Construction and validation of a predictive model for HBsAg loss in chronic hepatitis B patients treated with Peg-IFNα-2b.

BACKGROUND: Hepatitis B surface antigen (HBsAg) loss is the ideal treatment endpoint for chronic hepatitis B (CHB). Pegylated interferon alfa-2b (Peg-IFNα-2b) is currently the preferred drug for achieving HBsAg loss, but its efficacy varies significantly among different patients. METHODS: This multicenter, retrospective real-world study enrolled CHB patients treated with Peg-IFNα-2b for ≥48 weeks. Patients were divided into cured and uncured groups based on week-48 HBsAg status and randomly assigned to training and validation cohorts (7:3). Logistic regression identified independent predictors of HBsAg loss and constructed a predictive model. The model's discriminative ability, calibration, and clinical applicability were evaluated using receiver operating characteristic (ROC) curve analysis, calibration curves, Hosmer-Lemeshow test, and decision analysis curve (DCA) analysis. RESULTS: Among 859 patients, 133 (15.5%) achieved HBsAg loss at week 48. Multivariate analysis identified combination therapy with traditional Chinese medicine (TCM) (OR = 3.83), lower baseline HBsAg (OR = 0.32), lower baseline aspartate aminotransferase (AST) (OR = 0.97), and entecavir (ETV) regimen (OR = 0.33) as independent predictors. The model demonstrated strong discrimination and calibration performance, with AUCs of 0.884 in the training cohort and 0.866 in the validation cohort. Logit(P)=-0.87-1.12*HBsAg(log10)+1.34*group-0.03*AST-1.1×ETV. Note: group (1=no combination with TCM, 2=combination with TCM), ETV (0=non-ETV antiviral regimen, 1=ETV antiviral regimen). CONCLUSION: The combination of TCM therapy, low baseline HBsAg levels, lower baseline AST levels, and non-ETV antiviral regimens were significant predictors of HBsAg loss at week 48 in CHB patients receiving 48-week Peg-IFNα-2b treatment. The predictive model we developed demonstrated high accuracy and potential clinical utility. KEY POINTS: This study focused on the establishment and validation of a clinical prediction model for HBsAg loss based on a large two-center cohort including 859 patients with CHB. The model was constructed using rigorous statistical methods and further evaluated through both internal and external validation. The results demonstrated good discrimination, calibration, and clinical net benefit, suggesting that the model may serve as a practical tool to help clinicians estimate the probability of HBsAg loss and facilitate individualized antiviral treatment strategies.