Time-series Machine Learning Models to Support Emergency Department Operational Planning

Area of Science:

• Health Services Research
• Operations Research
• Data Science

Background:

• Effective emergency department (ED) utilization prediction is crucial for hospital resource management and staff scheduling.
• Existing forecasting methods have rarely been implemented in real-world operational settings.
• A user-centered design approach is needed to bridge the gap between predictive models and operational needs.

Purpose of the Study:

• To develop and implement an accurate ED utilization prediction model tailored for operational planning.
• To engage nursing operations managers in selecting key metrics, models, and prediction horizons.
• To create a production dashboard for ED operational leaders.

Main Methods:

• Employed a user-centered design approach involving nursing operations managers across multiple hospital sites.
• Evaluated various time series and machine learning models using Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE).
• Selected and implemented the Prophet model, an open-source forecasting tool, for its superior performance.

Main Results:

• The Prophet model demonstrated the best performance across multiple hospital sites based on MAE and MAPE.
• Daily, 14-day ahead predictions were generated for critical ED metrics including arrivals, admissions, sitter needs, and ED holds.
• The model's implementation and monitoring design were established for ongoing operational use.

Conclusions:

• A user-centered approach successfully integrated advanced forecasting (Prophet model) into ED operational planning.
• Accurate, short-term ED utilization predictions can significantly enhance resource allocation and staffing decisions.
• This methodology provides a scalable solution for improving ED management within integrated health systems.