ANALYTICAL a laarelse Maelesnie ta CHRISTOPHER BURGESS *Valid Analytical Methods and Procedures Christopher Burgess Burgess Consultancy, County Durham, UK RSeCISBN 0-85408.182.5 ‘A catalogue record foe his books available Frm the Bish Library {© The Royal Society of Chemistry 2000 Al rights reserved. Apart from any fai dealing forthe purposes of esearch o private study, oF ‘rticiam or review as permultedunder the terms ofthe UK Copyright, Designs and Parems Aet, 1988, ths publication nay no be reproduce, stored or transmitted, in any form or by ey meas, without the prior permizsion in writing of The Royal Society of Chemisty. in the cave of reprgrepie reproduction anly in accordance withthe frm ofthe licences isueby the Copyrigh Licensing Agency inthe UK, ‘arin accordance with the terms ofthe licences sued by the eppropriate Reproduction Righss Organization ouside the UK. Enguiies concerning ‘reproduction side the terms ated here shouldbe sent to The Royal Society of Chemiziry atthe addrers printed on this page. Published by The Royal Society of Chemistey, ‘Thoms Graham House, sience Park, Milton Road, (Cambridge CB4 OWF, UK For further information se our web ite at waws36 0% ‘Typeset by Paston PrePress Lid, Beccles, SulTolk Printed by Athnucu Prost Lid, Gateshead, Tyne and Wee, UK, Preface ‘This handbook has been long in the making. Since the original decision to write it in 1995, much has changee in analytical science and progress made towards harmonisation of procedures and practices. However, the need remains for practising analytical chemists to adopt a formalism for analytical method development and validation embracing the necessary and sufficient statistical tools. The proactive role of the statistcianjchemometrician ia providing effective and elficent tools has long becu recognised by the Analytical Methods Committee (AMC) of the Analytical Division of the Royal Society of Chemistry Analytical peaettioners should be ever mindful of Sic R.A. Fisher's stricture that "to callin the statistician after the experiment has been done may be ne more than asking him to pecforma post-mortem examination: he may be able 0 say what the experiment died of As the title suggests, the intent isto provide a best practice approach which will meet the basic needs of the bench practitioner and at the same time provide links to more exacting and specialist publications. [a this endeavour the author has enjoyed the support and active participation of the Chairmen and members of the AMC and of the Analytical Division throughout its long gestation, Partieulae thanks are due to past chaiemen of AMC, Dr Roger Woed, Mr Colin Watson and Dr Neit Crosby for their enthusiasm and guidance. in addition, { am indebted to Dr Crosby for much of the material concerning the history of the AMC, From the outset, Mr lan Craig of Pedigece Petfoods Ltd and Dr Peter Brawn of Unilever Research, Colworth Lahoritory have devoted considerable time tnd effort in scoping and shaping the handbook and, in particulur, for generating and providing material on sampling and namenctature. Without ‘heir unflayging support the project may well have foundered, On the statistical side, my thanks are due to Professor Jims Miller, President of the Analytical Division, who has been kind enough to read the manuscript thereby saving me from statistical errors and obscurities, and Professor Mike Thompson for providing the data set for the [UPAC collaborate trial example calculation. {thank Dr Dai Beavan of Kodak Lid for allowing me access to some of theie data sets for use as examples, Many other members of the AMC and the ‘Analytical Division have kindly given me support and input including Professor ‘Arnold Fogg, Professor Stan Greenfield, Or Dianna Jones, Dr Bob McDowall, Dr Gerry Newman, Mr Braxton Reynolds, Dr Diana Simpson, Mr John Wilson and Mr Gareth Weight. 1 am grateful (0 Professor J.D.R. Thomas and Dri Preface David Westwood for their efforts in helping me ensure consistency and clarity within the handbook, “The help of Ms Nicola Dest of LIC in Burlington House has been invaluable and my thanks are due also to Dr Bob Andrews and Dr Sue Askey of RSC publishing. My thanks are due to Paul Nash for producing the subject index. Finally, {wish to thank my wife and family for their forbearance during the preparation of this handbook and acknowledge financial support provided by ‘The Analytical Methods Trust Contents Inteoduetion LL Historical perspective 12. Overview of the handbook U3 Purpose and scope ‘Nomenclature: Terms and Parameters 2.1 Introduction 22. Terms 23° Parameters ‘Samples and Sampling 3.1 Introduction 3.2. Whatis a sample? 33. Homogeneity and concentration ranges Method Selection 4.1 Fitness for purpose 42. Sources and stcategies 43° Sampling considerations 44° Mattix effects Equipment Calibration and Qualification 5.1 Qualification approacher 5.2 A convergence of ideas ‘The Method Development Process 6.1 Mapping the analytical process and determining the key factors 6.2 Simple experimental design 6.3 Multifactor experimental designs ‘Method Valiéation 7.1 Recommended best practice for method validation. 72. Describing and writing analytical methods Data Evaluation, Transformation and Reporting 8.1 Exploratory data analysis 8.2 _Lineae calibration models 8.3. Recording and reporting of data0 mology Transfer Performance expectations and acceptance criteria 9.2. Transfer of published methods into. single laboratory 9.3 Comparison of two methods 9.4 Restricted inter-laboratary trials 9.5 Full collaborative tials Appendix: Statistical Tables Selected ublications ofthe AMC 10.1 Statistics Sub-committee 10.2 Instrumental Criteria Sub-committee References Subject Index Contents oT 37 6 8 8 B 9 83 Valid Analytical Methods and Procedures 1 Introduction 1.1 Historical perspective ‘The development of standard methods of analysis hus been a prime objective of the Analytical Division of the Royal Society of Chemistry and its precursors, the Society of Public Anolysis and the Society for Analytical Chemistry, since the earliest of days aad the results of this work have been recorded in the pages of The Analyst since its inception in 1876. An “Analytical Investigation Scheme" was proposed by A, Chasion Chapman in (902, This tater evolved into the Standing Committee on Uniformity of Analytical Methous and was charged ‘with developing slandard chemicals and Securing comparative analyses ofthese standard materials, In 1935, the Committee was renamed the Analytical Methods Committee (AMC) but the main analytical work was carried out by sub-committess composed of analysts with specialised knowledge of the particular application area, The earliest topics selecied for study were milk products, essential ols, soup and tlie determination of metals in food colourants, Later applications included the determination of fluorine, erude fibre, total solids in tomato products, tcade effluents and (eace elements, and vitamins fa anata feeding stulls, These later topics led to the publication of standard methods in a separate booklet. Al! standard and revommended methods were collated and published na volume entitled Bibliograplly of Standard. Tentative and Recommended or R -d Methods of cnalysis in 951. This bibliography was expanded to inelude full details of the ethod under the title Oficut, Standardised and Recommended Methods of Analysisin 1976 with a second edition in 1983 and & thicd edition in 1994. ‘The work of the AMC has continued largely unchanged over the years with ew sub-committees being formed as required and existing ones being dis handed as their work was completed. fa 1995, the Council of the Analytical Division setin place a strategic review of the AMC in view of the changing need for approved analytical methods and the need to develop future direction for the AMC as it moves into the next millennium, ‘The sim of the AMC was reaffirmed to be participation in national and international elforts to establish 4 comprehensive framework for the appro- priate quality in chemical measurements, which is to be cealised by achieving Sive objectives:2 Valid Analytical Methods and Procedures The development, revision and promulgation of validated, standardised and official methods of analysis. ‘¢ The development and establishment of suitable performance criteria for methods and analytical instrumentation/systems. '¢ The use and development of appropriate statistical procedures. ‘¢ The identification and promulgation of best analytical practices including. sampling, equipment, instrumentation and materials. ‘© The generation of validated compositional data of natural products for interpretative purposes. 1.2. Overview of the handbook ‘The objective for any analytical procedure is to enable consistent and reliable data of the appropriate quality to be generatcd by laboratories. Such procedures should be sufficiently well-defined and robust to ensure the best use of resources and to minimise the possibility of expensive large-scale collaborative trials yielding unsatisfactory results through lack of application of best practices. As part of achieving the objectives of the AMC it was felt that such a handbook ‘would enable a consistency of approach to the work of the sub-committees. Recently, major developments in statistical methods have been made parti cularly in the areas of collaborative studies and method validation and robustness testing, [n addition, analytical method development and validation hhave assumed a new importance. However, this handbook is not intended to be a list of statistical procedures but rather a framework of approaches and an indication of where detailed statistical methods may be found. Whilst it is recognised that much of the information required is available in the scientific literature, it is scattered and not in a readily accessible format. in addition, many of the requirements are written in the language of the statistician and it twas felt that a clear concise collation was needed which has been specifically written for the practising analytical chemist. This garnering of existing informa- tion is intended to provide an indication of current best practices in these areas. ‘Where examples are given the intentis to illustrate important points of principle and best practice. This handbook will be brief and pragmatic where possible. Inevitably, this will lead to contentious selections in parts. Consistency of a disciplined approach, however, is deemed more expedient chan always espousing total scientific rigour. 1.3 Purpose and scope ‘The AMC identified the following four main objectives that this handbook should try to satisfy: «Provision ofa unified and disciplined framework that covers all aspects of the validation process from sample and method selection to full collaborative tral Introduction 3 ‘¢ Compilation of a selected bibliography of more detailed and specialist works to be used when appropriate and incorporating the work of the Statistical Sub-committee. « Guidance in the use of the selected statistical procedures for the comparison of methods where circumstances and resources donot permit the meeting of the requirements of the IUPAC protocol. «s Illustration, by way of worked examples, of the main statistical procedures for the calculation, display and reporting of the results. ‘Analy chemists at by nature innovators and seeker ofimprovement. In I eeseeeettiea thee quis are invaluable in optimising method tn eo ae oo ofc, this dee for conimuous improvement spils rate arpnaion of methods for quality conto. Here we requ ae evaton and rigorous cont of procsss and procedures. sae or aeathema for many prctifonets ofthe att chem vfaethis may be sustainable (albeit undesirable) for some spplicstions oe ratory, iipine becomes a neces when methods Bave eee fab between laboratories nan organation When he ope ae eae cs diferent organisation, national boundaries, ee 8 aoe enero assent i cormparabe els arto be oid, Pr aetgocsnocome easily, ai eguires a contol framework, The (rea be considered ksome and unnecessary by some analytical framework ey ay howe from a rewareh envionment. Ic hoped (© cee Pathe doubt is necsity that the sccessel deployment of a we apptatin rly hey on sich an approach and hat Ti techn excelente lone ae inset “easton forthe conhdence nan analytical sult equi that the sample i epresentative and noinogeneous, 1 IRS RUG selene i based upon sound scent principles and has been ‘foun to be robust and reliable forthe sample type under te «Helwteumentaionased has ben quaifed and calibrate 1 SURSn iho ds both competent and adequately trained has caried out the analy, a the gi of the calculation wsed to arcive atthe resul i correct and Sttistialy