Early Prediction of Plastic Bronchitis in Pediatric MPP Patients via Interpretable ML

A retrospective cohort study of 307 pediatric patients with Mycoplasma pneumoniae pneumonia (MPP) who underwent bronchoscopy intervention from April 2023 to June 2025. The dataset was used to develop and validate machine learning models for the early prediction of plastic bronchitis (PB). The best-performing model, XGBoost, achieved an AUC of 0.905 on the test set, and a web-based risk predictor was deployed.

Use Cases

• Training binary classification models to predict plastic bronchitis risk based on clinical indices like retinol-binding protein 4 and D-dimer levels.
• Benchmarking model performance (e.g., XGBoost vs. logistic regression) for rare disease prediction in pediatric cohorts.
• Developing interpretable clinical decision support tools using SHAP analysis on key predictors such as fever duration and pleural effusion.
• Creating web-based clinical calculators for early risk assessment, as demonstrated by the provided online application.

Strengths

• Model performance is quantitatively reported, with the XGBoost model achieving an AUC of 0.905 (95% CI: 0.843–0.957) on the test set.
• The study includes 307 patient records, providing a basis for model development.
• Feature importance was analyzed using SHAP, identifying six key clinical predictors for interpretability.
• A functional, freely accessible online risk calculator is provided for clinical adoption.

Limitations

• Row count for the underlying clinical data is unknown, which may limit suitability assessment for other modeling tasks.
• Column-level documentation is absent; field semantics must be inferred from the research description after download.
• The dataset is very small at 21.1 KB, indicating limited raw data scope.

Provenance

Source

figshare, author Pei Wang.

Collection Method

Retrospective cohort study of pediatric patients who underwent bronchoscopy.

Time Range

April 2023 to June 2025.

Freshness

Last updated 2026-04-23 05:34:41; freshness should be verified.