Ghana survey finds anemia in 49% of children aged 6–59 months

Researchers analyzed 3,858 children aged 6–59 months from Ghana’s 2022 Demographic and Health Survey, with children grouped across 235 districts. They used a two-level random-intercept logistic regression and three machine-learning classifiers, with districts separated between training and test sets. Predictors were classified as high-confidence when they were statistically significant in the multilevel model and ranked among the top 15 by SHAP importance.

Anemia prevalence was 49.0% (95% CI, 46.4%–51.6%); district-level variation was significant, with an intraclass correlation coefficient of 6.2%. Nine predictors met the study’s high-confidence definition: child age and sex, mean temperature, maternal anemia and age, birth order, population density, stunting and fever. Elastic Net had the highest test-set AUC at 0.727 (95% CI, 0.702–0.755), compared with 0.707 for Random Forest and 0.685 for XGBoost. Four districts were identified as having the highest odds of anemia.

The findings describe population-level associations and potential geographic variation; they do not establish causes or show that targeted interventions will reduce anemia. The authors noted that several contextual variables ranked highly in machine-learning analysis but were not significant in the multilevel model, an unresolved discrepancy. Model discrimination was moderate, and the analysis does not establish clinical changes for individual children.