第一部分：关于T1-2期声门上型喉鳞状细胞癌术后最小淋巴结检验数目的研究

目的

T1-2期声门上型喉鳞状细胞癌（Laryngeal squamous cell carcinoma，LSCC）的淋巴结转移状态与患者术后治疗方案的选择以及预后情况密切相关。而阳性淋巴结的发现依赖于颈部淋巴结清扫术后足量的淋巴结检验数目（examined lymph nodes，ELNs）。本研究旨在探讨T1-2期声门上型LSCC患者接受根治性手术切除后，为了确保淋巴结转移状态评估的准确性以及使患者获得更加良好预后所需的ELNs最低阈值‌。

方法

本研究回顾性地纳入了来自美国监测、流行病学和最终结果（The Surveillance, Epidemiology, and End Results，SEER）数据库及中国医学科学院肿瘤医院/国家癌症中心（The National Cancer Center，NCC）的T1-2期声门上型LSCC患者数据。通过单因素和多因素回归模型分析ELN数量与转移性淋巴结检出率以及总体生存期（Overall survival，OS）之间的相关性。采用局部加权散点平滑拟合曲线来拟合不同ELN数量下发现转移性淋巴结的比值比以及发生死亡的风险比随ELN数量变化的趋势。根据上述趋势变化规律，并结合R语言“changepoint”包来确定ELNs的最佳截断值‌。

结果

       本研究一共纳入来自SEER数据库的429例患者以及来自NCC的53例患者‌。术后检出的转移性淋巴结数量与ELN数目呈显著的正相关（R²=0.783，P=0.046）‌。ELN数目与转移性淋巴结的检出概率存在显著的相关性（P<0.001），当ELNs数目>10时，发现阳性淋巴结的期望急剧升高。在病理诊断淋巴结为阴性（pN0）的患者中，ELN数目与OS显著相关（P=0.036），当ELN数目>10时，患者的死亡风险显著降低‌。以ELN数目=10为界能够将pN0期的患者划分为生存显著差异的两组（SEER队列，P=0.001；NCC队列，P=0.020）。亚组分析提示，ELN数目>10在术后未接受辅助治疗的患者中是一种生存的保护因素（P=0.007），但在接受辅助治疗者中却不是（P=0.135）。

结论

1. 足量的ELN数目有助于淋巴结转移状态判断的准确性。

2. 10个淋巴结可能是颈部淋巴结清扫术后检验淋巴结所需的最小数量。

3. 对于pN0期的患者，ELNs少于10枚可能是一个不利的预后因素。

4. 术后辅助治疗可能可以弥补因ELN数量不足而造成的不良预后。

关键词

声门上型喉鳞状细胞癌，淋巴结检验数目，颈部淋巴结清扫，淋巴结转移，生存，辅助治疗

第二部分：基于头颈部鳞状细胞癌组织病理学图像识别预测颈部淋巴结转移及预后的人工智能模型

目的

淋巴结转移状态对头颈部鳞状细胞癌（Head and neck squamous cell carcinoma，HNSCC）手术策略的选择以及患者的预后都至关重要。然而，目前现有的术前检查方式对淋巴结状态的判断均无法达到满意的效果。本研究的目标是开发一种基于HNSCC肿瘤组织病理学图像以预测淋巴结转移的人工智能（Artificial intelligence，AI）模型，并评估其临床应用价值。

方法‌

本研究共纳入了来自中国医学科学院肿瘤医院/国家癌症中心（The National Cancer Center，NCC）、癌症基因组图谱（The Cancer Genome Atlas，TCGA）以及临床蛋白质组肿瘤分析协作组（Clinical Proteomic Tumor Analysis Consortium，CPTAC）的HNSCC病理切片数字化图像，用于模型开发和验证。模型利用CPTAC数据集进行预训练，利用NCC数据进行模型开发以及内部验证，利用TCGA队列中福尔马林固定石蜡包埋（‌Formalin-Fixed and Paraffin-Embedded‌，FFPE）的切片图像进行模型的外部验证，而新鲜冷冻（Fresh frozen，FF）切片图片部分用于模型的微调并构建基于FF切片的预测模型，并用剩余部分进行模型的验证。研究的主要通过模型对淋巴结转移预测的准确率、灵敏度、特异度、精准率、受试者工作特征曲线下面积（The area under the receiver operating characteristic curve，AUROC）以及精准率-召回率曲线下面积（The area under the precision-recall curve, AUPR）等指标衡量模型的预测效果。

