Early prediction of plastic bronchitis in pediatric patients with Mycoplasma pneumoniae pneumonia by interpretable machine learning algorithms.
Mycoplasma pneumoniae pneumonia (MPP) can cause plastic bronchitis (PB), a rare, life-threatening condition. However, current diagnostic methods often fail to identify early-stage PB in children.The aim of our study was to develop machine learning algorithms to identify early-stage PB in pediatric patients with MPP.
This retrospective cohort study involved 307 pediatric patients with MPP who underwent bronchoscopy intervention from April 2023, to June 2025.Patients were randomly split into training and test sets (7:3). After feature selection using LASSO and Boruta algorithms, four algorithms, namely, extreme gradient boosting (XGBoost), logistic regression, random forest, and support vector machine, were employed to construct machine learning (ML) models through 5-fold cross-validation. Model performance was evaluated using the area under the curve (AUC), calibration curves, and decision curve analysis (DCA). The best-performing ML was selected using AUC, and feature importance in the model was ranked using SHapley Additive exPlanations (SHAP). Finally, a web-based risk predictor was constructed to facilitate user operability.
MPP children with PB demonstrated more significant abnormalities in inflammation- and nutrition-related indices compared to those without PB. The XGBoost algorithm exhibited the best predictive performance, surpassing other models (logistic regression, random forest, and support vector machine) with an AUC of 0.948 (95% CI: 0.919-0.973), a sensitivity of 0.904, and a specificity of 0.858 on the training set, and an AUC of 0.905 (95% CI: 0.843-0.957), a sensitivity of 0.812, and a specificity of 0.852 on the test set. This algorithm also presented good calibration and net clinical benefit. SHAP analysis identified the retinol-binding protein 4 level, M. pneumoniae cycle-threshold value, D-dimer level, fever duration before admission, C-reactive protein-to-albumin ratio, and presence of pleural effusion as key predictors. To facilitate the clinical adoption, a freely accessible online calculator has been developed (https://plasticbronchitis.shinyapps.io/plastic_bronchitis_risk_calculator/).
The developed interpretable ML models deployed in the network application can help clinicians identify children at high risk of developing PB earlier and tailor timely bronchoscopy intervention and nutritional support as well as anti-inflammatory therapy.
This retrospective cohort study involved 307 pediatric patients with MPP who underwent bronchoscopy intervention from April 2023, to June 2025.Patients were randomly split into training and test sets (7:3). After feature selection using LASSO and Boruta algorithms, four algorithms, namely, extreme gradient boosting (XGBoost), logistic regression, random forest, and support vector machine, were employed to construct machine learning (ML) models through 5-fold cross-validation. Model performance was evaluated using the area under the curve (AUC), calibration curves, and decision curve analysis (DCA). The best-performing ML was selected using AUC, and feature importance in the model was ranked using SHapley Additive exPlanations (SHAP). Finally, a web-based risk predictor was constructed to facilitate user operability.
MPP children with PB demonstrated more significant abnormalities in inflammation- and nutrition-related indices compared to those without PB. The XGBoost algorithm exhibited the best predictive performance, surpassing other models (logistic regression, random forest, and support vector machine) with an AUC of 0.948 (95% CI: 0.919-0.973), a sensitivity of 0.904, and a specificity of 0.858 on the training set, and an AUC of 0.905 (95% CI: 0.843-0.957), a sensitivity of 0.812, and a specificity of 0.852 on the test set. This algorithm also presented good calibration and net clinical benefit. SHAP analysis identified the retinol-binding protein 4 level, M. pneumoniae cycle-threshold value, D-dimer level, fever duration before admission, C-reactive protein-to-albumin ratio, and presence of pleural effusion as key predictors. To facilitate the clinical adoption, a freely accessible online calculator has been developed (https://plasticbronchitis.shinyapps.io/plastic_bronchitis_risk_calculator/).
The developed interpretable ML models deployed in the network application can help clinicians identify children at high risk of developing PB earlier and tailor timely bronchoscopy intervention and nutritional support as well as anti-inflammatory therapy.