# Evaluating the predictive value of clinical models for HBV-related hepatocellular carcinoma: A meta-analysis

## TL;DR

This study evaluates how well existing models predict liver cancer in patients with chronic hepatitis B, finding moderate accuracy but regional limitations.

## Contribution

The study provides a meta-analysis of clinical models for HBV-related hepatocellular carcinoma prediction, highlighting their performance and geographical limitations.

## Key findings

- Models like mPAGE-B, GAGHCC, and CAMD showed better discrimination with C-index values between 0.79 and 0.80.
- Most studies did not report model calibration, and subgroup analysis suggested ethnic and bias-related differences in model performance.
- Predictive performance of models was moderate in Guangxi, with limited 3 and 5-year risk discrimination.

## Abstract

Chronic viral hepatitis B (CHB) is a prevalent liver disease with primary hepatic carcinoma (HCC) as a severe complication. Clinical prediction models have gained attention for predicting HBV-related HCC (HBV-HCC). This study aimed to evaluate the predictive value of existing models for HBV-HCC through meta-analysis.

Meta-analysis.

Embase, PubMed, the Chinese Biomedical Literature Service System, and the Cochrane database were used for searches between 1970 and 2022.

A meta-analysis was conducted to assess original studies on HBV-HCC prediction models. The REACH-B, GAGHCC, and CUHCC models were externally validated in a Guangxi cohort. The C-index and calibration curve evaluated 5 years predictive performance, with subgroup analysis by region and risk bias.

After screening, 27 research articles were included, covering the GAGHCC, REACH-B, PAGE-B, CU-HCC, CAMD, and mPAGE-B models. The meta-analysis indicated that these models had moderate discrimination in predicting HCC risk in HBV-infected patients, with C-index values from 0.75 to 0.82. The mPAGE-B (0.79, 95% CI: 0.79–0.80), GAG-HCC (0.80, 95% CI: 0.78–0.82), and CAMD (0.80, 95% CI: 0.78–0.81) models demonstrated better discrimination than others (P < 0.05), but most studies did not report model calibration. Subgroup analysis suggested that ethnicity and research bias might contribute to differences in model discrimination. Sensitivity analysis indicated stable meta-analysis results. The REACH-B, GAGHCC, CUHCC, PAGE-B, and mPAGE-B models had average predictive performance in Guangxi, with medium to low 3 and 5 years HCC risk prediction discrimination.

Existing models have predictive value for HBV-infected patients but show geographical limitations and reduced effectiveness in Guangxi.