Uncovering spatially heterogeneous predictors of reported malaria cases in the Ashanti Region of Ghana: a triangulated framework integrating correlation analysis, random forest, and multiscale geographically weighted regression.

Abstract

Introduction: Malaria remains a major cause of illness and death in sub-Saharan Africa. In 2023, WHO reported 263 million cases and 597,000 deaths globally. Ghana records a substantial malaria burden, and the Ashanti Region is among the regions contributing a large share of reported cases. Despite successive interventions, reported malaria incidence remains high.

Method: This study used a triangulated analytical framework to examine environmental and anthropogenic correlates of district-level reported malaria case counts in 2023 (n = 43 districts). Spearman's rank Correlation, Random Forest, and MGWR models were fitted to raw district-level laboratory-confirmed case counts, and a population-normalized incidence rate was used only for the descriptive spatial map. Spearman's rank correlation characterized global monotonic associations, Random Forest (RF) with partial dependence plots (PDPs) examined nonlinear marginal patterns, and Multiscale Geographically Weighted Regression (MGWR) examined spatially varying local associations, following a variance inflation factor (VIF) screen for multicollinearity.

Results: Enhanced Vegetation Index (EVI) was the most consistently ranked correlate across the three analytical methods and was the only RF importance estimate distinguishable from the permutation null at the unadjusted 0.05 level (permutation p = 0.017; not adjusted for 11 comparisons). RF validation showed substantial overfitting and limited out-of-sample generalization (in-sample R² = 0.629; LOOCV R² = 0.056; MAE = 6897 cases; RMSE = 8851 cases; MAPE = 55.8%). Eight predictors had a bandwidth of 3 k-NN, suggesting strong estimated spatial non-stationarity under the MGWR model. With only 43 districts, these narrow bandwidths may also reflect coefficient instability rather than a confirmed spatial scale.

Conclusion: Environmental and anthropogenic correlates of district-level reported malaria case counts varied across the Ashanti Region. These ecological associations are exploratory and hypothesis-generating, not evidence of causal transmission mechanisms or intervention effectiveness. Future studies should use reliable population denominators, facility-level coordinates, multiyear surveillance, entomological information, and socioeconomic covariates.