 Invalid lag function argument. Check all DO statements for corresponding NEXT statements. A NEXT statement ends a DO loop. 641 . Program terminated. 10057 String length exceeds 256. Program terminated.y) was invalid. The argument for lag(x) was invalid. Check all NEXT statements for corresponding DO statements. Program terminated. X must be the column number of a defined variable. Program terminated. DO statements may only be nested 10 deep. Please check the model specification for problems. The argument for ATAN2(x. Program terminated. 10036 Invalid argument for ATAN2(X. The argument for SQRT(x) was invalid. 10053 Maximum model size is 64. Program terminated. A NEXT statement must end each DO loop. Program terminated. 10052 Invalid DO statement.

Either fewer than NPARM initial estimates were given. must precede DATA. The program was unable to read NPARM numbers on the INITIAL command. 10106 The number of parameter values. LOWER and UPPER commands. The number of observations to be input (NOBS). NOBS. Turning off of convergence checks is only supported for integrated equations. Using default units. No NPARAMETERS command was encountered. LOWER. was not given. were not given.2 . The preferred units were inconsistent with the default units or were not acceptable units. 10107 Error converting units for next value. 1. must be specified prior to INIT. 10110 Error encountered when reading the number of NOBS for each function. NPARM. Missing DATA command. OR UPPER NPARAMETERS must precede INITIAL.F WinNonlin User’s Guide 10101 The number of data observations. NPARM. 10111 The number of functions must be greater than 0. The number of data observations to be input (NOBS) was not defined. NOBS. was not defined.E-3). The program was unable to read NFUN numbers on the NOBS command. 10102 A DATA command was not encountered. 10105 CONV = 0 is only supported for integrated functions (NDER=0). 10103 The number of parameters to be estimated. The specified number of functions must be between 1 and 10. 642 . INIT. 10109 Error encountered when reading initial parameter values. 10104 The initial estimates of the parameters.g. must be specified prior to the DATA command. No INITIAL command was encountered. or one of the numbers was not carefully defined (e. 10108 The number of data observations.

execution continuing. execution continuing. Program defaulting to 100. 10115 The maximum number of iterations must be greater than or equal to zero. Method 3: Hartley's Modification of the Gauss-Newton Algorithm. 2 or 3. PRogram defaulting to 50. the number of differential equations in the system must be between 1 and 10. Program terminated. Program defaulting to 50. An invalid value was entered. If the system includes differential equations. Execution continuing.Warnings and Error Messages Compartmental modeling 10112 F The number of parameters to be estimated must be greater than 0. The specified METHOD must be 1. 643 . execution continuing. Method 1: Nelder-Mead. The specified number of parameters must be between 1 and 10. An invalid value was assigned to the CONVERGENCE command. The column numbers associated with X and Y (XNUMBER. 10119 The number of columns in the plots must be between 10 AND 100. YNUMBER) must be between 1 and 10.E-4. 10113 The number of derivatives must be greater than 0. 10118 The number of rows in the plots must be between 10 and 50. 2 or 3. 10116 The number of variables in the input file must be greater than 0. 10120 The specified convergence criterion must be greater than or equal to 0. A maximum of 10 variables (columns) can be read in from a data file. Program defaulting to method 2. Program defaulting to 1. Method 2: Hartley's and Levenberg's modification of the Gauss-Newton Algorithm (default). An invalid value was entered. 10117 The column numbers corresponding to X and Y must be greater than 0 and must be less than or equal to NVAR. Program terminated. execution continuing. 10114 The specified method for estimating the residual sum of squares must be 1. An invalid value was assigned to maximum number of ITERATIONS.

g.0 and 0. This can usually be attributed to an improperly defined model. 1. If LOWER limits are specified. Program terminated.F WinNonlin User’s Guide 10121 The increment for estimating partial derivatives must be between 0 AND 0.E-3). 10122 The variable names for X and Y cannot be equal. 1. Program defaulting to 0.2 . 10128 An improper call was made to the numerical integration routine. then UPPER limits must also be specified (and vice-versa). Check your input and model.E-3). If weights are to be read in from the data file.2 . or the number was not correctly defined (e.E-3). Either fewer than NPARAMETERS numbers were on the card or one or more invalid numbers were specified (e. execution continuing. 10123 The variable number corresponding to weight must be greater than 0. 1. The increment used for estimating the partial derivatives DINC must be between 0.1. 10126 Error encountered when reading in the upper or lower parameter limits.g. Fewer than NOBS observations were on the data file. An error was encountered when trying to read the (first) number from the preceding command. An error was encountered when reading the lower (or upper) parameter limits.2 . Execution continuing without limits on the parameter space. One or more illegal numbers (e. Execution continuing without limits on the parameter space. There are several ways this can happen: a. Check to see that the model was correctly specified. Fewer than NVAR numbers were on one or more rows on the data file.g. 10124 Error encountered when trying to read a number on the preceding command. 644 . 10127 If lower (or upper) limits are specified then upper (or lower) limits must also be specified. b.1. 10125 Error on data file An error was encountered when reading the data file. c. Either the program was expecting a number to be specified and none was given. An error was encountered when trying to numerically integrate a system of differential equations. they must have an assigned column number greater than one. X and Y cannot be assigned to the same column in the input data file.001.

If CONSTANTS are used. An error was encountered when trying to read the NCON values assigned to constants. 10130 The program is having to work very hard to obtain the desired accuracy. 10134 10135 Parameter estimate for <parameterName> is at or near boundary. Check the model for possible singularities. Program terminated. Check your model for possible singularities. the number of user input constants must be between 1 and 500. Check to see that the model was correctly specified.Warnings and Error Messages Compartmental modeling 10129 F Too much relative accuracy was requested for this problem. MODEL 7). Either fewer than NCON values were specified. 10138 A convergence failure occurred during a singular value decomposition. 10131 The number of constants must be greater than 0. 10132 The number of constants must be specified before assigning them values. Program terminated. An initial parameter estimate is outside of the specified upper and lower limits. Each MODEL statement in a WinNonlin library must also have a corresponding model number (e. 645 . 10133 An error was encountered when reading the values assigned to the constants. Parameter limits ignored .2 . Execution continuing.g. or one of the values was not correctly defined (e. 1.E-3). One or more of the initial parameter estimates is outside the specified upper and lower limits. The NCON command must appear prior to the CONSTANTS command. 10136 10137 An error was encountered when reading constant names. The referenced library contains a model statement without a model number. Make sure your model is correct.Execution continuing.g. Program terminated. The program is having difficulty obtaining the necessary precision when integrating a system of differential equations. Execution continuing A very large number of function evaluations are having to be made to obtain the desired precision. Relative error and-or absolute error tolerances were reset. Check the model for possible singularities.

The specified TITLE number must be 1. 10147 Scale values could not be computed for the next plot. Program continuing with default names. 10139 NCON was specified but the constants were not input. 10144 10145 The specified model number does not exist in the specified library. data or unit names. Make sure that the number of input names matches NPARAMETERS (or NSECONDARY) and that each name is enclosed in single quotes. NPOINTS reset to default. An error was encountered reading the parameter or secondary parameter names. 4. If secondary parameters are specified. 10142 NSEC must be greater than 0. 646 . OR 5. The NSECONDARY command must precede the SNAMES command.F WinNonlin User’s Guide Check to make sure that the model has been correctly defined. secondary.g. for example. 10143 An error was encountered when reading parameter. or 5 (e. Plot deleted. 10148 The number of constants (NCON) is inconsistent with the number of administered doses. 4. 10140 NSEC must be specified prior to the SNAME command. The specified number of points to be included in the output plot file must be between 10 and 20000. 10146 The specified number of plot points must be between 10 and 20000. TITLE 3). 2. The specified title number must be 1. An NCONSTANTS command was encountered but the constants were never defined (CONSTANTS). This could happen. if all values to be plotted were identical. 2. Axis scale values could not be computed for a plot. the number of secondary parameters must be between 1 and 50. 3. The number of constants (NCONSTANTS) is not consistent with the number of doses given. 3.

