Bayesian Nonparametric Methods for Precision Medicine

• Activity Number: 223
• Type: Invited
• Date/Time: Monday, August 1, 2016 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Qian Guan and Eric Laber and Dipankar Bandyopadhyay and Brian J. Reich*
• Companies: North Carolina State University and North Carolina State University and Virginia Commonwealth University and North Carolina State University
• Keywords: Bayesian ; nonparametric ; policy ; q-learning ; dynamic treatment regime ; biostatistics

Abstract

Precision medicine shows tremendous promise to improve health outcomes in many areas of medicine. In this talk we develop a policy for recommending the time until the next periodontal exam using the patient's characteristics and disease progression. Allocating more visits to the patients likely to benefit from them has the potential to both reduce costs and improve outcomes. We use a nonparametric Bayesian model to capture the complex disease dynamics and heterogeneity across subjects. While the optimal policy is completely determined by the fitted disease model, it is a highly-nonlinear function of the patient's characteristics and is thus difficult to compute and interpret. We therefore develop a decision theoretic method to identify the policy within an interpretable parametric class of policies that minimizes discrepancy with the optimal policy. We develop new Bayesian techniques to evaluate discrepancy between the parametric and non-parametric models, and to quantify uncertainty in the policy and contribution of each patient characteristic to the optimal policy.