Cardiovascular Medicine Department, Beijing Bo'ai Hospital, China Rehabilitation Research Center, Capital Medical University, Beijing 100068, China;Center for Coronary Artery Disease, Beijing Anzhen Hospital, Capital Medical University, Beijing 100029, China.
Respiratory Medicine Department, Beijing Friendship Hospital Affiliated of Capital Medical University, Beijing 100050, China.
Biomed Environ Sci. 2023 Jul 20;36(7):625-634. doi: 10.3967/bes2023.089.
We aimed to assess the feasibility and superiority of machine learning (ML) methods to predict the risk of Major Adverse Cardiovascular Events (MACEs) in chest pain patients with NSTE-ACS.
Enrolled chest pain patients were from two centers, Beijing Anzhen Emergency Chest Pain Center Beijing Bo'ai Hospital, China Rehabilitation Research Center. Five classifiers were used to develop ML models. Accuracy, Precision, Recall, F-Measure and AUC were used to assess the model performance and prediction effect compared with HEART risk scoring system. Ultimately, ML model constructed by Naïve Bayes was employed to predict the occurrence of MACEs.
According to learning metrics, ML models constructed by different classifiers were superior over HEART (History, ECG, Age, Risk factors, & Troponin) scoring system when predicting acute myocardial infarction (AMI) and all-cause death. However, according to ROC curves and AUC, ML model constructed by different classifiers performed better than HEART scoring system only in prediction for AMI. Among the five ML algorithms, Linear support vector machine (SVC), Naïve Bayes and Logistic regression classifiers stood out with all Accuracy, Precision, Recall and F-Measure from 0.8 to 1.0 for predicting any event, AMI, revascularization and all-cause death ( HEART ≤ 0.78), with AUC from 0.88 to 0.98 for predicting any event, AMI and revascularization ( HEART ≤ 0.85). ML model developed by Naïve Bayes predicted that suspected acute coronary syndrome (ACS), abnormal electrocardiogram (ECG), elevated hs-cTn I, sex and smoking were risk factors of MACEs.
Compared with HEART risk scoring system, the superiority of ML method was demonstrated when employing Linear SVC classifier, Naïve Bayes and Logistic. ML method could be a promising method to predict MACEs in chest pain patients with NSTE-ACS.
我们旨在评估机器学习（ML）方法预测非 ST 段抬高型急性冠脉综合征（NSTE-ACS）胸痛患者发生主要不良心血管事件（MACE）的可行性和优越性。
本研究纳入了来自北京安贞医院胸痛中心和中国康复研究中心的胸痛患者。使用 5 种分类器开发了 ML 模型。采用准确性、精密度、召回率、F1 度量和 AUC 评估模型性能，并与 HEART 风险评分系统进行比较，预测 MACE 的发生。最终，采用朴素贝叶斯（Naive Bayes）构建的 ML 模型预测 MACE 的发生。
根据学习指标，当预测急性心肌梗死（AMI）和全因死亡时，不同分类器构建的 ML 模型优于 HEART（病史、心电图、年龄、危险因素和肌钙蛋白）评分系统。然而，根据 ROC 曲线和 AUC，只有在预测 AMI 时，不同分类器构建的 ML 模型才优于 HEART 评分系统。在 5 种 ML 算法中，线性支持向量机（Linear SVC）、朴素贝叶斯和逻辑回归分类器表现出色，预测任何事件、AMI、血运重建和全因死亡的准确性、精密度、召回率和 F1 度量均在 0.8 到 1.0 之间（HEART≤0.78），预测任何事件、AMI 和血运重建的 AUC 均在 0.88 到 0.98 之间（HEART≤0.85）。基于朴素贝叶斯构建的 ML 模型预测，疑似急性冠脉综合征（ACS）、心电图异常、高 hs-cTnI、性别和吸烟是 MACE 的危险因素。
与 HEART 风险评分系统相比，当采用线性 SVC 分类器、朴素贝叶斯和逻辑回归时，ML 方法的优越性得到了证明。ML 方法可能是预测 NSTE-ACS 胸痛患者 MACE 的一种很有前途的方法。
