Randomized Phase II Trials: What Does Randomization Gain?

In this issue of the Journal of Clinical Oncology, Dark et al present the results of a trial of two intravenous schedules of a liposomal topoisomerase I inhibitor. The design they used, which was proposed by Simon et al, called for concurrent randomization between the schedules in what the authors describe as a “pick the winner statistical design.” They carefully spelled out the details of the design and followed it closely, representing a successful application of a randomized phase II trial. Because this trial design may not be familiar to some investigators, we believed that an additional discussion of the properties of this and other randomized phase II designs would be useful. Generally, the randomization for phase II trials is between two regimens, tumor response is the end point of interest, and the goal is to determine whether one experimental regimen should be brought forward for further testing. The subsequent trial would usually be a randomized phase III trial designed to compare the experimental regimen to a standard regimen, with a more traditional end point such as survival. Randomized phase II designs can also be used with more than two regimens; multiregimen trials are sometimes discussed in the context of selection trials. In this discussion, I will assume that only two regimens are being evaluated. In the abstract of their article, Dark et al refer to their design as a pick the winner design. In fact, the first part of such a design is to perform the equivalent of a two-stage, single-arm, phase II trial, such as the Simon design, for each regimen. The benefit of this approach is that there is randomization between the two treatments, allowing a greater degree of comparability than if there were two independent phase II trials in a seemingly similar population of patients. The pick the winner role only occurs if both regimens are efficacious. However, this trial design is not as simple as it seems and is a multistep process. Step 1 is to identify two experimental regimens that are to be evaluated in a phase II setting. In Dark et al, 1 the experimental regimens were two regimens involving the same therapies but with different schedules. Step 2 is to determine the historical response rate (R0), determine a response rate that would be sufficiently high to warrant further study of a regimen (R1), and identify a design for a two-stage, single-arm, phase II study that will have a small probability (usually P .05) of concluding that a regimen merits further study when the response rate for the regimen is no better than the historical rate (R0), but a high probability (usually P .80 or .90) of correctly concluding that the regimen merits further study when the true response rate is R1 (the power). The design will specify the number of patients to be accrued at each stage, the number of responses required to continue the study at the end of the first stage, and the number of responses required at the end of the trial to conclude that the regimen merits further study. Step 3 is as follows. When the trial opens, patients are randomly assigned to regimen A or B until each regimen has the desired number of patients for the first stage. In the Dark et al study, this number was 20. When response information has been obtained for all the patients entered onto the first stage, it is determined for each regimen whether enough responses have been observed to continue to a second stage (in the Dark et al trial, this number was one response). If both regimens fail to meet this criterion, the trial is stopped, and neither regimen has sufficient activity to warrant further study. If one regimen fails to meet this criterion, accrual to that regimen is stopped, but accrual is continued to the second regimen. If both regimens meet the criterion for continuation, patients are randomly assigned to both arms until the trial is completed. Step 4 is as follows. At the end of the study, if less than a certain predetermined number of responses are observed JOURNAL OF CLINICAL ONCOLOGY E D I T O R I A L VOLUME 23 NUMBER 9 MARCH 2