Determinants and disparities in skilled birth attendants during childbirth in Bangladesh: A study of machine learning and decomposition analysis

This study delves into the determinants and socioeconomic disparities influencing the utilization of Skilled Birth Attendants (SBAs) for childbirth in Bangladesh, deploying advanced statistical and machine learning (ML) methods. A Multilevel Binary Logistic Regression (MBLR) accounting for hierarchical data structures identified wealth, education levels, and adequate Antenatal Care (ANC) from qualified practitioners as meaningful predictors, while also revealing regional variances-with Khulna and Dhaka divisions and urban areas associated with higher SBA prevalence in contrast to rural areas and divisions like Mymensingh and Sylhet. Socioeconomic inequalities were evaluated using the Wagstaff-type decomposition method. The Concentration Index (CI) and Concentration Curve (CC) indicated that SBA usage was disproportionately higher among affluent households, with ANC visits, wealth index and, schooling contributing most to observed inequity. For prediction purposes, ML algorithms including Artificial Neural Network (ANN), Light Gradient Boosting Machine (LGBM), CatBoost (CATB), and Logistic Regression (LR) and 6 more were trained and gauged utilizing accuracy, precision, recall, F1 score, specificity, and, Area Under the Curve (AUC). ANN achieved the highest AUC (0.81). Model interpretability was enhanced through SHAP (SHapley Additive explanations) plots. The integration of MBLR, Wagstaff-type decomposition and, interpretable ML offers a comprehensive framework for comprehending and anticipating SBA utilization. Findings emphasize the necessity to address socioeconomic imbalances and ensure equitable access to quality ANC, providing actionable evidence for targeted maternal healthcare policies in Bangladesh.