177. Hierarchical Bayesian Model for 30-Day Hospital Census Forecasting and Resource Optimization at a Large Academic Medical Center
Invited abstract in session TC-1: Analytics and healthcare management, stream Sessions.
Tuesday, 13:30-15:00Room: NTNU, Realfagbygget R5
Authors (first author is the speaker)
| 1. | Thomas Kingsley
|
| Medicine, UCLA |
Abstract
Our team developed a hierarchical Bayesian model designed to accurately predict hospital census 30 days in advance across inpatient service lines for Mayo Clinic sites located in southeast Minnesota, Arizona, and Florida. Utilizing forecasts generated by this model, operational adjustments were made through the hospital command center to optimize staffing levels and elective surgical scheduling. This proactive approach effectively maintained inpatient census around the target level of 85%, avoiding critical overcapacity or inefficient undercapacity situations (below 70%). Initially critical during Mayo Clinic’s COVID-19 pandemic response, the model has since continued to provide substantial operational value and remains integral to daily decision-making processes within command centers across Mayo Clinic’s major healthcare facilities.
The forecast model was trained on existing census data collected in Mayo Clinic's electronic health record relational database. Elective surgical data was separately stored in a curated database created at Mayo Clinic over a decade prior to our model development. Several time-series modeling approaches were assessed and the hierarchal Bayesian model performed the best especially when considering inter-hospital transfers across sites - a large source of inpatient admissions at Mayo Clinic and most academic medical center's in the United States.
Keywords
- Capacity and network planning
- Analytics
- Forecasting
Status: accepted
Back to the list of papers