Abstract

Background childhood infectious diseases and related symptoms, such as fever, cough, and diarrhea among children constitute the leading cause of death in low and middle-income countries (LMICs). We examined the environmental predictors of double and triple burden of infection (D/TB) symptoms among under-five children using multilevel machine learning (ML) methods.
Methods We used Demographic and Health Surveys (DHS) data from 58 LMICs between 2000 and 2023. These data were merged with cluster-level particulate matter and nitrogen dioxide from the National Aeronautics and Space Administration and country-level data on political, social, and economic globalization from the World Bank report. We applied multilevel models to screen out the most important predictors of D/TB symptoms and applied machine learning algorithms to predict these symptoms among children across LMICs. We trained and validated the ML algorithms on (80,70, and 60%) of the data and tested on the remaining (20, 30, and 40%) with 2, 5 and 10 cross-validations.
Findings: Of 1,546,243 children;19.2%, 20.5% and 12.6% had fever, cough, and diarrhea, respectively; while the overall D/TB prevalence was 11.9% and 3.7%, respectively. The result revealed D/TB were associated with the location of a child, survey years, wealth index, family size, air pollutants, and environmental covariates. The estimated prevalence of both D/TB of symptoms substantially varies across districts [intraclass correlation (ICC =13.3%)] and countries (ICC=8.8%). We found that the Random Forest gave the maximum Area Under the Curve of 94% and 99% for D/TBs for the K10 protocol and 80:20 training and testing dataset splits.
Interpretations: The study found substantial variation in the prevalences of D/TB of illness among children under five and identified several environmental and sociodemographic predictors of these health outcomes. The Random Forest algorithm performed best in predicting these burdens.