Abstract
<title>Abstract</title> <p>Background. The 2014–2016 Ebola virus disease (EVD) epidemic was the largest on record, and Sierra Leone reported more confirmed cases than any other affected country. Although the outbreak generated extensive epidemiological data, few analyses have combined visual density mapping with formal statistical validation of spatial clustering at the district level. This study applies an integrated geographic information systems (GIS) framework kernel density estimation (KDE), global Moran’s I, and the Getis–Ord Gi* statistic to characterise the spatial structure of the outbreak and its relationship to health infrastructure across Sierra Leone’s fourteen districts. Methods. District-level confirmed case counts (n = 10,675) consistent with published World Health Organization and Ministry of Health and Sanitation cumulative figures were analysed together with the geographic coordinates of the principal referral hospital in each district. KDE was used to visualise transmission intensity. Global spatial autocorrelation was assessed with Moran’s I, and local clustering with the Getis–Ord Gi* statistic using a 100 km fixed-distance spatial weights matrix and 999 permutations. Healthcare accessibility was proxied by the distance from each district to its nearest neighbouring referral hospital and to the Freetown hub. Results. Confirmed cases were highly concentrated: the Western Area (Urban and Rural) accounted for 46.2% of the national total and, together with Port Loko, for 59.8%. KDE identified a dominant western transmission core, an eastern origin cluster around Kailahun and Kenema, and intermediate nodes at Makeni and Bo. Global spatial autocorrelation was weak and not statistically significant at the district scale (Moran’s I = 0.045; p = 0.19). Local Gi* analysis identified a statistically significant western hot spot Port Loko (95% confidence) with Western Area Urban and Rural (90% confidence) and cold spots at Bonthe (95%) and Bo (90%); Kambia and Moyamba reached hot-spot significance through spatial proximity (spillover) to the high-burden western cluster. Hot-spot districts lay closer to referral care (mean 41 km) and to the Freetown hub (mean 52 km) than non-significant and cold-spot districts (55–66 km and 155–213 km, respectively). Conclusions. The outbreak’s spatial signature was dominated by urban amplification and connectivity to the western coastal hub rather than by strong district-level autocorrelation. The weak global clustering indicates that the 14-unit district scale is too coarse for robust hot-spot inference; chiefdom-level (149-unit) analysis is required to validate localised clusters. The integrated KDE–Gi*–infrastructure framework nonetheless offers a transparent, replicable basis for spatially targeted epidemic preparedness in Sierra Leone.</p>