Power comparisons for group sequential tests with nonparametric statistics in case of nonproportional hazards

In clinical trials, it is important to set up a design to reach a decision on effectiveness of a drug in treating a disease with the loss of the minimum number of patients. Group sequential designs are very beneficial on this point. However, the proportional hazards assumption must hold to work under a group sequential design properly. A trial running under a group sequential design covers a long time period; therefore, assuming that hazards remain proportional over a long time period is somewhat unrealistic. We should examine and figure out the impact of nonproportional hazards over the hypothesis tests conducted under a group sequential design to set up more reliable designs and decide which test to use in which conditions. In this article, powers of group sequential tests with nonparametric statistics are evaluated under nonproportional hazards by a Monte Carlo simulation study. The simulation study covers different nonproportional hazards scenarios, censoring rates, survival distributions, sample sizes, and tied observations. With this study, we intend to be helpful for clinical trial designers to set up a more reliable group sequential design.