Application of spatial generalized linear models in predicting visceral leishmaniasis causing vector: a case of Turkana County

Abstract

Visceral leishmaniasis (VL) remains a public health concern in Kenya, especially in Turkana County, where control strategies often lack a multidisciplinary One Health approach integrating human, animal, and environmental health. This study investigates how environmental (NDVI, brightness), weather (temperature, evapotranspiration, wetness), and spatial factors (latitude, longitude) influence vector abundance. We applied the Spatial Autoregressive Moving Average (SARMA) model alongside machine learning algorithms XGBoost, Spatial Random Forest (RF), and k-Nearest Neighbors (KNN) to assess predictive accuracy. Model performance was evaluated using the Akaike Information Criterion (AIC), residual variance, spatial dependence tests, and classification metrics such as sensitivity, specificity, and Area Under the Curve (AUC). XGBoost outperformed SARMA, achieving the highest AUC (0.8997) and sensitivity (99.63%), and ranked Latitude (43.38%) and Longitude (29.12%) as the most influential variables, followed by NDVI (8.95%), Indoors Sticks (4.23%), Wetness (4.19%), and Humidity (2.13%). SARMA identified NDVI (β = 0.3638, p = 0.0079), Wetness (β = 0.4836, p = 0.0185), and ET (β = −0.0068, p = 0.0051) as significant, with strong spatial effects observed (ρ = 0.5440, λ = −0.7760). These findings underscore the value of spatial modeling in vector surveillance; while SARMA effectively captures spatial dependencies, XGBoost offers superior predictive power. The results support the development of data-driven, location-specific interventions for VL control in Turkana County, enhancing the precision and impact of public health efforts.

Keywords: Sandflies, Leishmaniasis, OneHealth, Social determinant, Transmission, Ecology