随机森林模型预测急性心肌梗死后急性肾损伤

目的:建立随机森林模型预测急性心肌梗死（acute myocardial infarction，AMI）患者并发急性肾损伤（acute kidney injury, AKI）。方法:使用温州医科大学附属东阳医院大数据平台，筛选出1 363例患AMI的病例，确定30个变量后，统计分析样本临床特点，将样本划分为75%的训练集建立随机森林模型，以及25%的测试集进行验证，使用R语言进行数据的筛选及模型的建立。最后根据特异性、敏感性、准确性、受试者特征工作特征曲线(relative operating characteristic curve, ROC曲线)等来评估模型性能，同时与其他三种常用的机器学习算法（神经网络，朴素贝叶斯，支持向量机）的模型性能进行比较。结果:AMI合并AKI的患者的人口学信息、心血管疾病的危险因素、入院时的生命体征、实验室检查等与未合并急性肾损伤的患者存在差异性。模型评估后得出测试集的ROC曲线下面积为0.893，特异度为0.791，灵敏度为0.866，其中入院首次肌酐、首次尿素、D-二聚体、年龄、机械通气是其最重要的影响因素。在本研究中，多种机器学习算法比较后，随机森林模型较有优势。结论:建立的随机森林模型具有帮助预测AMI并发AKI的潜力。

A random forest model to predict acute kidney injury after acute myocardial infarction

Objective:Our study aims to predict acute kidney injury (AKI) in acute myocardial infarction (AMI) by establishing a random forest model.Methods:By using the clinical database from affiliated Dongyang Hospital of Wenzhou Medical University, a total of 1 363 AMI cases were included. Then, 75% of participants were analyzed as training subsets and the remaining 25% were testing subsets. The CARET package in R was used to filter variables and build random forest. The prediction ability of established model was evaluated by specificity, sensitivity, accuracy, relative operating characteristic curve (ROC curve) in testing subsets. In addition, the performance of random forest was compared with other 3 commonly used machine learning algorithms (Artificial Neural Network, Naive Bayes, and Support Vector Machine).Results:In this study, 30 variables including the demographic information, risk factors of cardiovascular disease, vital signs at admission, laboratory tests were identified and used to establish our random forest prediction model. The area under the curve of the testing subsets ROC was 0.893. The specificity and sensitivity of prediction model was 0.791 and 0.866, respectively. And the first creatinine, first blood urea nitrogen, and D-dimer after admission, age, mechanical ventilation were the top-five factors in this model. After comparing various machine learning algorithms, random forest model had a better performance.Conclusion:The random forest model would be used to predict the occurrence of AMI with AKI.