基于机器学习构建甲状腺乳头状癌淋巴结跳跃性转移的预测模型  

Building a predictive model for the lymphatic skip metastasis of papillary thyroid carcinoma based on machine learning

作　　者：王帅 于耀程 尉志伟 王廷 WANG Shuai;YU Yaocheng;YU Zhiwei;WANG Ting(Department of Thyroid,Breast and Vascular Surgery,the First Affiliated Hospital of Air Force Medical University,Xi’an 710032,China)

机构地区：[1]空军军医大学第一附属医院甲乳血管外科,陕西西安710032

出　　处：《中国肿瘤外科杂志》2024年第4期324-329,共6页Chinese Journal of Surgical Oncology

摘　　要：目的探讨甲状腺乳头状癌淋巴结跳跃性转移的发生率和危险因素,并利用机器学习算法建立预测模型。方法回顾性分析2013年6月至2021年12月空军军医大学第一附属医院收治的730例甲状腺乳头状癌患者的临床病理资料,依据是否发生淋巴结跳跃性转移分为阳性组和阴性组。使用单因素及多因素分析确定跳跃性转移的独立相关因素。基于逻辑回归、k近邻、决策树、随机森林、极致梯度提升和支持向量机6种机器学习算法构建预测模型。结果甲状腺乳头状癌淋巴结跳跃性转移的发生率为11.51%(84/730);多因素分析提示,年龄>45岁、甲状腺微小乳头状癌、肿瘤累及上极、腺外侵犯、中央区淋巴结检出数和颈侧区淋巴结转移数是跳跃性转移的独立相关因素;受试者工作特征曲线显示,k近邻算法预测性能最佳,受试者工作特征曲线下面积在训练集和验证集分别为0.813和0.808。结论甲状腺乳头状癌淋巴结跳跃性转移与年龄、是否为甲状腺微小乳头状癌、肿瘤是否累及上极、肿瘤是否腺外侵犯、中央区淋巴结检出数和颈侧区淋巴结转移数相关;基于独立相关因素构建的机器学习算法模型具有较高的预测性能,可以指导临床决策。Objective To explore the incidence and risk factors of skip metastasis in papillary thyroid carcinoma and establish a predictive model using machine learning algorithms.Methods A retrospective analysis was conducted on the clinical and pathological data of 730 patients with papillary thyroid carcinoma admitted to the First Affiliated Hospital of Air Force Medical University from June 2013 to December 2021.The patients were divided into a positive group and a negative group based on the presence of skip metastasis.Univariate and multivariate analyses were used to determine the independent factors associated with skip metastasis.A predictive model was built using six machine learning algorithms including Logistic regression,k⁃nearest neighbors,decision tree,random forest,extreme gradient boosting,and support vector machine.Results The incidence of skip metastasis in papillary thyroid carcinoma was 11.51%(84/730).Multivariate analysis indicated that age>45 years,microcarcinoma,tumor involvement of the upper pole,extrathyroidal extension,number of central compartment lymph nodes,and number of lateral compartment lymph node metastases were independent factors associated with skip metastasis.The receiver operating characteristic curve showed that the k⁃nearest neighbor algorithm had the best predictive performance,with an area under the curve of 0.813 in the training set and 0.808 in the validation set.Conclusions Skip metastasis in papillary thyroid carcinoma is associated with age,papillary thyroid microcarcinoma,tumor involvement of the upper pole,extrathyroidal extension,number of central compartment lymph nodes,and number of lateral compartment lymph node metastases.The independent⁃factor⁃based machine learning algorithm model has high predictive performance and can guide clinical decision⁃making.

关 键 词：甲状腺乳头状癌 跳跃性转移 机器学习 预测模型 

分 类 号：R736.1[医药卫生—肿瘤]