Detecting Early Hepatocellular Carcinoma in Patients With Chronic Hepatitis B Using Longitudinal α-Fetoprotein Screening

Abstract

This study aimed to evaluate the parametric empirical Bayes (PEB) longitudinal α-fetoprotein (AFP) screening algorithm performance in patients with hepatitis B compared with AFP surveillance with a fixed threshold. The serum AFP of 588 patients was measured. Patients were screened at least once every 6 months with AFP and ultrasound or computed tomography/magnetic resonance imaging. Age, aspartate aminotransferase level, alanine aminotransferase level, platelet count, total bilirubin, prothrombin time, and hepatitis B virus DNA level were adjusted in the PEB algorithm. All variables were abstracted at the time of hepatocellular carcinoma (HCC) diagnosis for cases or last follow-up for controls and at months -6, -12, -18, -24, -30, -36, -42, -48, and -54, up to month -60. Overall, 62 (10.5%) HCC cases developed during a median follow-up of 52.7 months. Moreover, 55 (88.7%) cases were detected at Barcelona Clinic Liver Cancer stage 0 or A. The area under the receiver-operating characteristic curve of the patient-level true positive rate against the screening-level false positive rate was significantly higher in the PEB algorithm than that in AFP alone (area under the receiver-operating characteristic curve: 0.94 vs 0.86; P < .0005). At 80% specificity, the PEB algorithm significantly improved the patient-level true positive rate within 2 years prior to HCC diagnosis compared with AFP alone (80.6% vs 67.7%, respectively; P= .0485; adjusted P= .1663). The PEB algorithm more effectively enabled first positive screening. The longitudinal assessment of AFP by the PEB algorithm improved HCC screening performance compared to AFP alone in patients with hepatitis B. This algorithm may improve HCC screening without additional cost or inconvenience to patients.