基于随机森林模型法的AMI患者并发AKI预测模型的建立

Establishment of prediction model in patients with AMI complicating AKI based on random forest model method

目的建立基于随机森林模型法的预测模型对急性心肌梗死(AMI)患者并发急性肾损伤(AKI)进行预测,找出相关重要指标。方法选取2014年1月至2021年1月该院急诊科收治的AMI患者1362例作为研究对象,将合并AKI患者设为观察组(270例),未合并AKI设为对照组(1092例)。在确定30个变量后进行数据的相关统计和分析,随机选取75%的病例进行训练数据库的建立,25%的病例作为测试数据库,采用R语言进行数据的筛选和模型的建立,对其进行相关评估,并与其余3种机器学习模型进行对比。结果1362例患者中合并AKI 270例(19.82%)。两组患者除血小板、球蛋白、入院时体温、血钠、天门冬氨酸氨基转移酶、丙氨酸氨基转移酶比较差异无统计学意义(P>0.05)外;其余各指标比较,差异均有统计学意义(P<0.05);随机森林模型受试者工作曲线下面积为0.894,均高于其余3种模型,灵敏度为0.792,特异度为0.867;模型中变量重要性依次为首次肌酐、尿素值,机械通气、年龄和D-二聚体。结论基于随机森林模型对AMI患者是否发生AKI进行预测具有较好的预测效能,在实际临床工作中具有一定参考价值。

Objective To establish a prediction model based on the random forest model method to predict the patients with acute myocardial infarction(AMI)complicating acute kidney injury(AKI)for finding relevant important indicators.Methods A total of 1362 patients with AMI admitted and treated in the emergency department of this hospital from January 2014 to January 2021 were selected as the research subjects.The patients with AMI complicating AKI served as the observation group(270 cases)and those without complicating AKI as the control group(1092 cases).After determining 30 variables,the relevant statistics and analysis of the data were performed,and 75%of the cases were randomly selected to conduct the training database establishment,25%of the cases as the test database,the R language was used to conduct the data screening and model establishment,the related evaluation on them was conducted,then which was compared with the other three kinds of machine learning models.Results Among 1362 cases,there were 270 cases(19.82%)complicating AKI.The comparison of platelets,globulin,admission body temperature,blood sodium,glutamic oxalacetic transaminase and alanine aminotransferase between the two groups showed no statistical difference(P>0.05).The other indicators showed statistically significant differences between the two groups(P<0.05);the area under the receiver operating curve of the random forest model was 0.894,which was higher than those in the other three models,the sensitivity was 0.792 and the specificity was 0.867;the importance of variables in the model was as follows:first time creatinine,urea value,mechanical ventilation,age and D-dimer.Conclusion Based on the random forest model,predicting the occurrence of AKI in the patients with AMI has good predictive power,and has a certain reference value in actual clinical work.