Logistic回归模型和随机森林模型对AMI患者MACE风险的预测价值  

Predictive value of Logistic regression model and random forest model for risk of MACE in AMI patients

作　　者：尹海宁 张文杰[1] YIN Hai-ning;ZHANG Wen-jie(Department of Nursing,Affiliated Hospital of Jiangsu University,Zhenjiang,Jiangsu,212001,China)

机构地区：[1]江苏大学附属医院护理部,江苏镇江212001

出　　处：《心血管康复医学杂志》2022年第2期131-137,共7页Chinese Journal of Cardiovascular Rehabilitation Medicine

基　　金：江苏省医院协会医院管理创新研究课题(JSYGY-3-2020-427);江苏大学第19批大学生科研立项资助项目(19A451)。

摘　　要：目的:研究Logistic回归模型和随机森林模型对急性心肌梗死(AMI)患者介入术后6个月主要不良心血管事件(MACE)风险的预测价值。方法:于我院治疗的480例AMI患者被随机分为建模组(336例,含Logistic回归模型组和随机森林模型组)和验证组[144例,提供全球急性冠脉事件登记(GRACE)评分]。比较GRACE评分、Logistic回归模型及随机森林模型对AMI患者介入术后6个月MACE风险的预测价值。结果:多因素Logistic回归分析显示,年龄>70岁、女性、吸烟、糖尿病、房颤、总支架数>2、B型利钠肽峰值>400pg/ml、单核细胞/高密度脂蛋白胆固醇比值>0.76是AMI患者介入术后6个月发生MACE的独立危险因素(OR=2.152~4.247,P<0.05或<0.01)。GRACE评分、Logistic回归模型、随机森林模型受检者工作特征曲线下面积分别为0.746、0.859、0.857,灵敏度分别为58.33%、83.33%、66.67%,特异度分别为83.33%、63.33%、92.50%,准确率分别为79.17%、66.67%、88.19%。与GRACE评分比较,Logistic回归模型特异度、准确率均显著降低,随机森林模型特异度、准确率均显著升高(P<0.05或<0.01);与Logistic回归模型比较,随机森林模型特异度、准确率均显著升高(P均=0.001)。结论:随机森林模型对AMI患者MACE风险具有较高预测价值。Objective:To study predictive value of Logistic regression model and random forest model for risk of major adverse cardiovascular events(MACE)in AMI patients on six months after intervention.Methods:A total of 480 AMI patients in our hospital were divided into modeling group(n=336,contain Logistic regression model group and random forest model group)and validation group[n=144,offer Global Registry of Acute Coronary Events(GRACE)score].Predictive value of GRACE score,Logistic regression model and random forest model for risk of MACE in AMI patients on six months after intervention were compared.Results:Multivariate Logistic regression analysis indicated that age>70 years,female,smoking,diabetes mellitus,atrial fibrillation,total stent number>2,peal b-type natriuretic peptide>400pg/ml,monocyte/high-density lipoprotein cholesterol ratio>0.76 were independent risk factors for MACE in AMI patients on six months after intervention(OR=2.152~4.247,P<0.05 or<0.01).Area under ROC curve of GRACE score,Logistic regression model and random forest model was 0.746,0.859 and 0.857 respectively;sensitivity was 58.33%,83.33%and 66.67%respectively;specificity was 83.33%,63.33%and 92.50%respectively;and accuracy was 79.17%,66.67%and 88.19%respectively.Compared with GRACE score,there were significant reductions in specificity and accuracy in Logistic regression model,and significant rise in specificity and accuracy in random forest model(P<0.05 or<0.01);compared with Logistic regression model,there were significant rise in specificity and accuracy in random forest model(P=0.001 all).Conclusion:Random forest model possesses higher predictive value for risk of major adverse cardiovascular events in patients with acute myocardial infarction.

关 键 词：心肌梗死 LOGISTIC模型 预测 

分 类 号：R542.220.9[医药卫生—心血管疾病]