Adaptive group sequential test with changing patient population

Standard group sequential test assumes that the treatment effects are homogeneous over time. In practice, however, this assumption may be violated. Often, this occurs when treatment effects are heterogeneous in patients with different prognostic groups, which are not evenly distributed over the time course of the group sequential trial. In this article, we consider a setting where the inclusion/exclusion criteria for patient entry are relaxed at interim analyses. This triggers heterogeneous treatment effects over the enlarged patient population. In particular, we assume that the population change relates to some baseline covariates. Simulation results show that the type I error can be severely inflated if adjustment is not made to the statistical analysis. We consider a set of linear regression models. With these models, we make inference on the target population based on all data from the changed populations. The proposed method leads to unbiased inference.