Warnings and Error Messages Compartmental modeling 10149 F The WEIGHT command cannot be used when performing simulations. Check for proper specification of the model file. 10154 The specified library file does not exist. Program terminated. Notify Pharsight technical support for assistance at support@pharsight. Program terminated. it is inappropriate to use the WEIGHT command. REWEIGHT will be used.com. 10151 The algorithm is unable to converge. Program continuing with default estimation of time range. The Gauss-Newton algorithm was unable to converge. Program continuing with default missing value code. The variance-covariance matrix for the model parameters is singular. since no observed data exist. 10152 10153 The residual sum of squares is zero. When performing simulations. FNUMBER must be the number of a column on the input data set. These are the values for selecting the Lambda Z time interval. An error occurred when reading the parameter values for the PK model. 10161 Inadequate system resources available for this problem. Program terminated. Check to make sure NVAR is correct. 10156 An error was encountered when reading the missing value code. The WEIGHT command was changed to REWEIGHT. Try using different initial estimates. 10155 An error was encountered when reading beta time range. or try rerunning the problem using the Nelder-Mead algorithm. and surrounded by quotes. 10150 The parameter variance-covariance matrix is singular. Convergence assumed. Weight set to missing. One or more weights became negative. 647 . Make sure that the missing value code is eight characters or less in length. 10157 The value of FNUM is invalid. Check model definition statements to make sure WT is defined correctly. Program terminated. 10162 Error encountered when reading link model PK constant values.

com. IV dose <= zero. Use bolus model if start time equals end time. Error degrees of freedom less than or equal to zero. 10172 10173 10174 10175 10201 648 . Please modify them and rerun your model. PUNIT and SUNIT values cannot contain embedded blanks. The function variable specified by FNUMBER will be used to determine which observations belong to which functions. It is recommended to use a function variable for data with more than one function. and 17 do not support bolus dose <= zero. Check to see if a compiled model was specified by user. Internal parsing file error. or infusion length <= zero. try adding bounds so that the program can grid search for initial estimates. Infusion model should not be used if end time is less than or equal to start time. Initial estimates cannot be determined for this model. SNAME. 10170 Improper use of END statement. 10171 The lower and upper bounds were equal for one or more model parameters. See also “Model structure” on page 220. See “FNUMBER” on page 562. The number of observations must be greater than the number of parameters. 10166 10167 Command encountered when reading model statements. If no bounds were used. 10164 The NOBS values are inconsistent with the number of records in the data file. Models 15.F WinNonlin User’s Guide 10163 NOBS command ignored since FNUM specified. Use END statements only in user models. See “User Models” on page 217. DNAME. Initial estimates cannot be found for this model. Check whether the model is missing an EOM statement or any commands in the model are not inside a block. 16. Contact Pharsight Technical Support for assistance: support@pharsight. Please report to technical support. The sum of the NOBSERVATIONS values is not equal to the number of observations on the data set. Error variance and standard errors or parameters cannot be estimated. 10165 The PNAME. CNAME. Model statements were contained in the command file.

Automatic estimation of initial estimates is available for single dose data only. save your work in these applications. Initial estimates could not be found for the model.Warnings and Error Messages LinMix and Bioequivalence wizards 10202 F Curve stripping is only done for single dose data. A grid search will be used to obtain initial estimates. 10206 The initial estimates are outside of the lower or upper parameter limits. Initial estimates could not be determined for this model via curve stripping. 11051 11060 Internal error during calculations. try adding bounds so that the program can grid search for initial estimates. close them and try again. If no bounds were used. The initial parameter estimates are outside of the LOWER and UPPER limits. Insufficient memory. If other applications are running. Internal parsing error. 10205 Due to failure of the curve stripping method. Closing other WinNonlin windows would also help. Try fitting a less complex model. less compartments. 10998 Execution error. 10203 An error occurred during curve stripping. a grid search will be performed. Corrupted data set. 649 . 10204 The requested number of exponentials could not be fit. 11001 Memory allocation error. Note that this may occur when extracting PWO’s directly from zipped files.com. LinMix and Bioequivalence wizards When errors occur using the LinMix or Bioequivalence wizards a message box will appear with the error number and a description of the error. Execution continuing without limits on parameter space. Contact Pharsight Technical Support. Initial estimates for the number of exponential terms in the specified model could not be determined. Contact Pharsight Technical Support. 11002-11032 11050. Contact Pharsight technical support at support@pharsight. The files should be extracted first. Please try fitting a simpler model.

Information matrix is deemed singular. Model may be over-specified. Newton's algorithm converged with modified Hessian.. Seq+Subj(Seq)+Formulation+Period. Character variables in models must be class variables. Starting estimates supplied by user are invalid. Try the Satterthwaite option. Too many levels of a classification variable. Failed to converge in allocated number of iterations.F WinNonlin User’s Guide 11061 11062 11063 11064 11065 11066 11067 11068 11069 11070 11071 11075 11089 11090 Input data not rectangular. The residual variance is not modeled. There is no residual variance. Dependent variable must be numeric. Negative final variance component. Try a different model. Asymptotic covariance matrix not computed. Error in Variance Matrix. There are no residual degrees of freedom. Output is suspect. Output is suspect. Using method of moments. i. . Try random instead of repeated. Weight variable must be numeric. 11091 11092 11093 11094 11095 650 Negative final variance parameter. Consider omitting this VC structure. If this occurs when using the default bioequivalence model with Subj(Seq) as a random effect. Try using Residual DF option.e. No statistical tests are possible. it most likely indicates that the within-subject variance (residual) is greater than the between-subject variance. Consider FA0 structure instead of UN. There are no degrees of freedom for residual. Error in Satterthwaite DF. Error opening file. Bioequivalence statistics could not be computed for some test formulations. A more appropriate model would be to move Subj(Seq) out of the random model and into the fixed model.

Missing data for dependent variable causing an incomplete design. Program terminating. Fewer than 2 sequences had complete design. The column ColName contains alphanumeric value(s) that are greater than maxsize in length. Fewer than two periods. Ln-transform requested for data that is zero or negative causing an incomplete design. Program termininating. More than 10 sequences not allowed for population/individual bioequivalence.Warnings and Error Messages LinMix and Bioequivalence wizards 11096 F Negative final block term for CS. Do you want to continue with shortened values? Alpha variables are limited to a length of 128 characters. More than 10 periods not allowed for population/individual bioequivalence. We suggest creating a new variable with re-coded values that are less than or equal to this maximum length. More than 2 treatment formulations not allowed for population/individual bioequivalence. Negative final variance parameter for AR structure. More than 200 subjects not allowed for population/individual bioequivalence. 11097 11098 11099 11101 11102 11103 11104 11105 11106 11107 11108 11109 11121 11122 11201 11206 There is an error on the following tab(s): tab name(s). Negative final standard deviation for ARH/CSH structure. Fewer than two treatment formulations. Program termininating. Possibly the model is missing. This error occurs when there is an error on one of the tabs. Sequence < > had less than 2 complete subjects and was discarded. or the dependent variable is missing. Negative final variance parameter for TOEP structure. 651 . Subject <name> had incomplete design and was discarded. Consider omitting random model for this structure.

close them and try again. 12999 Memory allocation error Insufficient memory. Unable to find any DATA command in command filename. save your work in these applications. 11211 File data filename not found This error occurs when the data file referenced by the data statement in the command file is not found. Maximum variable name length is 256 characters.F WinNonlin User’s Guide 11207 11208 Failed to create model as specified in the model file. There are greater than 256 variables in your data set. 11212 11214 Second delimiter in filename for DATA statement not found. If other applications are running. 13000 13999 652 Floating Point error Memory allocation error . This error occurs when the user loads a model command file from the tool bar and this file does not contain any valid DATA statement. 11215 Table Wizard When errors occur with the Table Wizard a message box will appear with the error number and a description of the error. The LinMix module has a limit of using no more than 256 variables. Variable name too long for LinMix. Descriptive statistics When errors occur in the Descriptive Statistics module a message box will appear with the error number and a description of the error. variable will be temporarily truncated for use by LinMix. Please decrease the number of variables on the data set then rerun the LinMix module. 11210 Unable to find file command filename This error occurs when the specified command file does not exist. Closing other WinNonlin windows would also help.

using default units. save work. Internal parsing error.Warnings and Error Messages Noncompartmental analysis F Insufficient memory. Use Descriptive Statistics to compute mean data and perform NCA on the means. 653 . Incompatible units for summary table Amount values. Contact Pharsight Technical Support. or negative volume or concentration. Closing other WinNonlin windows may also help. No grid output for this subject. Incompatible units for summary table AUMC values. If other applications are running. Contact Pharsight Technical Support. Error opening file. Closing other WinNonlin windows would also help. and it was at dosing time. Warnings 14002 14021 14022 14024 14050 14061 14062 14064 14066 14501 14502 14503 Incompatible units for partial AUCs and summary AUCs. using default units. Insufficient random access memory. Invalid urine data: lower time >= upper time. No data in steady state dosing interval. Noncompartmental analysis Error messages 14001 Memory allocation error. close the applications and try again. Only 1 non-zero concentration value found. close them and try again. All volume values are zero. program terminating. using default units. Duplicate time values on data file. If other applications are running. All concentration values are zero. Data error: same subject cannot have multiple samples at same time. save your work in these applications. or select Sparse Sampling in NCA.

