Fig. 4

The risk score predicts poor survival in the training set. A Correlation analysis between expression levels of nine hub genes and Z-scores of key cancer hallmarks (including glycolysis and hypoxia) in the training set. B Patients who died during follow-up had a higher risk score than those who were alive. C Kaplan–Meier analysis showed that the high-risk score group exhibited worse overall survival. D Principal component analysis suggested that risk score could be used as a new dimension to evaluate the prognosis of HNSCC patients. E The ROC curve showed the prediction efficiency of the risk score for the survival rate in the training set (AUC > 0.66). F The tROC analysis showed that the predictive power of risk score was significantly higher than that of other clinical characteristics. G Univariate and multivariate Cox regression analysis showed that risk score was an independent risk factor for OS in patients with HNSCC. HR, hazard ratio; OS, overall survival; tROC, time-dependent receiver operating characteristics