结果‌

本研究共收集了来自699名HNSCC患者的1590张数字病理图像。在三个验证集中单因素检验结果表明，模型预测结果能有效提示淋巴结转移（三组均为P<0.001）。经多因素检验后，模型预测结果依然是淋巴结转移的独立预测因子（三组均为P<0.001）。在三组验证集中，模型的准确率、灵敏度、特异度、精准率多数能达到0.8甚至0.85以上的水平。再以受试者工作特征曲线（The receiver operating characteristic curve，ROC）检验模型的预测效能，结果显示，模型在内部验证集（AUROC：0.895，95%置信区间[Confidence Interval，CI]：0.856-0.934）、外部验证集（AUROC：0.853，95%CI：0.804-0.901）及FF验证集（AUROC：0.812，95%CI：0.744-0.881）中均能达到较好的预测效力。此外，模型的AUPR值在三个验证集中也均达到了0.8以上。在亚组分析中，预测模型在三个验证集的具有不同临床病理特征的患者亚群中均可达到良好的预测效能。进一步拓展将模型应用于预后的预测中，结果显示也体现良好的提示价值（内部验证集，P<0.001；外部验证集，P<0.001；FF验证集，P=0.006）。对模型的可解释性探究结果提示，阳性预测值的HNSCC肿瘤微环境总体表现出一种免疫抑制状态。

结论‌

1.基于肿瘤H&E染色切片的AI模型能够准确地预测HNSCC的淋巴结转移状态。

2.该模型能够利用肿瘤的FFPE以及FF切片对淋巴结转移进行预测。

3.该模型对HNSCC患者预后情况也有良好预测能力。

4.模型的预测机制可能与HNSCC免疫微环境状态密切相关。

关键词‌

人工智能‌，头颈部鳞状细胞癌，淋巴结转移，图像识别，预测

第三部分：肿瘤中CD276表达与头颈部鳞状细胞癌辅助及免疫治疗预后的相关性

目的

       ‌在头颈部鳞状细胞癌（Head and neck squamous cell carcinoma，HNSCC）中，术后辅助治疗的效果有限，而免疫检查点抑制剂（Immune checkpoint inhibitors，ICIs）的疗效尚未达到理想水平‌。CD276作为与PD-L1同家族的分子，已被证实是一种新型免疫检查点，可通过抑制抗肿瘤免疫活性在多种实体瘤中发挥作用‌。本研究旨在探索CD276对HNSCC预后以及辅助治疗、免疫治疗疗效预测的价值，并初步探究其与免疫微环境以及肿瘤分子特征之间的关系。

方法‌

本研究利用来自中国医学科学院肿瘤医院/国家癌症中心（The National Cancer Center，NCC）的216例接受了根治术治疗的HNSCC患者样本构建组织芯片，采用免疫组织化学（Immunohistochemistry，IHC）染色方法对肿瘤组织中的CD276表达以及浸润性免疫细胞、免疫分子的数量和功能进行计数及分析。同时整合癌症基因组图谱（The Cancer Genome Atlas，TCGA）和基因表达综合数据库（Gene Expression Omnibus，GEO）的基因组及表型数据用于对NCC队列中发现结果的验证，以及对潜在机制的研究‌。

‌结果‌

在三组队列中，CD276的高表达均表现出与患者不良的预后密切相关（探索队列：OS，HR=2.250，P=0.001；PFS，HR=1.798，P=0.007。NCC队列：OS，HR=3.007，P=0.001；DSS，HR=3.399，P=0.001。TCGA队列：OS，HR=1.489，P=0.004；PFS，HR=1.384，P=0.023）。在接受辅助化疗（Adjuvant chemotherapy，ACT）的患者中，低表达CD276者具有显著更优的生存率（NCC队列：P=0.042；TCGA队列：P=0.035）。但CD276的表达与辅助放疗的疗效并无显著的相关性。对于接受了抗PD-1/PD-L1免疫治疗的患者来说，CD276高表达亚组中对药物具有反应者所占的比例更低（P=0.046），并且呈现出显著不佳的预后（P=0.030）‌。联合CD276与现有生物标志物可优化对免疫治疗反应的预测效力‌。对肿瘤免疫微环境分析发现，高表达CD276的HNSCC微环境呈现出适应性免疫细胞浸润数量减少、CD8⁺ T细胞功能受损、巨噬细胞倾向于M2型分化以及免疫相关分子表达较少‌的表现。此外，CD276高表达的HNSCC呈现低人乳头瘤病毒（Human papilloma virus，HPV）感染率、高TP53突变频率的分子特征‌。