Steady state PK parameters could not be estimated. All concentration values are zero. AUC_TAU could not be estimated. All final parameters are set to missing. No response at dosing time. Adjusting for infusion has caused negative MRTinf(pred) value. using default units. close the applications and try again. No grid output for this subject. You may also try using a constant infusion model. If other applications are running. 14511 14512 14513 14514 14515 14520 14521 14522 14524 14523 14530 14531 14532 14533 14534 IVIVC 15001 Memory allocation error. Threshold cannot equal baseline. No parameters could be extrapolated to infinity. All final parameters are set to missing. Only one non-missing observation. 654 . The value at dosing time could not be determined by log-linear regression of the first two measurements. Insufficient random access memory. threshold will not be used. Baseline is set to 0. MRT parameters are adjusted for length of infusion. Adjusting for infusion has caused negative MRTlast value. Only 1 non-zero concentration value found. Closing other WinNonlin windows may also help. save work. Lambda_z could not be estimated. Less than two non-missing observations. and it was at dosing time. and was set to the first observed measurement.F WinNonlin User’s Guide 14504 Incompatible units for final parameter <default parameter name>. All volume values are zero. Therapeutic window calculations cannot be done for negative data. Adjusting for infusion has caused negative MRTinf(obs) value.

Number of non-missing observations for a function is zero. Cannot produce Wagner-Nelson table. save your work in these applications.) Memory allocation error. Duplicate time values on data file. Internal parsing error. Contact Pharsight Technical Support. LambdaZ could not be estimated by WNL.com. (Loo-Riegelman only. Cannot continue with Loo-Riegelman. Use Descriptive Statistics to compute mean data to perform analysis on the means. If other applications are running. 15010 15011 15101 15102 15103 15104 15105 Scripting language Errors that are generated when using the Scripting Language are written to the data log file. 19999 Memory allocation error Insufficient memory. Error opening one or more input files. Closing other WinNonlin windows would also help. Cannot have equal time values in input dataset. Insufficient random access memory. If no errors exist. Contact Pharsight Technical Support. close them and try again. Try specifying AUCINF value. this log file will not be created. Internal convolution error. Contact Pharsight Technical Support. Closing other WinNonlin windows may also help.Warnings and Error Messages Scripting language 15002 15003 15004 F Error opening file. Please consult this log for guidance in resolving problems with script files. Contact Pharsight technical support at support@pharsight. 655 . If other applications are running. General 20000 and above Internal error. close the applications and try again. Contact Pharsight Technical Support. save work. (only for Wagner-Nelson) LambdaZ could not be estimated by WNL.

F WinNonlin User’s Guide 656 .

Diletti. Hauschke. The bioavailability of erythromycin sterate versus enteric-coated erythromycin base taken immediately before and after food. • • • • • • “Bioequivalence” on page 657 “In vitro-in vivo correlation and formulation science” on page 658 “Pharmacokinetics” on page 659 “Regression and modeling” on page 660 “Sparse data analysis methods” on page 662 “Statistics” on page 663 Bioequivalence Anderson and Hauck (1983). Elze and Blume (1994). 2nd ed. Hsuan and Holder (2000). Commun Stat Theory Methods. Statist Medicine 19:2885-97. Int J Clin Pharm Ther 32(7): 376-378. Clayton and Leslie (1981). 657 . Presentation of the intrasubject coefficient of variation for sample size planning in bioequivalence studies. Steinijans. Schall. Int Med Res 9:470-7. Marcel Dekker. A small sample on confidence interval approach to assess individual bioequivalence.Appendix G References Further reading on modeling and related analyses See the following headings for an alphabetized listing of related references. Design and Analysis of Bioavailablilty and Bioequivalence Studies. 12:2663-92. A new procedure for testing equivalence in comparative bioavailability and other clinical trials. Inc. Hyslop. Chow and Liu (2000). Luus.

New York. J Pharcokinet Biopharm 15:657-680. Godfrey. Comparison of six deconvolution techniques. 3rd ed. J Pharmacokinet Biopharm 6(3):227-41. Cutler (1978). Population and Individual Approaches to Establishing Bioequivalence. AAPS Press. In Amidon GL. Gibaldi and Perrier (1975). Statistical Approaches to Establishing Bioequivalence. A comparison of the two one-sided tests procedure and the power approach for assessing the equivalence of average bioavailability. Robinson JR and Williams RL. Average. Pharmacokinetics. Gabrielsson and Weiner (2001). Modeling Strategies for In Vivo-In Vitro Correlation. New method for calculating the intrinsic absorption rate of drugs. US FDA Guidance for Industry (August 1999). 658 . 21. J Pharmacokinet Biopharm 15. Marcel Dekker. US FDA Guidance for Industry (October 2000). Hovroka R and Bates RA (1996). Programs 21. Daubechies (1988). Iman and Conover (1979). Loo and Riegelman (1968). Cutler (1978). Communications on Pure and Applied Mathematics 41(XLI):909-96. J Pharmaceut Sci 57:918. VA. Chappell MJ. Technometrics. Swedish Pharmaceutical Press. Gillespie (1997). US FDA Guidance for Industry (January 2001). J Pharmacokinet Biopharm 6(3):243-63.. Alexandria.499-509. 645-55. Numerical deconvolution by least squares: use of prescribed input functions. Orthonormal bases of compactly supported wavelets. 98-118. eds. Scientific Foundation for Regulating Drug Product Quality. J Pharmacokinet Biopharm 24:282. Inc. Math. Madden. In vitro-in vivo correlation and formulation science Charter and Gull (1987). Numerical deconvolution by least squares: use of polynomials to represent the input function.G WinNonlin User’s Guide Schuirmann (1987). Haskel and Hanson (1981). Stockholm. Bioavailability and Bioequivalence Studies for Orally Administered Drug Products—General Considerations. Pharmacokinetic and Pharmacodynamic Data Analysis: Concepts and Applications.

Kinetic Data Analysis: Design and Analysis of Enzyme and Pharmacokinetic Experiments. Alexandria. In Amidon GL. Swedish Pharmaceutical Press. J Pharm Sci 67:1687-91. A BASIC computer program for obtaining initial polyexponential parameter estimates. Cheng and Jusko (1991). J Pharm Bioph 10(5):551-8. Plenum Press. Holford and Sheiner (1982). Estimation of statistical moments and steady-state volume of distribution for a drug given by intravenous infusion. AAPS Press. Noncompartmental determination of mean residence time and steady-state volume of distribution during multiple dosing. 659 . Pharmacol Ther 16:143. Stockholm. (1981). Linear inverse--theory and deconvolution. Chan and Gilbaldi (1982). Robinson JR and Williams RL. Scientific Foundation for Regulating Drug Product Quality. Verotta (1990). ESTRIP.. J Pharmacokin Biopharm 21:457. Dayneka. 2nd ed. Comments on two recent deconvolution methods. IVIVC Examples. Gouyette (1983). Treitel and Lines (1982). J Pharm Sci 80:202. New York. New York. Jusko (1990). Kinetics of pharmacological response.References Pharmacokinetics G Meyer (1997). Garg and Jusko (1993). Scientific Foundation for Regulating Drug Product Quality. AAPS Press. J Clin Pharmacol 30:303. Koup (1981). Pharmacokinetics. ed. Alexandria. eds. Pharmacokinetic and Pharmacodynamic Data Analysis: Concepts and Applications. Geophysics 47(8):11539. Endrenyi. J Pharm Sci 70:1093-4. Pharmacokinetics: Statistical moment calculations. Comparison of four basic models of indirect pharmacodynamic responses. Corticosteroid pharmacodynamics: models for a broad array of receptor-mediated pharmacologic effects. Direct linear plotting method for estimation of pharmacokinetic parameters.. VA. Robinson JR and Williams RL. Marcel Dekker. J Pharmacokinet Biopharm 18(5):483-99. Analysis of In Vitro-In Vivo Data. eds. Arzneim-Forsch/Drug Res 33 (1):173-6. Gabrielsson and Weiner (1997). Gibaldi and Perrier (1982). In Amidon GL. VA. Pharmacokinetics Brown and Manno (1978). Polli (1997). 2nd ed.

