Machine learning prediction of functional cure to pegylated interferon-alpha therapy in two HBV populations: Advantaged populations and HBeAg-negative patients.

Despite advances in antiviral therapy, the rate of functional cure for chronic hepatitis B (CHB) remains unsatisfactory, and developing an applicable prediction model is pivotal to improving it. Thus, we aimed to identify key predictive factors and develop prognostic models for functional cure in HBeAg-negative patients and the advantaged populations. This retrospective study included 202 HBeAg-negative CHB patients (114 classified as advantaged populations) receiving pegylated interferon-alpha (PEG-IFNα) therapy for model derivation and internal validation, and 183 HBeAg-negative CHB patients (117 classified as advantaged populations) for model external validation. Using 48 routinely collected clinical indicators, we constructed prediction models through LASSO regression followed by multivariable logistic regression. Two nomogram-based models were developed: the SHAN model, based on four independent predictors - ln (HBsAg +1), age, neutrophil percentage (NE%), and sex - was tailored for HBeAg-negative patients. For the advantaged populations, two additional variables - alpha-fetoprotein (AFP) and lactate dehydrogenase (LDH) - were incorporated into the FLASH-N model. Both models demonstrated strong discrimination, with AUCs of 0.908 in the training set and 0.949 in the test set for the SHAN model and an AUC of 0.920 (bootstrap-corrected to 0.889) for the FLASH-N model in the advantaged populations. In external validation, SHAN model achieved an AUC of 0.861, and FLASH-N achieved an AUC of 0.800. Calibration plots and decision curve analysis further confirmed the robustness, accuracy, and clinical utility of both nomograms. By leveraging routinely available baseline variables, these models offer powerful tools for predicting functional cure in CHB, enabling refined risk stratification and more personalized clinical decision-making.