基于机器学习算法构建胸科患者术后并发症的预测模型  

Establishment of machine learning‑based prediction models for complications in patients after thoracic surgery

作　　者：许珍真[1] 李雪[1] 谢柯祺[2] 梁新全 刘庆浩[3] 陈冰璐 柳洁 闫婷[5] 王东信[1] Xu Zhenzhen;Li Xue;Xie Keqi;Liang Xinquan;Liu Qinghao;Chen Binglu;Liu Jie;Yan Ting;Wang Dongxin(Department of Anesthesiology,Peking University First Hospital,Beijing 100034,China;Department of Anesthesiology,Mianyang Cen-tral Hospital,Mianyang 618202,China;Department of Thoracic Surgery,Peking University First Hospital,Beijing 100034,China;Department of Anesthesiology,Cangzhou Hospital of Integrated Traditional and Western Medicine,Cangzhou 061001,China;Depart-ment of Critical Care Medicine,Peking University First Hospital,Beijing 100034,China)

机构地区：[1]北京大学第一医院麻醉科,北京100034 [2]绵阳市中心医院麻醉科,绵阳618202 [3]北京大学第一医院胸外科,北京100034 [4]河北省沧州中西医结合医院麻醉科,沧州061001 [5]北京大学第一医院重症医学科,北京100034

出　　处：《国际麻醉学与复苏杂志》2025年第1期59-67,共9页International Journal of Anesthesiology and Resuscitation

基　　金：北京大学第一医院科研种子基金(2013SF10);贝恩麻醉科学研究项目(2022)。

摘　　要：目的评估使用不同机器学习算法构建胸科手术术后并发症预测模型的效果并进行比较。方法该研究是一项针对既往两项随机对照研究数据的二次分析,共收集并分析了868例胸科手术患者的临床数据(术前、术中、术后资料)。根据患者是否出现术后并发症,分为有并发症组(255例)和无并发症组(613例)。以住院期间术后并发症(包括肺部并发症、心血管并发症、急性肾损伤、术后脑梗死、谵妄、外科并发症)为主要研究终点,采用最近邻(KNN)、随机森林(Random Forest)、支持向量机(SVM)、极致梯度提升(XGBoost)、正则化回归(Glmnet)和逻辑回归(logistic)等机器学习算法分别构建住院期间术后并发症的预测模型。通过5折交叉验证进行模型内部验证,计算受试者操作特征曲线下面积、校准度和准确率,选择表现最佳的模型进行特征重要性分析。结果术前资料方面,两组患者性别、年龄、白蛋白水平、吸烟史、饮酒史、高血压病、美国麻醉医师协会(ASA)分级、1秒率[第1秒用力呼气量(FEV1)/用力肺活量(FVC)]及最大分钟通气量(MVV)差异均有统计学意义(均P<0.05)。术中资料方面,两组患者围手术期预防性吸入戊乙奎醚、是否使用七氟醚/右美托咪定、术中舒芬太尼等效剂量、术中舒芬太尼剂量、术中出血量、液体正平衡情况、术中输血、术中是否出现低血压/高血压、手术时长、手术方式、手术种类、术后是否进入重症监护治疗病房(ICU)、术毕是否拔除气管导管差异均有统计学意义(均P<0.05)。术后资料方面,两组患者术后第3天的静息和术后第1~3天的运动数字分级评分法(NRS)评分、病理类型确诊恶性肿瘤、住院时长差异均有统计学意义(均P<0.05)。其余临床数据指标差异均无统计学意义(均P>0.05)。868例患者中有255例(约29.4%)发生术后并发症,肺部并发症发生率(154例,约17.7%)最高,心血管并�Objective To evaluate the efficiency of different machine learning algorithms in constructing prediction models for postoperative complications in thoracic surgery and compare their performance.Methods This study involved secondary analysis of data from two previous randomized controlled trials,collecting preoperative,intraoperative,and postoperative data from 868 tho‑racic surgery patients. The patients were divided into two groups based on the presence of postoperative complications: a complicationgroup (n=255) and a non‑complication group (n=613). Primary endpoints were postoperative complications during hospitalization, includingpulmonary complications, cardiovascular complications, acute kidney injury, postoperative stroke, delirium, and surgical complications.Machine learning algorithms such as K‑Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM), ExtremeGradient Boosting (XG‑Boost), regularized regression (Glmnet), and Logistic Regression were used to establish prediction models forpostoperative complications during hospitalization. The models were internally validated through 5‑fold cross‑validation, and their areaunder the receiver operating characteristic curve (AUC), calibration, and accuracy were calculated to select the optimal model for featureimportance analysis. Results Statistical differences were observed between the two groups in preoperative data, such as gender,age, albumin levels, smoking history, alcohol consumption, hypertension, American Society of Anesthesiologists (ASA) classification,forced expiratory volume in 1 second/forced vital capacity (FEV1/FVC) ratio, and maximal voluntary ventilation (MVV) (all P<0.05).There were also statistical differences in intraoperative data, such as prophylactic inhalation of penehyclidine hydrochloride, use ofsevoflurane/dexmedetomidine, equivalent doses of intraoperative sufentanil, intraoperative sufentanil dosage, intraoperative blood loss,fluid balance, intraoperative transfusion, intraoperative hypotension/hypertensio

关 键 词：胸科手术 术后并发症 机器学习算法 预测模型 

分 类 号：R614[医药卫生—麻醉学]