sound. “This guide is concerned with establishing a control framework for the development and validation ofInboratory-based analytical methods. Many of those pethods wil be employed in generating daa that could have profound egal Oe commercial impacts The validity of anabtical results should be established beyond reasonable doubt. Validation oan anajtzl method is ot single event defied itncrry and stopping places a wel sa inal destination, ‘The goal is method tha satis the original intent. A. dseplined route is Itis a journey with a‘4 Volid Analyicl Methods and Procedures Intradetion Spocinedas stabbed by ° rors seeeretne an i Suite 32 ge senate en | mein epee 25355 ano &3822 : gg538 —_ Figure 1 150 °V' model adopied for analytical method validation required which maps out the validation journey, more frequently called the validation process. The ISO °V* model for system development life cycle in computer software validation is a structured description of such a process. In this instance, the basic ‘V' model has been adapted for analytical method validation and is showa in Figure | Like all models, there are underlying assumptions, The main ones for analytical method validation include the areas of equipment qualification and the integrity of the calibration model chosen. If the raw analytical data are produceé by equipment that has not been ealibrated o¢ nat shown to perform reliably under the conditions of use, measurement integrity may be severely compromised, Equally, ifthe calibration model and its associated calculation methods chosen do not adequately describe the data generated then it is inappropriate to use it. These (Wo areas are considered in some detail in Chapter 8 Euch layer of the {SO *V' model is dependent upoa the layer below and represents stages in the process. Broadly speaking, the boxes in the leftchand ection ofthe V" model represent che aims and objectives ofthe validation, The boxes in the right-hand portion of the 'V" model contain the processes and procedures that must be earied out successfully and be properly documented to demonstrate that these specified aims and objectives have been met. At the fulcrum of the model is the development process itself, At each level of tke model there is a horizontal correspondence between the two boxes, Verification of the matching of these pairs provides a method of closing the oop at each level For example, at the highest level, conformance to the user requirements specification may be verified through data generated in house, through limited laboratory tals or through use of the full JUPAC harmonised protocol. What is critical here is the confirmation of the original user requirements under appropriate performance conditions (Figure 2) (osen REQUIREMENTS ‘SPECICATION6 Valid Anaytcel Methods and Procedures ‘One useful approach to visualising these relationships isto list bullet points for each of the pairs in the manner shown below. In this way key areas are identified although there are not corresponding relationships between indivi- dual bullet points. Individual elements of the model are covered more fully in Chapter 7 where method validation is considered as a whole, Specified as Established by ‘+ Method applicability + Sclectivity/specifcity ‘+ Analytes to be quantified « Linearity f= Ranges or limits specified # Accuracy # Methodology to be used # Repeatability ‘= Sampling considerations + Within-laboratory repeatability 1+ Matrices to be covered Reproducibility ete © Recovery # Robustness Chapter 8 outlines basic aspects of data evaluation and manipulation. The important topic of linear calibration models is covered in some detail Recommended procedures for comparing methads and for taking a single method through to a full IUPAC collaborative trial with the harmonised protocol are covered in Chapter 9. Chapter 10 is a bibliography of recom- ‘mended books and papers that should be consulted for more details in specific 2 Nomenclature: Terms and Parameters 2.1 Introduction To avoid confusion, the terms and parameters used in the validation of methods, for example, as used in Figure 3, must be clearly and unambiguously defined. This glossary contains the recommended definitions and corresponding descriptions and is based on the various standards and publications summarised in the Bibliography.’ This is not exhaustive and it is recommended that the IUPAC ‘Orange Book™ be consulted if require. 2.2 Terms 2.2.1 Analyte Component or group of components of which the presence/absence of mass Fraction/concentration is to be determined in the test sample. Nomenclature: Terms and Parameters 7 Lateatory Sample Measurement andlor ‘chemical operations Figure 3 Flowchart of sample nomenclature 2.2.2 Analysis ‘The method used in the detection, identification andfor determination of the analyte ina sample. 2.2.3 Laboratory sample ‘The semple or sub-sample(s) of the bulk of the material under consideration sent to ot received by the laboratory. 2.24 Test sample ‘A cepresentative quantity of material, obtained from the laboratory sample ‘hich is eepresentative for the composition of the laboratory sample.8 Valid Aveytical Methods and Procedures 2.2.5 Test portion ‘The representative quantity of material of proper size for the measurement of the concentration or other property of interest, removed from the les! sample, weighed and used in single devermination. 2.2.6 Observed value The result of a single performance of the analysis procedure/method, starting With one test portion and ending with one observed value or test result. Note that the observed value may be the average of several measured values on the test portion (2.2.5) ria the test solution (2.2.8) or aliquots (2.2.9). 2.2.7 Test result ‘The result of a complete test (frequently a combination of observed values). 2.2.8 Test solution The solution resulting from dissolving the test portion and treating it according to the analytical procedure. The test solution may be used directly to determine the presence/absence or the mass fraction or mass concentration of the analyte without attributable sampling ecror. Alternatively, an aliquot (2.2.9) may be used. 2.2.9 Aliquot A known volume fraction of the test solution (2.2.8) used directly to determine the presence/absence or the mays fraction/eoncentration of the analyte without attributable sampling error. 2.2.10 Detection “The determination of the presence of the analyte as a chemical entity. 2.2.11 Determination (quanification) ‘The determination of the absolute quantity ofthe analyte (mass, volume, mole) or the relative amount of the analyle (mass fraction, mass concentration) in the test sample. Nomenclaure: Teams and Parameters 9 2.2.12 Content mass fraction The fraction of the anaiyee in the test summple. The mass fraction is a dimension- less numiber, However, the mass fraction is usually reported as a quotient of two smass-units or mass-volume, Value Mass foction (Stunts) Nan Stunts (olqror mis) mgg-'smgmb-! g ke pm, parts per million ne mL~', yg kg"! ppb, pars per bilion pee”! pemi7!ngke™' ng L7 pt, parts per tilion 2.2.13 Mass concentration ‘The concentration expressed as the mass of the analyte in the test solution divided by the volume of the test solution. The term mass fraction should be used ifthe amount of the analyte is related to the mass ofthe sample. 2.114 Matrix All components of the test sumple excluding the analyte 2.3 Parameters 23.1 Standard deviation(s) ‘A measure of the spread in the observed values as a result of random errors (23.12). These observed valves all have the same expected value. The equation SEE] « in which x= individual measured value, = mean measured value, ‘n= number of measurements Equation (1) applies to the calculation of 5, (2.3.74, (2.3.8) and 4— (2.3.9) under the measurement conditions specified therein. La = 80 Vall Analytica! Methods and Procedires 2.3.2 Relative standard deviation(s) (RSD) ‘The standard deviation(s) expressed as a percentage of the mean value. The relative standard deviation is defined as: RSD = 100% @ x 2.3.3 Detection limit ‘The calculated amount of the analyte in the sample, which according (0 the calibration line, corresponds to a signal equal to three times the standard deviation of 20 representative blank samples. A blank sample isa sample which does not contain the analyte. Ifthe recovery (2.3.21) of the analyte is ess than 100%, ideally the detection limit should be corrected for the average recovery. However. this is a con- {entious issue and needs o be considered carefully for each method. 2.3.4 Limit of quantification ‘The minimum content of the analyte in the test portion that can be quantita- lively determined with a reasonable statistical confidence when applying the analytical procedure. © Report the limit of quantification either in absolute quantities of the analyte (mass, volume or moic) relative amount of the analyte [mass Fraction (2.2.12) or mass concentration; (2.2.13)} 4s The amount of test portion (For example in grams) .iust be reported as used in the determination. “The limit of quantification is numerically equivalent to six times the standard deviation of the measured unit when applying the analytical procedure to 20 representative blank samples. For recoveries less than 100% the limit of {quantification must be corrected for the average recovery of the analyte. 2.3.5 Sensitivity ‘The change of the measured signal as a result of one unit change in the content of the analyte. ‘The change is calculated from the slope of the calibration line ofthe analyte. 2.3.6 Rounding off The process of achiving agteement tetsen an observed valve and the repeatability (2.3.7) of the analytical procedure. The maximum rounding o ee a ge iargst decimal unit determined (oe smaller than bal Nomenclature: Terms and Parameters u the value of the standard deviation of the repeatability (2.3.7). See Section 8.3.1 for more details. 23.7 Repeatability (r) ‘The expected maximum difference between two results obtained by repeated application of the analytical procedure to an identical test sample under identical conditions. ‘The measure for epeatability (7 isthe standard deviation (s,). For seriss of measurements of a sufficient size (usually not less than 6), the repeatability is defined as 1 = 28 x 5, (confidence level 95%) ® Repeatability should be obtained by the same operator with the same equip- ‘ment in the same laboratory atthe same time or within a short interval using the ssame method. 23.8 Within-laboratory reproducibility (Rw) ‘The expected maximum difference between two results obtained by repeated application of the analytical procedure to an identical test sample under diffeent conditions but in the same laboratory. The measure for the within- laboratory reproducibility (R.) is the standard deviation (sa. For series of measurements of sufficient size (usually not less than 6), the Wwithin-laboratory reproducibility is defined as Re 8 x se, (Confidence level 95%) ® Within-laboratory reproducibility should be obtained by one or several opera tors with the same equipment in the same laboratory at different days using the same method, 23.9 Reproducibility (R) ‘The expected maximal difference between two results obtained by repeated application of the analytical procedure to an identical test sample in different laboratories. The measure for the reproducibility (R) is the standard deviation om) For series of measurements of suficient size (usually not lest than 6) the reproducibility is defined as, R= 8 x 54 (confidence level 95%) 0 Between-laboratory reproducibility should be obtained by different operators2 Valid Analytical Methods and Procedures with diffeent instrumentation in different laboratories on diflerent days using the same method. For a given method, the most important factors in the determination of repeatability and reproducibility are Laboratory, Time, Analyst and Insteu- mentation. OO Experimenval condition to determine Faclorso vary or control Repeatability Same LTA] Within-laboratory eproducitil Same Li diferent T; Land A may be diferent Betweee-aboratory reproducibilty —Diflereat LT. A, ew eO—OEoeorT Ii isnot possible to involve additional taboratories for the determination of the between-laboratory reproducibility, then the within-laboratory ceproduci- bility may be used to get an estimate ofthe between-laboratory reproducibility ‘The reproducibility of the method may be dependent upon the mass fraction of the analyte in the test sample. It is therefore recommended, when studying the reproducibility, to investigate whether a relation exists between concentration ‘and ceproducibility. The measurement series should be greater than 8 2.3.10 Trueness ‘The closeness of agreement between the average value obtained from a large series of test results and an accepted reference value. The measure of trueness is usually expressed in terms of bias. 2.3.11 Systematic error or bias “The difference between the average observed value, obtained from s large series of observed values (n >8), and the true value (2.13) (Figure 4) 2.3.12. Random error “The difference between a single observed value and the average v>lue of large rhumber of observed values (atleast 8), obtained by applying the same analytical procedure to the same homogeneous test sample. 