Robust inference from multiple statistics via permutations
*Jitendra Ganju, Amgen Inc.  Julie Ma, Gilead Sciences  Xinxin Yu, University of Wisconsin 

Keywords: Permutations, multiple hypothesis testing, multiple analyses, randomized trials

Formal inference in randomized clinical trials is based on controlling the type I error rate associated with a single pre-specified statistic. The deficiency of using just one method of analysis is that it depends on assumptions that may not be met. For example, the logrank test for comparing two groups may perform poorly compared to a particular weighted logrank test if the hazard ratio varies over time. We propose pre-specifying multiple test statistics and relying on the minimum of p-values for testing the null hypothesis. Rejection of the null hypothesis when the smallest p-value is less than the appropriate critical value controls the type I error rate at its designated value. Inference with the minimum of p-values is shown to be more robust than the p-value derived from a single test with application in a variety of settings.