Multiplicity Considerations in the Design and Analysis of Clinical Trials

Abstract

The need for efficient use of available resources in medical research has led to the increased appeal of clinical trial designs based on multiple responses, multiple treatment arms and repeated tests of significance. In recent years there has been considerable methodological work pertaining to these types of multiple comparison, with the common objective typically being the control of the experimental type I error rate. Here we reconsider the appropriateness of these objectives in a variety of contexts and suggest that multiple-comparison procedures are frequently adopted unnecessarily. In particular we argue that, provided that a select number of important well-defined clinical questions are specified at the design, there are situations in which multiple tests of significance can be performed without control of the experimental type I error rate. The primary restriction for this to be reasonable is that test results are interpreted marginally.