Toward Real-Time Discharge Volume Predictions in Multisite Health Care Systems: Longitudinal Observational Study.

Area of Science:

• Health Informatics
• Hospital Operations Management
• Predictive Analytics

Background:

• Emergency department (ED) admissions are critical, with 40% of Medicare patients hospitalized.
• ED boarding is a challenge due to slow patient movement to inpatient units.
• Predicting discharge volume can improve capacity management and reduce ED boarding.

Purpose of the Study:

• To predict discharges from general care units in a Midwest tertiary teaching hospital network.
• To evaluate the impact of inter-hospital data on discharge prediction model performance.

Main Methods:

• Two experiments were conducted using 174,799 discharge records from two hospitals.
• Random forest (RF) and linear regression (LR) models predicted discharges within 1 and 4 hours.
• System-aware (network data) and system-agnostic (internal data) models were compared.

Main Results:

• RF and LR models showed high accuracy (R² 0.76-0.89) in Hospital 1.
• RF models performed best in Hospital 2 (R² 0.68-0.84).
• No significant performance difference was found between system-aware and system-agnostic approaches.

Conclusions:

• Short-term discharge prediction in multisite hospital systems is highly accurate.
• Accurate predictions are achievable even when forecasting hours in advance.