Generative models for hospital census prediction with cumulative difference learning VAE
Pengyi Shi
Krannert School of Management, Purdue University
Date: Friday, March 17, 11:00 am – noon
Venue: GB120, Galbraith Building, 35 St George St, Toronto
Abstract: The pandemic has highlighted weaknesses in hospital care delivery on a global scale: hospitals worldwide are all facing shortages of beds, nurses, and medical resources due to demand surges. The core challenge is that hospitals need to plan for capacity and staffing based on accurate patient census prediction in real-time, i.e., the number of patients in each hospital unit at different time points. This census prediction problem becomes particularly hard in settings where the inflow and outflow of patients are driven by an exogeneous random environment, resulting in complex temporal correlations. To address the challenges in census prediction and subsequent resource planning, we introduce the generative modeling framework, in which a sequence of latent variables drives the daily arrivals, discharges, and census over the network of hospitals. To estimate the model, we develop a novel learning algorithm, DT-VAE, that specifically incorporates domain knowledge and is mathematically justified. We demonstrate the empirically superior performance of our algorithm on semi-synthetic and real datasets. We also highlight its advantage in integrating with subsequent decision problem via a case study on nurse staffing and deployment. This is a joint work with Tianchun Li, Eric Wu, and Joy Wang at Purdue University.
Bio: Pengyi Shi is an associate professor at the Krannert School of Management, Purdue University. She received her Ph.D. degree in Industrial Engineering from Georgia Institute of Technology before joining Purdue in 2014. Her research interests include data-driven modeling and decision-making in healthcare and service operations. She has collaborated with practitioners and faculty members from different healthcare organizations, including major hospitals in the US, Singapore, and China. Her research has won the first place of MSOM Responsible Research in OM Award in 2021, the first place of INFORMS Pierskalla Best Paper Award in 2018, and the second place of POMS CHOM Best Paper Award in 2019 and 2020.