Weighted Statistics for Testing Multiple Endpoints in Clinical Trials

Abstract

Bonferroni method, Holm [15] proposed a scheme based on the ordered p-values. Developing upon Holm’s idea, step-up and step-down methods for multiple testing have been developed for non-sequential [11,16-19] and most recently, sequential experiments [20-23]. These Holm-type methods (also called stepwise for testing marginal hypotheses in the order of their significance) allow to use higher levels of j α leading to increased power, while still controlling FWER. These stepwise methods and most of the other approaches to multiple tests do not account for different levels of difficulty of the participating tests, or proximity between null hypotheses and their corresponding alternative hypotheses. Why should we take this into account when designing statistical ABBA.MS.ID.000532. Abstract Bonferroni, Holm, and Holm-type stepwise approaches have been well developed for the simultaneous testing of multiple hypotheses in medical experiments. Methods exist for controlling familywise error rates at their preset levels. This article shows how performance of these tests can often be substantially improved by accounting for the relative difficulty of tests. Introducing suitably chosen weights optimizes the error spending between the multiple endpoints. Such an extension of classical testing schemes generally results in a smaller required sample size without sacrificing the familywise error rate and