Abstract:

Objective To construct and verify a risk prediction model of emergence agitation in patients undergoing thoracoscopic radical resection of lung cancer，and to screen the optimal model by using machine learning algorithm，so as to provide references for clinical formulation of a nursing risk management plan. Methods The convenience sampling method was used to retrospectively select 476 patients who underwent thoracoscopic radical resection of lung cancer in a tertiary hospital in Hangzhou，Zhejiang Province from January to December 2023 as a construction group. Logistic regression，decision tree，random forest and naive Bayesian model were constructed by SPSS 29.0 and R 4.3.0 software. The prediction performance of each model was compared by accuracy，precision，recall，F1 score and area under the receiver operating characteristic curve，and the optimal model was screened. From January to June 2024，204 patients in the unit were prospectively selected as the research subjects of an external validation group. The discrimination and calibration of the optimal model were evaluated by AUC value and calibration curve. Results A total of 680 patients completed the survey. All 4 models showed that multimodal analgesia，thoracic drainage tube type，pain score，tracheal intubation type，state anxiety and catheter indwelling time were the influencing factors of emergence agitation in patients undergoing thoracoscopic radical resection of lung cancer（P<0.05）. The 4 risk prediction models showed that the random forest prediction model had the best comprehensive performance. The external verification results showed that the AUC value was 0.913，and the calibration curve fitted well with the 45° ideal line. Conclusion Among the 4 risk prediction models，the random forest prediction model has the best performance，which is more suitable for the assessment of the risk of emergence agitation in patients undergoing thoracoscopic radical resection of lung cancer，and has good generalization and clinical application value.

Key words: Thoracoscopic Radical Resection of Lung Cancer, Emergence Agitation, Machine Learning, Risk Factors, Predictive Model, Nursing Care

摘要：

目的 构建并验证胸腔镜肺癌根治术患者苏醒期躁动风险预测模型，利用机器学习算法筛选最优模型，为临床制订护理风险管理方案提供借鉴。方法 采用便利抽样法，回顾性选取2023年1—12月在杭州市某三级甲等医院行胸腔镜肺癌根治术的476例患者作为构建组研究对象，运用SPSS 29.0和R 4.3.0软件构建Logistic回归、决策树、随机森林和朴素贝叶斯模型，通过准确率、精确率、召回率、F1分数及受试者操作特征曲线下面积比较各模型预测性能，筛选最优模型。前瞻性选取2024年1—6月同一单位行胸腔镜肺癌根治术的204例患者作为外部验证组研究对象，通过受试者操作特征曲线下面积和校准曲线评估最优模型的区分度和校准度。结果 最终680例患者完成调查。4种模型均显示，是否使用多模式镇痛、胸腔引流管型号、疼痛评分、气管插管型号、是否焦虑和导尿管留置时机是胸腔镜肺癌根治术患者苏醒期躁动的影响因素（P<0.05）。内部验证结果显示，随机森林模型综合效能最佳，外部验证结果显示，受试者操作特征曲线下面积为0.913，校准曲线与45°理想线贴合较好。结论 4种预测模型中随机森林模型性能最优，更适用于胸腔镜肺癌根治术患者苏醒期躁动风险评估，具有良好的泛化性和临床应用价值。

关键词: 胸腔镜肺癌根治术, 苏醒期躁动, 机器学习, 风险因素, 预测模型, 护理