GIS-based spatial analysis of fatal road traffic incidents in Medellin, Colombia: Road segment burden, facility buffer zone proximity, and spatial autocorrelation

Abstract

This study characterized the spatial distribution of fatal road traffic incidents in Medellín, Colombia, from January 2008 to August 2025. Of 702,540 records, 2,762 were fatal; 2,507 spatially eligible incidents were retained. Proximity to 2,065 education facilities and 78 strict hospital and emergency-care facilities was evaluated using 100-, 250-, and 500-m buffers. Each incident was assigned to the nearest segment of an all-road network (68,308 segments); segment burden and road-class distribution were assessed. Global and Local Moran’s I were calculated using k-nearest-neighbor weights (k=8) on segment centroids, with sensitivity analyses for repeated coordinates and alternative networks. 19.55%, 68.77%, and 94.22% of incidents were within 100, 250, and 500 m of an education facility, versus 3.91%, 14.84%, and 36.14% for hospital and emergency-care facilities. Only 1,331 segments (1.95%) received an incident. Residential roads had the largest burden, but several major-road and link classes held larger shares of incidents than of network length. Global Moran’s I was positive but weak (0.0245), whereas the unique-location hospital-proximity sensitivity analysis showed strong autocorrelation (0.7624–0.8875). Local Moran’s I identified 140 High–High and 1,866 Low–High segments. High-burden rankings were stable across network definitions. The findings provide a screening and prioritization framework, not causal or exposure-adjusted risk estimates.