肺癌胸腔镜术后病人肺部并发症列线图预测模型的构建

Construction of a line diagram predictive model for lung complications of patients with lung cancer after thoracoscopy surgery

目的:构建肺癌胸腔镜术后病人肺部并发症的列线图预测模型,并验证其预测效果。方法:便利抽取2021年1月—2022年4月在山东省某三级甲等医院胸外科行胸腔镜肺癌根治术的病人435例,以7∶3的比例将病人分为建模集(n=305)和验证集(n=130)。通过Lasso回归筛选预测因子,采用多因素Logistic回归分析建立肺癌根治术病人术后肺部并发症的列线图预测模型。采用Hosmer-Lemeshow检验判断模型的拟合度,采用受试者工作特征(ROC)曲线下面积(AUC)检验模型的区分度,Bootstrap重抽样法进行内部验证。利用验证集数据绘制ROC曲线、校准曲线,进行外部验证。结果:病人术后肺部并发症的发生率建模集为38.7%,验证集为32.3%。最终纳入模型的预测变量为年龄(OR=1.042)、美国麻醉医师协会(ASA)评分(OR=0.160)、慢性肺部疾病(OR=0.208)、病理分期(OR=11.418)。内部验证:AUC为0.854,95%CI[0.808,0.900]。Hosmer-Lemeshow检验,χ^(2)=7.350,P=0.500,灵敏度为78.8%,特异度为84.9%,准确率为82.6%。外部验证:AUC为0.848,95%CI[0.774,0.921],HosmerLemeshow检验,χ^(2)=7.672,P=0.466,灵敏度为70.2%,特异度为83.1%,准确率为78.5%。结论:基于Lasso回归构建的预测模型具有良好的预测效果,可为医护人员识别并发症高危病人提供依据。

Objective:To construct a line diagram predictive model for patients with lung cancer after thoracoscopy surgery and verify its prediction effect. Methods:A total of 435 patients with lung cancer who underwent radical resection of pulmonary carcinoma through thoracoscopy at a tertiary grade A hospital in Shandong province from January 2021 to April 2022 were conveniently extracted. The patients were divided into the modeling group(n=305)and the verification group(n=130)at a ratio of 7∶3. Prediction factors were screened through Lasso regression,and the line diagram prediction model of the lung complications of lung cancer patients underwent radical resection of pulmonary carcinoma was constructed by multi-factor logistic regression analysis;the fitting of the model was tested by Hosmer-Lemeshow test;the distinction of the model was tested by the area under curve(AUC)of receiver operating characteristic(ROC);the Bootstrap sampling method was used for internal verification;the data of verification group was used to draw and calibrate the ROC curve for external verification. Results:The occurrence rate of patients with postoperative lung complications was 38. 7% in the modeling group and 32. 3% in the verification group. The prediction variables finally included:age(OR=1. 042),American Society of Anesthesiology(ASA) class(OR=0. 160),chronic lung disease(OR=0. 208),and pathological staging(OR=11. 418). Internal verification included:AUC was 0. 854,95%CI[0. 808,0. 900],Hosmer-Lemeshow test showed χ^(2)=7. 350,P=0. 500,sensitivity was78. 8%,the specific degree was 84. 9%,and the accuracy was 82. 6%. External verification included:AUC was 0. 848,95%CI[0. 774,0. 921],Hosmer-Lemeshow test showed χ^(2)=7. 672,P=0. 466,sensitivity was 70. 2%,the specificity was 83. 1%,and the accuracy was78. 5%. Conclusions:This study is based on the prediction model of Lasso regression which had good prediction effects and could provide a basis for medical staff to identify high-risk patients with complications.