You are on page 1of 7

STIMULI TO THE REVISION PROCESS

Stimuli articles do not necessarily reflect the policies Pharmacopeial Forum


772 of the USPC or the USP Council of Experts Vol. 35(3) [May–June 2009]

Acceptable, Equivalent, or Better: Approaches for Alternatives to Official


Compendial Procedures
W.W. Hauck, PhD,a A.J. DeStefano, PhD, T.L. Cecil, PhD, D.R. Abernethy, MD, PhD, W.F. Koch, PhD, R.L. Williams, MD

ABSTRACT The General Notices in the US Pharmacopeia (USP) permit analysts to use acceptable (suitable) alternatives
to an official procedure. In a revision that will be effective May 1, 2009, the General Notices will require that the
alternative procedure be demonstrated to give results that are equivalent to or better than those obtained by the
official procedure. This Stimuli article discusses approaches for determining equivalent or better procedures. The
concepts and tests discussed in this paper may become one or more USP General Chapters. The preceding Stimuli
article speaks to approaches that determine an acceptable procedure without requiring a study to document
equivalent or better.
Stimuli to the Revision Process

INTRODUCTION od will yield results equivalent to, or better than, the results
generated by the conventional method [emphasis
In its General Notices, the US Pharmacopeia (USP) added]‘‘ (1, p. 681). USP does not provide a definition
states: ‘‘Compliance may be determined also by the or approaches for alternative procedures, whether they
use of alternative methods, chosen for advantages in ac- are deemed acceptable, equivalent, or better.
curacy, sensitivity, precision, selectivity, or adaptability to The Food and Drug Administration (FDA) and ICH em-
automation or computerized data reduction or in other ploy similar language regarding alternative procedures.
special circumstances. Such alternative or automated The Code of Federal Regulations, in the section that cov-
procedures or methods shall be validated (1).b The Gen- ers requirements for Equivalent Methods and Processes
eral Notices requirement for validation means the analyst for biological products, requires ‘‘that the modification
using the alternative procedure has determined the will provide assurances of the safety, purity, potency,
procedure is acceptable (suitable) for its intended use and effectiveness of the biological product equal to or
in the specified monograph test. The wording of the greater than the assurances provided by the method or
USP General Notices presumes that the alternative proce- process specified in the general standards or additional
dure possesses some property for which there is an ad- standards for the biological product [emphasis added]’’
vantage, however defined—otherwise why use it? But (4). FDA also uses a similar phrase in its draft guidance,
the General Notices statement does not require com- Analytical Procedures and Methods Validation: Chemistry,
parison to the compendial procedure on advantageous Manufacturing, and Controls Documentation (5). In dis-
properties. A priori, one would think that the alternative cussing alternative analytical procedures this guidance
procedure should be not be worse, however defined, notes, ‘‘A validated alternative analytical procedure
than the compendial procedure, but this is not stated should be submitted only if it is shown to perform equal
in the current General Notices. A revision to the General to or better than the regulatory analytical procedure [em-
Notices (3) that will be official May 1, 2009, calls for phasis added]’’ (5). ICH Q6A, 2.7 Alternative Procedures,
the alternative procedure to ‘‘give equivalent or better states, ‘‘Alternative procedures are those which may be
results’’ by comparison with the compendial procedure. used to measure an attribute when such procedures con-
In addition, elsewhere the General Notices state: ‘‘Propor- trol the quality of the drug substance or drug product to
tionately larger or smaller quantities than the specified an extent that is comparable or superior to the official
weights and volumes of assay or test substances and procedure [emphasis added]’’ (6). In MAPP 5310.7,
Reference Standards may be taken, provided the mea- FDA’s Office of New Drug Quality Assessment states the
surement is made with at least equivalent accuracy. . .[em- following policy for its reviewers: ‘‘If there is no USP/NF
phasis added]‘‘ (1). General Chapter Validation of monograph for an excipient, drug substance, or drug
Alternative Microbiological Methods h1223i states, ‘‘The product, and the applicant proposes to use an analytical
critical question is whether or not the alternat[iv]e meth- procedure from the BP, EP, or JP in a specification in lieu of
the corresponding analytical procedure in the General
a
Correspondence should be addressed to: W.W. Hauck, PhD, Senior Chapters of the USP/NF, the BP, EP, or JP procedure is
Scientific Fellow, US Pharmacopeia, 12601 Twinbrook Parkway, Rock-
ville, MD 20852-1790; tel. 301.816.8390; e-mail wh@usp.org. considered an alternative analytical procedure and may
b
Using terminology of the International Conference on Harmonization be used provided the analytical procedure in the BP, EP,
(ICH), the US Pharmacopeial Convention (USPC) uses the term proce- or JP is equivalent to or better than the corresponding ana-
dure to describe the steps to be followed by the analyst in conducting lytical procedure in the USP/NF [emphasis added].’’
a test. This is in agreement with the International Vocabulary of Metrology
(VIM) (2), which describes a measurement procedure as a detailed de- ‘‘Equivalent’’ is used in this paper rather than the ‘‘equal’’
scription in contrast to a measurement method, which is a generic de- of 21 CFR 610.9(c) in order to avoid the connotation of
scription (such as ‘‘indirect, a direct method, or an instrumental equal as meaning identical (7).
method’’). This Stimuli article thus will use the terminology comparison
of procedures rather than comparison of methods.

