A machine learning predictive model for staging chronic hepatitis B inflammation based on non-invasive metrics

Early screening and antiviral therapy for significant liver inflammation in chronic hepatitis B (CHB) remain significant barriers as the key to disease reversal. The aim of this study was to develop a machine learning predictive model based on clinically available hematological indicators to assess liver inflammation in patients with CHB. This multicentre retrospective study comprised patients with untreated CHB who underwent hepatic puncture biopsy from 2014 to 2022. The best features were screened by the Gini index and Lasso. The machine learning model with the highest diagnostic performance for apparent liver inflammation was selected by comprehensive evaluation of parameters such as area under the receiver operating characteristic curve (AUROC) and decision curve analysis. This study included a total of 1,592 patients with chronic hepatitis B (CHB). Four indicators were ultimately selected: aspartate transaminase (AST), gamma-glutamyl transpeptidase (GGT), alpha-fetoprotein (AFP), and albumin-globulin ratio (AGR), which were used to construct the AAAG model. The model's AUROC value ranged from 0.79 to 0.84, particularly in female patients, where the AUROC value reached 0.85 to 0.88. Additionally, a publicly available online tool was developed for assessing liver inflammation. A simple, non-invasive tool with good diagnostic value for assessing significant liver inflammation in patients with chronic hepatitis B for early screening was developed and validated.