Forecasting Mortality Associated Emergency Department Crowding with LightGBM and Time Series Data

Area of Science:

• Public Health
• Healthcare Management
• Data Science in Medicine

Background:

• Emergency department (ED) crowding is a significant global public health concern.
• Crowding is linked to increased patient mortality, particularly when occupancy exceeds 90%.

Purpose of the Study:

• To predict periods of ED crowding associated with increased mortality.
• To enable proactive interventions by forecasting demand using time series data.

Main Methods:

• Utilized retrospective time series data including weather, hospital bed availability, calendar variables, and ED occupancy statistics.
• Employed a LightGBM model for prediction in a large Nordic ED.
• Forecasted crowding for the entire ED and its specific operational sections.

Main Results:

• Afternoon crowding was predicted at 11 a.m. with an AUC of 0.82 (95% CI 0.78-0.86).
• Forecasting at 8 a.m. achieved an AUC of 0.79 (95% CI 0.75-0.83).
• Demonstrated the feasibility of forecasting mortality-associated ED crowding.

Conclusions:

• Forecasting ED crowding using time series data is achievable.
• Predictive models can identify high-risk periods, allowing for timely interventions.
• Proactive management of ED crowding can mitigate adverse patient outcomes and mortality.