Análise da acessibilidade geográfica às Unidades Básicas de Saúde (UBS) em Teresina, Piauí: uma rotina computacional aberta e reprodutível para o diagnóstico multimodal de iniquidades territoriais em saúde

Abstract

Primary Health Care constitutes the main gateway to the Brazilian Unified Health System (SUS), a role that presupposes the geographic accessibility of Basic Health Units (UBS) to the population. However, traditional measurement methods, such as the Two-Step Floating Catchment Area (2SFCA), rely on spatial simplifications that distort the diagnosis, particularly by representing population demand solely through the census tract centroid. In light of this, this research aimed to build an open and reproducible computational routine in Python to apply the Areal Enhanced Two-Step Floating Catchment Area (AE2SFCA) method to diagnose inequalities in access to UBS in the urban area of Teresina, Piauí, Brazil. Methodologically, the road network extracted from Overture Maps had its functional classification gaps filled by a Graph Neural Network. The network was then converted into time impedances, incorporating an urban friction factor to generate multimodal isochrones of 10, 20, and 30 minutes for both car and pedestrian modes. The AE2SFCA index was calculated based on 81 UBS and 474 primary care teams, and was compared with the traditional model and median household income using Spearman's rank correlation and bivariate local spatial association (LISA). The results indicated a comfortable aggregate supply (5.87 teams per 10,000 inhabitants), yet highly unequal in space. While 30-minute physical coverage reached 98.5% of the urban area by car, pedestrian reach was limited to 69.5%. Under the vehicular scenario, accessibility was positively associated with income (Spearman of +0.438), clustering critical exclusion zones in the periphery, whereas on foot this association reversed (−0.199), confirming the socioterritorial hypothesis (H_1). From a methodological perspective, the areal approach proved equivalent to the traditional one in the motorized mode (Spearman of 0.991), but corrected the misdiagnosis of 64 false health deserts in the pedestrian mode, confirming the methodological hypothesis (H_2).