基于TCGA-HNSC放疗队列的lncRNA预后模型构建与内部验证：一项生物信息学研究

Construction and Internal Validation of a lncRNA Prognostic Model Based on a TCGA-HNSC Radiotherapy Cohort: A Bioinformatics Study

摘要: 目的：头颈部鳞状细胞癌(HNSCC)放疗患者的总体生存(OS)存在显著异质性，急需特异性的分子风险分层工具。本研究旨在TCGA-HNSC队列的放疗(RT = Yes)人群中，筛选预后相关lncRNA并构建验证预后模型，为放疗后的个体化管理提供依据。方法：整合TCGA-HNSC转录组与临床数据，提取放疗患者(RT = Yes)样本。对表达数据进行过滤及标准化预处理后，依次采用单因素Cox和LASSO-Cox回归筛选特征基因，通过多因素Cox构建风险评分(RiskScore)模型。利用Kaplan-Meier曲线、timeROC、C-index、列线图(nomogram)及校准曲线等多维度指标，对模型的区分度与校准性能进行内部验证。结果：最终纳入25例放疗样本(含17例死亡事件)，筛选出2个关键lncRNA进入模型。多因素分析显示，HCFC1-AS1为显著保护因素(HR = 0.33, P = 0.008)，Lnc-ENSG257226呈保护趋势(HR = 0.47)；Bootstrap内部验证证实了模型系数的稳健性。模型C-index高达0.84，且RiskScore被证实为独立于年龄和分期的预后因子。模型能显著区分高、低风险组的OS差异(P = 0.0038)，列线图与校准曲线显示其在预测1年、3年生存率方面具有良好的一致性。结论：本研究在TCGA-HNSC放疗人群中构建了基于两个lncRNA的预后模型并完成了内部验证。该模型可有效用于放疗患者的OS风险分层及个体化预测，具备潜在临床价值，但未来仍需更大样本量及外部队列的进一步验证。

Abstract: Objective: There exists significant heterogeneity in overall survival (OS) among patients with head and neck squamous cell carcinoma (HNSCC) undergoing radiotherapy, highlighting an urgent need for specific molecular risk-stratification tools. This study aimed to identify prognosis-associated long non-coding RNAs (lncRNAs) within the radiotherapy-treated (RT = Yes) subcohort of The Cancer Genome Atlas Head and Neck Squamous Cell Carcinoma (TCGA-HNSC) project, and to construct and internally validate a corresponding prognostic model, thereby providing a basis for individualized post-radiotherapy management. Methods: Transcriptomic and clinical data from the TCGA-HNSC cohort were integrated to extract samples from patients who received radiotherapy (RT = Yes). Following filtering and normalization of the expression data, univariate Cox regression and LASSO-Cox regression were sequentially applied to screen feature genes. A multivariate Cox proportional hazards model was subsequently employed to construct a risk score (RiskScore) model. The model’s discriminative ability and calibration performance were assessed through multi-dimensional internal validation metrics, including Kaplan-Meier curves, time-dependent receiver operating characteristic (timeROC) analysis, concordance index (C-index), nomogram construction, and calibration curves. Results: A total of 25 radiotherapy-treated samples (including 17 death events) were ultimately included in the analysis, leading to the identification of two key lncRNAs for model construction. Multivariate analysis identified HCFC1-AS1 as a significant protective factor (Hazard Ratio [HR] = 0.33, P = 0.008), while Lnc-ENSG257226 exhibited a protective trend (HR = 0.47). Bootstrap internal validation confirmed the robustness of the model coefficients. The model achieved a high C-index of 0.84, and the RiskScore was validated as a prognostic factor independent of age and disease stage. The model effectively stratified patients into high- and low-risk groups with significantly different OS outcomes (P = 0.0038). The nomogram and accompanying calibration curves demonstrated good agreement between predicted and observed 1-year and 3-year survival probabilities. Conclusion: In the radiotherapy-treated subpopulation of the TCGA-HNSC cohort, this study successfully developed and internally validated a two-lncRNA-based prognostic model. This model can effectively stratify OS risk and facilitate individualized outcome prediction for HNSCC patients receiving radiotherapy, showing potential clinical utility. However, future validation in larger, independent external cohorts is warranted.