An Estimate-then-Optimize Approach to Improve Access to Healthcare Services
“Geographic inequalities in access to essential health services are well-documented, extending beyond a rural-urban divide to include socioeconomic, racial, and other disparities. Proximity to hospitals, clinics, healthcare providers, and pharmacies varies widely, posing a challenge in determining where to strategically locate such facilities. Demand for each service depends on the population health in the catchment area, individual preferences, provider capacity, and other factors. This study introduces a novel estimate-then-optimize framework combining structural (BLP) demand estimation with a choice-based optimal facility location model to maximize health service utilization. An advantage of this empirical approach is its reliance on aggregated data (e.g., market share) rather than individual choices or outcomes.
We illustrate our proposed methodology by examining the Federal Retail Pharmacy Programme — a historic public-private partnership that administered millions of COVID-19 vaccinations — in California. Our demand estimates reveal that residents of socioeconomically vulnerable communities are more sensitive to travel distances to pharmacy-based vaccination sites. Augmenting the existing set of pharmacies with 500 strategically selected retail stores serving lower-income communities could increase predicted vaccinations by 2.9 percent overall, translating to 770,000 additional vaccinations statewide, and by 5.4 percent in the least healthy neighborhoods. By combining a structural demand model with prescriptive analytics, our study offers policymakers and practitioners a systematic, data-driven framework for optimizing healthcare delivery networks. Key insights applicable across settings include: (1) demand estimates must account for socioeconomic heterogeneity, (2) optimization approaches outperform greedy policies, especially under spatial inequities, and (3) efficient network design prioritizes strategic site selection over simple expansion.”