Background and Aims: Hepatocellular carcinoma (HCC) is one of the most common malignant tumors with poor prognosis, and its etiology and pathogenesis are still elusive. Therefore, identification of reliable prognostic factors and survival biomarkers of HCC patients is of great clinical importance. This study was to screen the prognostic immune-related genes of HCC through bioinformatics approach, and then construct a prognostic risk score model based on an immune-related gene signature, so as to provide a basis for prognosis evaluation and individualized treatment decision-making regarding the HCC patients. 
Methods: The clinical information and RNA-seq data (377 HCC samples and 50 adjacent non-cancerous samples) of HCC patients were obtained from the TCGA database. The information of immune-related genes was downloaded from the Immport database, and the differentially expressed immune-related genes were selected from HCC tissues using the limma package of R software. The immune-related genes closely related to the overall survival (OS) among the HCC patients (344 cases with complete clinical record in the 377 HCC patients) were determined by univariate and multivariate Cox proportional risk regression models, and then the prognostic risk score model based on an immune-related gene signature was constructed by using the determined genes, by which the prognostic risks of the HCC patients were scored. Meanwhile, 50% cases (172 cases) were randomly picked up from the above model sample as an internal validation sample for internal validation. Kaplan-Meier method was used to analyze the survival status between patients with high and low risk score, and the accuracy of the risk score was evaluated by ROC curve and C-index analysis. Finally, the relations of the risk score with the clinicopathologic factors of HCC were analyzed, and the effectiveness of this risk score as an independent prognostic risk factor for HCC was ascertained by univariate and multivariate Cox regression analysis.
Results: A total of 329 differentially expressed immune-related genes between HCC tissue and tumor adjacent were identified, in which 24 were significantly associated with the OS of the HCC patients (all P<0.001), and 9 genes that included PSMD14, S100A11, FABP6, RBP2, LCNL1, FCN2, NDRG1, CSPG5 and NR6A1 were determined as high-risk genes for OS by using forward and backward selection algorithm in multivariate Cox proportional hazards regression analysis. According to the classification by the 9-gene signature based prognostic risk score model, the OS in patients with high risk score was significantly worse than that in patients with low risk score in the model sample (P=1.715E-08), and the same result was also obtained in the internal validation sample (P=2.222E-05). The 1- and 3-year values of the area under the ROC curve (AUC) were 0.790 and 0.733 in the model sample, and were 0.799 and 0.743 in the internal validation sample, respectively. The results of C-index analysis showed that the C-index values in the model sample and the internal validation sample were 0.715 (95% CI=0.683–0.829) and 0.756 (95% CI=0.668–0.762), respectively. Tumor grade, pathological stage, T stage and new tumor events were correlated with the risk score (all P<0.05). Univariate and multivariate Cox analysis showed that the risk score was an independent prognostic factor for HCC (univariate: HR=1.057, 95% CI=1.041-1.074, P<0.001; multivariate: HR=1.050, 95% CI=1.033-1.067, P<0.001).
Conclusion: Nine immune-related genes closely related to the prognosis of HCC patients are identified by TCGA database mining, and a 9-gene signature based prognostic risk score model is developed, which may help the clinicians to assess the prognosis of the HCC patients and design a personalized treatment plan for them.