Predicting the Environmental Suitability and Identifying Climate and Sociodemographic Correlates of Guinea Worm (Dracunculus medinensis) in Chad

Abstract

A comprehensive understanding of the spatial distribution and correlates of infection are key for the planning of disease control programs and assessing the feasibility of elimination and/or eradication. In this work, we used species distribution modeling to predict the environmental suitability of the Guinea worm (Dracunculus medinensis) and identify important climatic and sociodemographic risk factors. Using Guinea worm surveillance data collected by the Chad Guinea Worm Eradication Program (CGWEP) from 2010 to 2022 in combination with remotely sensed climate and sociodemographic correlates of infection within an ensemble machine learning framework, we mapped the environmental suitability of Guinea worm infection in Chad. The same analytical framework was also used to ascertain the contribution and influence of the identified climatic risk factors. Spatial distribution maps showed predominant clustering around the southern regions and along the Chari River. We also identified areas predicted to be environmentally suitable for infection. Of note are districts near the western border with Cameroon and southeastern border with Central African Republic. Key environmental correlates of infection as identified by the model were proximity to permanent rivers and inland lakes, farmlands, land surface temperature, and precipitation. This work provides a comprehensive model of the spatial distribution of Guinea worm infections in Chad 2010-2022 and sheds light on potential environmental correlates of infection. As the CGWEP moves toward elimination, the methods and results in this study will inform surveillance activities and help optimize the allocation of intervention resources.

Figure 1.

Location of villages reporting Guinea worm infection in all hosts from 2010 to 2022. (A) Subset of Chad highlighting locations of Guinea worm endemic villages. (B) Insert is the entire map of Chad, indicating Guinea worm endemic villages.

Figure 2.

Variogram plot showing spatial correlation of the Guinea worm surveillance data. The empirical variogram is represented by the black dots; the theoretical variogram is represented by the black solid line.

Figure 3.

Model performance comparison as measured by area under the receiver operating characteristic and true skill statistic values of all model algorithms. The points represent the mean estimates, and the solid lines represents the 95% CIs.

Figure 4.

Variable importance as measured by the percentage increment in mean square error (MSE) by variable permutation.

Figure 5.

(A–G) Marginal effects plots for covariates included in the ensemble. The y axis is the response (i.e., probability of detection of Guinea worm infection), and the x axis is the full range of covariate values. The black lines represent the mean marginal effects, and the grey shading indicates the 95% bootstrap CIs.

Figure 6.

(A) Mean predicted probability of Guinea worm infection at 1 km × 1 km resolution. (B and C) Lower and upper confident limits, respectively.

Figure 7.

(A) Mean predicted probability of Guinea worm infection stratified by district. (B and C) Lower and upper confident limits, respectively.

Figure 8.

Maps of mean predicted probability of Guinea worm infection. (A) Gridded 1 km × 1 km resolution maps overlaid with location of villages reporting Guinea worm infection in all hosts from 2010 to 2022. (B) Maps stratified by district overlaid with location of villages under surveillance in 2022. Note the arrow pointing to an area that was predicted as high risk that subsequently began reporting new infections in March 2023.