Artificial intelligence–enabled ECG-identification of occlusive myocardial infarction in non-ST-elevation acute coronary syndrome and association with infarct size and survival
M Unterhuber, L Donazzan, F Vitali, M Bonatti, A Cantone, F Nani, P Engl, A Ruzzarin, L Malfatti, D Bernardi, G Campo, R OberhollenzerAbstract
Background
Rapid identification of occlusive myocardial infarction (OMI) remains challenging in acute coronary syndrome (ACS) without ST-elevation. Artificial intelligence (AI)-assisted electrocardiography (AI-ECG) may enable early recognition before routine workup and allow access to timely reperfusion.
Methods
In this two-centre observational study, patients with suspected ACS were enrolled and managed according to current guidelines and underwent standard ACS pathway. ST elevation myocardial infarction (STEMI) was excluded. Serial high-sensitivity troponin T measurements were collected. If needed, the patients underwent coronary angiography (CAG) and OMI was defined as acute culprit lesion with reduced TIMI flow or apparently normal flow with >1000ng/l peak troponin. ECGs were analysed in a blinded core-lab using a commercially available, CE-certified AI-ECG algorithm to classify patients as AI-OMI or AI-NOMI (No signs of Occlusive Myocardial Infarction). Cardiovascular death was evaluated at twelve months using Kaplan–Meier analyses.
Results
Among 1,619 patients (mean age 69±17 years, 41% females) with suspected NSTE-ACS, 412 (25%) had confirmed NSTEMI and underwent CAG. Of those, 155 (37.6%) were classified as AI-OMI. AI-ECG reached a sensitivity of 83%, specificity of 99%, a positive predictive value of 98% and a negative predictive value of 89% for OMI detection. Baseline troponin levels were comparable among groups (61 [7; 205] vs 59 [8; 218] ng/L, p=0.8) with significantly higher peak troponin in AI-OMI patients (539 [122–1868] vs 219 [42–553] ng/L, p < 0.001). At 12 months of follow-up, 7 (2.7%) AI-NOMI and 12 (7.7%) AI-OMI patients died (log-rank p=0.019). After adjustment for GRACE score >140, AI-OMI was independently associated with increased 12-month mortality (HR 2.45; 95% CI 1.01–6.40; p=0.048).
Conclusions
Among NSTE-ACS, the prevalence of OMI is high and remains often clinically unrecognized. In this study, AI-ECG accurately identified OMI in a NSTE-ACS cohort. Compared to AI-NOMI, patients with AI-OMI are burdened with higher peak troponin values and with a higher 12-month cardiovascular mortality independently of GRACE score and may redefine NSTE-ACS early triage at first contact.Abstract illustration