Short-Term Occupancy Prediction at the Ottawa Hospital Using Time-Series Data for Admissions and Longitudinal Patient Data for Discharge

Abstract

The Ottawa Hospital cancels hundreds of elective surgeries every year due to a lack of beds, and has an average weekday occupancy rate above 100%. Our approach to addressing these issues, by way of informing administrators of resource needs, was to model the flow of patients coming and going from the hospital. We used administrative data from the Ottawa Hospital to build a time-series model of emergency department admissions, and studied models that would predict next-day discharge of patients currently taking up hospital beds. In the latter, we considered population-averaged models for groups of patients based on their primary medical condition, as well as subject-specific models. We included the random effects from subject-specific variation to improve on predictive accuracy over the population- averaged approach. The result was a model that provided more realistic probabilities of discharge, and stable predictive accuracy over patient length of stay.