A nonparametric test for the evaluation of group sequential clinical trials with covariate information

Group sequential design is frequently used in clinical trials to evaluate a new treatment vs a control. Although nonparametric methods have the advantage of robustness, most such methods do not take into consideration of covariate information that could be used to improve the test accuracy if incorporated properly. We address this problem using a two-sample U-statistic that incorporates covariate information into the test statistic. The asymptotic properties of the proposed estimator are presented. Simulations are conducted to evaluate the performance of the test. We then apply the proposed method to the analysis of data from a Parkinson disease clinical trial, and demonstrate that the significance of the effect associated with deprenyl could be detected at an early stage.