Play the winner for phase II/III clinical trials

Abstract

In comparing two treatments under a typical sequential clinical trial setting, a 50-50 randomization design generates reliable data for making efficient inferences about the treatment difference for the benefit of patients in the general population. However, if the treatment difference is large and the endpoint of the study is potentially fatal, it does not seem appropriate to sacrifice a large number of study patients who are assigned to the inferior arm. An adaptive design is a data-dependent treatment allocation rule that sequentially uses accumulating information about the treatment difference during the trial to modify the allocation rule for new study patients. In this article, we utilize real trials from AIDS and cancer to illustrate the advantage of using adaptive designs. Specifically we show that, with adaptive designs, the loss of power for testing the equality of two treatments is negligible. Moreover, the study patients do not have to pay a handsome price for the benefit of future patients. We also propose multi-stage adaptive rules to relax the administrative burden of implementing the study and to handle continuous response variables, such as the failure time in survival analysis.