A combination of α-fetoprotein, midkine, thioredoxin and a metabolite for predicting hepatocellular carcinoma.

INTRODUCTION AND OBJECTIVES: The heterogenous nature of hepatocellular carcinoma (HCC) motivated this attempt at developing and validating a model based on combined biomarkers for improving early HCC detection. PATIENTS/MATERIALS AND METHODS: This study examined 196 patients for an estimation study (104 patients with HCC, 52 with liver cirrhosis and 40 with liver fibrosis) and 122 patients for the validation study (80 patients with HCC, 42 with liver cirrhosis). All patients were positive for hepatitis C virus. Four markers were measured: Midkine and thioredoxin using ELISA, 1-methyladenosine and 1-methylguanosine using a gas chromatography-mass spectrometry (GC-MS). The results were compared with alpha-fetoprotein (AFP). The performance of the model was estimated in BCLC, CLIP and Okuda staging systems of HCC. RESULTS: The model yielded high performance with an area under ROC (AUC) of 0.94 for predicting HCC in patients with liver cirrhosis, compared with AUC of 0.69 for AFP. This model had AUCs of 0.93, 0.94 and 0.94 in patients who had only one single nodule, absent macrovascular invasion and tumor size <2cm, respectively, compared with AUCs of 0.71, 0.6 and 0.59 for AFP. The model produced AUCs of 0.91 for BCLC (0-A), 0.92 for CLIP (0-1) and 0.94 for Okuda (stage I) compared with AUCs of 0.56, 0.58 and 0.64 for AFP. No significant difference was found between AUC in the estimation and the validation groups. CONCLUSION: This model may enhance early-stage HCC detection and help to overcome insufficient sensitivity of AFP.