W&M Featured Events
This calendar presented by
William & Mary
[PAST EVENT] Mathematics Colloquium/CSUMS Lecture
March 23, 2012
3:30pm - 4:30pm
Title: Markov Decision Processes and Approximate Dynamic Programming Methods for Optimal Treatment Design
Abstract: Chronic diseases are the leading cause of death in the United States. For many chronic diseases there are treatment options to manage the disease and reduce the risk of adverse health events. Optimal control of these treatments can prolong lives, improve quality of life and reduce costs. In this talk we present two Markov decision processes (MDPs) related to prevention of adverse events, such as heart attack and stroke. The first model is a continuous-state MDP to determine the optimal timing of multiple medications over a patient's lifetime given uncertainty in changes to their metabolic profile over time. Approximate dynamic programming methods, including state aggregation and basis function approximation of the value function, are developed to solve the continuous-state MDP. Numerical results from this model, calibrated with a large longitudinal data set from the Mayo Clinic, are presented for the treatment of cholesterol and blood pressure of patients with type 2 diabetes. The second MDP model considers the optimal timing of adherence-improving interventions over the course of a patient's lifetime. Studies suggest that long-term adherence to common preventive medications is very poor. Interventions have been proposed as a means to improve medication adherence, but the optimal time to perform an intervention has not been well studied. Electronic health records (EHRs) are a valuable source of information that can be used to monitor patient adherence to medication. Our MDP model uses EHR data to determine when to perform adherence-improving interventions based on a patient's medication fill-rate. Structural properties about the optimal intervention policy are presented. Finally, we present results regarding the costs and benefits of implementing an EHR-based active surveillance system for interventions aimed at cardiovascular disease management.
Bio:
Jennifer Mason is a Ph.D. candidate at North Carolina State University in the Edward P. Fitts Department of Industrial & Systems Engineering. Her research interests focus primarily on stochastic models applied to medical decision making. She received her M.Sc. in Operations Research from North Carolina State University in 2009 and her B.Sc. in Mathematics from the University of South Carolina in 2007.
Abstract: Chronic diseases are the leading cause of death in the United States. For many chronic diseases there are treatment options to manage the disease and reduce the risk of adverse health events. Optimal control of these treatments can prolong lives, improve quality of life and reduce costs. In this talk we present two Markov decision processes (MDPs) related to prevention of adverse events, such as heart attack and stroke. The first model is a continuous-state MDP to determine the optimal timing of multiple medications over a patient's lifetime given uncertainty in changes to their metabolic profile over time. Approximate dynamic programming methods, including state aggregation and basis function approximation of the value function, are developed to solve the continuous-state MDP. Numerical results from this model, calibrated with a large longitudinal data set from the Mayo Clinic, are presented for the treatment of cholesterol and blood pressure of patients with type 2 diabetes. The second MDP model considers the optimal timing of adherence-improving interventions over the course of a patient's lifetime. Studies suggest that long-term adherence to common preventive medications is very poor. Interventions have been proposed as a means to improve medication adherence, but the optimal time to perform an intervention has not been well studied. Electronic health records (EHRs) are a valuable source of information that can be used to monitor patient adherence to medication. Our MDP model uses EHR data to determine when to perform adherence-improving interventions based on a patient's medication fill-rate. Structural properties about the optimal intervention policy are presented. Finally, we present results regarding the costs and benefits of implementing an EHR-based active surveillance system for interventions aimed at cardiovascular disease management.
Bio:
Jennifer Mason is a Ph.D. candidate at North Carolina State University in the Edward P. Fitts Department of Industrial & Systems Engineering. Her research interests focus primarily on stochastic models applied to medical decision making. She received her M.Sc. in Operations Research from North Carolina State University in 2009 and her B.Sc. in Mathematics from the University of South Carolina in 2007.
Contact
[[rrkinc, Rex Kincaid]]