基于LASSO回归对冠心病相关血脂指标的筛选  

Screening of lipid parameters in coronary artery disease based on LASSO regression

作　　者：张韶辉[1] 苏强[2] 赵永亮[3] 卓军[4] 刘立新[1] 杨国良[1] 陈雪英 戴雯[1] Zhang Shaohui;Su Qiang;Zhao Yongliang;Zhuo Jun;Liu Lixin;Yang Guoliang;Chen Xueying;Dai Wen(Department of Cardiology,Affiliated Hospital of Jining Medical University,Jining 272029,China;Health Management Center,Affiliated Hospital of Jining Medical University,Jining 272029,China;Department of Cardiosurgery,Affiliated Hospital of Jining Medical University,Jining 272029,China;Interventional Radiography,Affiliated Hospital of Jining Medical University,Jining 272029,China)

机构地区：[1]济宁医学院附属医院心内科,272029 [2]济宁医学院附属医院健康管理中心,272029 [3]济宁医学院附属医院心外科,272029 [4]济宁医学院附属医院介入放射科,272029

出　　处：《中国综合临床》2021年第2期148-153,共6页Clinical Medicine of China

基　　金：山东省高等学校科技计划项目(J16LL52);济宁市医药卫生科技项目(济科字[2015]57号-17、济科字[2016]56号-33)。

摘　　要：目的利用LASSO回归分析筛选出与冠心病密切相关的血脂指标。方法选取2013年5月至2015年11月在济宁医学院附属医院心内科住院并诊断为冠心病的患者3062例的临床资料进行回顾性分析。按照冠状动脉造影结果分为冠心病组(n=2427)和对照组(n=635)。统计分析用R语言。建立冠心病相关血脂指标的多元逻辑回归模型,评估模型多重共线性的严重程度。利用LASSO回归筛选出冠心病预测模型中具有代表性的血脂指标。结果研究对象共入选患者3062例,其中冠心病组2427例,对照组635例。将血脂指标同时纳入多元逻辑回归模型后导致了模型较严重的共线性,逐步回归(stepwise)只能够在部分地减少共线性的严重程度,而LASSO回归模型显著减少了共线性的严重程度。经过LASSO回归分析,发现低密度脂蛋白胆固醇(low density lipoprotein-cholesterol,LDL-C)、高密度脂蛋白胆固醇(high density lipoprotein-cholesterol,HDL-C)和非高密度脂蛋白胆固醇(non-high density lipoprotein-cholesterol,non-HDL-C)是预测冠心病的代表性的血脂指标。结论LASSO回归在处理多重共线性的样本数据时有优势。LASSO回归发现LDL-C、HDL-C和non-HDL-C是预测冠心病的代表性的血脂指标。Objective Using lasso regression analysis to screen out the blood lipid indexes closely related to coronary heart disease Methods The clinical data of 3062 patients with coronary heart disease who were hospitalized in the Department of Cardiology,Affiliated Hospital of Jining Medical College from May 2013 to November 2015 were retrospectively analyzed.They were divided into control group(n=2427)and coronary angiography group(n=635).R language was used for statistical analysis.Multiple logistic regression models were established for indicators of blood lipid related to CAD,and their multicollinearity severity was assessed.LASSO regression was used to screen out the representative lipid parameters in the CAD prediction model.Results A total of 3062 patients were enrolled,including 2427 patients in coronary heart disease group and 635 patients in control group.The inclusion of lipid parameters into multiple logistic regression model leads to serious multicollinearity.Stepwise regression can only partially reduce multicollinearity severity,while LASSO regression model significantly reduces multicollinearity severity.Low density lipoprotein cholesterol(LDL-C),high density lipoprotein cholesterol(HDL-C)and non-high density lipoprotein cholesterol(non-HDL-C)were found to be the representative lipid indexes for predicting coronary heart disease by LASSO regression analysis.Conclusion LASSO regression has advantages in processing multicollinearity data.LASSO regression showed that LDL-C,HDL-C and non-HDL-C were representative lipid indicators for predicting coronary heart disease..

关 键 词：冠心病 血脂指标 R语言 多重共线性 LASSO回归 

分 类 号：R541.4[医药卫生—心血管疾病]