Further Comments on the Alpha-Spending Function

Abstract

The use of alpha-spending functions has become a powerful tool in designing group sequential trials since 1990. In this manuscript, we present practical problems that we encountered and suggest possible solutions. Since group sequential methods deal with only one endpoint, they have certain intrinsic limitations in medical research. As a result, we often modify the design in conducting clinical trials since the sequential statistical framework is not sophisticated enough to handle many practical problems related to multiple endpoints in clinical trials. Topics discussed include the choice of a primary endpoint; the use of symmetric and asymmetric boundaries in interim analysis; data-driven interim analyses; overruling of a group sequential boundary; comparison of two means and the use of logrank tests in survival trials.