Peer-reviewed article
Diet, Diarrhoea and Maternal Factors Associated With Wasting in Ghanaian Children Aged 6-23 Months: A Machine Learning Predictive Approach Using the 2022 Demographic and Health Survey.
Luis Javier Sánchez-Martínez, Mohammed Bukari, Martin Nyaaba Adokiya ·
Summary
Written by AI, not yet AICS's view · claude-haiku-4-5-20251001This study analysed data from 1,506 Ghanaian children aged 6-23 months to understand wasting (being too thin for height) and what factors predict it. Wasting affected 9.4% of children in Ghana, with significant differences between regions. Children were less likely to be wasted if their mothers were heavier, employed, if they were female, or if they ate animal-source foods like eggs or meat; recent diarrhoea increased wasting risk. Using machine learning, researchers built a model that could identify which children were at risk with reasonable accuracy.
Key points
- Wasting prevalence in Ghana was 9.4% with marked regional heterogeneity
- Lower maternal body weight, lack of maternal formal employment, male sex of child, and lack of animal-source food consumption increased wasting risk
- Recent diarrhoeal episodes were strongly associated with higher odds of wasting
- An ensemble machine learning model achieved AUC of 0.79 with 85% sensitivity and 62% specificity for predicting wasting
- Wasting is associated with interacting biological vulnerabilities and ecological constraints during a critical developmental window
Limitations
This is a cross-sectional analysis, so it cannot establish causation or the direction of relationships between factors and wasting. The findings are specific to Ghana's context in 2022 and may not generalise to other countries or time periods.
Abstract
OBJECTIVES: This study estimates the prevalence of wasting and examines associated factors among children aged 6-23 months in Ghana from a machine-learning perspective, with a particular focus on predicting new cases to inform preventive efforts. METHODS: This study is a secondary analysis of the 2022 Ghana Demographic and Health Survey, which includes 1506 children. Wasting status was assessed using the weight-for-height z-score. A broad set of 150 child-, maternal-, household- and environmental-level variables was considered. Factors associated with wasting were first identified using survey-weighted logistic regression, whereas Multiple Correspondence Analysis was applied to explore structural relationships among predictors. Subsequently, an ensemble machine learning model was fitted to evaluate the combined predictive capacity of these factors in classifying wasting status, also comparing algorithm performance. RESULTS: Wasting prevalence in Ghana was 9.4%, with marked regional heterogeneity. Lower odds of wasting were associated with higher maternal body weight, maternal formal employment, female sex of child and consumption of animal-source foods, whereas recent diarrhoeal episodes were associated with higher odds. The exploratory analyses further showed that these factors clustered into coherent vulnerability profiles, distinguishing more clearly between wasting and non-wasting patterns. The ensemble model outperformed individual algorithms, achieving an AUC of 0.79, with high sensitivity (0.85) and moderate specificity (0.62), indicating moderate-to-good internal predictive performance. CONCLUSIONS: These findings suggest that wasting is associated with interacting biological vulnerabilities and ecological constraints during a critical developmental window. Integrating nutritional quality, infection control and maternal socioeconomic conditions is essential for addressing acute malnutrition in early childhood and for advancing a holistic understanding of child growth within human populations, guiding data-driven public policies.