The risk prediction of hepatocellular carcinoma (HCC) in patients with chronic hepatitis B (CHB) is a challenge especially in the era of antiviral therapy. The aim of this meta-analysis was to comprehensively evaluate the performance of existing HCC prediction scores in HCC prediction on antivirals.

We searched PubMed, Web of Science and Cochrane Library for relevant prospective studies from the inception to August 24, 2021. The areas under the receiver operating characteristics (AUROCs) and their relevant 95% confidence intervals (CIs) of the risk prediction models were calculated.

Nine eligible articles with 21561 patients (HCC developed in 947patients, 4.39%; mean follow-up duration: 5 years) and 14 predictive risk scores were included. The pooled AUROC of all included scores for 3-year and 5-year prediction of HCC was 0.72 (95%CI 0.68-0.76) and 0.80 (95%CI 0.76-0.83), with the corresponding sensitivity of 0.84 (95% CI 0.71-0.92) and 0.91(95% CI 0.86-0.95) and specificity of 0.46 (95% CI 0.30-0.63) and 0.48 (95% CI 0.37-0.59), respectively. All the 14 prediction models, as a whole, performed well in different populations, whether they include factor cirrhotic status or not; while those integrated viral load were less accurate (sensitivity 0.78, specificity of 0.57).

In patients with CHB on antivirals, the scores included in our meta-analysis have been proven to be useful for mid-long term HCC prediction. Viral load seems not useful, whereas cirrhosis and its objective surrogates remain the predominant components. These models are expected to translate clinical benefits if used in complementarity with regular HCC surveillance.