结论

‌1. CD276是HNSCC预后的可靠预测因子

2. CD276能够较好预测ACT及抗PD-1/PD-L1治疗的疗效。

3. CD276与免疫抑制性肿瘤微环境密切相关。

4. 高表达CD276的HNSCC具有一定分子表型特征。

‌关键词‌

CD276‌，头颈部鳞状细胞癌，免疫治疗，辅助化疗，免疫组化，预后

Part I：The minimal number of examined lymph nodes for accurate nodal staging and favorable prognosis in T1-2 supraglottic laryngeal squamous cell carcinoma

Background‌
The nodal metastasis status in T1-2 supraglottic laryngeal squamous cell carcinoma (LSCC) has a critical impact on postoperative treatment strategies and prognosis. Accurate diagnosis of metastatic lymph nodes depends on an adequate number of examined lymph nodes (ELNs) after neck dissection‌. This study aimed to determine the minimum ELN threshold required for precise nodal staging and improved survival outcomes in T1-2 supraglottic LSCC patients after radical resection.

‌Methods‌
A retrospective analysis was conducted using data from the Surveillance, Epidemiology, and End Results (SEER) database (429 patients) and the National Cancer Center (NCC) (53 patients)‌. Univariate and multivariate regression models were employed to analyze correlations between ELN count, metastatic lymph node detection rate, and overall survival (OS). Locally weighted scatterplot smoothing (LOWESS) curves and the R package “changepoint” were utilized to identify the optimal ELN cutoff‌.

‌Results‌

The number of identified metastatic lymph nodes positively correlated with ELN count (R²=0.783, P=0.046) ‌. And the probability of detecting nodal metastasis was significantly related to ELNs, since a sharp increase in the odds of metastatic lymph node detection was observed when the number of ELNs grew. In pathologically node-negative (pN0) patients, ELN count >10 significantly reduced mortality risk (P=0.036) and stratified survival outcomes (SEER cohort: P=0.001; NCC cohort: P=0.020) ‌. Subgroup Analysis showed ELN >10 was a protective factor for survival in patients without adjuvant therapy (P=0.007) but not in those receiving adjuvant therapy (P=0.135) ‌.

‌Conclusions‌

1. Adequate ELNs enhance the accuracy of nodal staging‌.

2. ELN ≥10‌ is the minimum threshold for reliable nodal staging after neck dissection‌.

3. ELN <10 in pN0 patients correlates with poorer prognosis‌.

4. Adjuvant therapy may mitigate adverse outcomes caused by insufficient ELNs.

‌Keywords‌
Supraglottic laryngeal squamous cell carcinoma‌, Examined lymph nodes, Neck dissection‌, Lymph node metastasis‌, Survival, Adjuvant therapy‌

Part II: Artificial Intelligence Model for Predicting Cervical Lymph Node Metastasis and Prognosis Based on Histopathological Image Recognition in Head and Neck Squamous Cell Carcinoma‌

‌Background‌

Lymph node metastasis status is critical for treatment selection and prognosis in head and neck squamous cell carcinoma (HNSCC). However, current preoperative evaluation methods for lymph node status remain unsatisfactory. The objective of this study was to develop an artificial intelligence (AI) model based on histopathological images of HNSCC tumors to predict lymph node metastasis and evaluate its clinical utility.

‌Methods‌

This study included digitized pathological section images of HNSCC from the National Cancer Center (NCC) of the Cancer Hospital of Chinese Academy of Medical Sciences, The Cancer Genome Atlas (TCGA), and the Clinical Proteomic Tumor Analysis Consortium (CPTAC) for model development and validation. The model was pre-trained using the CPTAC dataset, developed and internally validated using NCC data, externally validated using formalin-fixed and paraffin-embedded (FFPE) section images from the TCGA cohort. Fresh frozen (FF) section images were used for model fine-tuning to construct an FF-based predictive model, with the remaining portions used for model validation. The predictive performance was evaluated through metrics including accuracy, sensitivity, specificity, precision, area under the receiver operating characteristic curve (AUROC), and area under the precision-recall curve (AUPR).‌Results‌

