Answers are grounded in this paper and cite their sources.
Background/Objective Chronic hepatitis B (CHB) remains a major cause of hepatocellular carcinoma (HCC), and the risk persists despite potent nucleos(t)ide analogue therapy. Current surveillance relies on broad clinical categories and conventional risk scores, although HCC risk in CHB is heterogeneous and changes over time. This review summarizes current approaches to HCC risk stratification in CHB, critically appraises machine learning (ML)-based models for individualized prediction, and discusses requirements for clinical implementation. Methods We searched PubMed/MEDLINE from inception to July 1, 2026 for studies that developed or validated an ML-based model for incident HCC in adults with CHB, and appraised the included models with a structured approach informed by PROBAST+AI and TRIPOD+AI. Results Conventional models are generally static and do not fully capture non-linear relationships, time-updated risk, or multimodal information. ML-based models using clinical, virological, treatment-related, post-seroclearance, and imaging-derived data can refine risk prediction across stages of CHB care. However, most appraised models were at high risk of bias in the analysis domain, and none has undergone prospective clinical validation. Future models should move toward dynamic prediction using longitudinal data and multimodal integration of clinical variables, liver stiffness, imaging, and viral or molecular biomarkers. Key requirements include external validation, calibration, interpretability, fairness, cost-effectiveness, and prospective impact studies. Conclusion ML-based models may complement conventional risk scores and support individualized surveillance in CHB. Future research should prioritize dynamic, multimodal, externally validated, and clinically implementable systems that improve surveillance decisions and patient outcomes.
No takes yet. Share an insight, caveat, or question.
Hur et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: