P-value adjustment in sequential clinical trials

Many randomized clinical trials utilize a group sequential monitoring procedure for assessing treatment efficacy during the course of the trial. For a group sequential trial in which the null hypothesis is ultimately rejected, the nominal p-value generally overstates the statistical significance of the result and some adjustment is required. Several orderings of the sample space have been proposed for performing this adjustment, each with unique operating characteristics. In this article, we compare four proposed methods with respect to the degree to which each captures the strength of evidence against the null hypothesis implied by the data. We conclude that the ordering of the sample space induced by the observed z-score has the most desirable operating characteristics.