Monitoring Continuous Long-Term Outcomes in Adaptive Designs

Abstract

ABSTRACT Interim analyses offer the possibility to assess the efficacy of a new treatment prior to the originally planned end of a clinical trial. In addition, adaptive designs allow the statistician to modify the study at the time of the interim analysis without inflating the Type I error rate or compromising the overall power of the trial. In studies with the option of early stopping, overrunning occurs when patients have entered the trial but their data was not included in the test decision of an interim analysis while a stopping criterion has been reached. Especially in case of a long-term endpoint, overrunning is hardly avoidable. If overrunning happens, one is interested that the test decision made in the interim analysis remains stable if the delayed observations are included in the evaluation. The available adaptive designs use the data of those patients that have already completed the study for computation of the test statistic at the interim analysis. This approach exhibits a considerable risk of conflicting decisions. In this article, we consider a procedure that takes into account both the data at the endpoint and at an intermediate point of the treatment phase when calculating the test statistic for the interim analysis in an adaptive two-stage design. For known nuisance parameters, the test statistics for the two stages fulfill the p clud condition of a combination test which assures control of the Type I error rate. It is shown analytically that the probability of conflicting decisions in the case of overrunning is considerably reduced if the proposed procedure is used. For unknown nuisance parameters, simulation results demonstrate that the overall Type I error rate is not relevantly inflated and that the reduction of the probability for conflicting test decisions in the case of overrunning is maintained if the sample size in the interim analysis is not too small.