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

Abstract Introduction Most patients presenting with chest pain in the Emergency Medical Services (EMS) setting are suspected of non-ST-elevation acute coronary syndrome (NSTE-ACS). Distinguishing true NSTE-ACS from non-cardiac chest pain based solely on the electrocardiogram (ECG) is challenging. Recent advances in artificial intelligence (AI) models for ECG interpretation show promising results. Purpose The aim of this study is to develop and validate a Convolutional Neural Network (CNN)-based model to diagnose NSTE-ACS within the prehospital ECG and to compare it to current prehospital available diagnostic tools. Methods For this study, a training/internal validation cohort (n= 4891) and external validation cohort (n= 754) were used, both consisting of suspected NSTE-ACS patients, and the first prehospital standard 12-lead ECG at presentation was extracted. A CNN (ECG-AI) was trained and validated to detect NSTE-ACS. The training/internal validation cohort was extracted from suspected NSTE-ACS patients presenting to the emergency department. The external validation cohort was extracted from a prehospital, prospective cohort study in which subjects were assessed by EMS paramedics. A validated prehospital clinical risk score (preHEART) was calculated, which includes manual on-site ECG analyses by EMS paramedic (ECG-EMS) and a point-of-care (POC) troponin assessment, and compared to ECG-AI in diagnosing NSTE-ACS. Results In the external validation cohort, 202 (27%) subjects were diagnosed with NSTE-ACS. ECG-AI had a better diagnostic performance than ECG-EMS (AUC 0.70 (0.66–0.74) vs. AUC 0.65 (0.61–0.70), p=0.045) for diagnosing NSTE-ACS. The overall diagnostic accuracy of preHEART was AUC 0.78 (0.74–0.82) and superior compared to ECG-AI (p= 0.001). Incorporating ECG-AI instead of ECG-EMS into preHEART led to a significant improvement in diagnostic performance (AUC 0.83 (0.79–0.86,p < 0.001)). Conclusion Integrating AI in prehospital ECG interpretation improves NSTE-ACS diagnosis. However, clinical risk scores still yield the best diagnostic performance in the prehospital setting and their accuracy could be further enhanced through AI.ROC prehospital diagnostic toolsROC preheart and preheart-AI