Belmont. Nonlinear Parameter Estimation. O’Reilly and Levy (1969). Applied Regression Analysis. A method for fitting linear combinations of exponentials. Clin Pharmacol Ther 10:22. J Pharmacokinet Biopharm 23(3):307-22. Kuh and Welsch (1980). Regression and modeling Akaike (1978). Am J Physiol 235(2):E97-E102. New York. Allen and Cady (1982). Parameter Estimation in Engineering and Science. 660 . Regression Diagnostics. radioligand assay and physiological dose-response curves. Simultaneous analysis of families of sigmoidal curves: Application to bioassay. John Wiley & Sons. Kinetics of pharmacologic effects in man: The anticoagulant action of warfarin. Academic Press. Wagner (1975). New York. Technometrics. Nagashima. Annals of the Institute of Mathematical Statistics 30:A9-14. Bates and Watts (1988). John Wiley & Sons. Chapter 12 in Numerical Methods for Non-linear Optimization. 2nd ed. DeLean. Belsley. Cornell (1962). 18:31-8. Munson and Rodbard (1978). New York. New York. Davies and Whitting (1972). Analyzing Experimental Data By Regression. New York. Restricted maximum likelihood (REML) estimation of variance components in the mixed models. Biometrics 18:10413. Beck and Arnold (1977).G WinNonlin User’s Guide Kowalski and Karim (1995). J Pharmaceutical Sciences 70:733-7. John Wiley & Sons. John Wiley & Sons. Metzler and Tong (1981). Lifetime Learning Publications. A semicompartmental modeling approach for pharmacodynamic data assessment. Illinois. Corbeil and Searle (1976). Academic Press. Bard (1974). Fundamentals of Clinical Pharmacokinetics. NY. Computational problems of compartmental models with MichaelisMenten-type elimination. CA. A modified form of Levenberg's correction. Drug Intelligence. Nonlinear Regression Analyses and Its Applications. Draper and Smith (1981). Posterior probabilities for choosing a regression model.

Amer Stat Assoc Proceedings Statistical Computing Section 57-65. The use of nonparametric methods in the statistical analysis of the two-period change-over design. 69:819-41. J Pharmacokinet Biopharm 23:307-22. Arzneim Forsch 29:1894-8. Peck and Barrett (1979). Parsons (1968). Giesbrecht and Burns (1985). Nelder and Mead (1965). Computing Journal 7:308-13. PCNonlin. Foss (1970). A method of exponential curve fitting by numerical integration. AUTOMOD: A Polyalgorithm for an integrated analysis of linear pharmacokinetic models. An interactive curve fit programme for pharmacokinetic analyses. Nonlinear least-squares regression programs for microcomputers. Gill. 1: Unconstrained Optimization. Academic Press.. Practical Optimization. North Carolina. Biological problems involving sums of exponential functions of time: A mathematical analysis that reduces experimental time. Kennedy and Gentle (1980). A semicompartmental modeling approach for pharmacodynamic data assessment. Vol. Math Biosci 2:123-8. Practical methods of optimization. Biometrics 41:477-86. Scientific Consulting Inc. Jennrich and Moore (1975). Approximate f-tests of multiple degree of freedom hypotheses in generalized least squares analysis of unbalanced split-plot experiments. Murray and Wright (1981). Two-stage analysis based on a mixed model: Large sample asymptotic theory and small-sample simulation results. Comput Biol Med 9:39-48. Journal of the American Statistical Association. A simplex method for function minimization. Statistical Computing. Fletcher (1980). Kowalski and Karim (1995). Biometrics 26:815-21. Hartley (1961). John Wiley & Sons. STRIPACT. Maximum likelihood estimation by means of nonlinear least squares. J Stat Comp Sim 554:363-78. Longley (1967). Koch (1972). Leferink and Maes (1978). New York. New York. The modified Gauss-Newton method for the fitting of nonlinear regression functions by least squares.References Regression and modeling G Fai and Cornelius (1996). Biometrics 577-83. Marcel Dekker. USA. Technometrics 3:269-80. 661 . J Pharmacokinet Biopharm 7:537-41. Gomeni and Gomeni (1979).

Watts and Davenport (1976). CSTRIP. Miller and Ham (1979). J Biopharm Stat 8(2):317-28. Improved resolution in the analysis of the multicomponent exponential signals. Nonlinear Regression Modeling. Comments on Nedelman and Jia's extension of Satterthwaite's approximation applied to pharmacokinetics. J Pharm Sci 65:1001-10. Nedelman and Jia (1998). J Pharm Sci 87(3):372-8. Estimating the dimension of a model. Holder (2001). Tong and Metzler (1980). Marcel Dekker. Mager and Goller (1998). Tests of statistical hypotheses concerning several parameters when the number of observations is large. Satterthwaite (1946). Schwarz (1978). Transaction of the American Mathematical Society 54. Mathematical properties of compartment models with MichaelisMenten-type elimination. Resampling methods in sparse sampling situations in preclinical pharmacokinetic studies. The effect of azone on ocular levobunolol absorption: Calculating the area under the curve and its standard error using tissue sampling compartments. A FORTRAN IV Computer Program for Obtaining Polyexponential Parameter Estimates. New York. Sheiner. Pharm Res 12:124-8. Wald (1943). Annals of Statistics 6:461-4. Nuclear Instrum Meth 205:479-83. Sparse data analysis methods Bailer (1988). Shampine. Gibiansky and Lau (1995). Applying Bailer's method for AUC confidence intervals to sparse sampling. Testing for the equality of area under the curves when using destructive measurement techniques. An extension of Satterthwaite's approximation applied to pharmacokinetics. J Biopharm Stat 11(1-2):75-9. Vozeh. Nedelman. Solving nonstiff ordinary differential equations . Sedman and Wagner (1976). 662 . Biometrics Bulletin 2:110-4. Simultaneous modeling of pharmacokinetics and pharmacodynamics: application to d-tubocurarine.G WinNonlin User’s Guide Ratkowsky (1983).the state of the art. Smith and Nichols (1983). An approximate distribution of estimates of variance components. Stanski. J Pharmacokinet Biopharm 16:303-9. Mathematical Biosciences 48:293-306. SIAM Review 18:376-411. Clin Pharm Ther 25:358-71. Pharm Res 5:238-41. Tang-Liu and Burke (1988).

Koch (1972). The use of non-parametric methods in the statistical analysis of the two-period change-over design. Practical Nonparametric Statistics 2nd ed. 2nd ed. et al. Dixon and Massey (1969). Introduction to Statistical Analysis. McGraw-Hill Book Company. Linear Models.References Statistics G Yeh (1990). Biometrics 577-84. Private correspondence. New York. John Wiley & Sons. ASA Proceedings of the Biopharmaceutical Section 74-81. Statistics Allen (2001). Principles and Procedures of Statistics. Winer (1971). 3rd ed. Searle (1971). The American Statistician. McGraw-Hill Book Company. New York. 663 . Steel and Torrie (1980). 28(3):98-100. Kutner (1974). A Biometrical Approach. New York. Conover (1980). Hypothesis testing in linear models (Eisenhart Model 1). 2nd ed. New York. Estimation and Significant Tests of Area Under the Curve Derived from Incomplete Blood Sampling. Statistical Principles in Experimental Design. New York. John Wiley & Sons. McGraw-Hill Book Company.

G WinNonlin User’s Guide 664 .

pks. 87 creating. 106 *. 418 *. 15.wto.lml.psc. 15 *.pwo. 48 fill. 579 Akaike Information Criterion AIC defined. 16 placeable header information. 455 *. 15. 79 *. 363 in LinMix vs SAS PROC MIXED. 218 *.cmd.imp. 15. 103. 483 Actual clock time defined. 15 *. 418 *.pto. 395. 37 *.dll.wgo. 148. 15. 102 saving. 483 Absolute error bars. 396. 484 *.wmo.exp. 487 *.wsc. 350. 15. 484 665 .wdo. 213 in PK workbook output.tdf. 215 Alias PKS.xpt. 487 *. 483 AggregateVar.map. 14 *.pco. 360 *. 102.anv. 120 Accumulation index. 451 Alignment.dep. 87 *.wmf. 103. 350.lib. 193 *. 14 *. 15 *. 48 center across cells. 16 *. 15 *. 193. 493 saving. 483 Actual time defined.txt. 499 *. 48 table cells.Index Symbols *. 15 *. 16 *. 483 in LinMix. 15. 145 Alpha.bmp.wsp. 105 creating.pmo. 373 in PK text output. 15. 78 A A defined. 193 *. 16 *.jpeg. 19 *.

581 AUC. 657 Bioassay functions. 485 average. 354 Bar charts. 386 pval. 375. defined. 582 Automatic sorting. 233 reserved variable names. 483 half-life. 184 Audit trail. 222 comparison operators. 354 Banded Toeplitz variance. 231 weights. 485 Bibliography. 580 AppendFileType. 117 Autoregressive variance. 235 creating. 16 ASCII models block content. 235 Anderson-Hauck. 170 Beta B. 236 Bioequivalence. 396 Axes log. 582 Average bioequivalence. 557 NCA model 220. 395 classification variables. 213 defined. 213 AUCall. 485 AUClast. 485 defined. 484 AUCINF. 485 defined. 484 statistic. 379 fixed effects. 127 options. 396 . 484 ANOVA. 581 ARRANGE.Pharsight WinNonlin User’s Guide A. 183 calculation methods. 218 examples. 485 calculation formulas. 243 Aterms(). 235 operators. 233 programming language. 372 APPEND. 485 half-life. 220 variables. 581 ASCII data files. 470 AUMC AUMCINF. 221 comments. 649 fixed effects (average). 354 AutoSort. 485 AutoFileOptions. 580 AppendFilename. 375. 485 Banded no-diagonal factor analytic. 484 partial area calculations. 379 setting up. 397 dependent variables. 233 structure. 354 Banded unstructured variance. 579 AND. 397 errors. 123 B B defined. 580 AppendSheet. 126 log base. 396 Bioequivalence Wizard. 470 reason. 485 Balanced design. 177 compartmental modeling. 485 666 AUMClast. 117 Baseline command. 396 recommended models. 122 uniform. 484 AlphaTerms(). 226 functions. 232 statements.

