Abstract
BACKGROUND: Malnutrition among children under five remains a pressing public health issue in Bangladesh. Identifying its determinants is critical for designing effective interventions. This study aims to evaluate the suitability of statistical models that account for the ordinal nature of malnutrition categories, comparing Multinomial Logistic Regression (MLR) and the Proportional Odds Regression Model (POM) using data from the sixth round of UNICEF's Multiple Indicator Cluster Survey (MICS). METHODS: Child nutritional status was assessed using weight-for-age Z-scores (WAZ), categorized into severely undernourished, moderately undernourished, and nourished. MLR and POM were applied to model the relationship between malnutrition and various socio-demographic and health-related factors. Model performance was compared using Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). RESULTS: POM demonstrated superior model fit (AIC: 8788.996, BIC: 9099.4353) compared to MLR (AIC: 8844.849, BIC: 9451.617). Significant predictors of malnutrition were identified through POM which included geographical division, child's sex, mother's BMI, mother's education, prenatal care, birth size, and household wealth index. CONCLUSIONS: The Proportional Odds Regression Model outperformed Multinomial Logistic Regression by effectively capturing the ordinal structure of malnutrition categories. These findings underscore key determinants of child malnutrition and offer valuable guidance for targeted nutritional policies and development programs in Bangladesh.