Sponsoring Section/Society: ENAR

Session Slot: 8:30-10:20 Thursday

Estimated Audience Size: xx-xxx

AudioVisual Request: xxx

*Session Title: Multiple Endpoints and Decision Making in Clinical
Trials *

The essence of interpretation in a clinical trial is consistency from both the
clinical and the statistical perspective. The interpretation of and the
conclusions drawn from the results of a clinical trial are dependent on a
number of factors that include the disease under study, the patient
population, the endpoints, the study design, the conduct of the study, the
appropriateness of the statistical analysis for the given design and the
sensitivity of the chosen statistical test(s) to the scientific question(s)
the study is designed to investigate. One of the many factors that may make
such interpretation difficult and sometimes impossible is the presence of
multiplicity in a clinical trial that is unaccounted for in the design and
ensuing statistical analyses.

One of the incentives for wanting to identify more than one endpoint in a
clinical trial is the compelling argument that the conclusion of therapeutic
benefit would be strengthened if treatment effectiveness is demonstrated
consistently across multiple endpoints. This invited session examines the use
of exact, stable, resampling, and nested statistical tests to address the Type
I error rate and power issues during the design and analysis of clinical
trials with multiple endpoints.

Theme Session: Yes

Applied Session: Yes

Session Organizer: **Sankoh, Abdul**
Food and Drug Administration

Address: Division of Biometrics III, HFD-720 FDA/CDER/OEB 5600 Fishers Lane Rockville MD 20857

Phone: 301-827-3090

Fax: 301-443-9279

Email: Sankoha@cder.fda.gov

Session Timing: 110 minutes total (Sorry about format):

Opening Remarks by Chair - 5 First Speaker - 30 minutes Second Speaker - 30 minutes Third Speaker - 30 minutes Discussant - 10 minutes Floor Discusion - 5 minutes

Session Chair: **Sankoh, Abdul**
Food and Drug Administration

Address: Division of Biometrics III, HFD-720 FDA/CDER/OEB 5600 Fishers Lane Rockville MD 20857

Phone: 301-827-3090

Fax: 301-443-9279

Email: Sankoha@cder.fda.gov

*1. Characterizing the Multivariate Treatment Difference
in Clinical Trials by Multiple Nested Tests *

**Tang, Dei-In**,
Nathan S. Kline Institute of Psychiatric Research

Address: Statistical Sciences and Epidemiology Division Nathan S. Kline Institute of Psychiatric Research Orangeburg NY 10902

Phone: 914-356-2000

Fax: 914-359-7029

Email: Tang@rfmh.org

Abstract: Clinical Trials comparing two treatments usually use multiple enpoints to measure treatment difference. How to characterize the unerlying ymulitvariate treatment difference in a valid, efficient, and clinically meaningful way has been a subject of long term research interest. Various statistical tests have been proposed that are powerful for their respective inference purposes. However, in most cases the null hypothesis is either too small for ites rejection to be sufficient information for the purpose of treatment selection, or too large to be rejected with a reasonable power. Moreover, it seems that any single criterioon for treatment selection, however carefully designed, may turn out to be too loose or too stringent relative to the outcome of the study. Although a confidence set can be used to make a valid inference under any criterion, currently available confidence sets may not be powerful for specific inference purposes. Between conducting a single test of conducting a whole continuum of tests by using a confidence set lies the possibility of conducting multiple tests. The focus of this talk is on how available tests might be used in combination. The strategy is to formulate a nested sequence of null hypotheses that connect existing formulations so that successive rejections would lead to a sequence of conclusions of increased clinical significance.

*2. The Application of Exact Stable Multivariate Tests in
Clinical Research *

**Laeuter, Juergan**,
Otto-Von-Guericke-Universitaet Magdeburg

Address: Institut fuer Biometrie und Medizinische Informatik Otto-Von-Guericke-Universitaet Magdeburg Leipziger Strasse 44 39120 Magdeburg Germany

Phone: +49 391 6713535

Fax: +49 391 6713536

Email: Juergen.Laeuter@medizin.uni-magdeburg.de

Abstract: In 1995 we proposed a principle of multivariate testing which represents an alternative to the classical least squares mehtod. This principle is based on forming linear combinations of the given p variables, the so-called lienar scores, that are spherically distributed under the null hypothesis of the test. The spherical distribution enables us to perform exact t-tests, F-tests, etc. in the same way as traditional statistics. This strategy of testing is available for cases with a very small sample size n and, nevertheless, with a large number of variables p. A broad range of possibilities for the definition of the score coefficients allow data pre-processing that is suitable to the given application. Thus exploratory statistical methods become an essential part of the confirmatory procedures, not impairing the level of significance alpha of the test.In this talk, some tests for the comparison of mean values are considered which can be useful in clinical research in situations with multiple endpoints: the standardized sum test (SS test), the principal component test (PC test), and the covariance sum test (CS test), each without and with selection of variables. If the multiple endpoints fulfill corresponding assumptions, these tests have good properties of numerical and statistical stability, i.e. they possess a high power with a large number of variables p and with high correlation between the variables. Furthermore, we will show that the PC test is admissible in the sense of decision theory, i.e. that it cannot be surpassed by another score-based test with respect to power in all cases of application.

Moreover, the tests treated here may be modified by taking into account additional conditions of practical applications: symmetries between the variables, the directions of their effects, some special profiles of their effects, their mutual similarities, their relationship to certain latent variables, the possibility to eliminate redundant variables, and so on.

Based on the theory of spherical distributions, the claissical tests of the repeated measurements analysis can be replaced by certain more general exact multivariate tests which no longer need the assumption of compound symmetry. In these tests, a decomposition of the mean value curve into uncorrelated components corresponding to the treatment and time effects is utilized.

The methods will be demonstrated by some examples from neurological research.

*3. Muliple Endpoints in Clinical Trials: A Regulatory
Perspective *

**Jin, Kun**,
Food and Drug Administration

Address: Division of Biometrics I, HFD-710 Woodmont II FDA/CDER/OEB 5050 Rockville Pike Rockville MD 20857

Phone: 301-594-5302

Fax: 301-594-6593

Email: Jink@cder.fda.gov

**Chi, George**,
Food and Drug Administration

Abstract: Clinical trials often have multiple endpoints, and handling multiple endpoints is a difficult problem in evaulation of the trials. We propose first to define what we mean by primary, co-primary, and secondary endpoints. Then we discuss how clinical decision rules whould be defined in terms of these endpoints. The hypothesis should then be defined to reflect the clinical decision rule with appropriate testing procedure. To design an appropriate testing procedure with accurate Type I error, we must adequately account for correlations among test statistics. We will discuss how to assess correlations of different types of test statistics and how this information will be used in the evaluation of clinical trials.

Discussant: **D'Agostino, Sr., Ralph**
Boston University

Address: Department of Mathematics Boston University 111 Cummington Street Boston MA 02215

Phone: 617-353-2767

Fax: 617-353-8100

Email: ralph@math.bu.edu

List of speakers who are nonmembers: None