AI -ECG Results in Faster and More Accurate Identification of STEMI

By Christina Lalani on

Key Points:

    • The use of an artificial intelligence electrocardiogram (AI-ECG) support tool in the evaluation of patient ECGs was associated with a reduction in the door to coronary angiography time (43.3 minutes vs. 52.3 minutes) in patients who presented with an ECG concerning for ST-elevation myocardial infarction (STEMI).
    • The AI-ECG support tool has a high positive predictive value of 88.0 (81.8-94.1) and high negative predictive value of 99.9 (99.9-100.0).
    • AI technology helped to reduce the treatment waiting time for patients with STEMI from about 52 minutes to 43 minutes.

Prior studies have shown that acute ST-elevation myocardial infarction (STEMI) can be misdiagnosed on electrocardiogram (ECG) in more than 20% of cases when evaluated by emergency medicine physicians, general cardiologists, and interventional cardiologists. The rapid diagnosis of a STEMI is critical to minimize patients’ door-to-balloon times and improve patient outcomes. In a late-breaking presentation at the 2023 American Heart Association conference, Dr. Chin-Sheng Lin outlined the results of the Artificial Intelligence Enabled Rapid Identification of ST-Elevation Myocardial Infarction (ARISE) trial. 

The ARISE Trial was a randomized control trial based at Tri-Service General Hospital (Taipei City, Taiwan) that was designed to evaluate the impact of an artificial intelligence electrocardiogram (AI-ECG) support tool on the outcomes of patients who presented to the emergency department or inpatient department and had an ECG vs routine care. The key question evaluated in this study was whether the use of an AI-ECG support tool would improve the timeliness of receiving a coronary angiogram in patients who presented to the hospital with a ST-elevation myocardial infarction. The initial enrollment for the trial included 43,994 patients who had visited the emergency department or an inpatient department and received at least one ECG without a history of coronary angiogram within the preceding three days. Patients who were less than 18 years old were excluded. All patients were randomized to receive care that was supported by the AI-ECG tool or usual care. The providers caring for patients in the intervention group received real-time AI-ECG analysis at the time of the patient’s ECG. For this reason, patients who were randomized to the intervention group and had an ECG completed prior to the activation of the AI-ECG support tool were also excluded.

The primary outcome in this study was the time from ECG to coronary angiography. Patient characteristics were evaluated in the population overall as well as in the subset of patients who underwent a coronary angiogram for concern for STEMI. There were no statistically significant differences between patients in the control arm and intervention arm in terms of age and rates of co-morbidities such as coronary artery disease, diabetes, hypertension, hyperlipidemia, and chronic kidney disease. However, the patients who underwent a coronary angiogram due to concern for a STEMI were more likely to be male and older and less likely to have been evaluated as an inpatient. 

There was a statistically significant reduction in the primary outcome, time from ECG to coronary angiography, among patients in the intervention arm (43.3 minutes) compared to patients in the control arm (52.3 minutes). When patients were stratified by whether the AI-ECG support tool read the ECG as “potential STEMI” vs “potential not-STEMI”, only patients who had an ECG that was read as potential STEMI by the AI-ECG tool had a statistically significant reduction in their time to coronary angiography. Interestingly, the improvement in time to coronary angiogram seen with the AI-ECG support tool was only significant during regular working hours. In the secondary analysis, there were fewer STEMIs that were not validated on coronary angiography in the AI-ECG arm (6.5%) compared to the control arm (15.8%). Finally, in an analysis of accuracy, the positive predictive value of the AI-ECG tool was found to be 88.0 (81.8-94.1) with a negative predictive value of 99.9 (99.9-100.0), sensitivity of 88.8 (82.8-94.8) and specificity of 99.9 (99.9-100.0).  

In conclusion, the authors found that the use of the AI-ECG tool resulted in a faster time to coronary angiography with a lower rate of ECG-identified STEMI’s that were not validated with coronary angiography, suggesting greater efficiency and accuracy in the identification of true ST-elevation myocardial infarctions. The authors propose that future studies could be multi-center, larger scale and evaluate pre-hospital EMS application of the AI-ECG tool.