Please join us for our seminar with Mallikarjun (Arjun) Shankar of the Oak Ridge National Laboratory:
In Silico Modeling and Predictive Analytics for Healthcare Delivery Innovation: A Case-Study of Diabetes Progression with Agent-Based Models.
Volanakis Faculty Seminar Room (Buchanan 051, Tuck School)
June 17, 12:30-2:00pm
Abstract: A fundamental shift in national healthcare from fee-for-service to paying for performance and value is underway. As new rules and policies are implemented in a complex healthcare delivery environment, there are risks and consequences that we need to explore in a principled manner. In silico predictive simulation and analysis approaches allow exploration of possible responses of individuals and the overall system to individual behaviors, incentives, and risk management innovations. We discuss our construction of a large-scale high-fidelity model for various entities in a healthcare system. We then narrow our focus to a case study in which we use an agent-based simulation model to emulate disease states and behaviors critical to the progression of Type 2 Diabetes. We build our model to reproduce certain essential findings that were previously reported for a systems dynamics model of diabetes progression. We translate critical elements of this system dynamics model which mimics diabetes progression over an aggregated U.S. population. We track disease states and estimate impacts of elderliness factor, obesity factor, and health-related behavioral parameters. Although our agent-based simulation model contains over thirty adjustable parameters, we observe that our models are considerably less complex than the system dynamics model which requires several time series inputs to make its predictions. Three elaborations of our baseline model provide increasingly improved fits to the prior findings. The models also suggest that behavioral factors appear to contribute more than the elderliness factor. Our results illustrate a promising approach to translating complex system dynamics models into agent-based model alternatives that are both conceptually simple and expressive enough to capture the main effects of local interactions.
For details of other forthcoming events, please click here.
For a copy of the slide deck from James O’Malley’s workshop on techniques for social network analysis, please click here.