A machine learning model for predicting complete virological response in chronic hepatitis B patients receiving first-line nucleos(t)ide analogues therapy.

BACKGROUND

Complete virological response (CVR) is the primary goal of nucleos(t)ide analogues (NAs) therapy for chronic hepatitis B (CHB) patients. The objective of this study was to develop and validate a machine learning (ML) model to predict CVR in CHB patients receiving first-line NAs.

METHODS

We retrospectively analyzed 2394 CHB patients treated with first-line NAs, randomly divided into a training set and an internal validation set at a 7:3 ratio. Key predictors were identified through univariate analysis, followed by Lasso, SVM-RFE, and Boruta feature selection algorithms. Seven machine learning models were developed and compared, interpreting results with SHapley Additive exPlanations (SHAP) analysis.

RESULTS

A total of 2394 CHB patients were included, among whom 1597 (66.7 %) achieved CVR. Baseline HBV DNA, HBeAg, DBIL, Age, Cirrhosis, ALT, PLT, FIB4, and HBsAg were identified as significant influencing factors. Among the seven ML methods, the XGBoost model demonstrated the best performance with an AUC of 0.864 in the training set and 0.823 in the validation set. SHAP analysis identified baseline HBV DNA and HBeAg status as the top two predictors. This web-based calculator is designed to help predict the probability of achieving CVR at 48 weeks in CHB patients receiving first-line NAs treatment (http://yuanxq.shinyapps.io/CVR-prediction-app).

CONCLUSION

We developed and validated an interpretable machine learning model to predict CVR at 48 weeks in CHB patients treated with first-line NAs.