2.3.13 True value ‘The value that describes the content and is completely defined by the circum stances under which the content has been determined, 2.3.14 Precision A measure of the agreement between observed values obtained by repeated application of the same analytical procedure under documented conditions @3. Momenclaare: Terms and Parameters Bb Messued ase (Reauacy= maneaesvabe rosie sates et Figure 4 Accuracy. preelsin ond bus 23.15 Accuracy |A measure of the agreement between a single analytical result and the true value 23.16 Ruggedness Ruggedness of an analytical method is the insensibility of the method for varistions in the circumstance and the method variables during execution. It is important to note that statistically significant deviations are not always relevant. The purpose for which the measurement is made is a more important criterion for deciding on the ruggedness and the statistical method employed is merely a tool. 2.3.17 Noise ‘A phenomenon defined as fast changes in the intensity and frequency of @ measured signal irrespective ofthe presence or absence of the analyte. The speed fof change is significantly diflerent from the normally expected detector response. A measure of noise is the measured difference between the highest and lowest value of the measured signal with no analyte present, observed in a relatively short timespan, as compared to the time-span necessary for measure: ‘ment of the analyte 2.3.18 Selectivity ‘A measure of the discriminating power of a given analytical procedure in Uiffecentiating between the analyte and other components in the test sample.“ Valid Anotsvcat Metods and Procedures 2.3.19 Significant figures ‘Yalues which contain the information consistent with either the repeatability or eproducibility of the analytical procedure. Significant values are obtained by using the described method for rounding off (Section 8.3.1) 2.3.20 Specificity (see also Selectivity) “The property ofthe analytical procedure to measure only that which is intended 'o be measured. The method should not respond to any other property of the analyte or other materials present. 3.21 Recovery “The fraction of the analyte determined in a blank test simple oF test portion, ater spiking with a known quantity of the analyte under predefined conditions # Recovery is expressed as a percentage © The part of the analytical procedure in which recovery is involved should be reported. ‘© The critical stages/phases relating to instability, inhomogeneity, chemical conversions, difficult extractions, etc. should be reported. © Recovery must not be based on an internal standard unless work is undertaken to demonstrate identical behaviour under the conditions of test. 2.3.22 Scope ‘The collection of matrices and analyte(s) to which the analytical procedure is applicable 2.3.23 Range ‘The content mass fraction interval to which the analytical procedure is applicable. 2.3.24 Drift ‘The phenomenon observed as a continuous (increasing or decreasing) change (lowly in time) of the measured signal in the absence of the analyte. Samples end Sampling is 3 Samples and Sampling 3.1 Introduc ‘The importance of sampling in method validation and, in particular, inter- ‘comparison of methods cannot be overemphasised. If the tes! portion is not representative of the original material, it will not be possibie to relate the analytical result measured to that in the original material, no matter how good the anolytical method is nor how carefully the analysis is zerformed. It is essential that the laboratory sample is taken from a homogeneous bulk sample asa collaborator who reports an outlying value may claim receipt of a defective laboratory sample. It is important to understand that sampling is always an error generating process and that although the reported result may be depen- dent upon the analytical method, it will cays be dependent upon the sampling process. ‘The essential question in the inter-comparison of analytical methods is, "Ifthe ‘same sample (ora set of identical aliquots of a sample) is analysed by the same method in diferent laboratories, are the results obtained the same within the limits of experimental error”. I is apparent, therefore, thatthe selection of an appropriate sample or samples is critical to this question and that the sampiing stage should be carried out by a skilled sampler with an undeistanding of the overall context ofthe analysis and trial ‘Any evaluation procedure must cover the range of sample types for which the method under investigation is suitable, and details ofits applicability in terms of sample matrix and concentration range must be made clear. Similarly, any restrictions in the applicability of the technique should be documented in the method, For more details, the works listed inthe Bibliography should be consulted. In particular, Crosby and Patel's General Principles of Good Sampling Pract.” and Prichard* provide readily digestible guidance to current best practices in this area, 3.2, What is a sample? ‘The Commission on Analytical Nomenclature of the Analytical Chemistry Division of the International Union of Pure and Applied Chemistry has pointed out that confusion and ambiguity can arise around the use of the term “sample” and recommends that its use is confined to its statistical concept. When being used to describe the material under analysis, the term should be qualified by the use of laboratory sample’ or ‘test sample’, for example. ‘One of the best treatments of sampling terminology i given in recommenda- tions published by IUPAC? which describes the terms used in the sampling of bulk or packaged goods. In this example, the sampling procecure reduces the original consignment theough (ots or batches, increments, primary ot gross samples, composite or oggregate samples, subsamples ot secondary samples to a laboratory sample, The leborotory sample, i€ heterogeneous, may be further16 Volid Analyical Mesos and Procedures prepared to praduce te ‘es sample, Actival at either a laboratory sample or test sample is deemed to be the end of the sampling procedure. ‘Once received into the laboratory the laboratory samples or test samples be recorded and then be subjected to analytical operations, beginning with the measuring out of a test portion and proceeding through various operations to the final measurement and reporting of results/indings. ‘The IUPAC nomenclature for the sampling process is lustesced in Figuce 5 ‘This links with the sampling nomenclature diagram on Page 8 (Figure 3). The problems associated with sampling in many areas of chemical testing have been addressed and methods have been validated and published (see ref 3 for more details). Where specific methods are not avaiable, the analytical ‘chemist should rely upon experience or adapt methods from similar applica tions. When in doubt, the material of interest and any samples taken {eam it Testa Figure $ JUPAC umping process Samples and Sampling a should always be treated as heterogeneous. [is important when documenting a sampling procedure to ensure that all of the terms are clearly defined, 0 thatthe procedure will be clear to other users. The use of sampling plans may be appropriate and guidance is available for procedures based upon attributes or variables.* 3.3 Homogeneity and concentration ranges Extreme care must be taken to ensure that the bulk sample from which the laboratory or test samples are taken is stable and homogeneous—this is particularly important if'spiked” samples are provided. ‘The homogeneity should be eseablished by testing a representative number of laboratory samples taken at random using either the proposed method of analysis or other appropriate tests such as UV absorption, refractive index, etc. ‘The penalty for inhomogeneity is an increased variance in analytical results that is nat due to intrinsic method variability, For quantitative analysis the working range for a method is determined by examining samples with diffrent analyte concentrations and determining the concentration range for which acceptable accuracy and precision can be achieved. The working range is generally more extensive than the linear range, which is determined by the analysis of a number of samples of varying analyte concentrations and calculating the regression from the results (see Section 8.2 for more details), For a comprehensive study, which has been designed 10 evaluate the method fully, samples postessing low, medium and high concentea- tion levels ofthe analyte to be determined must be prepared. The only exception (0 this would be when the level of the analyte always falls within 2 ngceow range ‘oF euncenteations 4 Method Selection 4.1. ‘Fitness for purpose’ Far too often, method selection is carried out by Jeviding to apply the technique. that is most popular of familiar. Ifa laboratory has expertxe in a particular technique then it is empting (0 fet that expertise be the overtiding factor in riethod selection. Rarely is there a structured and considered approach to method selection. Whilst itis often possible to make inappropriate methods work within a single laboratory, the impact on the reliable transfer between laboratories can be very large. In the past, che transferability of methods has not been given the prominence it deserves. However, within the current climate of harmonisation and interchangeability, the technical requirements of mechod transfer and method performance have been addressed in some detail and are covered in Chapter 9. There are two areas which have received less attention and agreement, namely the inter-comparison of different methods for the same analytes in-house or within a few laboratories and the methods for describing and writing analytical methods. The former topic isthe subject of Section 9.318 Velid Anolytical Methods and Procedures “The latter is discussed in Section 7.2. No method is ‘fit for purpose" unless there are clear and unambjguous written instructions for carrying out the prescribed testing in accordance with the conditions laid down in the original method development cycle. ‘The literature contains examples of collaborative trials that only prove that the method was not fit for its intended purpose! The full IUPAC harmonised protocol is by its very nature an extensive and expensive exercise. From an Cconomie perspective such trials should only be undertaken when there is good and well-documented evidence that it is ikely thatthe method under evaluation is sufficiently robust. Investment of time and intelectual effort in method selection and the other aspects of the user requirements specification will pay treat dividends. Prevention is better and nearly always cheaper than cure. 4.2. Sources and strategies Once the User Requirements Specification has been drawn up and the method performance criteria set, the method development process can begin. Quite Often there are existing methods available within the literature or within trade and industry. On many occasions itis tempting to ignore the difficulties of 2 comprehensive literature search to save time, However, as a minimum, key word searches theough the primary literature and abstracting journals such as “Analytical Abstracts and Chemical Absinacts should be undertaken. For Standard of statutory methods, itis essential to scan international standards from Europe aad the USA as well as local sources and those deriving from statutory publications. Once existing methods have been identified, it is good practice to compare them objectively. One way to do this is to lst the Derformance criteria and relevant sections of the User Requirements Specifica- tion and tabulate the corresponding data. “An existing method may have a sufficiently good fit that adaptation is likely to lead to a suitable method, This relies upon professional knowledge and experience For methods that are likely to be widely used, other aspects of suitability need to beconsidered. Some areas for consideration are listed below. ‘= Can the method be written down sufficiently clearly and concisely to allow case of transfer? ‘+ Can all the critical method parameters be identified and controlled? This is particularly important where automated systems are involved. Is the equipment readily available to all the likely participants? This assumes a special importance for internationally distributed methods and may involve questions of maintenance and support. fe Are all the reagents and solvents readily available in the appropriate quality? © Do the staff have the requisite skills and taining to carry out the procedure? Method Selection 1» ‘e Are Health and Safety or environmental considerations likely to cause problems? 