# 2009 The United States Pharmacopeial Convention, Inc. All Rights Reserved.
STIMULI TO THE REVISION PROCESS
Pharmacopeial Forum Stimuli articles do not necessarily reflect the policies
Vol. 35(3) [May–June 2009] of the USPC or the USP Council of Experts 773
Based on this brief review, the various regulatory and two procedures that correspond to different methods
compendial statements appear to lack precision in defini- or between two different procedures for the same meth-
tion. Following internal discussions, USPC has identified od (2). The statistical and compendial issues are the same
four options (Table 1) to allow consideration of alterna- in either case. This paper assumes that the two proce-
tives to official procedures, even though the usual word- dures being compared are measuring the same attribute.
ing allows only ‘‘equivalent or better.’’ Comparison of two bioassays that measure different attri-
 The first is for USP to set minimum performance re- butes of an article, for example, is a topic not addressed
by this discussion. A similar problem is termed method
quirements for alternative procedures. Procedures
transfer (same procedure but different laboratories). That
meeting these requirements are termed Acceptable
is not a pharmacopeial issue but is statistically similar to
Procedures. For example, for chromatographic proce- the topic considered here (see Appendix for some refer-
dures, manufacturers could establish minimum re- ences).
quirements for repeatability and peak separation. This paper first addresses each of the four options indi-
The concept then would be that any alternative vidually with consideration of advantages and disadvan-
procedure that meets the applicable requirements tages and a brief overview of statistical considerations. In
would be acceptable, i.e., suitable for its intended the discussion the paper summarizes choices and ways
use. In effect, two procedures that both meet the forward. The purpose is to identify issues that should
performance requirements would be considered be considered and candidate statistical approaches with
‘‘equivalent’’ (and hence acceptable) under this op- the intent of initiating a dialog that could lead to com-
pendial recommendations or requirements in one or

Stimuli to the Revision Process


tion without any direct comparison to the official
more General Chapters. This Stimuli article includes an
procedure.

Appendix that cites some publications that are relevant
The remaining options directly compare the alterna-
to the discussion.
tive procedure to the compendial procedure and re-
quire ‘‘method-performance studies,’’ (8) i.e., OPTION 1: ACCEPTABLE PROCEDURES
procedure comparison studies, thus dealing more
explicitly with the requirement to show that the al-
ternative is equivalent to or better than the official General Comments
procedure. This article proposes three questions that The essence of Option 1 is to allow the analyst more
drive the subsequent three options. For the second than one—perhaps even many—acceptable alternative
option, termed performance equivalence, the ques- procedures without a requirement for comparison to an-
tion is: Are the two procedures equivalent or better other procedure. The comparison, instead, is to a refer-
ence material with known properties. This option may
with respect to validation characteristics? This is sim-
be particularly valuable in cases where the current com-
ilar to Option 1 because it considers validation char- pendial procedure is especially accurate or precise. It
acteristics but features a direct comparison to the would be difficult to demonstrate that the alternative
official procedure. For the third and fourth options, procedure is equivalent to or better than the official
the question is: Do the two procedures agree? Agree- procedure in such cases. Implicitly, the judgment of ac-
ment can be assessed by considering the numerical ceptability of a procedure would acknowledge that it is
result (Option 3) of the procedure or a decision (Op- ‘‘equivalent’’ in meeting compendial expectations. To
tion 4) relative to some acceptance criterion that is this end, USPC has proposed, in a related article in this
based on the numerical result. For the third and Pharmacopeial Forum, performance-based monographs
fourth options, USPC proposes the terms results in which one and perhaps more than one optional ac-
equivalence and decision equivalence, respectively, to ceptable procedure would be deemed acceptable for
use in a particular test (9). The key characteristic of a per-
delineate the two options. Each question leads to dif-
formance-based procedure in a monograph is that it
ferent statistical approaches to allow the com- would specify performance requirements for a procedure
parison. rather than provide a required compendial procedure,
Following the metrological distinctions between proce- although one could be deemed preferred (official) for
dure and method in VIM, the discussion of this paper ap- compliance purposes.
plies regardless of whether the comparison is between
Table 1. Comparison of Options

