文章摘要
基于机器学习的中老年患者胸腔镜肺部切除术后肺部并发症的危险因素分析
Risk factor analysis for postoperative pulmonary complications in middle-aged and elderly patients undergoing thoracoscopic lung resection surgery based on machine learning
  
DOI：10.12089/jca.2026.06.002
中文关键词: 术后肺部并发症  胸腔镜手术  危险因素  机器学习  随机森林  SHAP分析
英文关键词: Postoperative pulmonary complications  Thoracoscopic surgery  Risk factors  Machine learning  Random forest  SHAP analysis
基金项目:济仁慈善基金会课题（JRH000032）；东方英才计划拔尖人才项目（BJKJ2024039）
作者单位E-mail
戈弋 200093,上海理工大学健康科学与工程学院  
王委 上海市胸科医院麻醉科  
刘坤 上海市胸科医院麻醉科  
吴镜湘 上海市胸科医院麻醉科 wjx1132@163.com 
摘要点击次数: 392
全文下载次数: 64
中文摘要:
      
目的： 基于机器学习分析中老年患者胸腔镜肺部切除术后肺部并发症（PPCs）的独立危险因素。
方法：收集2020年8月至2023年8月行择期胸腔镜肺部手术的中老年患者围术期资料，年龄60～84岁，BMI 18.5～32.0 kg/m2，ASA Ⅱ或Ⅲ级。根据是否发生PPCs将患者分为两组：PPCs组和非PPCs组。采用单因素分析筛选相关危险因素，采用随机森林算法构建预测模型，并通过沙普利加性解释（SHAP）值解释模型决策过程，量化各特征的贡献度。
结果：共纳入患者392例，有104例（26.5%）发生PPCs。单因素分析显示，性别、年龄、身高、体重、预测体重、吸烟史、肺一氧化碳弥散量（DLCO）、加泰罗尼亚风险评估（ARISCAT）评分、手术时间、术中胶体液输注量、潮气量、OLV最佳呼气末正压(PEEP)值与PPCs相关（P<0.05）。随机森林模型预测PPCs的准确率为0.80，受试者工作特征曲线下面积（AUCROC）为0.88（95%CI 0.82～0.91）。SHAP分析显示，ARISCAT评分升高是预测PPCs的最重要因素，其后依次为手术时间延长、男性、高龄（≥75岁）、吸烟史、术中胶体液输注量偏大及潮气量偏高。亚组分析显示，对于ARISCAT评分≥43分或第1秒用力呼气容积(FEV1)处于50%～70%的中度肺功能受损患者，较高的个体化PEEP（9～13 cmH2O）与PPCs风险降低相关。
结论： ARISCAT评分升高、手术时间延长、高龄、男性、吸烟史、术中胶体液输注量偏大及潮气量偏高是中老年患者胸腔镜肺部手术发生PPCs的危险因素。
英文摘要:
      
Objective: To identify the independent risk factors for postoperative pulmonary complications (PPCs) in middle-aged and elderly patients after thoracoscopic lung resection surgery based on machine learning.
Methods: The perioperative data of middle-aged and elderly patients, aged 60-84 years, BMI 18.5-32.0 kg/m2, ASA physical status Ⅱ or Ⅲ, who underwent elective thoracoscopic lung surgery from August 2020 to August 2023 were collected. Patients were divided into two groups based on whether PPCs occurred: the PPCs group and the non-PPCs group. Univariate analysis was first used to screen associated factors, then a random forest model was built for prediction, and the Shapley additive explanations (SHAP) framework was applied to interpret the model's decision-making process, quantifying the contribution and direction of each feature.
Results: A total of 392 patients were included, and 104 patients (26.5%) developed PPCs. Univariate analysis showed gender, age, height, weight, predicted body weight, smoking history, diffusing capacity of the lungs for carbon monoxide (DLCO), ARISCAT score, duration of surgery, intraoperative colloid infusion volume, tidal volume, optimal positive end-expiratory pressure (PEEP) during one-lung ventilation were associated with PPCs (P < 0.05). The random forest model predicted PPCs with an accuracy of 0.80 and an area under the receiver operating characteristic curve (AUCROC) of 0.88 (95% CI 0.82-0.91). SHAP analysis indicated that the ARISCAT score was the most important factor for predicting PPCs, followed by extended surgery time, male gender, advanced age (≥ 75 years), smoking history, high intraoperative colloid infusion volume, and high tidal volume. Subgroup analysis revealed that for patients with an ARISCAT score ≥ 43 or moderately impaired lung function(FEV1 50%-70%), higher individualized PEEP(9-13 cmH2O) were associated with a reduced risk of PPCs.
Conclusion: An increase in ARISCAT score, prolonged operation time, advanced age, male gender, smoking history, high intraoperative colloid fluid infusion volume, and high tidal volume are risk factors for PPCs in middle-aged and elderly patients undergoing thoracoscopic lung surgery.
查看全文   查看/发表评论  下载PDF阅读器
关闭