Predicting emergency department admissions.

Area of Science:

• Healthcare Operations Research
• Health Services Research
• Predictive Analytics in Healthcare

Background:

• Emergency department (ED) presentations and hospital admissions exhibit temporal patterns.
• Accurate forecasting of patient flow is crucial for efficient hospital management.

Purpose of the Study:

• To develop and validate predictive models for emergency department (ED) presentations and hospital admissions.
• To assess model accuracy across different time scales and hospital types.

Main Methods:

• Utilized 5 years of historical data from two hospitals for initial model development.
• Validated models on data from 27 hospitals, covering 95% of state ED presentations.
• Assessed forecast accuracy using Mean Average Percentage Error (MAPE).

Main Results:

• ED presentations and hospital admissions are predictable, not random.
• Forecast accuracy decreased with smaller time intervals (hourly MAPE ~50%).
• Presentations were more predictable (daily MAPE ~7%) than admissions (daily MAPE ~11%).
• Urban facility forecasts were generally more accurate than regional ones.
• Subgroups with >10 daily events had similar forecast errors to the entire dataset.
• A software dashboard was developed for bed managers.

Conclusions:

• De-identified historical data can generate valid ED prediction tools.
• These tools can assist in elective surgery scheduling and hospital bed management.
• The study provides forecasting performance benchmarks for future research.