# Recurrence risk prediction model for hepatitis B virus-associated hepatocellular carcinoma patients: a systematic review and meta-analysis

## TL;DR

This study reviews and evaluates existing models for predicting recurrence in hepatitis B-related liver cancer patients, finding moderate accuracy but significant methodological flaws.

## Contribution

A systematic review and meta-analysis of prediction models for HBV-HCC recurrence, highlighting methodological issues and pooled predictive accuracy.

## Key findings

- 22 models were evaluated, with 20 using Cox regression and one using machine learning.
- The pooled C-index for model performance was 0.73 in validation cohorts.
- All models had high risk of bias, mainly from the analysis domain.

## Abstract

Hepatitis B virus-associated hepatocellular carcinoma (HBV-HCC) is characterized by high postoperative recurrence rates. Although numerous recurrence prediction models exist, their performance and clinical utility remain uncertain.

To systematically evaluate the performance and methodological quality of existing recurrence risk prediction models for HBV-HCC patients.

We searched PubMed, Web of Science, Embase, Scopus, and OVID databases. Data were extracted following the CHARMS checklist, and the PROBAST tool was used to assess the risk of bias. A meta-analysis of the C-index from validation cohorts was performed using a random-effects model.

A total of 22 studies, encompassing 22 models, were included. Regarding the modeling methodology, 20 models were developed using the Cox proportional hazards regression model, one used a logistic regression model, and one utilized machine learning (ML). All 22 studies exhibited a high risk of bias, predominantly originating from the analysis domain. The meta-analysis revealed a pooled C-index of 0.73 (95% CI: 0.70-0.75) in the validation cohorts. The most frequently used predictors were MVI, AFP, tumor size, tumor number, and HBV-DNA.

Existing recurrence prediction models for HBV-HCC demonstrate moderate predictive accuracy but are universally affected by a high risk of bias. This limits their reliability and applicability in current clinical practice. Future research should emphasize methodological rigor and conduct multicenter external validation before applying models in clinical practice.

https://www.crd.york.ac.uk/PROSPERO/, identifier CRD42025629973.