预测电视辅助胸腔镜肺叶切除术后肺部并发症的机器学习模型开发与验证

打开文本图片集

中图分类号 R734.2 TP181 文献标识码A 文章编号 2096-7721（2026）03-0536-07

Development and validation of a machine learning model for predicting pulmonary complications after video-assisted thoracoscopic lobectomy

JIANG Zhichao， MA Changyun， GENG Zhongwei， ZHANG Jianping， TANG Zicheng， MAO Kunpeng （DepartmentofCardiothoracic Surgery，Maanshan General HospitalofRanger-DureeHealthcare，Maanshan2430oo，China）

AbstractObjective：Todevelopandvalidateamachinelearning modelfor predicting postoperative pulmonarycomplications （C） aftervideo-aistedoracosopicrgery（A）ngacedtstabishplid，licallyoveentsst tol.Metods：195lungcancerpatientswhoweretreatedatourhospitalfomJanuary2O23toAugust2O25wereenolled.Theywere temporallydividedinto the trainingset（ n=147 ，January2023toDecember2024，）and thevalidationset（ n =48，January2025toAugust 2025）.Featureswereselectedusingtheleastabsoluteshrinkageandselectionoperator（LASSO）.Logisticregresion （LR）adom forest （RF），extremegradientbosting （XGBoost）and light gradientboosting machine（LightGBM）modelswere trainedand tuned viacross-validation.Modelperformancewasevaluatedusingtheareaunderthereceiveroperating characteristiccurve（AUC），Brier score，calibrationslope，anddecisionurveanalysis（DCA）.Tememetricswererecalulatedontheetealvaliationsetndisk stratificationwasperformed.Results：ThePPCincidencewasrelativelyhighandcorrelated withsurgicalburden.LASSOidentified sevennon-zroabsagrativeratioaomobidex （CC），ageIIiusinapacityffor carbon monoxide（DLCO），smokingandredbloodcelldstributionwidth （DW）.XGBoostpeforedtheest（AUC=0.861，ri0.145， calibratinslopO18），utpeforingseinemodelssuchasLR.Iteetealvaliatioset，XGBostmaitaiedtestUC （0.898），withaBriersoeofO.53ndcrationslopeofO979.CAidicatedmaiuneteefitatccaltresoldprbbli （pt） between O.10 and O.25. Risk stratification using XGBoost-predicted probabilities （ <10% ， 10% -<20%， ≥20% ）yielded observed PPC rates of 11.11% 17.65% ，and 53.85% ，respectively.Bothlength of hospital stay and ICU admissonrate increased significantly with higher riskstrata.Concusi：TeXBotodelostuctedfroseencliallcssblevarableshasgooddisciiationaation andclinicalnetbnefitinbothtetrainngsendtheextealtemporalvalidationset，makingitsuitableforearliskasestand stratified management of PPC in patients undergoing VATS for lung cancer.

Key WordsLung Cancer; Thoracoscopy; Postoperative Pulmonary Complications; Machine Learning

肺癌是我国发病率和死亡率最高的恶性肿瘤之一，电视辅助胸腔镜肺叶切除术是早中期患者的主要治疗方法[-2]。（剩余11514字）