喉鳞状细胞癌患者发生隐匿性淋巴结转移风险预测模型的开发和验证

Development and validation of a predictive model for the risk of occult lymph node metastasis in patients with laryngeal squamous cell carcinoma

摘要:

目的 对比分析喉鳞状细胞癌患者有隐匿性淋巴结转移和无隐匿性淋巴结转移患者之间的差异，开发和验证发生隐匿性淋巴结转移风险的预测模型。方法 采取病例对照研究方法，回顾性收集2005—2018年来山西医科大学第一医院耳鼻咽喉科就诊的喉癌患者的临床资料，所有患者在术中均进行了功能性或根治性颈部淋巴结清扫术，将符合条件的136例纳入本研究中，按是否发生隐匿性淋巴结转移分为有隐匿性淋巴结转移组（46例）和无隐匿性淋巴结转移组（90例），比较分析两组患者的年龄、病理分级、临床分期、肿瘤大小、浸润深度、切缘阴阳性、喉癌分型、会厌前间隙是否受侵、软骨是否受侵、前联合是否受侵等危险因素，将差异具有统计学意义（P<0.05）的危险因素进一步行二元Logistic回归分析，并由此建立一个可以预测其发生风险的预测模型。预测模型公式为：Logit （P）=0.262+1.010×X1+1.624×X2+1.205×X3-1.153×X4-2.270×X5（X1=前联合受侵，X2=肿瘤大小>4~6 cm，X3=肿瘤大小>2~4 cm，X4=中分化，X5=高分化）；计算工作特征（ROC）曲线下面积（AUC）评价模型准确度；利用bootstrap方法对模型进行内验证，用AUC评价模型的区分度，用校准图评价模型的一致性。结果 前联合受侵、病理分级低分化、肿瘤大小（>2~4 cm）是发生隐匿性淋巴结转移的独立危险因素（P<0.05），AUC为0.767，95%CI为（0.679，0.854），予以bootstrap法内验证：AUC为0.757，95%CI为（0.661，0.841），校准曲线与理想曲线较接近，可认为预测结果与实际结果较一致。结论 前联合受侵、肿瘤大小（>2~4 cm）、病理分级低分化是发生隐匿性淋巴结转移的独立危险因素，经评价和验证后，得知此风险预测模型对预测是否发生隐匿性淋巴结转移的准确性较高。

关键词:喉鳞状细胞癌;隐匿性淋巴结转移;危险因素;预测模型

Abstract:

Objective To compare and analyze the difference between laryngeal squamous cell carcinoma with or without occult lymph node metastasis, and to develop and validate a predictive model for the risk of occult lymph node metastasis.Methods To adopt the method of a case-control study, and retrospectively collect the clinical data of laryngeal cancer patients admitted to the Department of Otolaryngology, the First Hospital of Shanxi Medical University from 2005 to 2018.All patients underwent the operation of functional or radical neck lymph node dissection. In this study, 136 eligible patients were included. According to the occurrence of occult lymph node metastasis, they were divided into one group with occult lymph node metastasis(46) and another group without occult lymph node metastasis(90). The age, pathological grade, clinical stage, tumor size, depth of invasion, caner positive or negative of cutting edge, classification of laryngeal cancer, invasion of anterior epiglottic space, cartilage invasion, anterior commissure invasion and other risk factors were compared and analyzed between the two groups. The difference with statistically significant (P<0.05) of risk factors between the two groups was further analyzed by binary Logistic regression. The established prediction model can predict the occurrence risk. The prediction model formula was:Logit(P)=0.262+1.010×X1+1.624×X2+1.205×X3-1.153×X4-2.270×X5 (X1=anterior commissure invasion, X2=tumor size >4~6 cm, X3=tumor size >2~4 cm, X4=moderately differentiated, X5=well differentiated). At the same time, a scoring system was established to compare whether there was any difference between the two groups (P<0.05). The area under the ROC curve (AUC) was calculated to evaluate the accuracy of the model. Bootstrap method was used for internal validation of the model, AUC was used to evaluate the differentiation of the model, and calibration map was evaluated the consistency of the model.Results Anterior syndesmosis invasion, poor differentiation and tumor size (>2~4 cm) were independent risk factors for occult lymph node metastasis (P<0.05). The AUC was 0.767, 95%CI was (0.679, 0.854), which was verified by bootstrap method:AUC was 0.757, 95%CI was (0.661, 0.841). The calibration curve was close to ideal curve, which was considered as consistent with actual results.Conclusions Invasion of anterior commissure, tumor size (>2~4 cm) and poor pathological grade were independent risk factors for occult lymph node metastasis. After evaluation and verification, it was known that this risk prediction model had a high accuracy in predicting the occurrence of occult lymph node metastasis.

Key words:Laryngeal squamous cell carcinoma;Occult lymph node Metastasis;Risk factors;Prediction model