Background: Although a number of serum biomarkers for detection of hepatocellular carcinoma (HCC) have been explored, their exact diagnostic value remains unclear. We aimed to conduct a direct comparison of five representative serum biomarkers for detecting HCC and to derive multi-marker prediction algorithms. 
Patients and methods: In total, 846 patients were recruited from three hospitals in China, including 202 HCC patients, 226 liver cirrhosis patients, 215 chronic hepatitis B virus-infected patients, and 203 healthy volunteers. Serum levels of alpha-fetoprotein (AFP), lens culinaris agglutinin-reactive AFP (AFP-L3), des-gamma-carboxyprothrombin (DCP), squamous cell carcinoma antigen, and centromere protein F autoantibody were measured by ELISA. The diagnostic performances of individual biomarkers and multi-marker combinations were evaluated by receiver operating characteristics analysis. The bootstrapping method was adopted to adjust for potential overfitting of all diagnostic indicators. 
Results: DCP exhibited the best diagnostic performance, with areas under the curve (AUC) for detecting HCC of 0.82 (95% CI 0.64–0.80) and sensitivity of 65.2% (95% CI 63.3–82.1%) at 90% specificity. Of note, DCP showed similar diagnostic efficacy for detecting AFP-positive and AFP-negative HCC. After a comprehensive search for multi-marker combinations, a two-marker prediction algorithm including AFP and DCP was constructed and yielded an AUC of 0.87 (95% CI 0.68–0.84) for detecting HCC. In addition, the combination showed good ability in discriminating early-stage HCC and decompensated liver cirrhosis, with an AUC of 0.81 (95% CI 0.75–0.86). 
Conclusion: DCP could be a complementary biomarker in the early diagnosis of HCC. The constructed multi-marker prediction algorithms could contribute toward distinguishing HCC from non-malignant chronic liver diseases.
Keywords: early detection, liver cirrhosis, prediction model