Programa del congreso

Childhood immunization is a cost-effective public health intervention to reduce morbidity and mortality in children under five years of age; however, in low- and middle-income countries there are coverage gaps that limit their impact. In Honduras, these gaps are challenging, so analytical approaches are needed to detect populations at risk of incomplete schemes. The study sought to identify the determinants of not fully vaccinated in children aged 12 to 23 months, develop predictive classification models and explain the contribution of each variable with SHAP values. The ENDESA/MICS 2019 was analyzed, including 1,711 children aged 12–23 months. The outcome was defined as not fully vaccinated, without BCG, three doses of Polio, three doses of Pentavalent and one of MMR. Logistic regression, decision trees, Random Forest and Gradient Boosting models were trained and compared, evaluating performance with AUC-ROC and Brier Score, and interpretability was examined with SHAP. The prevalence of non-vaccination was 19.3%. The best performance was Gradient Boosting (AUC-ROC = 0.694; Brier Score = 0.138). According to SHAP, the main determinants were the age of the child (|SHAP| = 0.41), possession of vaccination card (|SHAP| = 0.31) and maternal education (|SHAP| = 0.10). The department of Gracias a Dios had the highest prevalence of non-vaccination (51.8%). The findings indicate that machine learning with SHAP allows for the precise identification and prioritization of factors associated with incomplete vaccination schedules in Honduran children aged 12 to 23 months