Title:

Learning public health systems: how data-driven optimization can help them?



Abstract:

Agency for Healthcare Research and Quality (AHRQ) defines a learning health system as a health system in which internal data and experience are systematically integrated with external evidence, and that knowledge is put into practice. As a result, patients get higher quality, safer care, health care is delivered more efficiently, cost-effectively, and in a patient-centered manner, and health care organizations become better places to work. Becoming a learning health system is increasingly an imperative in an era of health system transformation. There is growing recognition that “business as usual” is no longer a sustainable model. Among the many challenges in helping health systems learn is effective use of data. This becomes more prevalent as we come out from the COVID-19 pandemic. How data-driven optimization can help?

In this talk, we investigate its application in learning public health systems where online learning has become widely available with increasing implementation of point-of-care sensing devices and increasing understanding of controlled disease transmission mechanisms. We consider the context of multi-period location-specific resource allocation in epidemic outbreak control. We formalize the decision problem within the mathematical framework of stochastic dynamic optimization with mixed observability on system states and indeterminate parameters of the system dynamics model. We present two recent studies. In the first study, for an emerging cholera outbreak, we formulate the problem via nonlinear optimization on an ordinary-differential-equation model that governs location-specific transmission dynamics. We propose a data-driven optimization approach to determine the optimal strategy of intervention resource allocation at each period and each community in a rolling-horizon manner. At each period, we integrate single-period model parameter fitting and scenario-based stochastic programming to make decisions under uncertainty with newly acquired observational data in the system. In the second study, for the COVID-19 outbreak, we formulate the problem via mixed observability Markov decision processes under time-varying interval-valued parameters. We propose a novel transfer reinforcement learning based algorithmic approach, which integrates transfer learning into deep reinforcement learning in an offline-online scheme, to determine the optimal strategy of joint screening-intervention resource allocation at each period and each community. To accelerate the online re-optimization, we pre-train a collection of promising networks and fine-tune them with newly acquired observational data. The hallmark of our approach comes from combining the strong approximation ability of neural networks with the high flexibility of transfer learning through efficiently adapting the previously learned policy to changes in system dynamics. Our studies offer viable data-driven solutions to problems requiring “learning while optimization”, which can further nourish the application of OR/MS in learning public health systems.

 

Bio:



Dr. Nan Kong is Professor and Interim Head of the Weldon School of Biomedical Engineering at Purdue University. He is a Full Member affiliated with the Purdue Regenstrief Center for Healthcare Engineering and was the center’s former Associate Director for Health Systems. He graduated with a B.S. degree in Automation from Tsinghua University, China, in 1999 and a Ph.D. degree in Industrial Engineering from the University of Pittsburgh in 2006. He joined the BME faculty at Purdue in August 2007. His research is primarily focused on innovating data-driven analytics techniques and developing user-centered tools to address challenges arising in healthcare systems, particularly those related to system operations. He has published close to 100 peer-reviewed articles. His research has been funded by NSF, NIH, AHRQ, and AFSOR. He was the Program Chair of the 2021 Institute for Operations Research and the Management Science (INFORMS) Healthcare Conference. He recently won the 2nd-Place in the national competition “Building Bridges to Better Health: A Primary Health Care Challenge, supported by Department of Health & Human Sciences – Healthcare Resources and Services Administration.