1» Are standards and reference materials readily available to ensure that equipment and systems are properly calibrated and qualified? 4.3. Sampling considerations In the enthusiasm for a particular technique or method, itis sometimes the case that the appropriateness of sample size is overlooked. Even though the sampling, process outlined in Chapter 3 has been followed, itis essential thatthe sizeof the test sample and its relationship to the sizes of the test portion and the test solution are considered. These factors need to be considered during method selection. For example, sa 2g test sample from a 1000 kg consigament or bulk batch adequate for the purpose? It may be under appropriate circumstances. If not, hhow much material needs 1a be taken? Recent draft guidance to the pharma- ceutical industry from the FDA" recommends that for blend uniformity sample sizes no more than three times the weight ofan individual dose should be taken. Equally, consideration needs to be given to sample presentation. Is it more appropriate to test non-destructively o gain physical and chemical information oF by solution/extraction processes to separate the analyte(s) of interest? ‘The most important aspect here is that these questions have been asked and documented answers given as part of the User Requirements Specification. It is essentisl to remember that whilst any test result may be method- dependent its always sample-dependent. 4.4 Matrix effects [As far as is practically possible, the selection and preparation of samples must take into account all possible variations in the matrix of the material to be analysed. The apnlicability of the method should be studied using various samples ranging from pure standards to mixtures with complex matrices as these may contain substances that interfere to a greater or lesser extent with the quantitative determination of an analyte or the accurate measurement of a parameter. Matrix effets can both reduce and enhance analytical signals and ‘may also act as a barrier to recovery of the analyte from a sample. ‘Where matrix interferences exist, the method should ideally be validated using a matched matrix certified reference material. If such a material is not available it may be acceptable to use a sample spiked with a known amount of the standard material ‘The measurement of the recoveries of analyte added to matrices of interest is ‘used to measure the bias of a method (systematic error) although care must be taken when evaluating the results of recovery experiments as it is possible t0 ‘obtain 100% recovery of the added standard without fully extracting the analyte which may be bound in the sample matcx. “The whole question of recovery adjustment is a vered one. In theory, one2 Valid Analytical Methods and Procedures should always carry out this correction. However, the best anuytica) practice is to consider the question for each application and sample matrix combination and make and document tke decision. For more detailed information, the fecently published "Harmonised Guidelines for the Use of Recovery Informa: {ion in Analytical Measuresnents™ should be consulted 5 Equipment Calibration and Quatification ‘Analyceat practitioners pace great fuith in the seadings and outpu‘s from their nsteuments, When unexpected or out of specification results occur, the initial suspieian often fais on the sample, the preparation technique or the analytical Standard employed. Rarely is the equipment questioned. Indeed, the whole ‘underpinning of method validation assumes that the analytical equipment used to acquire the experimental data is operating cortectly and reliably. Some industries that are highly regulated, such as the pharmaceutical sector, have placed great emphasis on method validation in, for example, HPLC.”"" However, until cently, there has been itl direct requirement for assuring thet the analytical instruments are working properly “The major regulatory guidelines for Good Manufacturing Practice (GMP) and Good Laboracory Practice (GLP) are similarly vague. ‘Fiiness for purpose js the phrase that is commonly used, but what does this mean in practice? Primarily, the Phormacopotias"' and the Australian Regulatory Authority"? have been sufficiently worried by instrumental factors to give written require: zebis for instrument performance. Whilst these guidelines are not consistent at Teast they are attempting (© ensure consistent calibration practices between laboratories. In contrast, the ISO Guide 25 approach (updated in 1999 to 1SO Guide 17025), a8 expanded in ref. 14 heavily focuses on good analytical practices and adequite calibration of instruments with nationaily or internationally traceable standards wherever possible 5.1 Qualification approaches Equipment qualification is an essential part of quality assuring the analytical data on which our knowledge of the sample rests. The importance of this “data to information iceberg’ is illustrated in Figure 6, There are several approaches commonly employed. Sut The ‘bottom up’ approach ‘The ‘bottom up" approach ensures the quality of the end result by building up from the foundations rather ike a Lege model.fn testing cerms this is ttusecated in Figuce 7. These Lego bricks are equivalent to the individual modules in any measurement system. Each brick is quatified or conficmed as suitable for use before the next layer is bull, Ln this way, integrity is assured all the way to the Equipment Calibration ond Quetficaion auipment Cali a Quai a4 ‘itqens for rpc! Knowed Derived tem combining Vala information ali aformasion: Derived from Good Data from Vatdvtd Sortware 2008 bats: Derived trom Informatio pil Preeessgica ry SGood PRorrect Data.gy) Samal Figure 7 The ‘bottom up" approach ;op-most layer. If firm foundations are not built, she information generated will ot stand serutiny. By following tis approach qui is From the lowest soe othe in providing hint dain o ee te el racer can hve ath ne ela te Ha nae la win preeminent en ener ered wil be worse than ult, Reliab of te daa quality sho linked to pecformyace standards far both modules anc sas wel aa a regular maintenance prog, 5.1.2 The ‘top down’ approach ‘An alternative and increasingly applied approach, particularly from the regulatory bodies, is from the other dizection, i. “top down’, This approach is Known as the 4s mode!, DQ, 1Q, OQ and PQ which are:n Valid Analytica! Methods and Procedures Design Qualification Installation Qualification Operational Qualification and Performance Qualification ‘A detailed discussion of this approach may be found in refs. 15-19 and references therein. By way of example, however, the approach wil be illustrated swith respect tea typical analytical insteument, Design Qualification, DQ, is about specifying what the instrument or instrument system has to do. This would include documenting technical fequirements, environmental conditions, sample and sample presentation requifements, data acquisition and presentation needs, operability factors and ‘any Health & Safety issues. In addition a cost-benefit analysis would normally be performed. ‘The Instrumental Criteria Sub-committee of the Analytical Methods Com- mittee has been active for many years in producing Guidelines for the Evaluation of Analytical Instrumentation, Since 1984, they have produced reports on atomic absorption, ICP, X-ray spectrometers, GLC, HPLC, ICP- MS, molecular fluorescence, UV-Vis-NIR, IR and CE. These are excellent source documents to facilitate the equipment qualification process. A current fisting of these publications is given in Section 10.2. Having chosen the analytical instrument or system, Installation Qualifca~ tion, 1Q, should be carried out to ensure that the equipment works the way the vendor or manufacturer specifies it should. IQ should be performed in accordance with a written test protocol with acceptance criteria with certifica- tion from the installation engineer, who is suitably qualified. Full waitten records ofall testing carried out should be maintained as well as ensuring that adequate documentation and manuals have been supplied. The latter should include any Health & Safety information from vendor or manufacturer. ‘Once satisfed that the instrument is operating in accordance with its own specification, the end user should ensure that itis “ht for purpose’ for the Applications intended. This step is called Operational Qualification, OQ. This, process would include writing the Standard Operating Procedure (SOP) and training staff in ils use. Further testing may be required to ensure that the instrument performance is in accordance with National and Corporate stan- dards ifnot carried out in 1Q, Frequently, instruments are used with accessories for sub-systems, e.g. sipper systems or other sample presentation devices Challenge the analytical system with known standards and record what you id. Ieis necessary to ensure that they work in the way intended and that documented evidence is available to support thei use. ‘Calibration procedutes and test methods and frequencies need to be defined usually as part of an SOP. Ifyou intend to transfer data [rom the instrument to 1B software package, ensure that data integrity is preserved during transfer Don’t assume that the transfer protocols on ‘standard’ interfaces always work as intended, Itis good practice to ensure that the data have not been truncated of distorted during transfer. Equipment Calibration and Qualification B [AU this point in the process, the equipment/system is able to be put into routine use, The final Q in the model, Performance Qualification, PQ, is about on-going compliance. Elements of PQ include a regular service programme, performance monitoring with warning and action limits (as defined in OQ). All Of these elements need to be documented and individual log books for systems are useful for this purpose. PQ data should be subject to regular peer review. All instrument systems should be subject to a simple change procedure which may well be connected to the equipment log system. 5.13 Holistic approach Furman er al,'7 discussing the validation of computerised liquid chromato- ‘graphic systems, present the concept of modular and holistic qualification. Modular qualification involves the individual components of a system such a5. ump, autosampler, columa heater and detector of an HPLC. The authors make the point that “calibration ofeach module may be useful for trouble shooting purposes, such tests alone cannot guarantee the accuracy and precision of analytical results. Therefore the authors introduced the concept of holistic validation where the ‘whole chromatographic system was also qualified to evaluate the performance Of the system. The concept of holistic qualification is important as some Taboratories operate with a policy of modular equipment purchase. Here they select components with the best or optimum performance from any manufac turer. Furthermore, some of these laboratories may swap components when they malfunction. Thus, over time the composition of a system may change. ‘Therefe >, to assure themselves and any regulatory bodies that the system Continues to function correctly, holistic qualification is vital Most laboratory systems require maintenance and inclusion preventative maintenance programmes. Therefore any holistic testing should form part of Performance Qualification to ensure on-going compliance. 