Predicting the Unpredictable: A Data-Driven Machine Learning Model for Emergency Department Waiting Room Surge...

Area of Science:

• Artificial Intelligence in Healthcare
• Machine Learning for Operations Management
• Emergency Medicine Analytics

Background:

• Emergency departments (EDs) face challenges managing fluctuating patient volumes.
• Predicting patient surges is crucial for optimizing resource allocation and patient care.
• Existing methods for surge prediction often lack accuracy over extended forecast horizons.

Purpose of the Study:

• To develop and validate a machine learning algorithm for predicting ED waiting room surge status.
• To forecast surge categories (green, yellow, red) 4, 8, and 12 hours in advance.
• To assess the predictive performance of deep neural networks and XGBoost models.

Main Methods:

• Retrospective cohort study with prospective validation (July 2023 - April 2025).
• Analysis of 154,956 ED encounters using operational metrics and timestamps with 72-hour lagging data.
• Training of deep neural network and gradient boosted decision tree (XGBoost) models to predict surge categories.

Main Results:

• XGBoost model demonstrated strong predictive performance across all forecast horizons (up to 12 hours).
• Area Under the Curve (AUC) for green/yellow levels ranged from 0.87-0.91; red levels showed AUCs of 0.76-0.77.
• Operational accuracy of 68%-70% achieved for multiclass forecasting, indicating real-world applicability.

Conclusions:

• XGBoost models effectively differentiate ED surge states using operational data and timestamps.
• Accurate forecasting of high-volume risk enables proactive operational adjustments.
• This predictive capability can significantly improve ED workflow and patient management.