Stratifying tumor immune microenvironment in head and neck squamous cell carcinoma: from parsing out immune-subtype prognosis to clinically applicable CT-radiomics models.

backgroundTumor immunological heterogeneity contributes to prognostic variation in head and neck squamous cell carcinoma (HNSCC), yet non-invasive imaging tools capable of predicting immune-related gene expression patterns remain underdeveloped.

methodsThis large-scale investigation, comprising 1030 HNSCC patients, provides the most comprehensive radiogenomic characterization of tumor immunological heterogeneity to date. Differential expression Immune-related genes (IRGs) were identified through intersection analysis of HNSCC transcriptomic data from The Cancer Genome Atlas (TCGA). Prognostic IRGs were selected via univariate Cox regression, followed by LASSO and multivariate Cox regression for model refinement. A nomogram integrating clinical parameters and risk scores was developed. Model accuracy was evaluated using Kaplan-Meier survival analysis, time-dependent ROC curves, decision curve analysis (DCA). Radiomic features from TCIA-derived CT scans were extracted to construct a non-invasive biomarker for immune subtype prediction. We incorporate advanced harmonization (ComBat) and rigorous internal validation (nested cross-validation) to enhance reproducibility and generalizability beyond conventional best practices. Multivariate logistic regression combined radiomics and clinicopathological data into an enhanced radiogenomics nomogram, validated via DCA and ROC analysis.

resultsNon-negative matrix factorization (NMF) stratified HNSCC patients into two clusters based on IRG expression, revealing distinct immune cell infiltration patterns. The IRGs-based prognostic signature we developed demonstrated significant clinical utility as an independent predictor of HNSCC outcomes. When stratifying patients by median risk scores, the signature effectively distinguished two subgroups with markedly different survival outcomes in three Gene Expression Omnibus （GEO）databases. A Kaplan-Meier analysis showed consistently significantly worse overall survival in high-risk patients, especially in those with advanced HNSCC. Quantitative CT-based radiomic profiling can effectively discriminate between immunologically distinct tumor clusters, which correspond to significantly divergent clinical outcomes. ROC curves revealed nomogram (AUC = 0.81) superior predictive accuracy compared to conventional clinical parameters grade (AUC = 0.67). Clinical utility assessment through decision curve analysis (DCA) revealed that the radiogenomic nomogram provided significant net benefit over alternative treatment strategies across all threshold probabilities above 10%.

conclusionsThis study constructed a novel non-invasive radiogenomics marker for the prognostic stratification of HNSCC. This emerging interdisciplinary approach establishes quantitative correlations between non-invasive imaging biomarkers and underlying molecular alterations, providing valuable insights for HNSCC precision therapy.