Comparison Based on Number of


to Official Numerical Results Characteristics
Option Name Demonstrating Procedure? or Conclusion? Considered
1 Acceptable Procedures Acceptable Noa Results Many
2 Performance Equivalence Equivalent or Better Yes Results Many
3 Results Equivalence Equivalent Yes Results One
4 Decision Equivalence Equivalent Yes Conclusions One
a
There need not be an official method.

# 2009 The United States Pharmacopeial Convention, Inc. All Rights Reserved.
STIMULI TO THE REVISION PROCESS
Stimuli articles do not necessarily reflect the policies Pharmacopeial Forum
774 of the USPC or the USP Council of Experts Vol. 35(3) [May–June 2009]
For Option 1, the criteria for an acceptable procedure When comparing two procedures to understand if
could be based on validation characteristics such as spe- they are ‘‘equivalent or better,’’ the analyst must deter-
cificity, absence of bias, and precision. Criteria for accep- mine which of three outcomes (questions) are sought
tance could be based on the procedure’s operating in the comparison study: 1) the new procedure may be
characteristics, specifically on whether articles of known importantly worse than the current procedure; 2) the
properties pass or fail appropriately when compared to new procedure may be importantly better than the cur-
the relevant acceptance criteria. See (9) for an example rent procedure; and 3) there may be no important differ-
of this approach. ence, either better or worse, between the new and
current procedures. When outcomes 2 and 3 are found,
OPTION 2: PERFORMANCE EQUIVALENCE the results are ‘‘equivalent or better,’’ and only the first
outcome is excluded. In clinical trials this is referred to
as noninferiority. When outcome 3 is found in the com-
General Comments parison results, the procedures are referred to as ‘‘equiva-
Performance equivalence means demonstrating lent.’’ Thus in a comparison study, the analyst must be
‘‘equivalent or better’’ results on some subset of valida- clear about the intent of the study in order to derive use-
tion properties. General Information Chapter h1225i lists ful conclusions.
many candidate properties. By focusing on a subset, Op-
tion 2 is similar to verification in General Information Demonstrating Equivalent or Better—Means
Chapter h1226i, where determination of which proper- A starting point for comparing two procedures is to
ties for the procedure’s intended use are important. In compare only means. Ideally, though, this would be a
Stimuli to the Revision Process

