Comprehensive geographic data are essential to accurately model geographic accessibility to healthcare and to guide equitable health system planning and implementation. In low-income countries such as Madagascar, however, incomplete road and building data in global databases such as OpenStreetMap (OSM) limit the precision and operational applications of geographic accessibility models. In Ifanadiana District, we piloted a large-scale OSM mapping effort to address this gap and inform the activities of local health actors, mapping over 100,000 buildings and more than 20,000 km of footpaths via the HOT Tasking Manager.

These data enabled the development of a suite of novel, high-resolution, decision-support tools for health program managers and community health workers. First, we estimated shortest-path itineraries and travel times to primary healthcare facilities for every household in the district using routing algorithms combined with locally calibrated travel speed models derived from field GPS data and remote sensing inputs. This approach allowed us to move beyond traditional accessibility models based on Euclidean distance or friction surfaces, generating more precise, context-specific estimates of travel times that reflect the actual routes used by populations (Figure 1) [1,2]. These analyses revealed major geographic inequalities in access, with a large proportion of the population living more than one hour from primary healthcare facilities, and some areas exceeding 4-5 hours of travel time.

Building on these results, we showed that geographic barriers persist even at the level of community health programs. Using high-resolution spatial data linked with healthcare utilization records, we quantified a strong distance-decay effect in the use of community health services, with consultation rates decreasing by approximately 28% per additional kilometer between households and community health worker (CHW) sites [3]. We then developed methods to optimize the geographic configuration of community health systems (Figure 2). While relocating existing CHW sites yielded limited improvements, adding additional sites in geographically dispersed areas was predicted to substantially increase consultation rates, highlighting the importance of adapting community health strategies to local geographic contexts.

In addition, we designed routing tools to support proactive, door-to-door care delivery by community health workers. Using OSM data and route optimization algorithms based on the Vehicle Routing Problem with Time Windows (VRPTW), we developed models to generate optimal itineraries for CHWs visiting all households in their catchment areas (Figure 3) [4]. These tools enabled detailed estimation of workload and personnel requirements for different types of interventions, such as mass distribution campaigns or proactive community case management programs. Results showed large variability in resource needs depending on geographic dispersion, with required personnel-days ranging from fewer than 15 to more than 60 per CHW catchment for mass campaigns. These analyses were integrated into an interactive e-health platform allowing program managers to visualize optimal routes, schedules, and resource needs under different implementation scenarios, thereby directly supporting operational decision-making at the local level.

Building on this pilot, we have scaled the approach to seven additional districts in southeastern Madagascar, mapping a total area of approximately 30,000 km²—comparable to the size of Belgium—and adding nearly one and a half million buildings and approximately 200,000 km of transportation networks, the vast majority consisting of footpaths [5]. This large-scale effort was conducted through a combination of dedicated mapping teams and participatory approaches, using standardized workflows and tools derived from the initial pilot. Using these expanded data, we generated household-level estimates of geographic accessibility to both primary healthcare facilities and community health sites across a population of more than 2.5 million people. These analyses revealed substantial heterogeneity across districts, with between 24% and 65% of the population living within one hour of a primary healthcare facility, compared to much higher coverage for community health sites. These findings reinforce the critical role of community health systems in mitigating geographic barriers, while also highlighting their limitations in highly dispersed settings.

Beyond accessibility modeling, this scale-up provided key insights into the completeness of OSM data and the feasibility of nationwide mapping. By comparing mapped data with AI-generated building and road datasets, we found that most regions of Madagascar remain poorly mapped, particularly in rural areas where accurate data are most needed. We estimated that mapping the entire country at a level sufficient to support precise, household-level accessibility modeling would require between 220 and 350 person-years of effort, corresponding to an estimated cost of approximately one million US dollars depending on assumptions and mapping strategies. These results demonstrate that exhaustive national mapping is technically feasible but requires substantial coordination, investment, and the integration of multiple approaches.

This work highlights how OSM can serve not only as a mapping platform, but as a foundational geospatial infrastructure for public health. We emphasize the importance of integrating mapping, spatial modeling, and health system data to move from descriptive analyses to actionable decision-support systems. By enabling precise measurement of geographic inequalities and supporting the design of targeted interventions, OSM-based approaches can directly contribute to reducing disparities in healthcare access and advancing progress toward universal health coverage.