Combining machine learning and optimization for the operational patient-bed assignment problem.

Area of Science:

• Operations Research
• Health Informatics
• Machine Learning

Background:

• Hospital bed assignment is complex due to unpredictable emergency patient arrivals.
• Limited bed capacity necessitates accurate forecasting to manage patient flow and avoid shortages.
• Current methods struggle with the uncertainty inherent in emergency admissions.

Purpose of the Study:

• To develop an improved model for patient bed assignment using machine learning (ML) for emergency patient forecasting.
• To enhance the accuracy of predicting emergency inpatient arrivals by integrating diverse data sources.
• To create an advanced optimization heuristic for real-world patient bed-assignment problems.

Main Methods:

• Implemented ML models incorporating weather, temporal, event, and occupancy data to forecast emergency patient arrivals.
• Developed a novel hyper-heuristic combining a pilot method with a greedy look-ahead (GLA) heuristic for bed assignment optimization.
• Validated ML forecasting against baseline averages, achieving up to 17% better RMSE.
• Benchmarked the hyper-heuristic against existing methods, including a Genetic Algorithm.

Main Results:

• ML forecasting significantly improved accuracy for emergency inpatient arrivals compared to traditional methods.
• The developed hyper-heuristic demonstrated superior performance, increasing the objective function by up to 5.3% over benchmarks.
• The combined ML forecasting and hyper-heuristic optimization yielded a 3.3% improvement on a real-world problem.

Conclusions:

• Machine learning provides a powerful tool for enhancing the accuracy of emergency patient arrival predictions.
• Advanced optimization heuristics, like the proposed hyper-heuristic, are crucial for efficient hospital bed management.
• Integrating predictive ML with optimization techniques offers substantial improvements in hospital operational efficiency and patient care.