the context of alternative procedures, the most impor- comparison of biases (accuracy in ICH and USP terminol-
tant generally are specificity and accuracy. Although pre- ogy). A noninferiority hypothesis would be that the bias
cision (variability) is important, analysts can sometimes of the alternative procedure is equivalent to or better
address precision in different ways, such as the choice (less) than that of the official procedure. This would re-
of the number of replicates (discussed below). Even with quire a comparative accuracy (trueness) assessment of
focus on a subset of characteristics, a disadvantage of the two procedures. If bias cannot be assessed, a related
Option 2 is the multiplicity of statistical tests correspond- approach would be to suggest that the bias of the official
ing to the multiple validation properties for a given procedure is acceptable. This leads to a two-sided
procedure and the designated attribute of the measur- equivalence hypothesis of equivalence of the two proce-
and. One can easily envision an alternative procedure dures’ means for the alternative. Specifically, the alterna-
that is better on some properties, equivalent on some, tive procedure must be similar on average to the official
and worse on some. What should be the final decision procedure. Statistical procedures for one- and two-sided
regarding the alternative? Prioritizing the importance of equivalence hypotheses for means are well developed
the validation properties seems to be a key. An additional (12, 13).
question is whether the new procedure actually is better The problem with a simple comparison of means is
than the official procedure on some property. An affirma- that the difference may vary with the quantity being
tive answer does allow that the alternative procedure measured. Analysts may need to compare accuracy sepa-
may be preferable based on considerations not captured rately at each of a range of values rather than perform a
by validation characteristics (e.g., cost or time). single, composite accuracy comparison. Calibration (re-
gression) approaches have been developed and allow for
Equivalence Testing both systematic and proportional differences between
Comparing two procedures to show equivalent or bet- the results of the procedures; see the Appendix for refer-
ter performance for one or more validation properties ences. Most of the procedures in the literature determine
leads to a body of statistical work referred to as equiva- the regression for the relationship and then, inappropri-
lence testing. The basic idea is to properly align the sta- ately, test the null hypotheses of intercept equal to 0 (no
tistical hypotheses with the intent of the work. If the goal systematic difference) and slope equal to 1 (no propor-
is to positively demonstrate equivalent or better perfor- tional difference). If both are not rejected, equivalence
mance, then ‘‘equivalent or better’’ needs to be the sta- is claimed. This is the common error of wrong hypoth-
tistical alternative hypothesis. The statistical null eses for an equivalence problem. Equivalence hypotheses
hypothesis is then ‘‘neither equivalent nor better.’’ This should specify a range for an intercept sufficiently close
has not been common practice, although the literature to 0 and for a slope sufficiently close to 1. These are
in this field has evolved to equivalence testing. For com- two-sided equivalence hypotheses.
parison of means, for example, a common practice has
been to declare equivalence if the two means are not Demonstrating Equivalent or Better—Variances
found to be statistically different with a standard t-test. Variance comparisons are clearly one-sided (noninfer-
With a standard t-test, the alternative hypothesis is differ- iority); i.e., a procedure that results in an important in-
ence, and the null hypothesis is no difference. The papers crease in variance only would not be considered
by Limentani et al. (10) and Chambers et al. (11) contain performance equivalence. The general approach is al-
good examples of why the equivalence approach is pre- ready covered in USP’s General Information Chapter Ana-
ferred to the still commonly used standard t-test. lytical Data—Interpretation and Treatment h1010i. The
larger question relates to which variance(s) to compare.
If the context is a single laboratory, reproducibility is not

# 2009 The United States Pharmacopeial Convention, Inc. All Rights Reserved.
STIMULI TO THE REVISION PROCESS
Pharmacopeial Forum Stimuli articles do not necessarily reflect the policies
Vol. 35(3) [May–June 2009] of the USPC or the USP Council of Experts 775
relevant. For companies with multiple laboratories using Design of experiments for interprocedure agreement
a procedure and for compendial statements about pro- should focus on ensuring that the two procedures obtain
cedures that are ‘‘equivalent or better,’’ reproducibility similar results for each sample tested. If the testing is non-
is relevant. Where comparisons are confined to a single destructive, this is straightforward. For destructive test-
laboratory, analysts also must bear in mind that repeat- ing, this involves, e.g., preparing replicate aliquots from
ability is not a full characterization of the precision of a common preparations.
procedure. The question then becomes, Which inter- A difficulty with both intraclass and concordance cor-
mediate precision components should be assessed? relation approaches is the lack of a direct connection to
The other issue with variances is that, in a certain consequences with respect to producer risk (failing to
sense, the variance comparison sometimes can be made adopt an equivalent alternative procedure) and consu-
less important, at least for repeatability. That is, variance mer risk (adopting an inequivalent alternative proce-
may be overcome with increased sample size. If the new dure). With this consideration in mind, the second
procedure appears to be more variable than the official general approach considered here, the total error ap-
procedure, the sample size can be increased to make proach, was proposed for procedure transfer. The basic
the new procedure comparable. With respect to a proce- idea is to use tolerance intervals to control how often
dure being developed for use as an alternative proce- the two procedures give numerical results that differ by
dure, variance can be addressed during validation. The an important amount. What constitutes an important
‘‘sample’’ for determining what needs to be increased amount then depends on the acceptance criteria asso-
will depend on the source(s) of variability and the var- ciated with the procedure.
iance of interest. If repeatability is at issue, this may mean

