Predicting exclusive breastfeeding in maternity wards using machine learning techniques

Area of Science:

• Maternal and Child Health
• Health Informatics
• Artificial Intelligence in Healthcare

Background:

• Maternity ward support is crucial for breastfeeding success in the first year.
• Quality improvement requires identifying factors influencing hospital breastfeeding indicators.
• Machine Learning (ML) and Explainable Artificial Intelligence (XAI) offer predictive and explanatory capabilities.

Purpose of the Study:

• To predict exclusive breastfeeding during the in-hospital postpartum stay using ML algorithms.
• To explain the ML model's behavior to support decision-making in maternity care.
• To identify key predictors influencing in-hospital exclusive breastfeeding rates.

Main Methods:

• Utilized a dataset of 2042 mothers from 18 hospitals in Eastern Spain.
• Collected data on demographics, breastfeeding history, clinical variables, and hospital support.
• Employed 10-fold cross-validation and metrics like ROC AUC, PR AUC, accuracy, and Brier score.
• Applied Shapley additive values for predictor importance and explanation.

Main Results:

• XGBoost algorithm demonstrated superior performance (ROC AUC=0.78, PR AUC=0.86).
• Primary predictors included pacifier use, breastfeeding self-efficacy, prior experience, birth weight, neonatal unit admission, and skin-to-skin contact timing.
• Baby-Friendly Hospital Initiative accreditation was also a significant factor.

Conclusions:

• The ML model effectively predicted exclusive breastfeeding during the hospital stay.
• Results highlight opportunities to improve care for specific maternal groups and newborn conditions.
• XAI identified non-linear relationships and effect heterogeneity, explaining variations in individual risk.