Forecasting emergency department hourly occupancy using time series analysis.

Area of Science:

• Health Services Research
• Data Science
• Emergency Medicine

Background:

• Emergency department (ED) overcrowding is a significant challenge impacting patient care and operational efficiency.
• Accurate forecasting of ED hourly occupancy is crucial for effective resource allocation and management.
• Existing time series forecasting methods often lack the precision needed for real-time ED operational adjustments.

Purpose of the Study:

• To develop and validate a novel predictive model for emergency department (ED) hourly occupancy.
• To utilize readily available data for real-time occupancy predictions.
• To employ a time series analysis methodology for enhanced forecasting.

Main Methods:

• A retrospective analysis of 65,132 ED visits from a large academic center in 2012 was conducted.
• A Seasonal Autoregressive Integrated Moving Average with external regressor (SARIMAX) model was selected due to time-of-day and day-of-week effects.
• SARIMAX models were built hourly to predict ED occupancy up to 4 hours ahead, incorporating regressors like current occupancy, ESI, and boarding totals.

Main Results:

• The SARIMAX model demonstrated superior performance in predicting ED hourly occupancy.
• The model achieved a Mean Square Error (MSE) of 16.20 for 1-hour-ahead and 64.47 for 4-hour-ahead predictions.
• Compared to the rolling average method, the SARIMAX model showed a 60% improvement in MSE while maintaining similar prediction intervals.

Conclusions:

• A 24-SARIMAX model effectively predicts ED occupancy up to 4 hours in advance, outperforming other forecasting methods.
• The model's reliance on readily available ED data makes it a practical tool for real-time forecasting.
• This novel technique shows promise for improving ED operational management and patient flow.