the purpose
To identify early predictors of hepatitis B surface antigen (HBsAg) clearance at 48 weeks after pegylated interferon (Peg-IFN) therapy in inactive HBsAg carriers (IHC) and develop an initial machine learning-based model to support clinical decision-making.
method
This retrospective analysis was based on a multicenter prospective cohort that included 777 IHC patients who received at least 48 weeks of Peg-IFN therapy. Least absolute shrinkage and selection operator (LASSO) regression and Boruta algorithm were applied to select predictor variables. Nine machine learning models, including logistic regression (LR), decision tree (DT), and random forest (RF), were built and evaluated using 10-fold cross-validation. External validation cohort (n= 167) Data from three medical centers in Beijing were used to validate the model. SHapley Additive exPlanations (SHAP) values were used to interpret variable contributions.
result
The overall HBsAg clearance rate at week 48 was 29.9% (232/777). Key predictors include baseline HBsAg level, HBsAg decline > 1 log IU/mL at week 12, and alanine aminotransferase (ALT) to HBsAg ratio at week 12. The RF model showed the best performance with an area under the curve (AUC) of 0.829 (95% CI: 0.784 to 0.874) and a specificity of 0.774. The AUC on the training set was 0.838 (95% CI: 0.759 to 0.917) and the specificity on the external validation set was 0.968. SHAP analysis showed that baseline HBsAg had the highest predictive importance.
conclusion
RF-based models provide a promising tool for accurately predicting HBsAg clearance in IHC patients undergoing Peg-IFN therapy and for early identification of candidates for personalized treatment strategies.
