Machine learning-based prediction of bronchiolitis in children under two years: a multicenter study within a single
Raoof Nopour1, Saeed Barzegari2, Mostafa Shanbehzadeh3
1Social Determinants of Health Research Center, Semnan University of Medical Sciences, Semnan, Iran. nopour.r70@gmail.com.
BMC Infectious Diseases
|June 12, 2026
Summary
Machine learning accurately predicts infant bronchiolitis risk using clinical and socioeconomic data. Interpretable AI identifies high-risk infants for early intervention and resource allocation.
Area of Science:
- Pediatric Medicine
- Computational Biology
- Public Health
Background:
- Bronchiolitis is a major cause of infant hospitalization, posing a challenge for early risk prediction.
- Traditional models struggle with complex interactions and lack clinical transparency.
Purpose of the Study:
- Develop, validate, and interpret machine learning (ML) models for predicting infant bronchiolitis risk.
- Utilize clinical and socioeconomic determinants for enhanced predictive accuracy.
Main Methods:
- Retrospective analysis of 1,260 children across 5 pediatric centers.
- Applied 8 ML algorithms, including ensemble methods, with rigorous data preprocessing and imputation.
- Evaluated performance using AUC, confusion matrices, Decision Curve Analysis (DCA), and calibration plots.
- Employed Explainable AI (XAI) with Shapley Additive exPlanations (SHAP) for model interpretation.
Main Results:
- XGBoost model achieved an AUC of 0.863, outperforming base learners.
- DCA and calibration plots confirmed XGBoost's superior clinical utility and accuracy.
- SHAP analysis revealed overcrowding, malnutrition, and maternal smoking as key predictors.
Conclusions:
- Integrated XGBoost and SHAP offer a robust, transparent framework for early bronchiolitis prediction.
- The model supports early triage and resource allocation for high-risk infants.
- Facilitates a shift towards precision pediatric medicine through interpretable AI.