This study collected a total of 1590 digital pathological images from 699 HNSCC patients. Univariate analysis across three validation sets demonstrated that the model predictions effectively indicated lymph node metastasis (all three groups: P<0.001). After multivariate analysis, the model predictions remained independent predictors of lymph node metastasis (all three groups: P<0.001). In the three validation sets, the model achieved accuracy, sensitivity, specificity, and precision levels mostly above 0.8 or even 0.85. Receiver operating characteristic (ROC) curve analysis further validated the model’s predictive performance, showing robust results in the internal validation set (AUROC: 0.895, 95% confidence interval [CI]: 0.856-0.934), external validation set (AUROC: 0.853, 95% CI: 0.804-0.901), and FF validation set (AUROC: 0.812, 95% CI: 0.744-0.881). Additionally, the model’s area under the precision-recall curve (AUPR) exceeded 0.8 in all three validation sets. In subgroup analyses, the predictive model maintained strong performance across patient subgroups with different clinicopathological characteristics in all three validation sets. When extended to prognosis prediction, the model also demonstrated significant predictive value (internal validation set: P<0.001; external validation set: P<0.001; FF validation set: P=0.006). The interpretability investigation indicated that the tumor microenvironment of HNSCC with positive predictive value exhibited an immunosuppressive state.

‌Conclusion‌

1. AI models based on tumor H&E-stained sections can accurately predict lymph node metastasis status in HNSCC.

2. The model can predict lymph node metastasis using tumor FFPE and FF sections.

3. The model also exhibits strong predictive performance for the prognosis of HNSCC patients.

4. The predictive mechanism of the model may be closely associated with the immune microenvironment status of HNSCC‌.

‌Keywords‌

Artificial intelligence, head and neck squamous cell carcinoma, lymph node metastasis, image recognition, prediction

Part III：Intratumoral CD276 Expression Correlates with Clinical Outcomes of Adjuvant Chemotherapy and Immunotherapy in Neck and Neck Squamous Cell Carcinoma

Background‌‌

In head and neck squamous cell carcinoma (HNSCC), the efficacy of adjuvant therapies remains suboptimal, and the therapeutic benefits of immune checkpoint inhibitors (ICIs) have not yet reached ideal levels‌. CD276, a molecule belonging to the same family as PD-L1, has been identified as a novel immune checkpoint that suppresses anti-cancer immune activity in multiple solid tumors. This study aimed to explore the prognostic value of CD276 in HNSCC, its predictive potential for adjuvant chemotherapy (ACT) and anti-PD-1/PD-L1 therapy efficacy, and its associations with the tumor immune microenvironment and molecular characteristics‌.

‌Methods‌

A tissue microarray was constructed using 216 HNSCC patient samples from the National Cancer Center (NCC) who underwent radical surgery. Immunohistochemistry (IHC) was employed to quantify CD276 expression and analyze infiltrating immune cells and immune-related molecules‌. Genomic and phenotypic data from the Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) were integrated to validate findings from the NCC cohort and investigate underlying mechanisms‌.

‌Results‌

‌In all three cohorts, high CD276 expression was significantly associated with poor prognosis (exploratory cohort: OS, HR = 2.250, P = 0.001; PFS, HR = 1.798, P = 0.007; NCC cohort: OS, HR = 3.007, P = 0.001; DSS, HR = 3.399, P = 0.001; TCGA cohort: OS, HR = 1.489, P = 0.004; PFS, HR = 1.384, P = 0.023). Among patients receiving adjuvant chemotherapy (ACT), those with low CD276 expression exhibited significantly better survival (NCC cohort: P = 0.042; TCGA cohort: P = 0.035). However, CD276 expression showed no significant correlation with the efficacy of adjuvant radiotherapy. In patients treated with anti-PD-1/PD-L1 immunotherapy, the high CD276 expression subgroup had a lower proportion of responders (P = 0.046) and demonstrated significantly worse prognosis (P = 0.030). Combining CD276 with existing biomarkers improved the predictive accuracy for immunotherapy response. Analysis of the tumor immune microenvironment revealed that HNSCC with high CD276 expression displayed reduced infiltration of adaptive immune cells, impaired CD8⁺ T-cell function, M2-polarized macrophage differentiation, and lower expression of immune-related molecules. Additionally, HNSCC with high CD276 expression exhibited molecular features of low human papilloma virus infection rates and high TP53 mutation frequencies.

‌Conclusions‌

1. CD276 serves as a reliable prognostic biomarker in HNSCC‌.

2. CD276 effectively predicts therapeutic responses to ACT and anti-PD-1/PD-L1 therapy‌.

3. CD276 is closely linked to an immunosuppressive tumor microenvironment‌.

4. CD276-high HNSCC tumors exhibit distinct molecular phenotypes‌.

‌Keywords‌

CD276‌, Head and neck squamous cell carcinoma‌, Immunotherapy‌, Adjuvant chemotherapy‌, Immunohistochemistry‌, Prognosis‌