A Bayesian nonparametric collaborative filtering algorithm to improve health care decisions

Medical University Of South Carolina, Charleston SC

Investigators

Linked publications & trials

Abstract

Patients are often confronted with an array of treatment options, and the optimal choice will depend in part on their preferences for various aspects of care. What is experienced as a minor inconvenience by some patients may seem severely disabling to others. The goal of this application is to help patients answer a critical question, ?How do patients who are like me in terms of their pre-treatment preferences ultimately rate their experiences with various treatments?? A gap in clinical practice is the absence of a systematic approach to explore the link between pre-treatment preferences and patient- reported outcomes such as satisfaction. This application seeks to close this gap by developing an enhanced method for shared decision making that can be incorporated in a broad range of clinical settings. Each of our aims is tailored to achieve these goals. In Aim 1, we apply methods grounded in marketing research to develop an advanced Bayesian nonparametric statistical model to measure individual preferences for various facets of treatment, and to subsequently allocate patients to well-defined subgroups based on shared preferences. In Aim 2, we will enhance the Bayesian model to enable a fully operationalized ?recommender system? in which existing patient satisfaction data are used to make treatment recommendations for individual patients and for subgroups as a whole. Finally, in Aim 3, we will test and validate the developed method in the context of a current clinical trial. We will conduct this proposal in collaboration a randomized clinical trial exploring trade-offs between three different anticoagulants: Aspirin, Warfarin, and Revaroxiban. When the study is complete, we will have in place a novel Bayesian framework to improve to transform personalized patient decision making in a variety of health care settings. Given the potential gains and challenges of meaningful involvement of patients in medical decisions, it is critical to develop tools that may be integrated into both current methodologies of preference elicitations and electronic systems that provide real-time treatment recommendations. This research is highly consistent with NLM?s stated mission to support the development of novel informatics techniques that will impact biomedical, behavioral, or clinical research.