Stimuli to the Revision Process


simply more samples and/or replicates. If intermediate OPTION 4: DECISION EQUIVALENCE
precision or reproducibility is at issue, improving repeat-
ability is not sufficient and it may not be feasible to in-
crease, e.g., the number of analysts conducting the General Comments
procedure or the time over which the procedure is con- Comments for decision equivalence are similar to
ducted in order to reduce the contribution from those those for results equivalence. The difference between
components of intermediate precision. Options 3 and 4 is whether agreement occurs with re-
spect to numerical results (Option 3) or a decision rela-
OPTION 3: RESULTS EQUIVALENCE tive to some acceptance criteria (Option 4). The
distinction leads to different statistical procedures. Thus
the key characteristic of Option 4 is that it looks only at
General Comments the pass/fail results, not the actual numerical results that
When results equivalence is sought, the hypothesis of are the basis of Options 1, 2, and 3.
the comparison study is that the alternative and official
procedures obtain equivalent numerical results. Showing Demonstrating Essentially the Same Pass/Fail Results
that two procedures obtain similar results could be use- If the only concern is whether or not the two proce-
ful, e.g., for a procedure applied in stability testing when dures agree on meeting acceptance criteria—rather than
it is desirable to avoid a discontinuity in results with a the actual numerical value—different statistical proce-
change of procedure. Option 3 is thus similar to Option dures apply. A related application arises when a proce-
2 in that the focus is on the numerical results obtained by dure yields only an either/or result (meets/does not
the procedures. Options 2 and 3 differ in the nature of meet), as is the case for limit or identity tests. The primary
the comparison. In Option 3, there is a direct comparison statistical procedure here is the kappa coefficient. Other-
of results. In Option 2, the comparison is indirect; valida- wise the issues are the same as for the intraclass and con-
tion characteristics are evaluated and compared. cordance correlations. The kappa coefficient shares the
same disadvantage, namely no easy connection to con-
Demonstrating Essentially the Same Numerical Result sequences. A possible alternative is to consider the full
There are at least two general approaches that could operating characteristic curves for the two procedures.
be taken. For the first, a large body of literature examines The operating characteristic curves, by moving the com-
the general issue of inter-rater reliability. The general idea parison to the probability rather than correlation scale,
is to assess the extent to which two raters get the same may make it easier to consider which differences be-
result when using the same instrument. Many applica- tween the procedures are important.
tions of inter-rater reliability involve psychometric studies
in which the instrument is a questionnaire, but applica- DISCUSSION
tions are widespread. For continuous measures, the intra-
class correlation coefficient is the most common measure This paper has considered four options for what is
of agreement, and generally accepted terms describe the meant by ‘‘equivalent or better.’’ Although there is some
degree of agreement (10). This literature can be adapted similarity across the four, they do differ. An alternative
to comparison of compendial procedures if one inter- procedure that might be judged equivalent or better
prets ‘‘rater’’ as ‘‘procedure.’’ A more recent alternative by one option need not be judged the same by the other
is to use the concordance correlation coefficient. options. One important distinction between Option 2
and Options 3 and 4, is that Options 3 and 4 preclude
‘‘better’’ because the alternative procedure must be dif-

# 2009 The United States Pharmacopeial Convention, Inc. All Rights Reserved.
STIMULI TO THE REVISION PROCESS
Stimuli articles do not necessarily reflect the policies Pharmacopeial Forum
776 of the USPC or the USP Council of Experts Vol. 35(3) [May–June 2009]
ferent to be better. Another distinction is that Option 1, made and justified by the laboratory. In the future the
because it does not involve a formal procedure com- choice may be specified in the monographs. Should ana-
parison study, yields no guarantee that any two accepta- lysts, however, wish to demonstrate ‘‘equivalent or bet-
ble procedures will agree with respect to either numerical ter,’’ they must posit the correct hypothesis and conduct
results or decisions relative to some acceptance criterion, the appropriate statistical tests. In so doing, they can
though tight criteria about what constitutes ‘‘accepta- readily determine through careful design of experiments
ble’’ makes it likely two acceptable procedures will not whether an alternative procedure is equivalent or better.
differ greatly with respect to results. The general approach relies on good understanding of
Discussion regarding alternative procedures tends to the control space of the public monograph—the accep-
return to the decision made based on results from the tance criteria for the specified tests.
procedure. The various options differ in how they con-
nect to the accept/reject decision. With Option 4, agree- REFERENCES
ment of procedures with respect to decision is a direct
1. USPC. USP 31–NF 26, General Notices. Rockville, MD:
connection. For Options 1, 2, and 3 the connection must
USPC; 2008:7.
be indirect. A general issue, as for any equivalence assess-
2. Joint Committee for Guides in Metrology. International Vo-
ment, is: How close is close enough? A general approach
cabulary of Metrology—Basic and General Concepts and As-
is to link the procedure comparison or the definition of
sociated Terms (VIM). 3rd ed. Geneva, Switzerland: ISO;
acceptable back to its use in support of accept/reject de-
2007:2.6.
cisions: What is the impact on consumer and producer
3. USPC. USP 32–NF 27, General Notices. Rockville, MD:
risks? The analytic error is just one component of uncer-
Stimuli to the Revision Process

