Bayesian Multiple Biomarker Subgroup Selection

• Activity Number: 522
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Zhaowei Hua*+ and Mingxiu Hu and Chuanhua Xing
• Companies: Millennium: The Takeda Oncology Company and Millennium: The Takeda Oncology Company and Boston University
• Keywords: Bayesian ; Subgroup analysis ; FDR ; Biomarker ; Enrichment design ; Multiplicity

Abstract

Identification of biomarker-defined subgroups in which experimental therapy reveals higher response rates than standard care at phase II trials is beneficial in defining target population of predictive repsponive patients for future phase III trials. It increases the success chance of identifying effective agents than convential phase III studies with broad-eligibility. Standard frequentist inference procedure can have difficulty in making valid asymptotic inference when the sample size per biomarker subgroup could be as small as 5 patients. As an alternative, we propose a Bayesian inference method to select the promising biomarker subgroups of higher response rates to the experimental therapy. Bayesian binary probit model is used to allow information borrowing across biomarker groups for the same treatment regime, which circumvents lacking of patients to provide valid inference of treatment effects in a subgroup of very few patients. Multiplicity is addressed by controlling Bayesian false discovery rate. Simulation studies are conducted to assess performance and the methods are applied to data from biomedical studies.