Background: Our study aimed to compare the predictive performance of different hepatocellular carcinoma (HCC) prediction models in chronic hepatitis B patients receiving entecavir or tenofovir, including discrimination, calibration, negative predictive value (NPV) in low-risk, and proportion of low-risk.
Methods: We conducted a systematic literature research in PubMed, EMbase, the Cochrane Library, and Web of Science before January 13, 2022. The predictive performance was assessed by area under receiver operating characteristic curve (AUROC), calibration index, negative predictive value, and the proportion in low-risk. Subgroup and meta-regression analyses of discrimination and calibration were conducted. Sensitivity analysis was conducted to validate the stability of the results.
Results: We identified ten prediction models in 23 studies. The pooled 3-, 5-, and 10-year AUROC varied from 0.72 to 0.84, 0.74 to 0.83, and 0.76 to 0.86, respectively. REAL-B, AASL-HCC, and HCC-RESCUE achieved the best discrimination. HCC-RESCUE, PAGE-B, and mPAGE-B overestimated HCC development, whereas mREACH-B, AASL-HCC, REAL-B, CAMD, CAGE-B, SAGE-B, and aMAP underestimated it. All models were able to identify people with a low risk of HCC accurately. HCC-RESCUE and aMAP recognized over half of the population as low-risk. Subgroup analysis and sensitivity analysis showed similar results.
Conclusion: Considering the predictive performance of all four aspects, we suggest that HCC-RESCUE was the best model to utilize in clinical practice, especially in primary care and low-income areas. To confirm our findings, further validation studies with the above four components were required.
Keywords: Calibration; Discrimination; Predictive value of tests; Risk factors; Surveillance.
© 2023. BioMed Central Ltd., part of Springer Nature.