USPC; 2009:6.
tainty. Applied to a product that naturally varies from
4. 21 CFR 610.9(c).
batch to batch, a good analytic procedure is one whose
5. FDA. Analytical Procedures and Methods Validation: Chemis-
uncertainty is small relative to the acceptance interval so
try, Manufacturing, and Controls Documentation. Rockville,
that the procedure only minimally leads to wrong deci-
MD: FDA; 2000:3–4. Available at: www.fda.gov/cber/
sions.
gdlns/methval.htm; accessed December 17, 2008.
Options based on validation characteristics (Options 1
6. ICH. Q6A, Specifications for New Drug Substances and Pro-
and 2) share a problem of multiplicity, i.e., many charac-
ducts, 2.7 Alternative Procedures. Geneva, Switzerland:
teristics to be compared. It seems desirable to attempt to
ICH; 1999. Available at: www.ich.org/cache/compo/276-
reduce the number of comparisons. Partly, this is being
254-1.html, accessed December 17, 2008.
selective in the characteristics to compare. Further, one
7. Williams RL, Chen ML, Hauck WW. Equivalence ap-
can envision multivariate approaches at least for related
proaches. Clin Pharmacol Ther. 2002;72(3):229-237.
characteristics or a combination of some characteristics,
8. Horowitz W. Nomenclature of interlaboratory analytical
e.g., by use of mean squared error rather than bias and
studies. Pure Appl Chem. 199;66:1903–1911.
repeatability. A further problem is termed assay sensitiv-
9. Small Molecules Collaborative Group and USPC Staff. Per-
ity. That is, Is the procedure–equivalence study capable of
formance-based Monographs. Pharm Forum.
finding an important difference between the two proce-
2009;35(3):765–771.
dures if one exists?
10. Limentani GB, Ringo MC, Ye F, Bergquist ML, McSorley EO.
Regulatory and compendial statements have empha-
Beyond the t-test: statistical equivalence testing. Anal
sized the importance of ensuring that alternatives to pri-
Chem. 2005;77(11):221A–226A.
vate or public (compendial) procedures be ‘‘equivalent
11. Chambers D, Kelly G, Limentani G, Lister A, Lung KR, War-
or better.’’ As discussed in this article, USPC believes that
ner E. Analytical method equivalency: an acceptable
the requirement or recommendation should be framed
analytical practice. Pharm Technol. 2005;29(9):64–80.
somewhat differently. Option 1 simply states that within
12. Schuirmann DJ. A comparison of the two one-sided tests
the acceptance criteria delineated for a test, many proce-
procedure and the power approach for assessing equiva-
dures using the same or different methods may be accep-
lence of average bioavailability. J Pharmacokinet Biopharma-
table. No further studies are needed, and specifically it
ceutics. 1987;16:657–680.
isn’t necessarily sensible to require additional studies to
13. Wellek S. Testing Statistical Hypotheses of Equivalence. New
show equivalence (options 3 and 4) or better (option
York: Chapman & Hall/CRC; 2003.
2). For USPC this leads to ‘‘performance-based mono-
graphs.’’ Another Stimuli article in this issue of Pharmaco-
APPENDIX
peial Forum describes performance-based monographs,
some details regarding how acceptable might be defined, Following are procedure comparison references that
and USPC’s plans for implementation (9). are based on equivalence hypotheses and other useful
Although USPC is advancing consideration of a perfor- or interesting references. This does not claim complete-
mance-based monograph approach (an implementation ness; a large literature is available, and the authors wel-
of Option 1), the choice of option can be viewed as come suggested additions. More general statistical
monograph specific. For monographs for which the per- literature about equivalence testing is not included.
formance-based option does not apply, one of the other
options will be needed. At this writing, USPC is not spec- 1. Altman DG, Bland JM. Measurement in medicine: the ana-
ifying which option to use for implementing the General lysis of method comparison studies. Statistician.
Notices requirement to demonstrate equivalent or better 1983;32:307–317.
for a proposed alternative procedure. The choice is to be