399 sort variables. 115 error bars. 124 copy and paste. 118 intervals. 49 filling adjacent. 16 scatter plots. 398 weight variables. 582 CaseSensitive. 496 random effects. 48 Chart Wizard. 397 Bitmaps. 49 deleting values. 397 repeated variance. 115 Charts. 44 including or excluding. 25 Pharsight Chart Objects. 378 variance structure. 402 general options. 123 legend. 123. 117 axis options. 59 Carry along variables. 486 C C defined. 119 histograms. 486 CarryData. 403 population/individual. 397 sums of squares. 48 Center across cells. 46 formulas. 129 title. 217 Carry along data for like sort levels.wmf. 33 save as *. 197 defined. 582 Bounds. 46 formatting. 16 saving to graphics formats. 401 power. 486 Cell references. 15 pink. 119 templates. 395 types. 44 Cells alignment. 398 regressor variables. 486 BQL defined.Index C fixed effects (pop. 117 page setup. 44 center across cells./individ. 381 linear model. 117 sort key title. 122 sort variables. 583 Cavg. 16 Blinding type. 486 changing chart type. 582 CASCADE. 486 C++. 43 deleting formats. 122 667 . 400. 454 BOLUS. 117 overlay. 401 least squares means. 378 missing data. 122 log. 122 bar charts. 122 axis titles. 381 type. 120 grid lines. 48 cell references. 486 Box and whiskers. 60 inserting.). 117 box and whiskers. 378 options. 127 group variables. 485 variable name limitations. 402 output. 115 automatic sorting. 126 multiple X-Y variables. 43 merge. 48 deleting. 130 creating. 152.

223 Command file. 222 Comparison operators. 235 Compiling Using C++. 488 degrees of freedom. 349 LinMix. 585 Copying charts. 356 nonestimability. 348 Control stream. 44 deleting. 488 LinMix. 82 in ASCII models. 383 Classification variables bioequivalence. 487 LinMix. 454 defined. 487 CONFIDENCE. 90 ODBC export.Pharsight WinNonlin User’s Guide uniform axes. 487 Clss/F. 487 CloseAll. 487 CLR. 293 settings. 297 units. 352 Clock time. 223 Comments comment character. 583 Cmax. 238 Compound. 83 Convergence criterion. 496 univariate. 536 COND. 212 descriptive statistics. 14 worksheet cell fill-down. 352. 487 Cmin. 354 668 CON array. 397 . 487 CMT. 266. 584 Constants defined. 369 Convolution. 299 variables. 7 Contrasts defined. 213 Covariates. 102 ODBC import. 499 Westlake. 83 Coefficients. 346. 373 defined. 213 Configuration PKS. 152 planar. 212. 583 Columns cell references. 123 units display default. 53 COMMAND. 43 names. 295 COPY. 116 CL. 500 Confidence limits Univariate. 53 units. 365 COLUMN. 487 Clss. 46 Correlation matrix. 44 freezing. 221 Command block. 232 ConcCol. 262. 397 defined. 56 including or excluding. 487 Constrained optimization. 487 COMMANDS. 487 Compound symmetry variance. 584 Confidence intervals. 87 CONSEQDELIM. 130 workbooks. 60 inserting. 122 X-Y variables. 487 Condition number. 487 Classical and Westlake Confidence Intervals. 207. 31 use group headers. 584 Concentration-effect models. 451 Connection string.

586 D DATA. 44 rows. 64 Data variables compartmental modeling. 57 reset selections. 262 Loo-Reigelman. 13 inclusions and exclusions. 488 exponential. 269 Wagner-Nelson. 658 UIR. 33 *. 587 DELTA. 180 Defaults missing value indicator. 269 units and UIR. 489 Custom query builder. 587 Deleting cell formats. 370 DeleteFiles. 260 references. 42 DELIMITER. 585 CROSSVARS. 31 workbooks. 567 Data dependencies. 259 CROSSVAR( ). 31 units. 372. 397 LinMix. 161 Dates. 29 page breaks in tables. 97 Customer support. 278 Curve stripping. 35 Dependent variables bioequivalence. 57 sparse sampling. 349 LinMix versus PROC MIXED. 514 toxicokinetic. 60 merging. 49 cell values. 215 Cubic model.map. 453 DCP description. 10 licensing. 28 modeling output. 62 sorting. 49 cells. 261.Index D defined. 62. 14 file types. 254 separate columns. 11 DefinitionFile. 547 Cumul_Amt_Abs_V. 29 NCA calculation method. 188. 79 tables. 36 refreshing output. 587 Dependencies. 488 Crossover design. 11 CUT. 278 DefaultNCAParameterNames. 488 Crossed factors. 146 Deconvolution. 11 user feedback. 514 transformations. 373 Satterthwaite’s approximation. 35 saving. 279 numeric. 37 origins. 28 Defect reports. 44 worksheets. 258 stacked in one column. 461 Decimal places SAS Transport export.dep. 196 noncompartmental analysis. 352 NONMEM export. 43 columns. 33 file type limitations. 83 Description 669 . 586 Degrees of freedom LinMix contrasts. 165 defined. 585 CSS.

454 Descriptive statistics. 654 LinMix. 588 DestCol. 590 Engines. 489 Eigenvalues. 215 DIFFERENTIAL. 201 data variables. 655 tables. 432 noncompartmental analysis. 171 worksheet data. 28 Disable curve stripping. 231 DVID. 120 absolute. 488 E0. 203 units. 418 Enhancement requests. 213. 120 relative. 423 defined. 489 EC50 defined. 652 Errors. 467 DoseFile. 653 scripting. 589 Dosing ASCII models. 120 DTA. 364 Else. 171 NONMEM export. 4 user models. 652 IVIVC. 589 DoseUnit. 232 E E. 83 DZ array. 234 Document manager. 489 scripting language. 171. 649 noncompartmental analysis. 11 EntireRow. 165 DNAMES. 649 compartmental modeling. 200 scenario-specific. 210 ErrType. 489 END. 637 bioequivalence. 590 EQ operator.Pharsight WinNonlin User’s Guide PKS studies. 230 DTA array. 589 Eigenvalue defined. 358 Examples Examples Guide. 234 Emax defined. 588 Do. 225 Directories options. 202 dose normalization. 2 Excel 670 . 333 DF. 224 DnErr( ). 589 EndRow. 448. 84 PKS. 36 Determining estimability. 202 Down variable. 413 Tau. 364 LinMix. 221. 637 descriptive statistics. 235 Error bars. 488 ECOL. 202 compiled PK models. 201 file structure. 591 Estimates fixed effects. 430. 221 EndCol. 225 Differential equations. 120 Error messages. 349. 151 KS_pval. 472 regimen. 472 simulation. 157 DestArea. 588 Detach.

352 parameter estimates. 146 worksheet cells. 76 Word export options. 221. 231 FONT. 50 find next. 16 saving workbooks to Excel. 103 ExportAll. 46 Final parameters confidence intervals. 78 FileType.Index F Excel 4. 198 FNUMBER. 396 defined. 402 Fixed parameters. bioequivalence.0 files. 166. 102. 227. 591 Exclude profiles with insufficient data. 223 Function variable. 489 File types WNL and WNM compatibility. 594 FootnoteFont. 60 by selection. 77. 594 FontSize. 16 EXCLUDE. 217 user models. 223. 215 Schwarz Bayesian Criterion. 51 Fit Akaike Information Criterion. 16 data file limitations. 45 Formulation defined. 490 half-life. 60 based on criteria. 60 from Lambda Z calculations. 212 FIND./individ. 336 average bioequivalence. 235 G Gamma defined. 13 PKS files. 489 Functions. 17 FileMask. 85 S-PLUS. 594 Find. 14 data files. 490 671 . 441 SAS Transport. 231 Fabs defined. 102 SAS Transport files. 595 FootnoteFontSize. 215 Fixed effects. 593 Fill-down. 593 Files ASCII. 15 files. 62 Exponential curve stripping. 174 Exclusions. 44 operators. 51 replace. 592 F F. 591 Exporting NONMEM. 336. 78 EXTRA. 237 FUNCTION. 595 Formatting tables. 80 ODBC. 364 pop. 225. 489 Fdiss defined. 86 Word export. 286 reset selections. 489 LinMix. 78 SigmaPlot. 489 Export definition. 489 FORTRAN. 46 Formulas. 595 multiple FUNC blocks.

