Exact Tests of Monotone Dose-Response for Binary Outcomes

• Activity Number: 644
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): William Brady*+ and Gregory Wilding
• Companies: Roswell Park Cancer Institute and Roswell Park Cancer Institute
• Keywords: binary data ; monotonicity ; dose reponse ; exact test

Abstract

An issue of interest in many areas of research, including many clinical trials, preclinical animal studies, toxicology studies, and observational studies, is that of testing for monotonic dose-response in binary endpoints. For example, these methods may be of use in the analysis of efficacy parameters (e.g., tumor repose or other dichotomized efficacy or pharmacodynamic endpoints) in randomized, multi-dose phase II trials. We propose exact versions of two tests for monotonic dose-response for binary data (Dosemeci and Benichou (1998) (DB) and Peddada (2001)), which make the tests more appropriate for small samples. The exact versions are based on the E+M method of Lloyd (2008). We examine the Type I error control and power for the tests using the tests' proposed method (bootstrap or asymptotic) and using conditional, maximization, and E+M approaches. We also compare the exact versions of the DB and Peddada tests with an exact version of the Cochran-Armitage test due to Shan, et al (2011). The E+M approach controls the type I error probability for all three tests. Across the scenarios examined, the power for the E+M approach is generally higher than for the maximization approach.