An explainable machine learning model predicts pediatric varicella encephalitis.

Area of Science:

• Pediatrics
• Neurology
• Infectious Diseases

Background:

• Pediatric varicella encephalitis is a rare but severe complication of varicella (chickenpox).
• Early diagnosis is difficult due to non-specific symptoms and lack of biomarkers.
• Accurate identification is crucial for timely intervention and improved prognosis.

Purpose of the Study:

• To develop a predictive model for early identification of pediatric varicella encephalitis.
• To create a practical tool for clinicians to assess patient risk.
• To improve diagnostic accuracy and facilitate prompt treatment.

Main Methods:

• Retrospective analysis of 201 children with varicella.
• Utilized LASSO regression, XGBoost, and random forest algorithms for feature selection.
• Constructed and validated prediction models using 6 algorithms, including SHAP analysis for interpretation.

Main Results:

• Identified 6 key clinical variables for predicting pediatric varicella encephalitis.
• The random forest model demonstrated high predictive performance (AUC=0.950).
• Rash duration, headache, and vomiting were significant predictors; a web application was developed for risk stratification.

Conclusions:

• A robust random forest model accurately identifies children with varicella encephalitis.
• The model offers significant clinical utility for early detection and intervention.
• This tool supports timely management, potentially improving patient outcomes.