Clinical Trials with an Adaptive Choice of Hypotheses
- PDF / 479,080 Bytes
- 7 Pages / 504 x 720 pts Page_size
- 87 Downloads / 155 Views
0092-8615/2001 Copyright 0 2001 Drug Information Association Inc.
F’rinted in the USA. All rights reserved.
CLINICAL TRIALS WITH AN ADAPTIVE CHOICE OF HYPOTHESES GERHARD HOMMEL,PHD Professor, Institute for Medical Statistics and Documentation, University of Mainz, Mainz, Germany
SIEGFRIED KROPF,PHD Biometrician, Coordination Centre for Clinical Trials, University of Leipzig, Leipzig, Germany
In a clinical trial with an adaptive interim analysis it is possible to modify not only the design, but even the hypothesis(es) of interest, in a formally correct mannel: Two examples of clinical trials are described where modifications of hypotheses are based on substantial scientific reasons. Generally, it is emphasized that the danger of manipulation caused by flexible designs must be controlled by very restrictive guidelines. Key Words: Adaptive design; Group-sequential design; Closure test; Multiple endpoints; A priori ordered hypotheses
INTRODUCTION Adaptive Interim Analyses IN A CLINICAL TRIAL, it is often necessary and useful to perform at least one interim analysis. The necessity is implied by ethical reasons, since a patient must obtain a treatment if it has turned out to be the better one. For the sponsor of a trial, stopping is useful for financial reasons, either if a treatment has proved to be superior or if the chance of obtaining a significant result in the future course of the trial becomes low. In general, no pure sequential design (1) is planned for economical reasons; instead, a group sequential design, usually with between one and four interim analyses, is planned. The basic principles of group se-
Presented at the DIA Workshop “Statistical Methodology in Clinical R&D,” April 2-4, 2001, Vienna, Aushia. Reprint address: Gerhard Hommel, Institut fur Medizinische Statistik und Dokumentation, Langenbeckstrasse 1, D-55101 Mainz, Germany. E-mail: h o r n e l @ imsd.uni-mainz.de.
quential designs and tests are described in Pocock (2); furthermore, designs with the possibility of stopping for futility have been developed (3). A comprehensive survey of group sequential methods can be found in Jennison and Turnbull (4). Further developments of group sequential methods with the aim of obtaining more flexibility were initialized by Bauer (5). In contrast to the aforementioned procedures, his basic idea was not to work with specific assumptions for the test statistic but rather to directly combine the p-values of the individual stages. Using the p-values is a simple way to obtain distribution-free procedures. In particular, it is possible to change the design after an interim analysis. First of all, one can reassess the sample size needed to obtain the required power; moreover, it is possible to change the layout and the statistical test for the next stages, and even modify the number of the interim analyses originally planned (Brannath W, Posch M, Bauer P, unpublished data, 1999, 6). The basic concepts for the combination of p-values are found in Bauer and Kohne
1423
1424
Gerhard Hommel and Siegfried Kropf
Data Loading...