38 items logged. 597 GroupVarBreak( ). 598 Hessian eigenvalues. 463 Group variables. 541 defined. 354 compound symmetry. 38 worksheets. 60 reset selections. 52 GOTO. 37 I IConsumer interface. 43 worksheets. 235 H HALT. 60 selected cells. 60 based on criteria.Pharsight WinNonlin User’s Guide Gauss-Newton algorithm. 490 HEADERS. 600 Independent variable. 490 Go to. 597 GT operator. 119 GroupVar( ). 292 linear trapezoidal with linear/log. 364 Heterogeneous autoregressive variance. 292 . 600 InData( ). 539. 601 Inserting cells. 60 by selection. 568 Initial parameter estimates. 599 If Then Else. 6 GE operator. 111 672 ID variables. 454 Indirect response models. 197 InitialDose. 600 Including based on criteria. 62 Increment for partial derivatives. 43 columns. 16 IncludeVar( . 233 Green scenario icon. 491 Interpolation linear trapezoidal. 207. 490 Hartley modification. 87 INCLUDE. 123 History "cannot be processed". 490 charts. 600 InputLag. 60 Inclusions. 491 Indication. 18 Hill model. ). 9 Levenberg-type modification. 42 Interaction. 596 GroupVar( . 354 Hidden windows. ). 542 Individual bioequivalence. 6. 235 Geometric mean defined. 376 INITIAL. 6. 542 parameters. 599 IncludeVars( ). 177. 491 IDVar( ). 597 GroupVars( ). 599 Include placeable header information. 233 Importing ODBC. 177. 548 Histograms. 598 Harmonic mean. 597 GroupVars. 491 InData. 9 Levenberg Hartley modification. 598 IDVars. 118 intervals. 491 Intercept. 491 parameter estimates and bounds. 43 rows.

307 use raw data. 302 JoinCrossVars. 602 JoinIDVar( . 208 history. 492 K01. 315 Wagner-Nelson. 476 673 J Join variables. ). 143 JoinCrossVar( . 492 LinMix. 123 IPR. 601 . 164 range file structure. 310 dissolution data and models. 157 L LABEL. 492 merge data sets. 344. 381 defined. 292 INTERVAL. 430. 157 KS_pval. 315 output options. 278 wizard. 307 Levy plots. 4. 306 dissolution models. 16 K K half-life. 277 convolution. 492 K0. 300 loading projects. 492 half-life. ). 305 Loo-Reigelman. 431 LamzFile. 162 defined. 602 JoinGroupVars. 492 K10. 491 IQueryBuilder interface. 603 KM. 242. 304 model validation. 604 LegendFontSize. 547 error messages. 359 Legend( ). 347 JPEG files. 654 fixed parameters. 292 linear/log trapezoidal. 198 fraction absorbed. 604 LE operator. 300 Libraries. 492 not estimable. 279 minimizing the wizard. 601 Intervals. 110 Iterations. 293 correlation models. 603 Joint Contrasts. 178.Index J linear up/log down. 287 predicting PK. 57 tables. 492 Kolmogorov-Smirnov normality test. 2 unit impulse response fit. 307. 365 procedure. 492 least squares fitting. 491 IVIVC. 122 Levy plots. 178. 277 toolkit introduced. 604 LegendFont. 602 JoinGroupVar( . 492 Lambda z. 603 JoinIDVars. 5 Iterative reweighting. 235 Least squares means. 604 Legends. 310 options. 233 Lag time defined. 492 KE0. ). 305 toolkit. 304 saving projects. 313 project status.

479 Logarithmic axes. 536 simultaneous. 336 parameters. 361 output parameterization. 177. 292 Linear/log trapezoidal rule. 360 Hessian eigenvalues. 336. 366 repeated effects. 355 regressor variables. 337 variable name limitations. 352 text output. 359 674 limits and constraints. 341. 349. 352 coefficients. 478 Licensing customer support. 364 general options. 350 linear mixed effects model. 350 Newton’s algorithm. 605 Loading ODBC import query. 353 weight variables. 74 LOAD. 352 REML. 361 transformation of the response. 605 Locked. 364 hypothesis tests. 360 iteration history. 96 LoadTemplate. 369 crossed factors. 605 LinMix. 344. 346. 337 sort variables. 352 fixed effects estimates. 370 Schwarz’s criterion. 323 Akaike’s criterion. 206 Linear trapezoidal rule. 184 Linear kinetics. 493 Linear models. 356 convergence criterion.Pharsight WinNonlin User’s Guide add or update. 352 workbook output. 333 estimates. 365 partial tests. 292 Linear up/log down. 178. 351 variance structure. 123 interpolation rule. 372 computations. 334 output. 292 Link models. 328 LLOQ. 184 Logarithms transformations. 352 errors. 356 residuals. 3. 3 Linear regression. 358 fixed effects. 279 . 545 LinkFile. 338. 368 non-estimable functions. 363 classification variables. 339. 372 compared to PROC MIXED. 361 X matrix. 649 estimability. 546 defined. 493 use when less than LLOQ. 459 defined. 62 Loo-Reigelman. 477 loading from PKS. 364 Satterthwaite’s approximation. 366 contrasts. 365 least squares means. 326 LinMix Wizard. 184 trapezoidal rule. 31 Linear interpolation rule. 362 singularity tolerance. 177. 364 random effects. 493 minimization algorithm. 477 status. 183. 363 sequential tests. 365 command file. 488 dependent variables. 362 predicated values. 362 initial variance parameters. 493 compared to ANOVA. 292 with linear/log interpolation. 430. 11 expiration warning. 178.

208 Model variables compartmental modeling. 204 output intermediate calculations. 207 iterations. 197 bounds. 499 Micro rate constants. 606 Makoid-Banakar model. 493 Mean square. 239 Microsoft Word export. 173. 4 mdllib_mdl602. 498 MaintenanceDose. 76 Minimization algorithm. 494 Microsoft Visual C++. 208 minimization. 4 Mm. 212.lib. 207 exclude profiles with insufficient data. 196 dosing. 550 Marking valid/invalid data. 167 partial areas. 174 PK settings. 493 stripping dose. 208 number of predicted values. 233 Maximum likelihood estimates. 549 mdllib_mdl603. 206 model size. 549 mdllib_mdl604. 200 Model size. 418 Methods defined. 162 threshold. 208 weighting for NCA. 206 title for NCA text output. 203 convergence criterion. 206 units. 175 Model parameters. 235 M Macro rate constants. 178. 167 Model options. 197 initial estimates. 207 output. 493 Merge cells. 493 Michaelis-Menten. 206 propagate final estimates. 288 workbook output.Index M output options. 193 675 . 196 noncompartmental analysis. 515 baseline. 221 Model 220. 208 Mean square variance. 59 Method scripting language. 176 transpose final parameter table. 550 Mean defined. 35 Mathematical operators. 208 mean square. 198 title. 288 LOWER. 48 Merging. 606 Missing values default indicator. 198 Model properties compartmental models. 542 defined. 57 carry along for like sort levels. 199 ParmFile. 174 increment for partial derivatives.lib. 543 Vmax. 28 in NCA parameter estimates. 568 LT operator. 287 weighting. 182 Mm. 161 Modeling. 493 equation. 206 MISSING. 543 MODEL. 167 slopes data ranges.

519 Pharsight Model Objects. 196 default output. 494 Multiple linear regression. 200 error messages. 328 No Scaling of Weights. 8 defined. 197 iterations. 218 cubic. 207 options. 352 Newton’s algorithm. 536 creating ASCII models. 545 types. 549 Modified likelihood. 207 data variables. 542 noncompartmental. 196 model size. 491. 546 link. 494 Name objects. 206 model options.Pharsight WinNonlin User’s Guide ASCII model number. 548 indirect response. 206 Nobs. 373 MRT. 471 NCA model 220 baseline. 547 double Weibull. 193 Makoid-Banakar. 208 weights. 223 NDERIVATIVES. 53 Native format. 223 Nelder-Mead simplex algorithm. 209 units. 86 start. 223 Nmiss. 15 sigmoidal/dissolution. 33 output options. 3 Multiple profiles in script files. 428 N N modeling output. 204 PK settings. 170 model 220 threshold. 210 output dependencies. 208 NONMEM export. 368 Next. 210 initial parameter estimates. 205 S-PLUS export. 494 No Intercept. 494 MRTlast. 637 errors. 209 stop. 494 . 504 number of predicted values. 472 Naming columns. 206 refreshing output. 550 676 Michaelis-Menten. 536 loading. 160 noncompartmental analysis. 203 model settings. 205 Models concentration-effect. 539 linear. 217 Weibull. 208 linear models. 234 NFUNCTIONS. 193 user models. 209 troubleshooting. 494 Nested factors. 29 dosing. 29 output. 494 MRTINF. 208 minimization algorithm. 547 simultaneous link. 170 NCONSTANTS. 505 pharmacokinetic. 80 notation and conventions. 35 scaling of weights. 549 Hill. 3 mean square. 6. 494 Nested variables. 221 convergence criterion.

