The Prediction of Clinical Outcome in Hepatocellular Carcinoma Based on a Six-Gene Metastasis Signature

Purpose: The dismal outcome of hepatocellular carcinoma (HCC) is largely attributed to its early recurrence and venous metastases. We aimed to develop a metastasis-related model to predict hepatocellular carcinoma prognosis. Experimental Design: Using microarrays, sequencing, and RT-PCR, we measured the expression of mRNAs and lncRNAs in a training set of 94 well-defined low-risk (LRM) and high-risk metastatic (HRM) HCC patients from a Shanghai cohort. We refined a metastasis signature and established a corresponding model using logistic regression analysis. The validation set consisted of 567 HCC patients from four-center cohorts. Survival analysis was performed according to the metastasis model. Results: Using relative expression of tumor to para-tumor tissues, we refined the metastasis signature of five mRNAs and one lncRNA. A generalized linear model was further established to predict the probability of metastasis (MP). Using MP cutoff of 0.7 to separate LRM and HRM in Shanghai cohort, the specificity and sensitivity of the model were 96% [95% confidence interval (CI), 85%-99%] and 74% (95% CI, 58%-86%), respectively. Furthermore, HRM patients showed a significantly shorter overall and recurrence-free survival in validation cohorts (P < 0.05 for each cohort). Early HCC patients also have a poorer outcome for multicenter HRM patients. Finally, Cox regression analysis indicated that continuous MP was an independent risk factor and associated with the recurrence and survival of HCC patients after resection (HR 2.98-16.6, P < 0.05). Conclusions: We developed an applicable six-gene metastasis signature, which is robust and reproducible in multicenter cohorts for HCC prognosis. (C) 2016 AACR.