Early Prediction of Hospital Admission and Specialty for Proactive Bed Allocation in the Emergency Department

Purpose : Emergency department (ED) overcrowding is frequently driven by exit block, where patients with an admission decision wait for an inpatient bed. Early prediction of hospital admission and target specialty department may support proactive bed allocation. Methods : We analyzed approximately 160,000 ED visits (2020 – 2024) from the University Medical Center Schleswig-Holstein. CatBoost models were trained to (i) predict hospital admission (binary) and (ii) forecast the next specialty department (14-class). Results : Admission prediction achieved 77.3% accuracy after triage; department prediction reached 73.4%. Retrospective simulations on early 2025 data indicate that a conservative high-confidence operational policy of 70% could reduce mean boarding times by approximately 69 minutes and ED occupancy by 2.13 (8.15%)–5.37 (15.00%), depending on hour. Conclusions : Early prediction of admission and department enables proactive bed allocation and, in retrospective simulations, may reduce boarding times and mitigate ED crowding, supporting more efficient hospital resource management.