506 parameter names. 181 default calculation method. 223. 207 O Object defined. 182 model selection. 174 Nonlinear kinetics defined. 102 677 . 421 Occasions. 160 therapeutic response window. 83 occasions. 102 building an export definition. 173 output text. 495 Objects Scripting Language. 418. 168 therapeutic window calculations. 30 default parameter names file. 494 disable curve stripping. 3.Index O NOBSERVATIONS. 495 Nonlinear regression. 174 page breaks in text output. 5 methods. 174 computational rules. 161 models. 495 ODBC export. 354 Nominal time defined. 495 Nonlinear models. 83 comment character. 160 model variables. 3. 180 Lambda z. 83 problem. 8 NONMEM export. 157 NPARAMETERS. 174 include parameter in output. 83 dependent variables. 102 creating an export. 173 output intermediate calculations. 83 dosing. 85 ODBC defined. 175 workbook output. 174 parameter calculations. 85 other data items. 245 Normality test. 181 data exclusions. 159 AUC formulas. 494 Noncompartmental analysis. 178 weighting. 223 Number format. 567 No-diagonal factor analytic. 165 dosing. 84 DVID. 162 Lambda z file structure. 653 exclude profiles with insufficient data. 186 units. 439 precision of output. 183 chart output. 505 output. 82 control stream. 568 NSECONDARY. 179 slope data ranges. 47 Number of predicted values. 80 CMT. 16 Nonparametric superposition. 103 export definition. 180 defined. 104 connection string. 515 errors. 184 partial areas. 162 sparse sampling data. 102 definition. 431 missing parameter values. 82 Non-numeric data troubleshooting. 174 output options. 182 data pre-treatment. 29. 179 partial area computations. 171 drug effect data. 90. 167 partial areas file structure.

5. 10 Pharsight Chart Objects. 97 custom query builder example. 92 saving export settings. 31 WinNonlin options. 93 connection string. 31 Page setup. 89 software developer’s kit. 36 Output intermediate calculations. 15 Pharsight Text Objects. 198. 15 Pharsight Workbook Objects. 184 Partial derivatives. 36 Other data items NONMEM export. 173 tables. 180 ParmFile. 9 contact information. 28 Word export. 436 fixed. 360 models. 28 bioequivalence. 430 Partial areas. 491 Partial tests. 174 Overlay chart. 434 file structure for link models. 92 save password. 173 NCA output. 9 scientific and consulting services. 33 refreshing. 519 Pharsight. 7 increment. 31 LinMix. 362. 87. 467 Pharmacodynamic models. 33 PK settings. 117 P P array. 401. 198 LinMix output. 97 Operators. 8. 78 workbooks. 533 Pharmacokinetic models. 207. 35 unlocking. 452 save ODBC password.Pharsight WinNonlin User’s Guide loading a saved connection. 83 Output compartmental models. 365 names for NCA. 235 Origins. 87 loading a saved connection. 495 Password PKS. 11 products. 402 directories. 21 Parameters file structure. 28 license expiration warning. 90 custom query builder. 10 training courses. 105 ODBC import. 235 Options. 180 preferred names. 206 . 15 Pharsight Model Objects. 430 preferred default names. 179 ParmFile. 152 Performance PKS. 10 technical support. 77. 107 Watson DMLIMS. 10 customer feedback. 174 table defaults. 232 678 Page breaks NCA output. 28 OR. 29 NCA model. 4. 111 import settings file. 210 pink. 430 PartFile. 89 saving. 167 computation of. 400. 89 Percentiles descriptive statistics. 87 building a query. 15 Pink output windows.

447 scenario-specific dosing. 449 load study or scenario. 450. 373 Profile. 447. 536 defined. 514 Preferred units. 479 libraries. 448. 449 save as. 453 time. 496 PUNIT. 451 create study. 447 user name. 450 password. 454 Propagate final estimates. 449 scenario search. 451 append/update study. 156 Queries building ODBC queries. 212 Planar confidence interval. 475 status. 82 PROC MIXED. 450. 450. 465 DCP description. 401 data requirements. 449 creating scenarios. 496 Programming language. 179 SAS Transport export. 441 green icon. 21. 93 679 . 446 study data requirements. 613 PrintMultiAcross. 477 logging on. 198 Properties defined. 445 PLANAR. 178 Printing. 450. 224 Q Quantiles computation. 464 scenarios. 545 PKS. 613 Problem NONMEM. 463 loading libraries. 613 PrintMultiDown. 233 Project. 452 study metadata. 452 using with WinNonlin. 452 performance. 401 fixed effects. 450 studies. 467 red icon. 21 print all. 466 other documents. 1 accessing. 461 dosing. 386 Precision NCA output. 211 PK/PD link models. 188.Index Q PK text output. 477 alias. 79 Pre-clinical data. 224. 476 library and share status. 454 Power. 25 PrintMulti. 450 add or update library. 229 Population bioequivalence. 472 file format. 496 simultaneous. 27 print preview. 478 load. 463 save. 402 Portfolio PKS studies. 376 Population/individual bioequivalence. 463 information. 496 PNAMES. 451 non-WinNonlin documents. 25 page setup. 450 share. 449 add columns. 450 configuration. 451 sessions. 472 server.

368 Rich text files. 370 . 368 680 REMOVE. 35 Regression methods. 616 Replace. 60 inserting. 16 Rows cell references. 497 Relative time. 479 Reason. 120 Relative nominal end time. 616 Repeated bioequivalence. 617 S S. 497 RESPONSE. 373 LinMix. 352 Regressors defined. 43 Rule sets. 617 Restricted maximum likelihood LinMix. 496 Random effects bioequivalence. 213 defined. 497 Relative nominal time. 615 RemoveCol. 496 Rank transformations. 497 Relative actual time. 44 freezing. 355 RANK. 213. 497 estimate of residual standard deviation. 232 SameErrCol( ). 397 LinMix. 79 export. 463 References. 222 REML. 341. 617 RunWordExport. 92 QueryBuilderInterfaces. 615 RenameCol. 615 RemoveRow. 475 SAS Transport files data set name. 66 Residuals. 78 Satterthwaite’s approximation. 399 Repeated effects LinMix. 215 S array. 109 R Random effect defined. 618 Sample number. 211 Rank. 497 RegVar( ). 56 including or excluding. 366 Newton’s algorithm. 68 Ratio tables.dll. 398 LinMix.Pharsight WinNonlin User’s Guide ODBC import queries. 406 Read only. 356 REPLACE. 614 Relative actual end time. 8 Regressor variables bioequivalence. 614 RegVars. 51. 458 SAS share. 8 Gauss-newton algorithm. 497 REMARK. 9 Nelder-Mead simplex. 71 RUN. 657 Refresh. 44 deleting. 470 Red scenario icon. 497 Relative error bars. 339. 366 Newton’s algorithm.

337. 478 Show formulas. 426 SD. 620 SortVar( ). 619 Scaling of weights. 165 SMOOTHING. 57 SortLevel. 224 SORT. 545 Singular or near-singular equations. 79 tables. 464 Schwarz Bayesian Criterion in LinMix. 579 engines. 655 file paths (Filename property). 619 SaveTemplate. 415 dosing. 413 model parameters. 619 SECONDARY. 498 SDK. 397 charts. 242 Scatter plots. 146 Simplex algorithm. 374 Slopes disable curve stripping. 57. 107 SE. 226. 221. 466 creating. 621 SortVar( . 620 SNAMES. 28 ShowUnits. 209 Semicompartmental modeling. 620 SigmaPlot export to. 145. defined. 6 Singularity tolerance.PSC). 495 PKS. 122 Sort variables. defined.Index S SAVE. 205. 496 script file (. defined. 621 681 . 179 SAS Transport export. 440 property. 352 Sorting worksheet data. 212 Simulation. 620 Sequential tests. 214 user models. 618 SaveAllPrefix. 498 SearchArea. 361. 621 SortVars. ). 451 Setting preferred units. 415 Simultaneous PK/PD link models. 208 Share. 226 Select Data for Modeling dialog. 85 version. 418 script file components. 489 error messages. 409 comparing dose schemes. 618 SaveAll. 466 adding non-WinNonlin documents. 213 in PK workbook output. 465 defined. 227 Secondary parameters. 475 status. 250 SEQUENCE. 85 Significant digits NCA output. 117 Scenarios. 498 bioequivalence. 620 Sort key title. 373 in PK text output. 119 LinMix. 498 search. 215 SBC defined. 497 Scientific notation. 593 method. 48 Scripting commands. 362 Server PKS. 493 object. 6 Singular value decomposition. defined. 415 standard error computation. 214 simulation. 412 secondary parameter estimates. 447 adding columns. 363 in LinMix vs SAS PROC MIXED. 6.