5.2 A convergence of ideas Much in the way of harmonisation of procedures and practices in analytical chemistry has been going on outside these activities. Many of these initiatives, are now coming to fruition. CITAC (Co-operation on International Tracs- ability in Analytical Chemistry) have produced an International Guide 10 Quality in Anaiylical Chemisiry'® which attempis to harmonise the following regulatory codes of practieefor the analytical laboratory: ISO Guide 25 (revised in December 1999 to ISO Guide 17025), ISO 9001 and 9002 and GLP. A VAM Irsrumenaton Working Group hs puted Gere or Exumen Quali. fication of Analytical Instruments: High Performance Liquid Chromatography featon of Arai igh Perfo iquid Chromarography IF there is one compelling reason for equipment qualification, i ies within theu Volid Analytical Methods and Procedures need to transfer methods between laboratories. Why are so many of our collaborative trials a failure? One answer lies in the fact that the key analytical variables are not always identified and controlled through specification and/or procedural practice. These may le within the method but more often are due to the operating parameters ofthe equipment or system. If, for example, tempera ture is a key lactor, how can it be specified if there is no assurance that instrument A’s temperature readout is operating within known accuracy and precision limits? Furthermore ifa laboratory is transferring a method involving an HPLC gradient separation and there is no equipment specification at the level ofthe pump, there may be problems in the technology transfer. Considera- tion needs to be given to the effects of choosing high pressure versus tow pressure solvent mixing and differences in dead volume between the pump and column which can affect the gradient formation. These factors are likely to affect che quality of the separation achieved. Without specification there cun be ‘no reliable control. Another reason may be that the overall analytical process capability is affected by one or more instrumental factors. Methods developed fon unqualified equipment or systems may well lack the robustaess and reliability needed. Calibration is often confused with qualification. As pointed out by Parriot with reference to HPLC methods: 2 “The term calibration implies that adjustments can be made to bring a system into a state of proper function. Such adjustments generally cannot be performed by chromatographers and are best left o trained service engineers Who work for, oF suppor, the instrument manufacturers.” Calibration is, therefore, inextricably linked to equipment qualification and preventative maintenance, Whenever calibration involves adjustments of the type described above, it is important to document the activity and where appropriate re-qualify the instrument cone. ne 6 The Method Development Process ‘The overall process (rom concept to validated method is illustrated on Page 4 (Figure 1). Once an appropriate analytical principle has been selected and the method performance eriteria defined, the actual method development process can begin. Usually, this phase is carried out using pure materials and limited samples that are known, or assumed, to be homogeneous. ‘The purpose of this process is to confirm the viability of the method chosen. ‘and show that the procedure is sufficiently analytically robust to allow a preliminary validation to be carried out. The AOAC collaborative study uidelines* explicitly state "Do not conduct collaborative study with an unoptimized method. An unsuccessful study wastes a tremendous amount of collaborators” time and The Method Development Process ‘analytleal factors PaOcess eye Figure 8 Method developnValid Analytical Methods and Procedures 400 10>? (0.1%) 566 tov 300 ton ret 10-* (opm) 1600 to"? 26 m0 535 6400 50st 12800 When the value of RSD and the mean-value of the analyte concentration have been established, the Horwitz ratio, HORRAT, can be calculated. as) Technology Transfer 9° For example, ifa collaborative trial produced a value for RSDeterecs 019.21 For a mean analyte concentration of 0.00953% then RSDonu 9.21 RSDine ~ 06 ~ HORRAT = I is now generally accepted practice that HORRAT values of 2 or less indicate that the method is of adequate precision. Hence, in our example, this is clearly 9.2 Transfer of published methods into a single laboratory The Nordic Committee on Food Analysis has published guideline on the “Validation of Chemical Analytical Methods which differentiates between ‘extecnal validation work carried out on published methods and that work required to transfer it into the working Inboratory to confirm its suitability for use. Table 20 is adapted from this document. This document is most easily accessed from a recently published book.” I certified ceference materials are available they should be used to confirm the verification of trueness. These guidelines could also apply to intra-laboratory training programmes ‘The overall validation requicements given in the guideline of the Nordic Committee on Food Analysis on the “Validation of Chemical Analytical Methods” ace listed in Table 12, ‘Table 20 Technology transfer validation requirements for publizhed anelytical Degree of external validation ‘Recommended internal validation ‘The method is externally validated ina Verification of trueness and precion ‘method-performaace study “The method is externally validated ina Verification of teveness and prestion and Imethod-performance study buts to be where appropriate the detection limit ‘sed with anew sample matrix or using The method i wel established but More extensive valida tion/verifation of untested ey parameters The method is published in the sciemlife. More extensive validaion/verifcation of Titerature and gives important key parameters petTormance characteristics ‘The method is published in the sciatic The method should be subject toll Tterature but without important validation performance characterisesry Volid Analytical Methods an Procedures 9.3 Comparison of two methods (One of the most common tasks in the analytical laboratory is to determine whether or not two methods give comparable mean results within a predeter- ‘mined confidence level. The test usually chosen for this task isthe (est ‘There are two different ways of cartying out this test. The first one involves taking a single sample and analysing it by both methods a number of times. The sual procedure is (o undertake a number of analyses (prelerably not les than 6) for the chosen sample with both methods and calculate the value of the ¢- slatistic. This is then compared with the tabular value for the appropriate degrees of freedom at the selected confidence level, I the calculated value is less than the tabulated 1 value then the mean values, and hence the methods, are accounted equivalent. This method has the advantage thot the number of replicates undertaken for each method does not have to be equal. However, it is not always recognised that for this test to be valid the precision of the two methods should be equal. The method used to compare the precisions of methods isthe F-ratio test and is carried out as part of the procedure, ‘The process may be illustrated with the following example. Table 21 gives some data from twa methods, B and C, for estimating the silver content in photographic media. Method Bis based upon an X-ray Auorescence procedure and Method Cis a controlled potentiat coulometrie method. ‘The steps and calculations areas follows, (1) Calculate the differences and the square of the differences for each of the ‘mean values of the data (n, data poinis from C and m; data points from B) feom Methods C and B respectively Table 21 t-Test comparison of means from two methods usivig a single sample Experioenial result ‘Squares of deviations Sunple no, Method C__Method B Mettod Methort @ t 03037 osm 0.000669 .a00010 2 oa ons 0.000843 o.000006 3 oxsrz oss 9.000006 0,000000 4 0.8556 one 0.900016 0.090008 5 o.ay2 o.sans (0900859 .000s01 ‘ 0.3862 O41 ‘0.000710 0.000384 7 oars gan 0.000175 0.000512 8 09678 o8s7s 0.000068 © 0.000535 9 08517 O08 0.000062 0.000380 0 0.8532 assez 0.000040 0.000234 u 0.99 0.900360 2 0.8958 000619 Som B96 104507 0.002947 0.003850 0.8596 0.8709 0.0003275 __0.0003227 Mean Variance Teclwology Transfer 6 (2) Calculate the sums of these squares. soa {G) Galeulate the variances of Methods C and B (s] and 33) by dividing the ssum of squares by the degrees of freedom (ny ~ land nz ~ | respectively) (4) Caloolate the F ratio by dividing the larger variance by the smaller. tn this instance variance of C divided by variance of B. (5) Compace this value with the tabular valve (one-sided) for F at 95% confidence for 9 and 11 degrees of freedom. (6) Calculate the pooled varianee, allt = DH = DL 6 utn= 3 . (1) Caleutate the absolute value for ¢ from a where 1 has (m+ tg ~ 2) degrees of freedom. (8) Compare this value with the tabuluted values of ¢ for (1 +m —2) degrees of Freedom (given in Tuble 22) Aste tabula valucenceeds the caleulated value, then we an conclude that chemeans are no sinifleualy diferent a 95% conlence Soe ey ot miking this comparison visual ist generate pictures of Jitliaye fom the mean and werane dts for Mehous and C and sanerinposig their date points This assumes that the data re normally SiReibuter but it hasbeen shown tha hs generally the cose These are saat gure). The depres of overnp high, an, importantly, the shapes SI aleibtions are very se, Th visual representation confirms the ead supports the requirement for approximately equal ¥ae- istic fi However, there are 0c ‘Another method, D, was used to determine the silver content of the photo- sasions when vasiance equivalence is not observed. ‘Table 22. Results from (test comparison of means ‘From 10 methods using a single sample vee F ratio of variances ealeulated “ solos P0053 { onesies » Posi vanances? oni Scndrd dont. coven ane # Statistic ‘Tabular 40.05.20) 2.09e Valid snalyncal Methods and Procedures Comparison of Methods 8 and C | Siletan oo, Frequeey Concentration at anya fg mt") Figure 33 Doro deributions for Methods B and Cand weir dava pains graphic materials which was based on high precision automated entulsion densitometry. The equivalent picture to Figure 33 for this comparison is shown in Figuce 34 ‘An F ralio (est is hardly needed to decide whether the variances are different! A eesti still applicwule but needs to be modified to take into account the variance differences. This is done by calculating the effective number of degrees of freedom using Sutterthwaite's method.®” This is still an area which is controversial and a number of differing approuches and equations fave been proposes. ‘Table 23 shows the data tabulations for Methods C and D (n= 10 and 2, respectively), analogous to Table 21 for Methods and C. Note the large value Tor the F ratio indicating highly significant differences between the variances ofthe two data sets. ‘The steps and calculations For the unequal variance calculation areas Follows. (1) Cateulate the sums and means for the fist two eolunins. {Q) Caleulate the squared differences between the mean for each method and the individual values and the sums for the second (wo columns. 0) Calculate the variances for Methods C and D, Ye and Yo, by dividing the sums for columns 3 and 4 by (the umber of data points ~ 1), in this example, 9 and 11 (4) Calculate the F eatio by dividing the laegee vari Yel¥n nce by the smaller, Techuology Transfer 6 Comparison of Methods C and O aan for meted 0 0.881 Sis devaton 0.0026 Freaveney Mean for mato ¢ 0.860 Concentration of analyte (mg ri) Figure 34 Dave ustributions for Meds Cond D an reir dare points ‘Table 23 Test calculations where the variances are known 10 be unequal (Satterthwoite's method) Experimental result Squares of deviations Sorple no. Method Method 8 Method Method B t 08337 08788 o.00086874 0.000026 2 08342 08788 0.00068313_0.00000361, 3 O57 64791 0,00000557 _0.00000256 4 08556 0.8791 .00001568 0.00000256, 5 08832 047 O.00055H8S _0.00000169 ‘ 0382 9.8795 0,00070969 _o.00000146 1 08728 0.8800 o.00017530 _0.00000089, 3 O67 8.8800 0,90006790_0.00000089 9 08517 08862 0.00006178 .0.00003025 0 0.8532 08862 o.00004085 "000003025 u o3s07 0.00000000 a 0.3806 0.00000001 eee Son 25956 0.684 0.002947 | 0.00007696 0.8596 08807 3.2743 x 10~* 69964 x 10-* Men ‘Variance Fratio exteatated 4680 Tabular F0.089,11) one-sided 2916 Valid cinaytcal Methouts aon Procedures (3) Caleutawe the difterence in the mean e~ Xo. (6) Caleulate the standard error between the method means from SO Tic F Vohno} (ay (1) Cateutate te vate foe eran ues of Methods © and D, Velie + Vojno) C2 @ Caleviate the effective number of degees of freedom from (Wejne + Volo? Velma - Voinoyt (20) mp1 noting that this is rarely a whole number and the degrees of freedom fesult usually requires rounding. inthis instance the value is 9.39. (9) Look up the tabulated value of forthe effective number of deprees of freedom, 9, at 95% confidence, (10) Compare this value with the calculated vatue of found in step 7, asin Table 24, - AS the bul 7 ae 2.26 it than the absolute clelsted value (he ‘dul (Gan wc on conta ane eames gon Uc Up il now we have been conecned with comparing ethos wih singe sample. However on many oceasons isi va possible For capes ney Beltha sump quanttss donot peat sueh esting he eequey ot too, sSoming may bow sth tenting hs ake paceaveranaucnies poset i ra rang of edn oben by vo mea nh that aibough the requirement