# 2009 The United States Pharmacopeial Convention, Inc. All Rights Reserved.
STIMULI TO THE REVISION PROCESS
Pharmacopeial Forum Stimuli articles do not necessarily reflect the policies
Vol. 35(3) [May–June 2009] of the USPC or the USP Council of Experts 777
2. Bland JM, Altman DG. Statistical methods for assessing 20. Lung KR, Gorko MA, Llewelyn J, Wiggins N. Statistical
agreement between two methods of clinical measure- method for the determination of equivalence of auto-
ment. Lancet. 1986;1(8476):307–310. mated test procedures. J Automated Methods Manage
3. Bland JM, Altman DG. Measuring agreement in method Chem. 2005;25:123–127. Note: Considers only equiva-
comparison studies. Stat Methods Med Res. 1999;8:135– lence of means.
160. 21. Magari RT. Bias estimation in method comparison studies. J
4. Carstensen B. Comparing and predicting between several Biopharmaceutical Stat. 2004;14:881–892.
methods of measurement. Biostatistics. 2004;5:399–413. 22. Mantha S, Roizen MF, Fleisher LA, Thisted R, Foss J. Com-
Note: assumes replicates. paring methods of clinical measurement: reporting stan-
5. Chambers D, Kelly G, Limentani G, Lister A, Lung KR, War- dards for Bland and Altman analysis. Anesth Analg.
ner E. Analytical method equivalency: an acceptable analy- 2000;90:593–602. Note: Survey of the anesthesia litera-
tical practice. Pharm Technol. 2005;29(9):64–80. Note: ture.
Based on a 2003 PhRMA workshop. 23. Martin RF. General Deming regression for estimating sys-
6. Dewe W, Govaerts B, Boulanger B, Rozet E, Chiap P, Hubert tematic bias and its confidence interval in method-com-
Ph. Using total error as decision criterion in analytical parison studies. Clin Chem. 2000;46:100–104.
method transfer. Chemometrics Intelligent Lab Syst. 24. Martı́nez A, Riu J, Rius FX. Multiple analytical method com-
2007;85:262–268. parison using maximum livelihood principal component
7. Dewitte K, Fierens C, Stöckl D, Thienpont LM. Application analysis and linear regression with errors in both axes. Ana-
of the Bland–Altman plot for the interpretation of method- lytica Chimica Acta. 2001;446:145–156. Note: Bivariate
comparison studies: a critical investigation of its practice least squares; good method of analysis but wrong hypoth-

