Forecasting models of emergency department crowding.

Area of Science:

• Healthcare Operations Research
• Applied Statistics
• Health Informatics

Background:

• Accurate short-term forecasting of emergency department (ED) bed occupancy is crucial for efficient hospital resource management.
• Traditional methods like historical averages may not capture complex temporal patterns in ED demand.
• Evaluating advanced time series models for ED bed occupancy prediction is essential.

Purpose of the Study:

• To investigate the accuracy of time series models for short-term emergency department (ED) bed occupancy forecasting.
• To compare the performance of seasonal Autoregressive Integrated Moving Average (ARIMA) and sinusoidal models against traditional historical average models.

Main Methods:

• Collected hourly ED bed occupancy data from three tertiary care hospitals over a one-year period.
• Developed and compared three models: hourly historical average, seasonal ARIMA, and sinusoidal with an autoregression (AR)-structured error term.
• Evaluated forecast accuracy using root mean square (RMS) error for 4- and 12-hour predictions and assessed model goodness-of-fit using AIC.

Main Results:

• Seasonal ARIMA models demonstrated superior goodness-of-fit compared to historical averages.
• Both AR-based models (seasonal ARIMA and sinusoidal) significantly outperformed historical averages in 4- and 12-hour forecast accuracy (ANOVA p < 0.01).
• Model prediction errors showed minimal sensitivity to training data duration beyond 7 days.

Conclusions:

• Seasonal ARIMA and sinusoidal models with AR-structured error terms provide robust and accurate short-term forecasts of ED bed occupancy.
• These advanced models require only historical occupancy data for prediction, simplifying implementation.
• The findings support the use of sophisticated time series methods for improving ED operational efficiency.