Interpretable machine learning model for prediction functional cure in chronic hepatitis B patients receiving Peg-IFN...

Area of Science:

• Hepatology
• Machine Learning in Medicine
• Virology

Background:

• Functional cure is the primary goal for treating chronic hepatitis B (CHB).
• Predictive models are needed to identify CHB patients likely to achieve functional cure with treatment.
• Pegylated interferon (Peg-IFN) is a key treatment modality for CHB.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting functional cure in CHB patients undergoing Peg-IFN therapy.
• To identify optimal time points and clinical variables for predicting functional cure.
• To create a practical tool for clinicians to assess treatment efficacy.

Main Methods:

• Retrospective analysis of 534 CHB patients for model construction and 269 for external validation.
• Development of seven ML models using baseline, week 12, and week 24 data.
• Variable selection via Boruta and LASSO regression, with performance evaluation using AUC, sensitivity, specificity, and F1 score.

Main Results:

• The Support Vector Machine (SVM) model at week 24 demonstrated superior predictive performance (AUC = 0.981) compared to baseline and week 12 strategies.
• Key predictors included age, ALT ratio at week 12, and HBsAg/HBsAg ratio at week 24.
• A web-based dynamic nomogram was developed for convenient clinical prediction.

Conclusions:

• An SVM model effectively predicts functional cure in CHB patients treated with Peg-IFN, particularly when assessed at week 24.
• The developed web tool enables individualized prediction of Peg-IFN therapy efficacy for CHB patients.
• This approach enhances clinical decision-making for CHB management.