Data-driven distributionally robust optimization for resilient healthcare resource planning and crisis mitigation

Healthcare systems face significant challenges during public health emergencies, particularly in efficiently allocating limited resources while maintaining service quality under uncertainty. This paper addresses the critical issue of resource allocation during pandemics, focusing on facility location, capacity planning, and workforce management under demand uncertainty and operational constraints. Two distributionally robust optimization (DRO) models are developed, incorporating the L1-norm and the joint L1- and L∞-norm ambiguity sets to capture uncertainty. To enhance robustness, the Conditional Value-at-Risk (CVaR) criterion is employed to more accurately account for some of the worst realizations of random future demand scenarios. The models incorporate the strategic deployment of alternate care sites, multi-category patient demand with heterogeneous service requirements, and allocation of different resources, including general and intensive care unit (ICU) beds, ventilators, and healthcare personnel. Workforce planning is enhanced by considering cross-training and staff mobility. Furthermore, varying penalty costs for unmet demand across patient categories ensure priority of critical care needs. The models are validated through a case study focused on the COVID-19 pandemic in Southern California, demonstrating superior stability and robustness in out-of-sample testing compared to the traditional stochastic programming approach. The findings provide valuable insights for healthcare administrators in designing resilient and efficient healthcare logistics networks during public health emergencies.