of vatanceequivatect sve nol ede he Table 24 Results from t-test calculations where the variances are known (0 be unequal Difference in rans ~oa2it Standard error bevween means 0.0088 {Stauste ~166 Men variances 32745 x 107" 6.9964 « 10-* ‘Efectine unber of degrees of freedom 9.39, (Velie of 9 efective derers of freedom 226 Oe Testoology Transfer 65 assumed that for each method the variance is constant over the concentration range. . Consider the requirement for comparing two methods covering 2 number of ‘sample types or matrices. Here the paired (lest is very useful. In the following example, eight different sanples have been analysed by both methods with one determination each, The steps aud ealeulations are similar to those curried out previously. (1) Calculate the menn for the dl ferences between the data paics, fe (2) Calculate the squaced ditferences between the mean of the differences and the differences for each data pai. (9) Calculate the variance of the differences by dividing the stim of squares and dividing it by (the number of data points — 1), in this example, 7. (4) Calculate the standard deviation of the differences from the square root of the variance, fy. (8) Caleulate the value for «from ay IS an (6) Look up the tabulated value of : for the number of degrees of freedom (a= 1), Tin this instance, at 95% confidence (two-tailed test) (% Compare this value with the calculated value of 1 found in step 5. ‘The ealevlations are shown in Table 25. As the tabulated value is much greater than the ealeutated valve the mean values derived from the to methods are not statistically significantly different Table 28 Paired test calevlation (Difference — mean Smplepait Method | Atethod? Difference _‘8fferenee)* 1 oss 0.540 oor 0.000356 2 1202 1a? =0035 0.000736 3 1360 (a oor —o.00in.6 4 1698 Vos7 0.037 0.002014 5 661 Las 002 0.000260 6 hin Lite 0.007 0.000221, 7 ren 1680 =0.053 0.002036 a L201 ia =0.033 0.000631 Mean 1309 17 0.008 Variance 0.001067 Standard deviation 0.03267 ‘ealevlted 0.68 10.05,7) 2066 Yolid dnulyncal Methods at Procedres 9.4 Restricted inter-Iaboratory trials Alter many years of debate and discussion, there is now an international consensus about the statistical approach and method to be employed in full collaborative tials.” This topic is discussed further in Section 9.5. The TUPAC protocol referred to above cequites the analysis of duplicate test samples of the same material (a minianuin of five materials or, exceptional, the) in eight oF more laboratories. However, there are many occasions where interlaboratory studies are needed which, for various reasons, cannot achieve the prescribed criteria, In instances where the TUPAC criteria cannot be achieved, itis recommended that the Youden Matched Pairs procedure is used. The statistics of the method have been recently updated and a new procedure described.” ‘The procedure requires that there are two diferent but similar samples of| the material to be analysed. Euch sample is analysed once by each iaboratory in the tral. The method is illustrated by using a data set of % aluminium in twvo limestone samples ( and ¥) for ten laboratories taken {rom ref. $8 and shown in Table 26, The Youden plot” of these dala is shown in Figure 35. The graph is obtained by plotting ¥, against ¥, esults for each of the ten Inboratories. The axes are drawn such that the point of inteesection is at the mean values for X, and YAS a single method is used in the trial, the circle represents the standard deviation of the pooled ¥ and: ¥ data, The plot shows the predominance of systematic error over random error. Ideally, for bins-ee data (Le, containing no systematic error) the points would be clustered around de mid-point with approximately equal numbers in each of the four quadrants Formed by the axes. In practice the points lie scattered around a AS" Tine, This pattern has been observed with many thousands of collaborative trials. Table 26% Aluminiam in limestone samples rial data Test samples Las, Yi % ' 135 Ls? 2 13a cn 3 135 on 4 13a te 5 130 X60 5 32 62 ? 139 us 1 so 190 HW 10 136 2 ie a Mean 140 130 Technology Transfer 0 1. Figure 35. Youlen pla of % cluminion dare for sorples Xan ¥ ‘The calculations for the Youden matched pairs procedure are less compli- cated than those for the IUPAC protocol and do not involve outlier detection and removal. For the example, the results are shown in Table 27 ‘Table 27 Basic calewlarions forthe Youden matched pairs procedure Test somes : Lab. KY Mean (Mean 2) (X,—Z)?_ (4-2? 1 13s 187 14600001 ‘o.009604 o0Lasae 2 ae hy ies ogee O.u0s628 0.1426 3 135133 30 oon .009608 0013925 4 ihe 1a 1k05 oot O.olisss 000484 5 30 160 1380 ooto# 0.00270 0.023104 ‘ 132 12 137 ool O.onsiee 0029588 7 1391321435 0.0000 0.000364 0.005184 : 130190700 0.0635 0.902708 0.204304, " 130136 1330 00139 Dorisoe 9.007744 2 te rss a2s 0.0005 Ooi6ie4 0.006724 Sum 19s 1501 Mas GUS? —__.087TM0 0.420180 Mean 1395 1sol Laat 1 Nove lsbs 10 Overall sample mein, Z Nove samples 2ss eld Analytical Methods and Proceduces ‘The steps for the calculation procedure are as follows. (1) Cateulate the sums and means for the frst \wo columns for samples X and Y. (2) Caleulzte the laboratory mean for exch ‘mean for all samples, Z. Q) Caleulate the squares of the dillerences between the laboratory mean and the overall mean, Z, and their sum (Colunia 4). (4) Calculate the squares of the differences between the individual values and the overall mean, Z, and their sum (Columns $ and 6)., (Column 3) and the overalt From these simple calculations the ANOVA table can be constructed together with the HORRAT ratio. An iinportant difference {rom the conven- tional ANOVA table is that it allows for the sample variation to be accounted for. The general ANOVA table and calculated values for $ samples are shown in Tabies 28 and 29 but the example uses only two. However, if more than two samples are desicedjavailable the procedure can be extended, (5) Calculate the sums of squares and mean squares indicated in the ANOVA table. (6) Calculate the repeatability from 5,» RIS. (7) Caleutate the reproducibility feom an ss (3) @ i Laboratori) E=1 $5. = SE (Aeon ~ 2 ws, = $8 SHRED WAH -2 +. = SS Ewor(e) (La KS) $5, # $54 ~S54— 85. Ms, Se Tov ws SS = DW MA MS Sample(S) St T Tecluology Transfer Table 29 Caleutate! values fron the aluminium in limestone deta set Degrees o . Source Pee” valve Sim ofuyares Mean sere aboraton 5 9 0316 003515, Smet i easea Soseis Error (E) bat oops 0.01305 Toval 2 1905019 Toot ent Reproducitity Sq = (HSER ME 4 ars, = 0.154 Repeatability Se = JAS = 04227 RSD woos Horwitefunetion 3.78, HORRAT iio 289 (9) Calculate the Horwitz RSD from equation (14), RSD = 20-8560 where C = 2/100 (10) Caeulate the HORRAT ratio from equation (15) RSDau/ RSDeie It is readily apparent that the HORRAT ratio of 2.89 indicates that the method is nol sufficiently precise lor inter-laboratory work. However, inspec: tion of the ANOVA table shows a liege value for the mean square due to the samples, MSs, This i an indication that they may not have been homogeneous and that the method may in fact be satisfactory. This demonstrates & major advantage in partitioning the variances between laboratory, sample and error rather than the (raditional within- and between-laboratories methou. 9.5 Full collaborative trials ‘As mentioned in the previous section, analytical methods intended for inter- laboratory use should be subjected to a collaborative trial using the [UPAC protocol. This is a prescriptive procedure and outlier testing/removal is mandatory, Before the process begins the data need to be examined to ensure that only valid data are input to the ealculation process. This process is best shown as.a flow chart (Figure 36). . : The process looks very complicated but itis very structured aad mechanical ‘and is best shown by a worked example. The example chosen to demonstrate the0 Valid Avolyicnl Merbode and Procedures START No — Sina ¥ nate Isboratony Sates ‘Snide rates | si ‘would enceed 29 a ‘iminated END. Reportorignat Figure 36 Flow chart of the [UPAC collaborative tral protocol (Reprinted from J. AOAC Ine, 1995, 78, LDA. © (1998) AOAC INTER. NATIONAL) ccaleulation process is from an aflatoxin (ppb level) in peanut butter trial? These data are (rom 21 laboratories with one sample analysed in duplicate Clearly in a full IUPAC trial there would be theee or mare samples, The basic dita set and calculations are shown in Table 30. For each laboratory, the within-laboratory varianes is calewlated from (8 — Ry) and the sim of the varianees caleulated. The first task isto identify Tectnology Transfer n Table 30 Aflaroxin in peanut butter data obs 21 and 5 Altlobt, Lab, 21 removed removed Wiivteb, — Webi la, Watietob Lab, Replicate | Replicate? variance varie variance 1 130) 0.66 hse Ls hse 2090 030 oor 001 oot 30 10s 095 003 005 4 0 0 9.00 0.00 0.00, S330 0.00, ne be 6 090 ety 0.09 009 0.09 7 900 000 0.00 0.90 0.00 810 180 001 bor oor 5 000 120 ta 1a rer 10 060 090 009 0.09 0.09 Ho 130 090 0.6 D6 016 12 om 10 00 1.00 1.00 Ota ro 0.00 0.00 0.00 Loo 020 Dot O64 0.64 13 ol 10 0.09 009 0.09 ‘6 0.00 0.00 0.00 0.00 0.00 7 20 0.0 225 225 225 18 000 0.00 0.00 0.00 0.00 19 080 110 O04 O04 004 10. 130 oot oor oot 730 a B09 Sum ofthe varinnces $9.20 at 742 Cochesa statistic (4) 57.10 ores 3o.34 Critical value (%) 41.50 240 30 ny single replicate which is too variable using Cochran's test. The Cochran Matistic is the ratio of the highest laboratory variance to the total variance expressed as 2 percentage. In our example, Laboratory 21 is the highest with 28.09 whieh is 57.1% of the total. The critical value from Table Al on Page 76s, 41.5 and therefore Laboratory 21's data are removed. ‘The next step is to examine the remaining data to see ifa laboratory average or pairare too extteme in comparison with the rest ofthe data, This is somewhat ‘more complicated and involves the calculation of Grubbs statistics but can be easily done with a spreadsheet, The process is as follows: (1) cateulate the standard deviation forthe remaining 20 labs, @) caleulate the standacd deviation for the data exchiding the highest average valve, 2 (9) calculate the standard deviation forthe data excluding the lowest average value, 5;n Volid snolytical Methods and Procedures () caleulate the % decreases in standard deviation fom 100(1 —!2) aed 100(1 =) (5) the larger ofthese two numbers isthe Grubbs single statistic; (6 repet the procets again removing the second highest and seond lowes (7) the larger of these two numbers isthe Grubbs pair statistic; (6) look up the criticat vaiues from Column | in Table A2 on Page 77 for Grubbs single statistic; (9) look up the critical values from Column 2in Table A2 for the Grubbs pair statistic, Table 31. First cycle Grubbs test calculations for aflatoxin dara Tecluotogy Transfer n ‘The results of this process for our example are shown in Table 31, Neither of the calculated Grubbs statistics exceeds the critical value and therefore no laboratory is excluded. Y IC is now necessary (0 repeat the entire cycle with the remaining 20 laboratories. The neat task is lo recalculate the Cocfran test statistic. In the example, Laboratory 5 has the highest remaining variance of 64.86, which is 57.1% of the new total variance of 21.11 (see Table 30). . ‘The critical valve Tom Table At is 42.8 and therefore Laboratory 5's data is removed. Note that this is allowed because the number of laboratories removed s0 far does not exceed 2/9th. In a similar manner, the Grubbs stacstics are ‘ecaleulated for this second eycle or the remaining 19 laboratories, ‘The results are shown in Table 32. Once again, neither of the critical values are exceeded s0 no new Inboratories are excluded. As with the previous cycle the ‘next task is (0 recalculate the Cochcan test statisti. In the'example, Laboratory ‘Yable 32 Second eyele Grubbs test ealeulesions for aflatoxin dato ab Mean Grabs eat (ft ele) 4s Oe te Mean Grubbs rea (secon yee) 1 1128 129 128 128 2 ous as 0385 ass 03s i 128128 Lat 12 28 3 hie 16 016 116 116 2 08s 08s os 085 08s 4 20120 120 120 20 5 Ne 16 1s 16 16 5 185 135 135 4 s30 0 a20 20 120 120 ‘ 10s 4.05, ros Los ros 6 10s tos 105 cos os 7 000 0.00 8.00 + 000 000 0.00 8 Mes bas 165 165 165 a Les ks 16s 165 ° 060 0 060 0.60 0.60 5 a6 060 0.69 06 0.60 0 033 ons 075 07s 075 10 a7 075 073 075 075 u Mio 110 rary 10 No i Tio Lo 10 ‘0 ho R 120120 20 120 120 a fe te 20 120 120 B tao 10 Uo 0 0 3 tao kao 10 10 fo 4 08 O80 2 o60 i 060 Oe 060 O00 oo 5 08s 035 095 035 re 095095 095 095, 035 id 000 8.00 000 500 i 000 000 0.00 0.00 0.90 0 135 135, 133 0 133138 135, 3s 8 000 000 0.00 0.00 & 000 600 200 0.00 0 00 1.00 1.00 1.00 » 001.00 1.00 1.00 1.00 | 20 