Construction and Validation of Risk Model

The “limma” R package was used to distinguish RBPs with different expressions between immunotypes. With a 1.4-fold difference and corrected p less than 0.05 as the screening conditions, 238 immune-related RBPs were obtained. Subsequently, 47 immune-related RBPs associated with prognosis were obtained through univariate proportional hazards regression (p < 0.05). The “glmnet” package was then utilized to perform LASSO Cox regression analysis (35). After 1,000 times of cross-validation, 15 immune-related RBPs and the correlation coefficients of the corresponding risk genes were obtained to construct a risk model at the same time. Risk score=ΣinExpi×Coefi , in which Expi is the expression of each risk gene and Coefi is its correlation coefficient. All patients were divided into a high-risk group and a low-risk group characterized by the median risk score of patients with HNSCC in the training set. The Kaplan–Meier curves were used to compare the overall survival (OS) difference of patients in the high- and low-risk groups. Receiver operating characteristic (ROC) curves were generated to evaluate the effectiveness and accuracy of the risk score in predicting the prognosis of patients with HNSCC. Next, the “ggExtra” R package was used to calculate the correlation between the risk score and the OS of patients with HNSCC. The independent correlation between the risk score and the prognosis of patients with HNSCC was then evaluated by univariate and multivariate proportional hazards regression analyses. Subsequently, a nomogram that could predict the prognosis of individual patients with HNSCC was constructed based on the stage, T stage, N stage, and risk group of patients with HNSCC through the “rms” R package (36). The C index was then used to assess the ability of the nomogram to distinguish prognosis, and a calibration chart was drawn to evaluate the accuracy of the nomogram. In addition, GSEA and gene set variation analysis (GSVA) were used to compare the differences in KEGG pathway enrichment between risk groups.