Designs for group sequential tests

Abstract

Several authors have proposed group sequential procedures to satisfy the ethical need in clinical trials for interim analyses. We propose here an alternative procedure that offers a good opportunity for early termination when initial results are extreme, while essentially maintaining the power provided by the procedure that applies when the corresponding test statistic is computed only at the predetermined time of final analysis. The new design is also sufficiently flexible to readily allow one to increase the maximum number of analyses for certain reasons such as unexpectedly slow accrual into the study, although not for reasons arising from consideration of outcomes observed at interim analysis.