A reviewer's perspective on multiple endpoint issues in clinical trials

Multiplicity issues due to clinical endpoints frequently arise in clinical trials. Conducting tests of significance separately for each endpoint in a univariate manner, or ignoring the issue, could lead to inflation of the type 1 error probability in making treatment effect claims. This is of concern because inflation of the type I error probability could lead to approval of inefficacious therapies. Therefore, one generally requires that this error probability be controlled at some prespecified alpha-level. At the same time the method employed for this purpose should be one with optimal efficiency so as to be able to detect clinically meaningful treatment effect with high probability. In this presentation, we give a clinical and statistical background to the problem with a few examples and show some simulation results that illustrate the impact of ignoring multiplicity due to multiple endpoints on the type I error probability. This is then followed by an overview and discussion of some global methods in the literature and how they can be used to make endpoint specific tests of significance. Finally, we will introduce a Monte-Carlo simulation and resampling approach (with examples using real data) for controlling the type I error probability.