Resampling-based methods for the analysis of multiple endpoints in clinical trials

We consider multivariate tests for comparing two treatments with multiple endpoints. The test decision is drawn from the simultaneous consideration of the univariate tests for the single endpoints. A general class of these tests, called cut-off tests, can be given, which, however, can lead to highly conservative procedures because the dependencies among the endpoints are not taken into account. In applying resampling-based methods considerable improvements for these tests can be achieved. Resampling-based cut-off tests are proposed which are sensitive against a treatment difference in a single endpoint, in a subgroup of endpoints, or in all endpoints. The results of Monte Carlo simulations demonstrate that a remarkable gain in statistical power as compared to the crude simultaneous consideration can be reached. In particular, for the multivariate one-sided test situation the proposed tests can be recommended. As an example the application of the tests is demonstrated by data from a clinical trial.