Enhancing Hepatocellular Carcinoma Surveillance: Comparative Evaluation of AFP, AFP-L3, DCP and Composite Models in a...

Background/Objectives: Biomarkers such as lens agglutinin-reactive alpha-fetoprotein and des-gamma-carboxy prothrombin, as well as biomarker- and/or clinical-parameter-derived composite models (GALAD, GAAP, ASAP, aMAP, Doylestown), may improve detection in addition to alpha-fetoprotein, yet comparative data across diverse populations remain limited. Methods: In this biobank-based case-control study, we evaluated 562 adults (120 healthy controls, 277 chronic liver disease, 165 hepatocellular carcinoma) from January 2019 to 2024. Diagnostic performance for any-stage and early-stage hepatocellular carcinoma was assessed across three thresholds: Youden-index-derived optimal cut-offs, research-established cut-offs, and cut-offs ensuring 90% specificity. Receiver operating characteristic analysis was performed. Subgroup analyses were stratified by etiology and alpha-fetoprotein status. Results: At optimal cut-offs, GALAD showed the highest sensitivity for any-stage (90.3%) and early-stage (89.1%) hepatocellular carcinoma, with 70-80% specificity. Using established cut-offs, GALAD retained the highest sensitivity for any-stage (75.8%) and early-stage (57.8%) hepatocellular carcinoma, with 93.5% specificity. GALAD demonstrated the best performance in non-viral hepatocellular carcinomas (area under the curve 0.872), whereas GAAP and ASAP showed similarly high area under the curve values in viral etiology (area under the curve 0.955-0.960). Conclusions: Our results demonstrate the consistent performance of the GALAD score across diverse populations and underscore its superiority over individual biomarkers and other composite models. Notably, the GAAP and ASAP scores-which use one less biomarker (AFP-L3)-exhibited comparable performance, particularly in viral etiology. These findings support the integration of the composite biomarker models into tailored hepatocellular carcinoma surveillance strategies.