Abstract:Background Correct interpretation of ECG recordings plays a crucial role in cardiology. ECG recordings are performed immediately in order to guide treatment of patients with suspected acute coronary syndrome (ACS). Non-cardiologists usually rely on automated manufacturer-derived ECG interpretation. The main purpose of these methods is to detect lifethreatening ECG patterns with high sensitivity, at cost of specificity. However, as yet, the accuracy of these automatic annotations has not been tested in investigator-initiated (IIT) clinical studies. Purpose To test the accuracy of automated ECG annotation provided by the largest ECG manufacturer in Germany in patients undergoing coronary angiography (CA), because of biomarker-positive suspected ACS. Methods We retrospectively identified patients presenting with suspected ACS to two large tertiary hospitals between 2014 – 2021. Inclusion criteria were defined as indication for CA because of biomarker-positive suspected ACS and availability of ECG raw-data. Patients with ST-elevation myocardial infarction were excluded from this analysis. All ECGs were classified as non-indicative or indicative for ischaemia, according to automatic annotations provided by the ECG-manufacturer. The primary endpoint of the study was type 1 myocardial infarction requiring revascularization. Secondary endpoints were identification of coronary artery disease (CAD) with or without revascularization and intrahospital mortality. The association of ECG-abnormalities with the primary and secondary endpoints was tested using logistic-regression analysis. Confidence intervals were calculated using bootstrapping. Results We screened 9,598 patients for eligibility and finally included 2,073 patients meeting the inclusion und exclusion criteria. The median age was 73 years (IQR 61 – 80), 661 (32%) of the patients were females. Among those patients 1,299 (63%) were diagnosed with type 1 myocardial infarction requiring revascularization, 475 patients (23%) didn't show signs of CAD and 153 patients (7%) died during the index hospital stay. Automatic annotation of ECG-abnormalities was significantly associated with the primary endpoint (OR 1.36; 95% CI 1.14 – 1.63; p < 0.001), and secondary endpoints (OR 1.72; 95% CI 1.39 – 2.13; p < 0.001 for CAD and 1.61; 95% CI 1.16-2.24; p = 0.005 for mortality). The overall sensitivity, specificity and negative predictive value of the method was 47% (45-50%), 60% (57-63%), and 41% (38-43%), respectively. The sensitivity and specificity for different subgroups is illustrated in Figure 1. Conclusion Current manufacturer algorithms for automatic ECG-annotation show a statistically significant association with CAD requiring revascularization. However, 59% of the patients requiring revascularization are not identified with the annotated ECG-abnormalities. Application of artificial intelligence might improve these algorithms and allow safe application of the method in clinical practice.Figure 1

The diagnostic accuracy of a shock advisory algorithm in automated external defibrillators in the presence of real-world artifacts

Clinical record analysis of 54 cases with automatic external defibrillator in public of mainland china

Bench study of the accuracy of a commercial AED arrhythmia analysis algorithm in the presence of electromagnetic interferences

Clinical performance of AED shock advisory system with integrated Analyze Whilst Compressing algorithm for analysis of the ECG rhythm during out-of-hospital cardiopulmonary resuscitation: A secondary analysis of the DEFI 2022 study

A Shock Advisory Algorithm Based on Hybrid Network for ECG Disturbed by Cardiopulmonary Resuscitation

Enhancing the accuracy of shock advisory algorithms in automated external defibrillators during ongoing cardiopulmonary resuscitation using a cascade of CNNEDs

Artificial intelligence for predicting shockable rhythm during cardiopulmonary resuscitation: In-hospital setting

Testing the accuracy of automatic ECG-annotation for suspected acute coronary syndrome

Added value of new acute coronary syndrome computer algorithm for interpretation of prehospital electrocardiograms

A Novel Algorithm for Improving the Prehospital Diagnostic Accuracy of ST-Segment Elevation Myocardial Infarction

A new ECG-based Automated External Defibrillator system

The General Public's Ability to Operate Automated External Defibrillator: A Controlled Simulation Study.

A Comparison of 5 Algorithmic Methods and Machine Learning Pattern Recognition for Artifact Detection in Electronic Records of 5 Different Vital Signs: A Retrospective Analysis

Robust Artificial Intelligence Tool for Atrial Fibrillation Diagnosis: Novel Development Approach Incorporating Both Atrial Electrograms and Surface ECG and Evaluation by Head‐to‐Head Comparison with Hospital‐Based Physician ECG Readers

Convolution Neural Network Algorithm for Shockable Arrhythmia Classification Within a Digitally Connected Automated External Defibrillator

An artificial intelligence–enabled ECG algorithm for comprehensive ECG interpretation: Can it pass the ‘Turing test’?

Diagnostic accuracy of apple watch ECG outputs in identifying dysrhythmias: A comparison with 12-Lead ECG in emergency department

Improving prehospital risk stratification using deep learning ECG interpretation in suspected acute coronary syndrome

Deep Learning Strategy for Sliding ECG Analysis during Cardiopulmonary Resuscitation: Influence of the Hands-Off Time on Accuracy

The impact on users of an unfamiliar AED following a recent training experience: A randomized cross over simulation study

Prediction of Shock-Refractory Ventricular Fibrillation During Resuscitation of Out-of-Hospital Cardiac Arrest