WHAT OPEN GEOSPATIAL DATA CAN AND CANNOT TELL US ABOUT GEOGRAPHIC ACCESS TO RURAL PUBLIC PRIMARY CARE: A VILLAGE-LEVEL CASE STUDY FROM SINDH, PAKISTAN

Abstract

Open geospatial data is now used as a matter of routine in accessibility analysis for low- and middle-income countries, but how complete that data actually is for rural health infrastructure has never been checked against an authoritative registry. This study measures the deficit, and what the deficit does to the results, for the province of Sindh, Pakistan. Travel time from 5,159 villages to the nearest health facility was computed over the OpenStreetMap road network using multi-source Dijkstra routing, with independent routing verification agreeing to within three percent. Two facility layers were compared while every other input to the model was held constant: open data alone (n=2,725), and open data combined with 616 facilities recovered from the register of the province's primary-care operator and geocoded by name within district (n=3,308). Only 5.2 percent of the register facilities fall within 250 metres of any facility in the open dataset, so the two layers are describing different populations of buildings rather than the same buildings at different levels of completeness. Open-data-only results overestimated the population-weighted median travel time by a factor of 2.01, and the population more than thirty minutes away from care by a factor of ten, 6.95 million people against 0.68 million. Facility counts per district rose by as much as thirty-four-fold, and only in rural districts, while the metropolitan districts gained nothing at all. District ranking priorities differed between the two facility layers, and so did the district carrying the maximum population-weighted underservicing burden. Satellite imagery validation of a stratified random sample confirmed a consistent facility structure at 84 percent of geocoded points. In this setting open geospatial data leaves out the vast majority of rural public health facilities, and analyses built on top of it risk confusing the populations that are actually underserviced with the ones that are merely misreported.