Data-driven Inpatient Bed Demand Forecasting Model for Surge Mitigation

Hospitals have been experiencing more frequent medical surges, the state at which inpatient beds are near or at full capacity, as both healthcare expenditure and demand have been increasing. Currently, there is no surge prediction tool designed for the short-term horizon, resulting in a short lead time to prepare for any upcoming bed demand. This study proposes a data-driven bed demand forecasting model composed of Long Short-Term Memory network, Gamma mixture model, and Discrete Event Simulation model, that intakes only high-level patient visit information and other exogenous variables, such as public holiday status and day of the week, to forecast inpatient admissions, estimate their length of stay, and simulate the upcoming bed usage to identify a potential surge at a specific program level. Given the forecast, the hospital can proactively plan and allocate its resources and reduce the likelihood of poor inpatient bed utilization.M.A.S