| Abstract: | When designing a phase III superiority trial, we often compute the sample size to detect a treatment effect observed in a previous trial. In this article, we propose to take into account uncertainties around the estimated treatment effect and the estimated variability in patients’ responses jointly. We argue that it will be better to base the sample size decision for a confirmatory trial on the probability that the trial will produce a positive outcome than the traditional statistical power. We show that sample sizes computed under the traditional method are often too small to yield a desirable success probability. We extend this argument to the “conditional probability” concept that is often the basis for futility decision and sample size reestimation in an adaptive design. We argue that in the latter case, the concept of predictive power is likely to provide a more realistic measure on the prospect of the trial. In our opinion, a trialist should be concerned when the required sample size for a confirmatory trial based on the chance of a positive trial is substantially higher than that calculated under the traditional approach, either at the design stage or at the interim sample size reestimation decision point. |
