基于XGBoost的COPD患者肺癌发生预测模型的建立与评价  

Establishment and evaluation of the model for predicting lung cancer occurrence in COPD patients based on XGBoost

作　　者：杨靖[1] 焦童 董宇娇 姚晨雨 孔群钰 石婕[2] 杨拴盈[2] YANG Jing;JIAO Tong;DONG Yujiao;YAO Chenyu;KONG Qunyu;SHI Jie;YANG Shuanying(Internet Hospital,The Second Affiliated Hospital of Xi’an Jiaotong University,Xi’an 710004;Department of Respiratory and Critical Care Medicine,The Second Affiliated Hospital of Xi’an Jiaotong University,Xi’an 710004;Chongqing Nanpeng Artificial Intelligence Technology Research Institute Co,Ltd.,Chongqing 401123,China)

机构地区：[1]西安交通大学第二附属医院互联网医院,陕西西安710004 [2]西安交通大学第二附属医院呼吸与危重症医学科,陕西西安710004 [3]重庆市南鹏人工智能科技研究院有限公司,重庆401123

出　　处：《西安交通大学学报(医学版)》2025年第2期345-352,共8页Journal of Xi’an Jiaotong University(Medical Sciences)

基　　金：陕西省重点研发计划资助项目(No.2017SF-173,No.2022JM-509)。

摘　　要：目的 利用慢性阻塞性肺疾病(COPD)患者的临床特征数据构建XGBoost预测模型,并评价预测模型对COPD患者肺癌发生风险早期预测的效能。方法 本研究为回顾性横断面研究,采用整群抽样的方法,对2018年1月1日至2022年12月31日在西安交通大学第二附属医院住院的经临床确诊的COPD患者进行筛选,共收集4 008例有完整数据的患者。首先对各特征基线进行分析,再利用XGBoost构建COPD患者肺癌发生风险预测模型,并利用SHAP(SHapley Additive exPlanation)值对各特征重要性进行量化和归因;决策曲线分析(DCA)曲线评价临床应用价值。结果 使用28个变量构建COPD患者肺癌发生风险模型之后,按照变量重要性排序及临床经验,筛选8个变量,重新构建预测模型,模型效能在训练集和测试集中分别为0.948(0.938,0.958)、0.797(0.738,0.856)。SHAP图显示CEA、CA125、FIB、嗜酸性粒细胞、PLT、D-二聚体升高和TT缩短均会增加COPD患者肺癌发生风险,DCA曲线显示该预测模型具有临床应用价值,可以帮助医师做出更准确的预后预测和治疗决策。结论 基于XGBoost成功建立了预测模型,以特征子集实现了对COPD患者肺癌发生风险的早期预测。Objective To construct an XGBoost predictive model using clinical characteristic data from patients with chronic obstructive pulmonary disease(COPD)and evaluate the efficacy of the predictive model in early risk prediction of lung cancer occurrence in COPD patients.Methods In this retrospective cross-sectional study,cluster sampling was used.We selected clinically diagnosed COPD patients admitted to The Second Affiliated Hospital of Xi’an Jiaotong University from January 1,2018,to December 31,2022.A total of 4008 patients with complete data were included.First,the baseline of each characteristic was analyzed,and then XGBoost was used to construct the lung cancer risk prediction model for COPD patients,and SHAP(SHapley Additive exPlanation)value was used to quantify and attribute the importance of each characteristic.DCA curve was used to evaluate the clinical application value.Results After constructing a lung cancer risk model for COPD patients using 28 variables,eight variables were selected according to the importance of the variables and clinical experience,and the prediction model was reconstructed.The model efficacy in the training set and the test set was 0.948(0.938,0.958)and 0.797(0.738,0.856),respectively.SHAP diagram showed that elevated CEA,CA125,FIB,eosinophils,PLT and D-dimer and reduced TT all contributed to an increased risk of lung cancer in COPD patients.DCA curve showed that the prediction model had clinical application value,which could help doctors make more accurate prognosis prediction and treatment decisions.Conclusion The successful establishment of an XGBoost predictive model,utilizing a subset of features,enables early prediction of lung cancer occurrence in COPD patients.

关 键 词：慢性阻塞性肺疾病(COPD) 危险性评估 预测模型 XGBoost SHAP 

分 类 号：R563.9[医药卫生—呼吸系统]