From strategic planning to virtual clinics operational support to face emergencies: A COVID-19 case study

Health care systems are fundamental assets in our society, especially during emergencies such as epidemics. Novel technologies are asked to contribute to their effective functioning. To this aim, we propose three mathematical models for prediction and optimization tasks, using machine learning and mathematical programming methodologies. Their goal is to tackle emergencies at different stages: to relief pressure on the system by prevention policies, to provide efficient resource allocation to hospitals and finally to manage non-critical patients in virtual clinics, whenever fruitful. We design the three models to work in an interactive pipeline, and we embed them as tools of a decision support system, for which we sketch the overall structure. We detail the resolution method of each model, validating them with simulations based on real data in Northern Italy during the COVID-19 pandemic of 2020, and which includes, at the finest level, 86 hospitals of Lombardy and tens of thousands of patients. Our models capture more features of emergency response applications than those proposed in the literature, allowing to analyze more detailed response scenarios and policies. For instance, our experiments on simulated scenarios indicate that (a) blocking only 15% of the connections between provinces in an optimal way, produces epidemic dynamics similar to those of full province isolation (b) adding as few as 2% of resources to hospitals, optimally relocating resources, and optimally discharging mild patients to home care, would be enough to fully satisfy the health care demand in a baseline scenario (c) patients severity can be classified with few attributes, reaching an expected accuracy ranging from 72.1% to 84.5%, depending on the disease observation point in time. • A DSS integrating ML and mathematical programming (MP) models to manage epidemics. • Containment policies are assigned to geographical areas with MP embedding SIR model. • Health resources are relocated between hospitals with MP to meet the expected demand. • Disease progressions are predicted with ML to make better decisions for the patients. • The DSS is efficient and consistent with real system expectation.