Generalized Mixture Commensurate Priors for Combining Nonexchangeable Functional Data Sources with Applications in Oncology

• Activity Number: 222
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Thomas Murray*+ and Brian P. Hobbs and Bradley P. Carlin
• Companies: University of Minnesota and MD Anderson Cancer Center and University of Minnesota
• Keywords: Bayesian Methods ; Commensurate Prior ; Evidence Synthesis ; Survival Analysis ; Hierarchical Modeling ; Historical Controls

Abstract

Conventional approaches to statistical inference preclude structures that facilitate incorporation of partially informative supplemental information acquired from similar circumstances. In this presentation, we detail a prior specification framework that enables investigators to implement fully Bayesian statistical models for functional data that use all available information while accommodating important between-source features. To do so, we extend commensurate prior methods developed in parametric settings so that our proposed methods flexibly borrow strength from the supplemental data to a degree that reflects the relative coherence between the shapes of the unknown curves. We illustrate the properties of the proposed framework using an application involving data obtained from perfusion computed tomography to characterize functional imaging biomarkers in cancerous regions of the liver, while supplementing inference with information from non-cancerous regions.