Limited geographic access to obstetric care prevents some pregnant people from receiving timely and risk-appropriate services. This challenge is especially acute in rural areas, where rural residents often travel far distances to obstetric care. Furthermore, obstetric access is worsening due to the growing number of closures of rural hospitals’ obstetric units, often due to financial concerns. In response, government organizations have initiated programs to prevent rural hospital closures, but there are limited tools to identify where best to allocate these resources. In this work, we propose the Maximum Choice-based Expectation Facility Location Problem (MCE-FLP), a generalization of the maximum capture facility location problem, to forecast which obstetric unit closures would have the worst impact on expected travel distance to care. In this problem, we model patients’ obstetric care-seeking behavior using a multinomial logit discrete choice model, which we show is more accurate than assuming patients seek their nearest obstetric unit. We show that the MCE-FLP’s objective function is not convex or concave in general, and thus the state-of-the-art maximum capture solution methods are not applicable. We present two linear reformulations of the MCE-FLP and design branch-and-cut approaches using analytical solutions to obtain the cuts. In our case study, we apply the MCE-FLP to the obstetric care system in the state of Georgia, and we find that the projected worst-case closures are isolated from other obstetric units and would impact above-average rates of marginalized groups. In this work, we propose a novel generalization of the maximum capture problem, which maximizes an expected value that is dependent on patient/consumer choice, and we demonstrate that this problem can be applied to determine worst-case obstetric closures that should be prevented to maintain critical access to care.