Machine learning model for HBsAg seroclearance after 48-week pegylated interferon therapy in inactive HBsAg carriers: a retrospective study.

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