Bayesian Approach for Proof of Concept and Dose-Finding Under Model Uncertainty for Binary Response

• Activity Number: 271
• Type: Contributed
• Date/Time: Monday, August 1, 2016 : 3:05 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Yuanyuan Tang* and Chunyan Cai and Jianghua He and Liangrui Sun
• Companies: Saint Luke's Health System and The University of Texas at Houston and University of Kansas Medical Center and
• Keywords: Bayesian ; proof of concept ; dose finding ; operating characteristics

Abstract

In literature, there are a few unified approaches to test Proof-of-Concept and estimate a target dose based on modeling including MCPMod (Bretz et al., 2005) for normal distributions, generalized MCPMod (Pinheiro et al., 2014) with generalized parametric models and permutation approach (Klingenberg, 2009) for binary response. We discuss and compare the operating characteristics of these unified approaches and further develop an alternative Bayesian approach. Our Bayesian approach is much more flexible to handle linear or non-linear dose-response relationships and is more efficient than permutation approach proposed by Klingenberg (2009). The operating characteristics of the Bayesian approach are comparable to and sometimes better than both approaches in a wide range of dose-response relationships. Our Bayesian approach can be easily extended to continuous, categorical and time-to-event responses. We illustrate our method with extensive simulations and Phase II clinical trial data examples.