Artificial intelligence-driven forecasting and shift optimization for pediatric emergency department crowding.

Area of Science:

• Artificial Intelligence in Healthcare
• Machine Learning Operations (MLOps)
• Healthcare Systems Engineering

Background:

• Pediatric Emergency Departments (PEDs) face overcrowding challenges impacting patient care and operational efficiency.
• Traditional methods for forecasting and staffing are often reactive and may not adapt to dynamic patient volumes.
• Optimizing physician schedules is crucial for managing workload and ensuring adequate patient coverage.

Purpose of the Study:

• To develop and evaluate an AI-driven system for forecasting PED overcrowding.
• To optimize physician shift schedules using machine learning operations (MLOps).
• To enhance the accuracy of overcrowding predictions and improve workforce distribution.

Main Methods:

• Analysis of 352,843 PED admissions from January 2018 to May 2023.
• Development and comparison of twenty time-series forecasting models, including deep learning architectures.
• Implementation of an MLOps simulation for automated data updates and model retraining.
• Optimization of physician shifts using integer linear programming based on forecasted patient volumes.

Main Results:

• Advanced deep learning models achieved R² scores up to 75%, with MLOps improving median R² from 44% to 60%.
• Shift optimization adjusted staffing in 69 out of 84 shifts, increasing physician allocation during peak hours.
• The AI system reduced the patient-to-physician ratio by an average of 4.32-4.40 patients during specific shifts.

Conclusions:

• The AI and MLOps integrated system effectively forecasts PED overcrowding and optimizes physician shifts, outperforming traditional methods.
• The system demonstrated resilience to data drift and improved workforce distribution without increasing staff numbers.
• Future research should focus on multicenter validation and real-world implementation for broader impact.