We integrate AI with a digital twin to optimize ambulance diversion policies in emergency departments during healthcare crises. In particular, our approach creates a data-adaptive decision rule that can improve mortality outcomes by dynamically adjusting diversion thresholds based on predicted patient surges, while addressing the computational challenges of real-time optimization on the digital twin through AI-based metamodeling.
The simulation.ipynb notebook contains a step-by-step guide on how to run the simulation and generate the results. Models for forecasting arrival rates and predicting thresholds are too big to store in Git; please open an issue in this repository with your contact information (i.e., email address) and they can be provided.