Abstract
Objective: Immune landscape is a key feature that affects cancer progression, survival, and treatment response. Herein, this study sought to comprehensively characterize the immune-related genes (IRGs) in oral squamous cell carcinoma (OSCC) and conduct an immune-related risk score (IRS) model for prognosis and therapeutic response prediction. Methods: Transcriptome profiles and follow-up data of OSCC cohorts were curated from TCGA, GSE41613, and IMvigor210 datasets. An IRS model was established through univariate Cox, Random Survival Forest, and multivariate Cox analyses. Prognostic significance was evaluated with Kaplan-Meier curves, ROC, uni- and multivariate Cox, and subgroup analyses. A nomogram was conducted and assessed with C-index, ROC, calibration curves, and decision curve analyses. Immune cell infiltration and immune response were estimated with ESTIMATE and ssGSEA methods. Results: An IRS model was constructed for predicting the overall survival and disease-free survival of OSCC, containing MASP1, HBEGF, CCL22, CTSG, LBP, and PLAU. High-risk patients displayed undesirable prognosis, and the predictive efficacy of this model was more accurate than conventional clinicopathological indicators. Multivariate Cox analyses demonstrated that this model was an independent risk factor. The nomogram combining IRS, stage, and age possessed high clinical application values. The IRS was positively associated with a nonflamed tumor microenvironment. Moreover, this signature enabled to predict immunotherapeutic response and sensitivity to chemotherapeutic agents (methotrexate and paclitaxel). Conclusion: Collectively, our study developed a robust IRS model with machine learning method to stratify OSCC patients into subgroups with distinct prognosis and benefits from immunotherapy, which might assist identify biomarkers and targets for immunotherapeutic schemes.