144 templates. 215 STACKED. 625 Tau. 160 S-PLUS export to. 131 cell alignment. 652 excluded data. 623 Status code transformations. 147 units display default. 147 formatting. 134 TableNum. 454 locked. 146 error messages. 145 formatting. 188 models. 498 description. 224 Temporary variables. 143 missing data. 453 PKS. 144 Summary variables. 129 tables. 145 column alignment. 133 options. 69 LLOQ. 622 Sparse data computations. 454 SUBJECT. 146 summary statistics. 133 font. 148 templates. 381 682 SUNIT. 625 Table Wizard. 31 page break defaults. 86 SS/df (mean square). 514 references. 74 rule sets. 221. 151 tables. 134 units. 622 Standard error. 446 start and end dates. 203. 224 T TAB. 146 data only. 212 START. 147 significant digits. 86 version. 498 Studies balanced design. 623 STARTING. 453 type. 35 SourceCol. 224 . 146 decimal places. 662 use sparse sampling. 624 SummaryVars. 498 SummaryVar( ). 146 significant digits. 493 SSR. 134 TEMPORARY. 131 cell alignment. 146 join variables. 624 Summary statistics.Pharsight WinNonlin User’s Guide SortVars( ). 625 Tables. 499 Technical support. 221 StartCol. 623 STATS. 145 summary statistics. 31 TAU. 145. 10 licensing. 479 name. 144 table definition files. 485 defined (PKS). 622 Source changed. 624 Sums of squares. 71 Stripping dose defined. 31 page numbers. 11 Templates charts. 225 StartRow. 454 design.

575 NCA model 220. 31 prefixes. 33 Therapeutic response window. 176. 629 Uniform axes. 497 relative. 269 automatic generation. 309 Units assigning. 168 calculations. 497 TimeCol. 626 TerminalUpper. 123 UNIT. 462 Troubleshooting. 494 relative actual end time. 222. 66 equations. 66 LinMix response. 626 Tests of hypotheses. 483 actual time. 499 TileHorizontal. 178 options. 221. 354 TOLERANCE. 54 683 . 209 loading tabular data. 299 dose.. 244 Transformations. 497 relative actual time. 629 Unit impulse response See also UIR. 24 Text windows Pharsight Text Objects. 29 Treatment PKS mapping. 55 convolution. 193 U UIR. 626 Time actual clock time. 54 summary statistics. 54 noncompartmental analysis. 15 pink. 213 default method for NCA. 67 filter. 171 file. 178 Trapezoidal rule. 29 compartmental modeling. 497 relative nominal time. 170 TI.Index U TerminalLower. 53 compartmental modeling. 206 TitleFont. 430 mass. 50 UNIFORM. 625 TerminalPhase. defined. 153 supported unit types. 10 TRANSFORM. 627 Title. 309 ULOQ. 66 status codes. 497 relative nominal end time. 627 TITLE. 627 TitleFontSize. 627 TimeRepeat. 487 nominal time. 64 rank. 206 non-compartmental modeling. 459 Undo. 208 converting. 186 Threshold command. 68 replace. 16 Trust region. 69 Transpose final parameter table. 483 clock time. 62 action. 362 Text page setup. 626 TileVertical. 64. 51 Training courses. 628 Tolerance. 628 Toeplitz variance. 337 multiple. 7 Types of models.

161 names. 226 starting values block. 231 Variance autoregressive. 224 transform block. 315 Use sparse sampling. 54 volume.Pharsight WinNonlin User’s Guide time. 491 independent. 354 defined. 224 weight variable. 568 Use group headers. 499 heterogeneous autoregressive. 630 UPPER. 54 Univariate command. 119. 378. 354 Variance Inflation Factor. 499 Variance structure. 351 684 Variables carry along. 120 Update Reason dialog. 354 banded unstructured. 487 function variable. 499 Valid data. 486 charts. 217 ASCII examples. 353 types. 354 mean square. 353. 225 structure. 119. 497 reserved names. 232 sort. 378 LinMix. 36 Unstructured variance. 500 WinNonlin variables. 338. 212 Univariate confidence interval. 307. 354 Up variable. 492 model variables. 196 model variables for NCA. 223 creating an ASCII model. 354 unstructured. 493 no-diagonal factor analytic. 225 FORTRAN. 398 LinMix. 222 WinNonlin variables. 111 . 231 User name PKS. 300 Variable names Bioequivalence. 35 Validation IVIVC. 398 bioequivalence. 354 Visual Basic custom ODBC query example. 354 heterogeneous compound symmetry. 212. 354 Banded Toeplitz. 354 Toeplitz. 499 UNLOAD. 452 V V. 116 classification. 354 components. 237 function block. 232 regressors. 490 ID. 354 compound symmetry. 415 VIF defined. 217 secondary parameters block. 220 temporary variables. 160 User models. 226 block content. 354 banded no-diagonal. 489 group. 498 summary variables. 225 libraries. 498 temporary. 122 Use raw data. 218 differential equation block. 221 commands block. 491 join. 630 Unlocking derived data. 471 UpErr( ).

21 Pharsight Workbook Objects. 1 file compatibility with WNM. 97 WCSS. 50 formulas. 242. 591 WordExportFilename. 87. 16 show formulas. 157 Weighting. 78 ExportAll command. 241 Weighted summary statistics. 637 Watson DMLIMS. 215 Weibull model. 549 double. 631 Worksheets cell references. 242 Westlake confidence interval. 241 compartmental modeling. 500 VZ/F. 500 VZ. 631 Workbooks copying. 418 WinNonMix file compatibility with WNL. 175 Wagner-Nelson. 205 uniform. 175 scaling. 241 weights from data file. 17 options. 15 pink. 243 iterative reweighting. 14 operators. 64 dependencies. 76 export options. 549 WEIGHT. 631 WinNonlin. 499 Vss. 16 unlocking. 397 variables (LinMix). 288 weighted least squares. 231 WinNonlin object files. 36 WorkingDir. 44 freezing rows and columns. 1 variables in user models. 287 weighting. 77. 44 deleting. 630 Weight variables (bioequivalence). 42 fill-down.Index W Vmax. 14 copying with transformations. 288 noncompartmental analysis. 491 least squares. 288 workbook output. 500 Windows hiding. 45 options. 35 saving to Excel. 33 missing values. 500 Weighted least squares. 18 Windows metafiles. 56 685 . 352 variables. 631 Word export. 288 Warnings and errors. 205 in ASCII models. 28 Professional. 33 refreshing. 16 WindowState. 28 out of date. defined. 35 page setup. 28 status codes. 17 WITH. 278 output options. 28 new. 46 find. 500 W Wagner-Nelson. 175 Loo-Reigelman. 15 WinNonlin Script File. 242 scaling of weights. 69 troubleshooting. 1 Enterprise.

633 XTitleFontSize. 634 YLabelsFontSize. 632 XLabelsFontSize. 634 Y YLabelsFont. 37 inserting. 635 YTitleFont.Pharsight WinNonlin User’s Guide go to. 636 Z Z array. 634 YTitle. 57 transformations. 633 XVar( ). 48 merging. 215 WT. 500 WSSR. 51 scientific notation. 64 units. 500 loading. 632 XTitle. 232 X matrix. 232 YTitleFontSize. 632 XTitleFont. 42 merge cells. 35 Wrap text. 42 replace. 52 history. 635 YVar. 62. 53 wrap text. 635 686 . 19 saving dependencies. 232 X X. 48 Workspaces. 48 sorting. 328 XLabelsFont. 633 XUnits. 19 defined. 635 YUnits. 48 WRSS. 20 new. 57 naming. 20 saving.

- Exposicion El Sujeto
- Acanthodactylus erythrurus
- Origen y Evolución de Las Membranas Celulares
- El Tercer Fausto de Salvador Novo
- Clarinete metodo MAGNANI (completo).pdf
- asturias.pdf
- Arguelles Hugo - Los Cuervos Estan de Luto
- Catalogo de Placas11
- 291.Full Anomalia Pelger Huet Pseudo
- Pphi Curso Bpc
- 25 Frases Que Te Harán Sonreir Al Trabajar