ss 138 uss 135 20 58s 135 135 Noofiabe 2918 9 6 8 Neots 9B 6 io " Highest -- Lowest" 2ndhighelt 2nd lowest Highest Lowest” Indhighese Ind fowest Overailmean 098 _eemoved__cemoved removed removed Overailimean 9.991 remaved __eemoved —_temoved”——_ removed Calestd.dev. 05290501040 0.508 04 Cale, sd.dev, 0501 | 48 OMS 0.469 0402 Ss Su & Se x SSH S. Su 5. Grubbssinglestastic Grubbs par statisie ‘Grubs single statistic Grubbs pir sirisie 532% 146% wom 33% 810 sar war Grvbbsiesteritial values Single - 23.6 ana ‘Grubbs test rvea values Single 246 Pair MS4 ‘Table 33. Caleulavions for one way ANOVA for oflvoxin data after rt olid Anotytcal Serbods and Procedures Technology Transfer 7 Table 34 ANOVA table and final coleulated values for oflatoxin date a Devel fceion Sunafeqarey Meonnqarr Witiniabortoy 8 Joaze sox Baeerinvastey 18 toe ines Tout 7 waa Repoaicbity ya [ERE se, oseas Reversi Sc JST oasis a6 Ser Forwi aue rd HORRAT io 9 17 has the highest comsining variance of 2.25 which ig 30.3% of the new total variance of 7.43 (see Table 30). The eritical value from Table AL is 44,3 for 19 laboratories and 2 roplicaies, and therefore Laboratory 17's data ace not removed. As no more changes fave taken place the thied eyele for the Grubbs testing does not need to be done and the process is coniplete. Allof the foregoing calculations are simply to trim the data set by removing outliers in accordance with the set criteria, AIL that cemains is to calculate the mean value for the sample, the repeatability and reproducibility ofthe trialled method and the HORRAT ratio, These values can be calculated using a one way ANOVA method similar to that described in Table 29 and givenin Table 33 and the final values in Table 34 ‘The HORRAT ratio of 14 confirms that the method is suffciently precise for Postscript ‘An Excel spreadsheet, Handbook tables.xls, containing all of the caleutation ‘examples in thie handbook, is to be made available for download, for teaching and information purposes only, fcom the Analytical Methods Committee home page. This can be accesed via the Books home page on the Royal Scciety of Chemistry web site hup:/ivww.rse.ore/is/books/vamp.htm, ‘The author and the Royal Society of Chemisicy do not assume any lability in connection with its use.% Vel Analyte) Melbods and Procedhret Tectnology Transfer Appendix: Statistical Tables Table A2 Critel values for Grubbs extreme deviation soar Take AL Gro! ales Colas mann varie ai ee vf Wo. of One highest Tvs highest ‘replicates per laboratory luboracories orlowest or ic lowest No.of laboratories = 2 es aa ens eo 4 85.1 929 5 ns 503, ‘ 943 Bo m5 B54 625 6 60 a3 5 Be ns 645 sh 339 1 S10 Da ‘ 52 658 383 322 a3 8 Sta 5 7 2 oz 52 7 2) ° a3 610 8 na 556 a4 ‘30 385 0 a8 564 5 93 se 23 23 383 u 383 525 0 655 Be BS 362 36 2 383 we uv 622 458 a D6 303 B 318 $61 2 394 et 380 313 283 ‘a air OS 3 S64 03 32 »2 265 5 Bs «3 1 538 383 Ms. 23 380 16 283 392 15 33 364 > 237 Ba 0 269 a4 6 oy Maa Bs wa no is 353 359 0 aa 333 am Da 22 9 246 34s 8 460 3a 359 ns 204 2» D6 332 6 a3 m0 2a as 95 a 23 39 20 428 23 2b 207 189 2 29 303 2 as 22 ns 199 0 2 2 297 n 403, m2 no 192 3 u 20s 283 2 3 63 212 8s 166 25 193 280 379 255 03 me 169 26 11 art 25 367 243 193 m2 155 n Ba 262 6 355 a1 wa 16s 150 2B 8 254 u sas Be i? 161 rs B 4 443 FH 33 ni 18 1s tai 30 1a 241 » BI ni 5 Ba Br 0 ba. 99 30 ns 216 169 143 B35 so nas 162 3s 22 19s 153 ns 16 ‘0 260 0 Bs Ns oa “The yale for“ Inboriode” a mittng fom ie engine 30 216 3 tee 3 36 iba Th va at obi me air te Se Reprint J. AOAC Int 385, 78, 138A. (0998) AOAC INTERNATIONAL. Hepa tam J. AOAC In, 1995, 78, 159A. 09 (1998) AOAC INTERNATIONAC,% Votit Anayticul Mestads and Procedhees 10 Sclected publications of the AMC CA. Watson, ed. Official and Standardised Methods of ana ‘of Chemistry, Cambridge, 3ed edn., 1994 fs, Roys 1 Society 10.1 Statistics Sub-comn ‘= ‘Recommendations for the Definition, Estimation and Use of tie Detection Limit’, dnuiyst (Condon), 1987, 112, 199. + *Rerommendations for the Conduct and Interpretation of Cooperative Trials" analyst (London), 1987, 112, 679, 4 ‘Uses (Proper and Improper) of Correlation Coefficients’, Analyse (London), 1988, 113, 1469 + "Report on an Experimental Test of “Recommendation for she Contact and (nieepretation of Cooperative Trials™', Analyst (Cambridge), 1989, 14, 1489, 4s ‘Principles of Data Quality Conteol in Chemical Analysis, analyse (Camtyridge), 1989, 114, 1497, = "Robust Statistics ~ How not to Reject Qutlers ~ Part 1 Basie Concepts’, Analyt (Cambridge), 1989, 114, 1693. + ‘Robust statistics - How not to Reject Outliers ~ Pact 2 Inter-laboratory Teials', Analyst (Cambridge), 1989, 114, 1699. ‘s ‘Proficiency Testing of Analytical Laboratories: Organisational and Statistical Assessment’, Analyst (Cambridge), 1992, 117, 97 + "Is my Calibration Linear’, sInalyse (Cambridge), 1994, 119, 2363 “loternal Quality Conteol of Analytical Data’, Analyst (Cambridge), 1998. 129, 29, 4 ‘Uncertainty of Measurement: Implications for its Use in Analytical Science’, Analyst (Cambridge), 1995, 120, 2303 + “Handling False Negatives, False Positives and Reporting Limits in Analytical Proficiency Tests’, Analyst (CavrSridge), 1997, 122, 495, 1s ‘Measurement Near Zeeo Concentration: Recording and’ Reporting, Results that Fall Close To or Below the Detection Limi’, in peep + “Dayesian Statistics ~ Can it Help Analytical Chemists, in preparation. tee 10.2 Instrumental Criteria Sub-committee Papers on the evaluation of analytical instrumentation: «Parti “Atomic Absorption Spectrophotometers, Primarily for use with Flames’, dno. Proc., 1984, 21, 45. Revised in Analyse (Combridge), 1998, 123, 1407, «Part 1 “Atomic Absorption Spectrophotameters, Primacily for use With Electrothermal Atomiers', Anal, Proc., 1985, 22, 128. Revised in einalyst (Combridge), 1998, 123, 1425. Selected publications of the AMC » © Par II ‘Polyckcomators for use in Emission Spectrometry with ICP Sources’, Anal, Proe., 1986, 23, 109. + Part IV *Monochromators for use'in Emission Specteometry with ICP Sources’, Anol. Proc., 1987, 24, 3 + Por V “Inductively Coupled Plasma Sources for use in Emission Spectrometey’, nal. Pror., 1987, 24, 266. = Part VI “Wavelengeh Dispersive X-ray Spectrometers’, Anal, Prot. 1990, 27, 324, = Part Vil “Energy Dispersive X-ray Spectrometers’, Anal. Proc., 1991, 28,312, f= Pact VILE “Instcumentation for Gas-Liquid Chromatography’, Proc., 1993, 30, 296. Part IX ‘Instrumentation for High-performance Liquid Cheomatos- raphy’, Analyst (Cambridge), 1997, 122, 381, # Part X “Instrumentation for Inductively Coupled Plasma Mass Spectrometry’, Analyst (Cambridge), 1997, 122, 393. + Pact Xt “Instrumentation for Molecular Fluorescence Spectrometey’, Analyst (Cambridge), 1998, 123, 1649, = Part XIL_ “Instrumentation for Capillary Electcophoresis', Analyst (Cambridge), 2000, 125, 361 ‘© Part XILL_ “Instrumentation for UV-Visible-NIR Spectrometry’, Ana- lyst (Cambridge), 2000, 125, 367, # Pact XIV “Instcumentation for Fourier Transform Infrared Spectrom- etry’, Analyst (Cambridge), 2009, 125, 375. © Part XV “Tostrumentatinn for Gas Cheomatography-lon.trap Mass Spectrometry’, in preparation '* Part XVI_'NIR instrumentation for Process Contral’, in preparation. # Pact XVIL_ (NMR Lostrumentation’, in preparation, Pact XVIII “Instrumentation for inductively Coupled Plasma Emission Spectrometcy’, revision and update of Parts II, IV and V, in preparation. Anal. References 1. (@) BS 150 $725-1: 1994, ‘Accursey (trueness and precision) of measurements, rathods and results, Part [s General principles and deintions, (0) BS 150 5725-1 1998, Accuracy (ieness 2nd precision) of measurements, methods and results. Part. 2A ‘basie method for Ue determination of repeatability and reproducibility of & standard measurement method’ (€) BS 150 $7253: 1994, “Accuracy (trueness and precision) of measurements, methods and results, Part 3 intermediate measures of the precision of a test method. (2) BS ISO 5725-4: 1994, "Accuracy (teueness and precision) of measurements, methods and cesulls. Part Basie methods for fEnlimating the trueness of & est method", (6) DS ISO 3725-6: 1994, “Acrursey {troeness sad pression) of measurements, method and results. Parl 6 Use in practte of accuracy value’. (2) TUPAG, "Nomenclature, symbols, units aed their Giage in speceochciniesl analysis I, Data interpretation’ anal, Chem. 1996, 45 2294. (g) ‘Protocot Tor the design, Conduct ad interpretation of collaborativeto Is. 16 1 oy ts. 2». 21 22, DB. 1. Valid Analytical Methods and Procedures studies, Puce Appl Chem. 1988, 60,855. (8) RSC Analytical Methods Committe, ‘Recommendations for the definition, eatimation and tse of the detection lini’ “Aneiys (Conlon), 1987, 112,199.) Rc W. Stephany, ‘Quality asurance aad control inthe aniyais of foodstulls, residue analyis i gucticulr’, Belgian J. Food Chern, 1989, 44 199. (j) (SO Guide 33: 1989, "Use of coed cefeence materials (K) British Standard 1957: 198) (Confirmed 198%), "The presentation of numerical values. (I) TUPAC, Nomenctature for Sampling in Analytical Chemisty. Pure Appl Cent, 1990, 82,1193. I incédy, T: Lengyel and AM. Ure, [UPAC Compenelian of Analytical Nomenela- tire Definitive Rules 197, Blackwell Science, Oxford, id edn, 1998, NT, Crosby and L. Patel, General Principle of Good Sampling Practice, Royal Souiety of Chemistry, Cambridge, 1995, f . F.E, Prichard (co-ordinating author), Quality in che Analytical Chemisiry Labora tory, John Wiley, London, 1995,Ch. 2, W. Horwit, "Nonenclatre for Simpliag in Analytical Chemistey’, Pure App. Chen, 1990, 62, 1193 (@) Belish Standard 6000; 1972, "Guide to she use of BSGOUI, Sampling procedures {nd tables for inspection by airbutes,(1SO 2889 part O- (8) Beish Standard 6001 198, Pars {to 3, Sampling proceduces for inspection by atiibutes (ISO 2889 pacts 1 3). (@) British Standare 6002: 1979, “Sampling procedures ané chaets (or inspection by variables for percent defective” (ISO 3951). Food & Drug Administration, Center for Drug Evaluation and Research, Draft Guidance for Industry, August 1999, ANDAs: Blend Uniformity Analysis M. Thompson, SLR. Blison, A. Fajgelj P. Wille and R. Wood, Pure Appl Chem. 1999, 71,337 ICH, 1995, Note for guidance on validation of analytical procedures: methouoiogy {CPMPIICH 281/95} FDA Cenier for Drug Evaluation ad Research, Reviewer Guidance, Validation of Chromatographic Methods’, 1994 US Pharwscopocia 24, 2000, (621) cheomatogesghy, (TI4) dissolution (esting, (GI) refractive inder, (SU spcettophotomelry and ight seattering, ec 2. BP 1999, Appendix Ui Infrared specirophotomeicy, Uliavioet and Visible spectto- photometey and Pharm. Eve, ed edn, 1997 ¥.6.19 Absorplion Spectrophotometty, ‘Australian Code of GMP for Therapeutic Goods, “Medicinal Products; Appendix D, Guidelines for Laboratory instrumentation’ 191, NIS 45, Edition 2, May 1996, Accreditation for Chemical Laboratories (EUR> ACHEM. Guidance Document No. I/WELAC Guidance Document No. WGD 2 (Edition b. C.Burgess, Laboravory Autontion nd Information Management, 1995, 31,38. E-BurgessD.G, Jones and RD, MeDowall,dlyer (Cambridge), :998, 123, 1879. W.B. Fueman, T-. Laylomfand RT. Teta, J. AOAC Int, 1994, 77,1316 CITAC Guide 1: 1595, "International Guide 1 Quality in Aniycieaf Chenistey, An Aid to Accreditation’, Eution{. VAM Insicumentation Working Group, "Guidance on the Equipment Qualification of Analytical Insruments’, High Performance Liquit Cheomatogeaphy (HPLC), 1998, D. Parriot, LC-GC, 1994, 12, 1320 Collaborative Study Guidelines, J. AOAC fn, 1995, 78, 143A-160A. G.T. Wernimont, Ut of atic 1o develop ond evaluate analytical methods, AOAC, Ailingion, VA, 1985 EURACHEM, “The Fitness for Purpose of Anelytcal Methods’, Edition 1.0, LOC. (Teddington) Limited, 1998, EURACHEMICITAC Guide CG2, “Quality Assurance for Research and Develops ment and Noa-routine Analysis Edition 1.0, LGC (Teddington) Limited, 1998 References, a 2s. 26. 2. 2 ». x. a 2. 2 M, 4, 36, 7 38. os. 40. au 2 a. “ 4s 46 » “ ”. 0. su 32 3, 5s. 38. 