Supervised machine learning for pre-hospital electrocardiographic assessment

Cardiovascular diseases rank among the leading causes of mortality globally. Acute Coronary Syndrome (ACS) is a condition primarily characterized by reduced blood flow to the heart. The management and symptoms of ACS closely resemble those of acute myocardial ischemia (AMI), making early recognition vital for preserving cardiac function. The electrocardiogram (ECG) is a powerful tool for diagnosis of many heart-related disorders. Its characteristic signal holds abundant information regarding electrical functionality of the heart, some of which are evident just by visual inspection, but others are hidden. The ECG enables the identification of specific waveforms, segments, and intervals that vary according to distinct pathologies. Recent advancements in machine learning (ML) algorithms have significantly enhanced the speed and accuracy of ECG interpretation. In particular, the combination of serial electrocardiography and machine learning has shown great promise in the early detection of cardiac diseases. By comparing an acute ECG recorded during the pre-hospital phase, such as one taken in an ambulance (AECG), with a previously acquired reference ECG (RECG), the triage process for patients presenting symptoms related to AMI and ACS can be expedited. The current study aims to investigate the potential of Supervised ML classifiers in detecting ACS and AMI starting from ECG trace characteristics. The dataset comprises 1,425 non-traumatic chest pain patients, each with an AECG and an earlier RECG recorded under stable conditions. From the AECG, the first feature set, consisting of 18 direct measurements, is extracted. The second feature set contains in addition also the 28 serial features derived from the differences between each ECG pair. The data, divided into training (70%), validation (10%), and testing (20%) subsets, are then used as inputs for several binary Supervised ML classifiers: Linear Discriminant Analysis (LDA), Decision Tree (DT), Support Vector Machine (SVM), Artificial Neural Networks (ANN), Logistic Regression (LR), Naïve Bayes (NB), and K-Nearest Neighbour (KNN) algorithms. The performance of these classifiers is evaluated using sensitivity, specificity and the area under the curve (AUC) of the receiver operating characteristic curve. For the identification of ACS, the linear SVM achieved an AUC of 78%, closely followed by LDA at 77% and KNN at 75%. In contrast, for AMI detection, LDA emerged as the most effective model, reaching an AUC of 86% on the testing dataset, with LR and KNN also demonstrating powerful performance at 85% and 80%, respectively. Using the first feature set or the second one led to a marginal improvement in performance, indicating that the AECG contains all essential information required for optimal screening. The findings underscore the efficacy of Supervised ML classifiers in detecting cardiac diseases from acute pre-hospital ECG readings. These innovative tools can assist clinicians in the timely identification of hearth pathologies, optimizing the triage system and improving patient outcomes.