Detection of acute coronary occlusion in bradycardic patients using ECG: comparison of STEMI criteria versus artificial intelligence–enhanced interpretation

Background

The detection of acute coronary occlusion (ACO) in patients presenting with chest pain remains challenging when relying solely on conventional ST-elevation myocardial infarction (STEMI) criteria on the electrocardiogram (ECG). Artificial intelligence (AI)-enhanced ECG analysis may improve diagnostic accuracy, particularly in bradycardic patients, in whom ST-segment alterations may reflect either a demand–supply mismatch or true coronary occlusion. Accurate differentiation is crucial to appropriately guide emergency treatment.

Purpose

To evaluate the diagnostic performance of an AI-based ECG model for the detection of ACO in bradycardic patients presenting with chest pain on admission ECG, in comparison with conventional STEMI criteria, within a Chest Pain Unit (CPU) setting.

Methods

All consecutive bradycardic patients (<60 bpm) presenting with chest pain to the CPU (Nov 2018 to Jun 2021) were included. Clinical management followed current European Society of Cardiology guidelines. The endpoint was ACO, adjudicated by an independent physician blinded to the AI predictions, using coronary angiography, biomarker kinetics, and echocardiography. Diagnostic performance was assessed by receiver operating characteristic (ROC) analysis and compared with STEMI criteria.

Results

Among 4,440 patients treated in the CPU, 485 presented with bradycardia and were included in the analysis (mean age 57.7 ± 18.4 years; 71.8% male; mean heart rate 53.9 ± 4.8 bpm). Observed rhythms were sinus bradycardia (92.6%), atrial fibrillation (2.3%), idioventricular rhythm (0.4%), and other rhythms (4.7%). The prevalence of ACO was 2.3% (n = 11). Diagnostic metrics are illustrated in the accompanying figure. The area under the curve was 0.968 (95% CI 0.944 – 0.991) for the AI-ECG model and 0.607 (95% CI 0.411 – 0.792) for STEMI criteria.

Conventional STEMI criteria identified 3 cases of ACO (27.3%), whereas the AI-ECG model detected 5 cases (45.5%). STEMI criteria yielded 33 false-positive findings (6.8%), compared with 6 (1.2%) using the AI-ECG model. Both approaches demonstrated high specificity.

In the low-risk subgroup (“rule-out” according to the troponin-based algorithm), the AI-ECG model produced only one false-positive result and, compared with strict application of STEMI criteria, could potentially have avoided fifteen emergency coronary angiographies.

Conclusion

In bradycardic patients presenting with acute chest pain, a purely ECG-based strategy is insufficient for the reliable detection of ACO. Neither approach achieved sufficient sensitivity to safely exclude ACO; however, the AI-ECG model substantially reduced false-positive findings compared with conventional STEMI criteria. The low prevalence of ACO in the analyzed cohort warrants further prospective validation of the investigated AI-ECG model.Figure