摘要
目的构建食管异物术后食管穿孔的预测模型,预测食管异物术后食管穿孔的发生几率。方法回顾性收集天门市第一人民医院耳鼻咽喉头颈外科2012年1月~2022年3月1137例食管异物患者的病例资料,按照入院时间顺序分为建模组(810例)和验证组(327例),所有患者均行食管镜食管异物探查术,根据术中食管有无穿孔,分为无食管穿孔组和食管穿孔组。对建模组进行lasso回归筛选变量以及通过随机森林对自变量重要性进行排序,纳入不同自变量构建模型1和模型2,两个模型在区分度、校准度、临床适用度三个方面进行对比,并且进行赤池信息准则比较,内部验证采用Bootstrap法10重交叉验证1000次进行验证,模型区分度采用受试者工作特征曲线评估Logistic回归模型预测价值,采用Hosmer-Lemeshow来评价模型校准度,临床适用度采用绘制临床决策曲线,最后绘制nomogram图展示预测结果。结果1137例食管异物患者均顺利取出异物,248例出现食管穿孔,食管穿孔的发生率约为21.8%(248/1137)。模型1在区分度、校准都及临床适用度三个方面较模型2优势,并且赤池信息准则较小,最终模型1为本次研究的预测模型。建模组Logistic回归显示居住地区、年龄、民族、嵌顿时间、手术时间是食管异物术后食管穿孔的独立危险因素。建模组中预测概率ROC曲线下面积为0.71(95%CI:0.664~0.756),验证组预测概率ROC曲线下面积为0.703(95%CI:0.634~0.771);当约登指数最大时,约登指数为0.4,此时灵敏度=0.82,特异度=0.58,曲线下面积为0.7,此时对应的P=0.208,根据P=1/(1+e-y),可计算出此时Y=-1.34,对建模组进行内部验证,采用Bootstrap法10重交叉验证1000次进行内部验证,准确性为0.778,kappa一致性为0.126,AUC为0.699;建模组模型1中Hosmer-Lemeshow拟合优度检测显示χ^(2)=0.004,P=1,验证组Hosmer-Lemeshow显示χ^(2)=9.13,P=0.10,表明模型具有良好的辨别能力,最后展示临�
OBJECTIVEETo construct the prediction model and predict the incidence of esophageal perforation after removal of the esophageal foreign body.METHODS The clinical data of 1137 patients were collected retrospectively from the Department of Otolaryngology Head and Neck Surgery of the First People's Hospital of Tianmen City from January 2012 to March 2022,and the patients were divided into the modeling group(810 cases)and the validation group(327 cases)according to the time sequence of admission.All patients underwent esophagoscopy for removal of the esophageal foreign body.The patients were divided into the non-esophageal perforation group and the esophageal perforation group according to whether the esophagus was perforated during the operation or not.The modeling group was screened by lasso regression method and the importance of independent variables was ranked by random forest method.The model 1 and model 2 were constructed by incorporating different independent variables.The two models were compared in three aspects of discrimination,calibration and clinical applicability,and the Chichi information criterion was compared also.The internal validation was verified by bootstrap method with 10 times of cross-validation for 1000 times.The model differentiation was evaluated by the prediction value of the logistic regression model using the working characteristic curve of the subjects,and the model calibration was evaluated by the Hosmer-Lemeshow method.The clinical applicability was to draw the clinical decision curve,and finally draw the nomogram diagramto show the prediction results.RESULTS The esophageal foreign bodies were removed successfully in all of the 1137 patients.Esophageal perforation occurred in 248 cases,and the incidence of esophageal perforation was 21.8%(248/1137).Model 1 was superior to model 2 in three aspects of discrimination,calibration and clinical applicability,and the Chichi information criterion was small.Finally,model 1 was chosen as the prediction model in this study.Logistic regression
作者
廖红明
秦亮
何本超
颜风波
LIAO Hongming;QIN Liang;HE Benchao;YAN Fengbo(Department of Otolaryngology Head and Neck Surgery,Tianmen First People's Hospital,Tianmen,Hubei,431700,China)
出处
《中国耳鼻咽喉头颈外科》
CSCD
2023年第4期251-256,共6页
Chinese Archives of Otolaryngology-Head and Neck Surgery
关键词
危险因素
回归分析
异物
列线图
临床决策曲线
Risk Factors
Regression Analysis
Foreign bodies
Nomogram
Decision Curve Analysis
