Group Sequential Monitoring of Multi-Armed Clinical Trials
- PDF / 584,256 Bytes
- 8 Pages / 504 x 720 pts Page_size
- 20 Downloads / 193 Views
0092-8615/99
Copyright 0 1999 Drug Information Association Inc.
GROUP SEQUENTIAL MONITORING OF MULTI-ARMED CLINICAL TRIALS NANCY
L. GELLER,MICHAELA.
PROSCHAN, AND
DEANA. FOLLMANN
Office of Biostatistics Research, National Heart, Lung, and Blood Institute, Bethesda, Maryland
Monitoring of two-armed clinical trials using group sequential methods has been implemented for over 20 years. This paper considers analogous monitoring procedures for multi-anned clinical trials. Monitoring the trial based on repeated tests of a global null hypothesis, monitoring of comparisons of several treatments to a control, and monitoring of all pairwise comparisons are considered. The procedure of choice for a given situation will depend on how the application at hand is best formulated. Key Words: Group sequential monitoring; Multi-armed clinical trials
INTRODUCTION
GROUP SEQUENTIAL DESIGNS are now established in two-armed clinical trials. An overall preassigned significance level is maintained by choosing the so-called nominal significance level for each interim analysis, so that the probability of falsely rejecting the null hypothesis in at least one of the “looks” is no more than a.Such designs initially were suggested by Pocock (I),who proposed a small number of looks, at most five, each undertaken at the same nominal significance level. Others, notably O’Brien and Fleming (2), suggested increasing sequences of nominal significance levels so that it would be more difficult to stop the trial early and the final nominal significance level would be close to the overall significance level. For example, if three analyses were planned after equal sample sizes were accrued, the Pocock nominal significance levels each would be .022, whereas the O’Brien and Fleming nominal significance
Reprint address: Dr. Geller, Office of Biostatistics Research, National Heart, Lung, and Blood Institute, Rockledge Center 2, Room 8210, 6701 Rockledge Drive, Bethesda, MD 20892-7938.
levels would be .0005, .014,and .045.Lan and DeMets (3) generalized the group sequential theory so that analyses could be undertaken at times which were not fixed in advance and equal increments of information between analyses would not be required. In this paper, the phrase “equal increments of information” means equal sample sizes between looks when the data are continuous and have a constant variance; in general, information means the inverse of the variance of the test statistic. It is clear that the need to monitor clinical trials as data accumulate is not restricted to two-armed trials, yet the issues of multiple comparisons and monitoring have rarely been addressed simultaneously. Bauer (4) gave a good review of multiple testing in clinical trials. This paper concentrates on aspects of statistical stopping rules specifically for treatment efficacy in multi-armed trials, although it is recognized that these are only part of the overall medical decision making. There are difficulties with certain “obvious’’ generalizations of the two-armed theory. Consider a three
Data Loading...