Stimuli to the Revision Process


[letter]. Clin Chem. 2002;48:799–801. Note: Commenting eses.
on use of Bland–Altman and other methods in Clin Chem. 25. Nguyet ANM, van Nederkassel AM, Tallieu L, et al. Statisti-
8. Dunn G, Roberts C. Modeling method comparison data. cal method comparison: short- and long-column liquid
Stat Methods Med Res. 1999;8:161–179. chromatography assays of ketoconazole and formalde-
9. Gerbet D, Richardot P, Auget J-L, et al. New statistical ap- hyde in shampoo. Analytica Chimica Acta. 2004;516:87–
proach in biochemical method-comparison studies by 106. Note: Interesting HPLC example with sophisticated
using Westlake’s procedure, and its application to continu- statistics and wrong hypotheses.
ous-flow, centrifugal analysis, and multilayer film analysis 26. Nix ABJ, Dunstan FDJ. Maximum likelihood techniques ap-
techniques.Clin Chem. 1983;29(6):1131–1136. Note: In- p l ie d t o m e t h o d c o m p a r i s o n s t u d i e s . S t a t M e d .
difference zone for bias and calibration but does not ad- 1991;10:981–988 [Discussion: 988]. Note: For relative er-
dress variance comparison. ror of two methods; no gold standard required but need to
10. Hoffman D, Kringle R. A total error approach for the valida- know response–error relationship for proper weighting.
tion of quantitative analytical methods. Pharm Res. 27. Petersen PH, Stöckl D, Blaabjerg O, et al. Graphical inter-
2007;24:1157–1164. pretation of analytical data from comparison of a field
11. Kringle R, Khan-Malek R, Snikeris F, Munden P, Agut C, method with a reference method by use of difference
Bauer M. A unified approach for design and analysis of plots. Clin Chem. 1997;43:2039–2046.
transfer studies for analytical methods. Drug Inf J. 28. Quiroz, Jorge. A method for determining equivalence in in-
2001;35:1271–1288. dustrial settings: defining and testing the equivalence of
12. Lao CS. Equivalence in test assay method comparisons for two methods or two laboratories. Quality Eng.
the repeated-measure, matched-pair design in medical de- 2006;18:5–14. Note: Uses a population bioequivalence
vice studies: statistical considerations. J Biopharmaceutical approach to compare means and variances simulta-
Stat. 2000;10:433–445. Note: Has hypotheses wrong but neously.
does stress confidence intervals for equivalence. 29. Rousson V, Gasser T, Seifert B. Assessing intrarater, interra-
13. Liao JJZ, Lewis JW. A note on concordance correlation coef- ter, and test–retest reliability of continuous measurements.
ficient. PDA J Pharm Sci Technol. 2000;54:23–26. Stat Med. 2002;21:3431–3446.
14. Liao JJZ. An improved concordance correlation coefficient. 30. Schepers U, Watzig H. Application of the equivalence test
Pharm Stat. 2003;2:253–261. according to a concept for analytical method transfers
15. Liao JJZ, Capen RC, Schofield TL. Assessing the concor- from the International Society for Pharmaceutical Engi-
dance of two measurement methods. Proc Biopharmaceuti- neering (ISPE). J Pharm Biomed Anal. 2005;39(1–2):310–
cal Section, ASA. 2006; 660–667. 314.
16. Liao JJZ, Capen RC, Schofield TL. Assessing the reproduci- 31. Schepers U, Watzig H. Application of the equivalence test
bility of an analytical method. J Chromatogr Sci. for analytical method transfers: testing precision using the
2006;44:119–122. Note: Bridging study to evaluate repro- United States Pharmacopoeia concept (1010). J Pharm
ducibility. Biomed Anal. 2006;41(1):290–292.
17. Liao JJZ, Capen RC. Multiple evaluators. In: D’Agostino A, 32. Shao J, Zhong B. Assessing the agreement between two
Sullivan L, Massaro J, eds. Wiley Encyclopedia of Clinical quantitative assays with repeated measurements. J Biophar-
Trials. Hoboken, NJ: John Wiley & Sons; 2008. maceutical Stat. 2004;14:201–212. Note: Addresses re-
18. Limentani GB, Ringo MC, Ye F, Bergquist ML, McSorley EO. gression with errors in both methods; uses population
Beyond the t-test: statistical equivalence testing. Anal equivalence approach for hypotheses; no variance com-
Chem. 2005;77(11):221A–226A. parisons; assumes repeated measures on each item.
19. Lin LIK. A concordance correlation coefficient to evaluate
reproducibility. Biometrics. 1989;45:255–268.

# 2009 The United States Pharmacopeial Convention, Inc. All Rights Reserved.
STIMULI TO THE REVISION PROCESS
Stimuli articles do not necessarily reflect the policies Pharmacopeial Forum
778 of the USPC or the USP Council of Experts Vol. 35(3) [May–June 2009]
33. Stöckl D, Dewitte K, Thienpont LM. Validity of linear regres- 37. Walter SD, Eliasziw M, Donner A. Sample size and optimal
sion in method comparison studies: is it limited by the sta- designs for reliability studies. Stat Med. 1998;17:101–110.
tistical model or the quality of the analytical input data? Note: For test of intraclass correlation.
Clin Chem. 1998;44:2340–2346. 38. Zhong B, Shao J. Testing the agreement of two quantita-
34. Stöckl D, Cabaleiro DR, Van Uytfanghe K, Thienpont LM. tive assays in individual means. Commun Stat Theory Meth-
Interpreting method comparison studies by use of the ods. 2002;31:1283–1299. Notes: Sets a measure of
Bland–Altman plot: reflecting the importance of sample agreement; hypotheses wrong (fixed in their 2003 pa-
size by incorporating confidence limits and predefined er- per—see next entry); does not assume linearity and does
ror limits in the graphic [letter]. Clin Chem. 2004;50:2216– not require replication.
2218. 39. Zhong B, Shao J. Evaluating the agreement of two quanti-
35. Tan CY, Iglewicz B. Measurement-methods comparisons tative assays with repeated measurements. J Biopharmaceu-
and linear statistical relationship. Technometrics. tical Stat. 2003;13:75–86. Notes: Method of their 2002
1999;41:192–201. Keywords: Confidence interval; equiva- paper with equivalence hypotheses; requires replication.
lence test; errors in variables; measurement error.
36. Westgard JO. Points of care in using statistics in method
comparison studies [editorial]. Clin Chem. 1998;44:2240–
2242.
Stimuli to the Revision Process

# 2009 The United States Pharmacopeial Convention, Inc. All Rights Reserved.

You might also like