Determination of power and sample size in the design of clinical trials with failure-time endpoints and interim analyses

An important but difficult task in the design of a clinical trial to compar e time to failure between two treatment groups is determination of the numb er of patients required to achieve a specified power of the test. Because p atients typically enter the trial serially and are followed until they fail or withdraw from the study or until the study is terminated, the power of the test depends on the accrual pattern, the noncompliance rate, and the wi thdrawal rate in addition to the actual survival distributions of the two g roups. Incorporating interim analyses and the possibility of early stopping into the trial increases its complexity, and although normal approximation s have been developed for computing the significance level of the test when the log-rank or other rank statistics are used, there are no reliable anal ytic approximations for evaluating the power of the test. This article pres ents methods, based on Monte Carlo simulations and recent advances in group sequential testing with time-to-event responses, to choose appropriate tes t statistics, compute power and sample size at specified alternatives, chec k the adequacy of commonly used normal approximations of the type I error p robability, and assess the performance of different interim analysis strate gies. It also presents two computer programs implementing these methods. (C ) Elsevier Science Inc. 1999.