Multidimensional Determinants of Childhood Wasting among Children Under Five in West Africa

Technical Summary: Multidimensional Determinants of Childhood Wasting among Children Under Five in West Africa

Problem Statement
Childhood wasting, defined as acute undernutrition, remains a critical public health challenge in West Africa, with the region accounting for over 27% of the global burden. While previous research has identified socioeconomic factors (e.g., maternal education, household wealth) and isolated environmental issues as drivers, there is limited evidence regarding the combined influence of climate change, environmental degradation, water resources, and health system factors on wasting prevalence across the sixteen West African countries. Furthermore, existing studies often rely on standard logistic regression or focus on individual nations, failing to capture regional patterns or the bounded nature of wasting data (proportions between 0 and 1).

Methodology
This study utilized panel data from the World Bank's World Development Indicators (WDI) covering the period 2000–2023 for sixteen West African countries. The dependent variable was the prevalence of wasting (% of children under five). The authors employed a dual-methodological approach to address the limitations of traditional statistical models:

1. Beta Generalized Linear Mixed Model (BGLMM): Given that wasting is a proportion bounded between 0 and 1, the authors utilized a BGLMM with a logit link function. This model included fixed effects for various covariates (climate, environmental, water-resource, socioeconomic, and health system factors) and a country-level random effect to account for unobserved heterogeneity. Average Marginal Effects (AMEs) were computed to interpret the coefficients as changes in prevalence.
2. Random Forest (RF) with SHAP: To capture non-linear relationships and interactions without strict distributional assumptions, a Random Forest model was trained (70:30 train-test split). Model performance was evaluated using RMSE, MAE, and R². To ensure interpretability, Shapley Additive Explanations (SHAP) were applied to rank predictor importance and determine the direction of influence for the most significant variables.

Data preprocessing included handling missing values using the MissForest algorithm, and multicollinearity was assessed via Variance Inflation Factors (VIF), all of which were found to be within acceptable limits.

Key Results
The study identified distinct yet overlapping sets of determinants through both statistical and machine learning frameworks:

• BGLMM Findings: The model revealed statistically significant associations between wasting and several factors. Agricultural water withdrawal and N₂O emissions were positively associated with higher wasting prevalence. Conversely, health expenditure, access to basic drinking water, forest area, and the year variable (indicating a general decline over time) were negatively associated with wasting. Interestingly, basic sanitation showed a positive association, which the authors attribute to confounding factors like rapid urbanization and persistent food insecurity in areas where sanitation improvements are occurring.
• Random Forest & SHAP Findings: The RF model ranked agricultural water withdrawal as the most informative predictor, followed by urban population, average precipitation, PM2.5 exposure, access to basic drinking water, and water productivity. SHAP dependence plots confirmed that higher agricultural water withdrawal and PM2.5 exposure correlate with increased wasting, while higher urban population, precipitation, and access to basic drinking water correlate with reduced wasting.
• Convergence: Both models consistently identified agricultural water withdrawal and access to basic drinking water as critical priority areas.

Key Contributions

• Methodological Advancement: The paper contributes to the literature by applying Beta regression (specifically BGLMM) to childhood wasting data, a more appropriate analytical approach for proportion-based outcomes than standard logistic regression.
• Hybrid Modeling: It integrates traditional econometric modeling with machine learning (Random Forest) and explainable AI (SHAP) to provide a robust, multi-perspective view of the determinants of wasting.
• Regional Scope: Unlike previous studies focusing on single nations, this research provides a comprehensive analysis across the entire West African region, highlighting regional patterns driven by climate and environmental shocks.

Significance and Claims
The authors claim that their findings underscore the necessity of integrating environmental and climate factors into nutritional strategies. The study posits that wasting in West Africa is not merely a result of dietary intake but is driven by a "vicious cycle" of environmental and structural shocks.

Specifically, the paper argues that:

1. Water Management is Central: The consistent identification of agricultural water withdrawal and drinking water access as top predictors suggests that policies must prioritize efficient water use in agriculture (e.g., drip irrigation, rainwater harvesting) and the maintenance of community water supply systems.
2. Climate-Nutrition Link: There is a direct link between climate action (SDG 13) and zero hunger (SDG 2). Reducing N₂O emissions and managing air pollution (PM2.5) are presented as indirect but vital strategies for improving child nutrition.
3. Policy Implications: The authors suggest that governments should promote climate-smart agriculture, drought-tolerant crops, and improved cookstoves to reduce household exposure to smoke. They emphasize that addressing wasting requires integrated approaches that simultaneously strengthen food security, environmental sustainability, and water resource management (SDG 6).

The paper concludes modestly, acknowledging limitations such as the use of country-level data which may mask within-country variations (e.g., specific feeding practices) and the lack of spatial random effects, suggesting these as areas for future longitudinal and spatial modeling.