86. £.0, Sandel, Errors ip Chemical Aastysisin Treaise of dnalpicel Chemistry, LM. Koliiotrand P. J. Elvng, eds, Pat, Vol I John Wiley, New York, 1959, E, Morgan, Chennmeies: Experimental Design, ACOL Sera, John Wiley, Chichs ester, 1991 MCM.W.0. Hendriks, 1.4, De Boer and A.K. Smile, els, Robusress of Analytical Chenicel Methods and Pharmaesutiol Techwologcel Products, Eleeviee, AMSIGE- ‘dam, 1996. CoM, Riley and TW, Rosanske, eds, Development and Validetion of Analytical Methods, Peeganson Elsevier Science, Oxford, 1996. S.N. Deming and S.L. Morgan, Experimental Design: a Chemometrie Approach, Elseviee, Amsterdam, Ind ed, 1993. D.C. Montgomery. Design and Analyse of Experimens, John Wiley & Sons, New York, aed, 1997. G.EP. Box, W.G. Hunter and 15, Hunter, Stubs for Experinenters, John Wiley, Londen, 1978. M, Mulholland and J. Waterhouse, Chromavogrephia, 1988, 25, 769, LC. Miller and JN. Millet, Statics for Analytical Chemizny, Elis Horwood, Chichester, ed dn. 1993. Dab. Massari, B.G-M. Vandeginste, LM.C. Budyens, S. De Jong, PJ. Lewi and J. Sincyers-Verbeke, Handbook of Chemaneries und Qualimetricg, Parts A and By Elsevice, Amsterdam, 1997/1998, NMLK’ Procedure No. 4, 1996, "Validation of Chemical Analytical Methods’, February 1997, US Pharmacopocia, 2, 2000, (1225). R, Kellner, Mt Mermet, Mi. Oo and H.M. Widower, es, Analytical Chemistry, Wiley-VCH, Weinheim, 1998 W. Fonk, V. Dammann and G. Donnevert, Quilty A Chamiziry, VCH, Seiten, 195. GA. Watson, ed. Ofeial ond Sonlardived Alethods of Analysis, Rnysl Society of Chemisiey, Cambiidge, 3rd eda. 1996 IM Sargent and G. MacKay, Guidelines for Achieving Quay in Trace Anupsi, Raya Society of Chemistey, Cambridye, 1995, 1, Tukey, Exploratory Data dnulyts, Addison Wesley, Reading, PA, 1977 F.3. Anscombe, sn, Startician, 1979, 27,17, Graphic display from reference 4 ERR Tutte, The Visual Display of Quomitaive Information, Graphic Press, Cheshire, cri93 EAR. Tule, Envisimingfnformotion, Graphics Press, Cheshire, CT, 1990. ERR Tuli, Vial Explinotons, Graphics Press. Chesbiee, CT, 1997. NR. Draper and H. Smith, Applied Repression raya, fon Wiley & Sons, New York, Sed eda, 199, J. Neiet, MH, Kutnet, CJ. Nachisheim and W. Wasserman, pelle Linear Regreston Sacl. lewi, Chicago. IL, 3 edn 1996 Re Cauleuet and 1. Uoday, Stristies Jor sinalyicel Chemists, Chapman and Hill Condon, 1993. Ve llaraet sid T, Lewis, Outliers in Statisial Dara, John Wiley, London, Jed edn. 1994 EURACHEM, Quantifying Uncertainty i Analytical Meesurenent, LGC (Tedding- ton) Limited, 1995, W. Honvite, J. AOAC Int, 1998, 81, 185, Analytical Methods Conimisee, Analyst (Cambridge), 1995, 120, 2203, ‘W. Horwite, tual. Chem, 1982, 54, 67A. R- Wood, A. Nilson and H. Wallin, Qolity in she sou Analysis Laboretory, Royal Society of Chemistey, Crnbride, 198 M.Thompson and RJ, Howacth, Analyst (Lov), 1980, 105, L188, vance in Analytical2 37. 38 3. 6t a Valid Anatyicel Meshodvand Procedures EW. Sattcthwaite,Biom, Bll, 1946, 2,110. FD, MeCute, J. AOAC Int, 1999, $2,375. W4. Youden and EM. Steiner, Statistical Mamal of the AOAC, AOAC tater ‘ational, Arlington, VA, 1975 FM, Worner, A.L, Paley and R, Wood, J. Assoe, Public Analysts, 192, 28, J-AQAC Int, 1995, 78 158A, D.AOAC Inn, 1995, 78,1590. Subject Index (Note: Figures and Tables are indicate (in thisindes) by italic page mumbers Acceptance criteria, 57-89 ‘Accuracy, deiition of parameter, 13 ‘Ageregate samples, meaning of tm, 15, 16 Aliquot definition of term, 8 Analysis, definition of erm, 7 [Analysis of variance (ANOVA) ‘of eifecizin HPLC example, 32.22, uM for Flt collaborative wal, 75 for restieted interlaboratory til, 63 Aaniyte accuracy of quaniiiation, 6 definition of ter, § Analytieal methods ‘oplieabilty. & ‘comparison of two methods, 60-85 visual repesentation(s) 82,63 evelopment of, M37 selection of, 17-20 transfer of, 37-75 validation of, 37-41 Anaiyieal Metnods Conunittee (AMC) sims and objectives, 2 history, | Instrumentat Criteria Sub-committee, publications, 22, 78-79 publications, 40, 78-79 andardised methods, 40 Statist Sub-comtiee, publications, 78 uncertainty estimation method, 37 Analytical process, mapping of, 26-27 AAnsconibe’s Quartet dataset, 2 plot), Aoac collaborative study guidelines, 24 Guide on staisizal procedures, 26 Batches, meaning of erm, 15,16 Deer-Lambert Law, Bias, definition of parameter, 12,17 Bottom up’ approach for equipment qualification, 20-21 for uncertainty quantifiation, 56-57 Box-and-whisker pots), 46, 48 crample, 7 Principles, «8 Calibration of equipment, 20, 24 Calibration pots (or spe=trophotometce say) least squaces egresion line, 30 with 95% confidence limite, 3, $4 Co-operation on Iaterational Trace ‘ability in Analytical Chemistry (CITAC), guide to quality, 23 Cochran's (oaximun variance ratio) slats, 7 critical vals sted, 76 iaborative tials, 69-75 TUPAC protocol, 18,56, 69-75 Composite samples, meaning of rm, 15, 16 Computer packages Tor data plotting, 44,48 For lneae regression analysis, 48 Concentration ranges, 17 Confidence (i analytical result), factors allecting, 3 Confidence mis, 52-53 Confimatory data analysis (CDA), 42 Content mass fraction, deiition of term, 9 Correlation coeticient in ‘spectrophotomeiri assay example, st ceu Data distributions, 62, 62 Data evaluation, 42-4 Data recording and reporting. $5-57 “Data to information’ iceberg model, 20, 2,8 Data weansfer protocols, 22 Decimal places, and significant figures, $8 Degrees offecdomefective numberof in unequal variance calelations 6 Design Qualifeation (DQ), 22 Detection, definvion of term. 8 Detection limit, definition of paraniter, 10 Determination, defisiion of tre, 6 Diteiplne, as necessity, 3 Documentation ‘of analytical methods, 39-81 fof method development, 26 cof sampling proceduce, 17 ot plows). #3 examples, 15,56 with 2seores, 56 Drift, deiition of parameter, 14 Equipment calibration, 20 meaning of term, 24 Equipment qualiestion, 20-23 “pottom vp" approach, 20-21 holistic approach. 23 “top down’ approsch, 21-23, Error budget approach, 56-57 Exploratory data analysis 8 Foraio test, 60,61, 62 Factorial design se Full factorial? design lanes for purpose mesning of tem in equipment calibeation, 20 and method selection, 17-18 and user requirements specification, 45 Food & Deug Administeation (FDA), on sample size for pharmaceuticals, 19 Force through zer0" model, 49 40s eradet (or equipment qualification), U2 Subject index Ful (ssorat 2 design ‘analysis of variance of effects, 32, 33 elect ealeulations, 30-31 HPLC example, 2-35, Probability plot of effets, 32,33 ‘anking of effects, 32 vial representation, 30 Glossacy, 14 Good Laboratory Practice (GLP) tudelines, 20, 23, Good Manufacturing Practice (GMP) guidelines, 20 Gross samples, meaning of trm, 15, 16 Grubbs exteme deviation outlier tes 1-75 critical value sted, 77, Harmonisation of analytical pesctices and Procedures, 23-24 High-performance liquid chromatography (HPLC) equipment calibration, 24 ‘equipment qualification, 24 simple experimental design example, 2S VAM Insirumene Working Group vide, 23 Histodeal perspective, -2 Holistic approach (for equipment ‘ualifeation), 23, Homozencity of sample, 17 HORRAT ratio, 8, 69,75 Hoewitsfeneion, $7 ‘ilevlation of 57, 69, 75 values listed, 8 Horwitz trumpet, $7, 8 ICH3 method validation guideines, 39, 40 Increments of ssmples, meaning of erm, 15,16 Installation Qualication (1Q),22 Interactions, 35 plots, 35-36 Inter-comparison of analytical methods, and sampling proceduce. 15 Inersisboruory tals Full collaborative ils, 69-75 reatieted tras, 66-69 Subjet Inder International Union of Pure nnd Applied Chemisiey (UPAC) analytical nontenelatre, 6, 15 collaborative ial protocol 18, $6, 69= 7 nomenclature for seeping process, 15= 16 180 Guide 25 (17025), 20,23 180°" model, + Laboratory sample ‘definition of term, 7 and sarnping procedure, 8, 15, 16 Least squates estimates, 48-99 caleulation of, 49-50 ‘Least square reression method, 449 Limit of quantifenton, definition of prametee, 10 Lincar tegtesion analysis Giterature search, 18 os, meaning of tem, 15, 16 85 Mass concentration, definition of term, 9 Mat (action, meaning of erin, 9 Mates definition of erm, 9 tnd method selection, 19-20 in square error (MSE), calculation of, st i spectrophotometrc assay example, 3 hod applicability, 6 we Method devetopment process. 24-37 existing methods considered, 18-19, ‘nisl evatuation st 1S0"V" model ¢ key fetors ented, 26-27 for mulifactor experimental desigas, 36 Sandel's Venn diagram approuch, 27,30 fr simple experimental design, 27-36 Metitod selection, 17-20 and fness-forpurpose, 17-18 rats effect, 19-20 sampling considerations, 19 sources and seateges, 18-19 Method validation, 7 extent and scope, 37 ICH guideines 39, 0 insportanee of sempling. 15° 180 °V" model adupted for, 25.26 85 NMLK No.4 guidetnes, 39,40 pretiminary validation, 38 recommended best prastice, 37-39 USP guidlines, 39 Modular quolifieztion, 23 ‘Multiaetor experimental designs, 36 NMLK No.4 method validation Buidlines, 39,40 Noise, defuition of parameter, 13 ‘Nomenclature of parameters, 9-14 of terms, 69 Nordic Committee on Food Analysis {idelines on method validation, 39,40, 39 seealso NMLK, Operationst Qualification (00), 22 Optimisation methods, 36 Outliers investigation of by Grubb test, 72-75 by potting, 43, v4, 56 by escort, 56| reasons fot ejecting, 35-56 Overview (ofthis book), 2 Priced test, 64-65 Parameters, Ufntions, 9-14 Peeformance expectations, $7 Performance Qualifeation (PQ). 23 Phaemaceuteals ‘method validation guidelines, 39 simple ae foe 19 Pictorial approach to data analysis, 43 Plackett-Burman designs, 36 Precision, Ufiniion of parameter, 12, (3 Primacy samples, meaning of term, LS, 16 Probability plot, of effects, 32,33 Purpose (ofthis book), 2-8 Quanication definition of etn, 8 Timit of deBition of parameter, 10 Random ereot, definition of parameter, 12 Range, definition of parameter, Ranking, of ees, 32 Recording of data, 556 Recovery, definition of parameter, Lé Recovery of analyte from mats, 19 ‘adjustment for, 19-20 Relative standard deviation ‘alelation in Youden matched pairs procedure, 68 Aefinition of parameter, 10 in Morwite fonction, 8, 69 variation with concentration, $7, 58 Repeatability aleultions in fll collaborative teal, 75 in restricted interlaboratory teins, 68,69, efiiion of parameter, 11 Reporting of data, $5-57 Reproducibility ‘alevations in fll collaborative teats, 75 in estrcted inter laboratory til, 68.99 Aefiition of parameter, 1-12 Residual error, evaluation of, 32, 2 Residcal variance estimation of, 3 ‘notmal uissibution, 36 Residuals in pectrophotometic assay example, 51 Root mean aque error (RMSE), in ‘spectraphotometric assay example, 3 Rounding of numbers, 55 Rounding of, definition of parameter, Toot Ruggedaes, definition of parameter, 13 Samples definition of tees 7,15, nomenclature 7, 15-16 size, 19 Sampling, 15-16 considerations in method selection, 19 documentation of procedure, 17 ‘Sande's Venn diageans approach, 27, 28 Satterthwaite’s method (for tes calculations), 62,63, Scope, definition of parameter, 14 Scape (ofthis book), 2-6 Secondary samples, meaning of term, 15, 1 ‘Suber tex Selectivity ‘debntion of parameter, 13 meaning of teem, 39 Sensitivity, definition of parameter, 10 Sipnicant figuees definition of parameters) 14 in rounding of umber, 35 Simple experimcntal design, method development proces (or, 27-36 Specificity , definition of parameter, meaning of ter, 39 Spreadsheets, 26 ‘we adavess for examples, 75 Standard deviation(s), definition of paramete(s).9 Standaed eceor between method means 64 of itercept of egeesson Hine, 82 Df slope af regression ine, 82 Standard Operating Peoeedire (SOM), 22 Stem-andsleat plots), 48,45 Subsamples, meaning of rm, 15,16 Som of squares ‘caleulation in least aquaces regression sppconch, 49, 50 ofeffects, 32, 2 in Youden matched paies procedure. 68 Sum of squares ($SQ), of effects. 32,3 Syatematicerror definition ofparamcter,12 1 Stine, caleultion of, 61, 64 Test, paired version, 64-65 ‘Technology taster, 37-75 ‘and equipment qualification, 23-24 ‘of published methods into single aboratory. 59 validation requirements for, 39 ‘Teems definitions, 6-9 “Test portion, defition of term, 8 “Test sult, defirition of term, 8 ‘Test ample ‘efntion of teem, 7 and sampling procedure, 7,16 ~ Test solution, deeition of ter, 8 “Threelevel factorial design, 36 “Top down’ approach for equipment qualifation, 21-23 for uncerainty quantification, $7 Subject Indes True value, dfsition of poeameter, Trucness, definition of parameter, 2 Tukey, Joha W., on exploratory dala analysis, 8 Twoseve fetorial design, 16 Uncectsinty estimation approaches, 56-57 Unite State Pharmacopocia (USP) assay categories, 37,19 rettod validation pudelines. 59 User cequientent specification and fness for purpose, #, 5,18 and peeformance criteria of analytic! method, 18 and sampling considerations, 19 Valid Amaiytient M programme, 16 Validation of analytical methods 1SO"V" model adapted for. ¢ see olto Meth validation coments (VAM) a VAM ostrumentation Working Group, vide on equipment qualifeation, D Variance eato (F vale) taleulalion of, 32,3, 34, 6,82 fee also Patio est Within-loboratory reproducibility, efinition of parameter, LT Working cange of analytical method, Factors determining, 7 wriuen procedures, 39-41 tack of guidelines on, 40 suggested framework for, / see alto Documentation Youden matched pairs procedure, 66 calevlations, 67, 62 plot. 67 Score, 56