Abstract / Summary
Background: We evaluated breast cancer-derived hypoxia and radiosensitivity signatures in head and neck squamous cell carcinoma (HNSCC), then developed and externally evaluated HNSCC prognostic signatures from hypoxia-related genes. Methods: This retrospective study used TCGA HNSC for development (514 patients; 217 deaths) and GSE65858 for external evaluation (270 patients; 94 deaths). LASSO Cox regression selected models from 414 hypoxia-related genes measurable in both cohorts. The 32-gene primary model and 13-gene sensitivity model were locked before external survival testing. External analyses assessed continuous scores, with additional adjustment for age, stage, tumour site, HPV DNA status and smoking. Cross validation with standardisation within training folds and bootstrap gene selection stability were examined in subsequent checks. Results: The main analyses did not establish prognostic associations for the breast-derived signatures in HNSCC. The primary HNSCC model had a mean development cross-validation C index of 0.628. In GSE65858, univariable hazard ratios per score standard deviation were 1.29 (95% confidence interval 1.05 to 1.59; p = 0.016) for the primary model and 1.28 (1.03 to 1.59; p = 0.024) for the sensitivity model. Their C indices were 0.566 and 0.581. After clinical adjustment in 265 patients with 92 deaths, the corresponding hazard ratios were 1.24 (0.99 to 1.55; p = 0.065) and 1.20 (0.95 to 1.51; p = 0.117). Neither signature reached statistical significance in the analyses using detailed HPV information. Five-year median-split log-rank p values were 0.095 and 0.204. Conclusions: The signatures showed unadjusted associations with overall survival in an external cohort, but discrimination was modest and independent prognostic value was not established after clinical adjustment. These exploratory prognostic signatures require further evaluation alongside clinical factors. The findings do not establish a quantitative measure of tumour hypoxia or utility for treatment selection.