High-resolution mapping of onchocerciasis risk in Ghana using spatial machine learning models

Onchocerciasis, caused by Onchocerca volvulus and transmitted through repeated bites of infected blackflies, remains endemic in several sub-Saharan African countries, including Ghana. Despite long-term ivermectin (IVM) mass drug administration (MDA), transmission persists, particularly in ecologically complex regions. Although geospatial models have advanced the identification of high-risk areas, nationwide fine-scale gridded maps for high-risk areas do not yet exist in Ghana.

Communities with documented infections were extracted from published studies and subsequently geocoded. Raster datasets representing environmental, climatic, and healthcare accessibility variables were generated using remote sensing platforms and global data repositories. The predictive models employed encompassed logistic regression, random forest, and gradient boosting machines.

Traditional and penalized logistic regression models yielded balanced accuracies of 63.17% and 53.56%, respectively, with areas under the receiver operating characteristic (ROC) curve (AUC) of 71.11% and 70.90%, respectively. The random forest model achieved a balanced accuracy of 70.30% and the highest AUC of 76.40%, along with a favorable sensitivity of 58.97%. Gradient boosting machines achieved a balanced accuracy of 61.69%, an AUC of 71.60%, and a sensitivity of 50.00%. Among all models, the random forest performed best. Based on predictions from the random forest model, high-risk regions were consistently identified in the central-west, the Volta Lake area, the southwest, and the central north. River density (relative risk [RR] = 7.99; 95% confidence interval [CI]: 3.16-20.19; p < 0.0001) was associated with increased transmission risk, whereas higher land surface temperatures (RR = 0.06; 95% CI: 0.02-0.20; p < 0.0001) and dense vegetation cover (RR = 0.19; 95% CI: 0.06-0.61; p = 0.005) correlated with reduced transmission risk.

These findings underscore the need for targeted vector control, improved healthcare access, and integrated social interventions. Generated risk maps provide valuable tools for adaptive strategies toward the World Health Organization's 2030 elimination targets.