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

Area of Science:

• Hepatology
• Machine Learning in Medicine
• Biomarker Discovery

Background:

• Early detection of significant liver inflammation in chronic hepatitis B (CHB) is crucial for disease reversal but faces screening barriers.
• Antiviral therapy effectiveness is linked to timely intervention, highlighting the need for accessible predictive tools.

Purpose of the Study:

• To develop and validate a machine learning predictive model for assessing liver inflammation in CHB patients using routine hematological indicators.
• To create a non-invasive tool for early screening of significant liver inflammation in CHB.

Main Methods:

• A multicentre retrospective study of 1,592 untreated CHB patients who underwent liver biopsy (2014-2022).
• Feature selection using Gini index and Lasso, followed by machine learning model selection based on AUROC and decision curve analysis.
• Development of the AAAG model using aspartate transaminase (AST), gamma-glutamyl transpeptidase (GGT), alpha-fetoprotein (AFP), and albumin-globulin ratio (AGR).

Main Results:

• The developed AAAG model demonstrated good diagnostic performance for significant liver inflammation, with an AUROC ranging from 0.79 to 0.84.
• The model showed particularly high diagnostic accuracy in female patients, achieving an AUROC of 0.85 to 0.88.
• A publicly available online tool was created to facilitate the assessment of liver inflammation.

Conclusions:

• The AAAG model, utilizing four readily available hematological indicators, provides a simple, non-invasive method for assessing significant liver inflammation in CHB patients.
• This validated tool can aid in the early screening of CHB patients, potentially improving disease management and outcomes.
• The developed online tool offers a practical solution for clinicians to assess liver inflammation risk non-invasively.