Estimating and Testing Treatment Effects on Two Binary Endpoints and Association Between Endpoints in Clinical Trials

Often, two or more binary endpoints are used to assess the impact of a new treatment in a randomized phase III clinical trial. In this article, we focus on two binary endpoints. In assessing treatment effects, we are interested in positive treatment effects on the endpoints, and a positive correlation between the two endpoints, i.e., we expect to see positive treatment effects on the same patient. Four estimation methods are reviewed and contrasted for this problem: weighted least squares (WLS), generalized estimating equations (GEE), maximum likelihood (ML), and Mantel-Haenszel (MH) type estimator. In order to test the efficacy of the new treatment, four testing strategies are proposed: a global test and three multi-stage tests. Simulation studies are conducted to study the performances of the four estimation methods and the four testing strategies. Different combinations of significance levels for multi-stage tests are also evaluated. Empirical test size, test power, and estimation precision (i...