中晚期子宫内膜癌患者术后近期复发的机器学习模型及其预测价值  

Machine learning model of postoperative recent recurrence in patients with middle or advanced endometrial cancer and its predictive value

作　　者：莫紫文 李培源 张应亮 MO Zi-wen;LI Pei-yuan;ZHANG Ying-liang(Department of Gynecology,No.9 People′s Hospital of Nanning,Nanning 530409,Guangxi,China;the Forth Department of Internal Medicine,No.9 People′s Hospital of Nanning,Nanning 530409,Guangxi,China)

机构地区：[1]南宁市第九人民医院妇科,广西南宁市530409 [2]南宁市第九人民医院内四科,广西南宁市530409

出　　处：《广西医学》2022年第12期1363-1367,1401,共6页Guangxi Medical Journal

基　　金：广西壮族自治区卫生健康委员会自筹经费科研课题(Z20210450)。

摘　　要：目的探讨机器学习模型对中晚期子宫内膜癌患者术后近期复发的预测价值。方法纳入接受手术治疗的260例中晚期子宫内膜癌患者,并将其分为训练集(144例)与测试集(116例)。收集患者的临床及病理资料,采用单因素Cox回归模型分析影响中晚期子宫内膜癌患者术近期复发的危险因素。以得到的危险因素为基础,采用R 4.0.2软件构建5种机器学习模型[随机生存森林(RSF)、梯度提升机、支持向量机、K最近邻、Cox回归];采用一致性指数(C-index)评估5种机器学习模型预测中晚期子宫内膜癌患者术后近期复发的准确性;采用10折交叉验证法进行模型训练和内部验证;采用受试者工作特征曲线分析5种机器学习模型预测中晚期子宫内膜癌患者术后近期复发的效能。结果年龄≥60岁、国际妇产科协会分期Ⅲ~Ⅳ期、组织学分级G3级、肌层浸润深度>1/2、有淋巴结转移、术前雌激素受体阴性表达是5种机器学习预测模型同时筛选的中晚期子宫内膜癌患者术后近期复发的危险因素(均P<0.05)。5种机器学习模型的预测结果与实际结果均呈中度一致性(C-index为0.710~0.862),其中RSF的C-index最高,Cox回归的C-index最低;RSF的曲线下面积相对较大(0.875),且敏感度、特异度及准确度最高(分别为0.977、0.810、0.900)。结论机器学习模型筛选出6个临床病理特征,能对中晚期子宫内膜癌患者术后近期复发进行有效预测,其中RSF的预测效能相对较好。Objective To explore the predictive value of machine learning model on postoperative recent recurrence in patients with middle or advanced endometrial cancer.Methods A total of 260 patients with middle or advanced endometrial cancer receiving operation for treatment were enrolled and assigned to training set(144 cases)or testing set(116 cases).The clinical and pathological data of patients were collected,and univariate Cox regression model was used to analyze the risk factors for the postoperative recent recurrence in patients with middle or advanced endometrial cancer.The R 4.0.2 software was employed to establish 5 machine learning models(random survival forest[RSF],gradient boosting machine,support vector machine,K-nearest neighbor,and Cox regression)based on the acquired risk factors.The concordance index(C-index)was used to evaluate the accuracy of the 5 machine learning models for predicting postoperative recent recurrence in patients with middle or advanced endometrial cancer.The 10-fold cross-validation method was employed to conduct model training and internal verification.The receiver operating characteristic curve was used to analyze the efficiency of the 5 machine learning models for predicting postoperative recent recurrence in patients with middle or advanced endometrial cancer.Results The risk factors for postoperative recent recurrence in patients with middle or advanced endometrial cancer simultaneously screened by the 5 machine learning prediction models were listed as follows:age≥60 years old,the International Federation of Gynecology and Obstetrics stage inⅢ-Ⅳ,histological grade G3,myometrial invasion>1/2,lymph nodes metastases,and preoperative estrogen receptor in negative expression(all P<0.05).There was a moderate consistency between the predictive results of the 5 machine learning models and the practical results(with C-index of 0.710-0.862),thereinto the highest C-index went to the RSF,whereas the Cox regression exhibited the lowest C-index.The area under the curve of RSF depicted r

关 键 词：子宫内膜癌 中晚期 机器学习 手术 近期复发 预测效能 

分 类 号：R737.33[医药卫生—肿瘤]