Estimating health facility-level catchment populations using routine surveillance data and a Bayesian gravity model

Accurate estimates of health facility catchment populations are crucial for understanding spatial heterogeneity in disease incidence, targeting healthcare interventions, and allocating resources effectively. Despite improvements in health facility reporting, reliable catchment population data remain sparse. This study introduces a Bayesian gravity model-based approach for estimating catchment populations at health facilities, with a focus on Zambias routine malaria surveillance data from 2018-2023. Our method integrates health-seeking behavior, facility attractiveness, and travel time, allowing for the development of probabilistic catchment areas that reflect the treat-seeking and facility selection process. We developed an open-source R package to implement this method, and we apply this model to Zambian health facilities and compare the results to reported headcount data, highlighting improvements in stratification of malaria incidence rates. Additionally, we validate the models sensitivity using real-world treatment-seeking data from household surveys in Southern Province, Zambia, demonstrating its utility in enhancing sub-district-level health facility data for strategic planning. Validation of model facility selection rates compared to the treatment-seeking data showed a model sensitivity of 0.72 overall, with sensitivity reaching 0.89 for households within 2 kilometers of their preferred facility. This validation supports the models ability to closely estimate treatment-seeking behavior patterns, offering a scalable, accurate tool for enhancing local-level decision-making for health interventions, contributing to improved targeting and understanding of healthcare access patterns.