Connectivity-based risk score for hepatocellular carcinoma prognosis


  • Potential conflict of interest: Nothing to report.

To the Editor:

Kim et al.[1] proposed a 65-gene-based risk score classifier of overall survival in hepatocellular carcinoma (HCC). The risk score, derived by multiplying the expression level of a gene by its Cox coefficient, could robustly predict overall survival of HCC patients. Its clinical usefulness was further confirmed in a second test cohort. There were some minor defects in Fig. 1A and Table 2. The article adopted a previous method[2] by simply using Cox's coefficient from univariate regression analysis, ignoring the inherent correlation between genes. However, as mentioned in the literature,[2] nonlinear relationships may exist between genes, that is, the potential interaction between signature genes. Furthermore, although three cohorts are used for survival analysis in the study, none of the survival data are publicly available. It is worth noting that the microarray data analyzed were from different platforms, including single- and double-channel arrays, but no details about data manipulation were provided, which further complicated the prediction model. To examine whether these signature genes plus univariate coefficients can consistently stratify patients into low- or high-risk groups, chip data GSE10141,[3] the gene-expression data set of an HCC cohort with survival data, was reanalyzed. The original expression-matrix data of intensity was utilized directly. Patients were similarly dichotomized into groups at high or low risk using the risk-score classifier, but the two groups of patients showed indistinguishable prognostic outcome (P = 0.13). To test whether the correlation between signature genes accounts for prediction power, the sum of Pearson's correlation (connectivity) with other signature genes was used, instead of the univariate Cox coefficient. Patients were regrouped, and the low-risk group had a better prognosis (P = 0.025). The different stratification result was possibly because of having taken gene interaction into consideration. However, it should be noted that most of the samples from Kim et al.'s study were frozen specimens, whereas samples from Hoshida et al.'s study[3] were formalin-fixed paraffin-embedded tissues. Moreover, only 52 of the 65 signature genes were analyzed because of platform differences. Further large-scale evaluation of the two risk-score methods as HCC overall survival prediction in prospective studies among multicenters are needed.

  • HUA YE, M.D.1

  • WEI LIU, PH.D.2

  • 1Department of Gastroenterology, Lihuili Hospital, Ningbo, China

  • 2Department of Pathology, Human Centrifuge Medical Training Center, Institute of Aviation Medicine of Chinese People's Liberation Army Air Force, Beijing, China