基于CT图像的评分模型预测食管鳞癌喉返神经旁淋巴结转移风险

目的:基于原发肿瘤及淋巴结CT特征建立评分模型预测食管鳞癌患者喉返神经旁淋巴结(RLN-LN)转移风险。方法:回顾性收集2014年1月至2019年12月于北京大学肿瘤医院行食管癌根治术并清扫RLN-LN的92例食管鳞癌患者。根据术后淋巴结病理结果分为RLN-LN转移组(n=37)和非转移组(n=55)。评估术前CT图像,记录食管癌患者年龄、性别、分化程度、肿瘤位置、肿瘤大小(肿瘤长度、肿瘤厚度、厚度/长度)、RLN-LN大小(淋巴结短径、长径、短径/多平面重建(MPR)最长径]。采用多元logistic回归筛选独立预测因子并建立评分模型,采用ROC曲线评估评分模型及独立预测因子诊断RLN-LN转移的效能,采用Z检验比较曲线下面积(AUC)的差异。应用Hosmer-Lemeshow检验和校准曲线评估模型拟合度。结果:肿瘤位置、肿瘤长度、RLN-LN短径、短径/MPR最长径是RLN-LN转移的独立预测因子,其诊断RLN-LN转移的AUC分别为0.586、0.705、0.831、0.777。基于以上4个CT特征建立评分模型,评分模型诊断RLN-LN转移的AUC为0.903(95%CI 0.846~0.959),优于各单一CT特征(Z=5.812,P<0.001;Z=2.161,P=0.030;Z=2.929,P=0.003;Z=4.052,P<0.001)。拟合优度Hosmer-Lemeshow检验结果显示P=0.555,校准曲线提示评分模型预测RLN-LN转移风险与实际转移风险之间具有良好的一致性。结论:基于CT图像的评分模型有助于食管鳞癌RLN-LN转移状态危险分层。

Developing a CT scoring system to predict the metastasis of recurrent laryngeal nerve lymph nodes in esophageal squamous cell cancer

Objective To develop a scoring model based on CT feature of primary tumor and lymph node to predict the risk of recurrent laryngeal nerve lymph node(RLN-LN)metastasis in patients with esophageal squamous cell cancer(ESC).Methods A total of 92 ESC patients who received radical resection of esophageal carcinoma and RLN-LN dissections during January 2014 to November 2019 in Peking University Cancer Hospital were retrospectively reviewed,and were divided into metastatic group(n=37)and non-metastatic group(n=55)according to the postoperative pathological results of RLN-LN.Pre-operative thoracic CT imaging features were analyzed,including age,sex,differentiation,tumor locations,tumor sizes(length,thickness,the ratio of thickness to length),and RLN-LN sizesshort diameter,long diameter,the ratio of short diameter to longest diameter in multi-planner reformation(MPR)].The stepwise regression of multivariate logistic regression analysis was used to identify the useful features and establish scoring model.The diagnostic efficacy for RLN-LN metastisis of CT features and scoring model were evaluated using ROC curves.The differences of area under the curve(AUC)were compared using the Z test.The Hosmer-Lemeshow test and calibrating curve were used to evaluate model fitting.Results The tumor location,tumor length,short diameter of RLN-LN and short diameter/MPR long diameter were independent predictors of RLN-LN metastasis,and the AUC of diagnosis of RLN-LN metastasis was 0.586,0.705,0.831 and 0.777,respectively.A scoring model was established based on the above 4 CT features,and the AUC of the scoring model in diagnosing RLN-LN metastasis was 0.903(95%CI 0.846-0.959),which was better than each single CT feature(Z=5.812,P<0.001;Z=2.161,P=0.030;Z=2.929,P=0.003;Z=4.052,P<0.001).The Hosmer-Lemeshow test results showed P value of 0.555,and the calibration curve indicated that there was good consistency between the predicted risk of RLN-LN metastasis and the actual value.Conclusion The scoring model based on CT image can help to predict the risk stratification of RLN-LN metastasis in patients with ESC.