Objective Serum α-fetoprotein (AFP) is the most commonly used biomarker for screening hepatocellular carcinoma (HCC) but fails to detect about half of the patients. Thus, we investigated if circulating microRNAs (miRNAs) could outperform AFP for HCC detection.

Design A retrospective cohort study.

Setting Two clinical centres in China.

Participants The exploration phase included 96 patients with HCC who received primary curative hepatectomy, and the validation phase included 29 hepatitis B carriers, 57 patients with HCC and 30 healthy controls.

Main outcome measures Expression of miRNAs was measured by real-time quantitative reverse transcription–PCR. Areas under receiver operating characteristic curves were used to determine the feasibility of using serum miRNA concentration as a diagnostic marker for defining HCC. A multivariate logistic regression analysis was used to evaluate performances of combined serum miRNAs.

Results In the exploration phase, miRNA profiling on resected tumour/adjacent non-tumour tissues identified miR-15b, miR-21, miR-130b and miR-183 highly expressed in tumours. These miRNAs were also detectable in culture supernatants of HCC cell lines and in serum samples of patients. Remarkably, these serum miRNAs were markedly reduced after surgery, indicating the tumour-derived source of these circulating miRNAs. In a cross-centre validation study, combined miR-15b and miR-130b demonstrated as a classifier for HCC detection, yielding a receiver operating characteristic curve area of 0.98 (98.2% sensitivity and 91.5% specificity). The detection sensitivity of the classifier in a subgroup of HCCs with low AFP (<20 ng/ml) was 96.7%. The classifier also identified early-stage HCC cases that could not be detected by AFP.

Conclusion The combined miR-15b and miR-130b classifier is a serum biomarker with clinical value for HCC screening.