Population, provider, spatial and friction-surface representation choices alter modelled walking access to primary care in Uasin Gishu County, Kenya

Abstract

Background. Gridded estimates of geographic access to primary care are increasingly used to identify underserved areas, but each estimate rests on several representation choices whose influence is rarely reported together. We quantified how population allocation, provider eligibility, destination restriction, facility attachment and friction-surface land-cover values change modelled walking access in one Kenyan county. Methods. We conducted a historical, public-data sensitivity analysis across 30 wards of Uasin Gishu County using WorldPop 2016 population (389,628 cells of about 100 m), an archived inventory of government dispensaries and health centres, and the Malaria Atlas Project (MAP) 2019 walking friction surface. Travel cost was computed on an eight-neighbour graph of about 1-km cells. Contrasts were evaluated against one baseline: government providers, WorldPop allocation, local facility attachment and the unmodified friction surface. The outcome was the share of represented population more than 60 minutes' walk from the nearest eligible facility. Results. The baseline share beyond 60 minutes was 7.17%. Admitting faith-based and non-governmental facilities lowered it to 4.83%; zero-cost facility attachment lowered it to 5.61%; and within-ward reallocation using Global Human Settlement population lowered it to 6.68%. Restricting destinations to county-labelled facilities doubled the share beyond 5 km straight-line distance (1.52% to 3.34%). The largest single effect came from 76 friction cells carrying the surface's slowest walking value (1 km/h). They held 5.2% of represented population and were concentrated in dense Eldoret wards. Replacing them with neighbourhood values lowered the county share to 4.52% and reduced the modelled share beyond 60 minutes in Huruma, Langas and Kapsaos from 16.5–17.5% to zero. Conclusions. Modelled access indicators in this county were more sensitive to friction-surface land-cover values and provider eligibility than to population allocation. Analysts should report and test these choices, and inspect friction values in densely populated cells, before using ward indicators to target services. Preprint, not peer reviewed. Companion papers: measurement sensitivity, decision analysis, and external check of inputs (Uasin Gishu County, Kenya). Code and derived data: S1 zip.