All Times EDT
Keywords: causal inference, instrumental variables, principal strata, Rubin Causal Model
Diseases typically have multiple symptoms, and drug approvals accordingly require evaluations of treatment effects on multiple endpoints. Standard testing procedures for multiple endpoints, designed to guarantee that Type I familywise error rates do not exceed some level, tend to yield lower Type I error rates than nominal, leading to a sacrifice in power and pharmaceutical productivity. We shall discuss new perspectives on creating more powerful procedures that maintain the desired significance level by means of randomization tests and Bayesian methods. Straightforward randomization tests for co-primary endpoints that are more powerful than the standard intersection-union test, and step-down tests for secondary endpoints that maintain strong control of Type I familywise error rates, will be given. These tests will be extended to the case of treatment nonadherence in clinical trials by means of Bayesian methods. Our discussion will shed new light on powerful Bayesian multiple comparison procedures that can satisfy frequentist criteria for clinical trials, and can help the pharmaceutical industry increase the number and quality of novel treatments for serious diseases.