Sex-specific validation of an artificial intelligence–enabled electrocardiogram for early triage of acute myocardial infarction
H S Lee, S Kang, J M Kwon, J M Son, K S Kim, M S Lee, K KimAbstract
Background
Timely recognition of acute myocardial infarction (AMI) remains challenging, particularly in women who more often present with atypical symptoms and less overt electrocardiographic findings. Artificial intelligence–enabled electrocardiogram (AI-ECG) analysis offers automated risk estimation based solely on raw waveform data, but prospective evidence supporting sex-consistent diagnostic performance and safe rule-out capability is limited.
Purpose
To assess sex-stratified discrimination, rule-out safety, and phenotype robustness of an AI-ECG model for AMI in a prospective, multicenter emergency department cohort.
Methods
In a prospective validation study conducted across 18 emergency departments, consecutive adults presenting with suspected AMI were enrolled. AI-ECG analyzed the initial 12-lead ECG and generated a continuous probability score using prespecified thresholds for low- and high-risk classification.1 The primary endpoint was adjudicated AMI during index hospitalization. Diagnostic discrimination was evaluated using the area under the receiver operating characteristic curve (AUROC). Rule-out safety at the predefined low-risk threshold was assessed by sensitivity, negative predictive value (NPV), and missed AMI rate. Analyses were performed separately in women and men, with additional evaluation across ST-segment elevation (STEMI) and non–ST-segment elevation myocardial infarction (NSTEMI) phenotypes.
Results
Among 8,493 patients, 3,186 (37.5%) were women. AMI prevalence was lower in women than men (10.5% vs 23.6%). AI-ECG demonstrated comparable discrimination in women and men (AUROC 0.875 vs 0.871). (Table 1) At the prespecified low-risk cutoff, sensitivity was 98.8% in women and 99.8% in men, with high NPV (99.2% and 99.1%, respectively) and missed AMI rates below 1% in both sexes. Performance remained stable across STEMI and NSTEMI, without systematic attenuation in women. In addition, AI-ECG enabled a larger proportion of women to be classified as low risk at presentation. (Figure 1)
Conclusion
In a prospective, multicenter emergency department cohort, AI-ECG demonstrated sex-consistent discrimination and preserved rule-out safety for AMI. Stable performance across myocardial infarction phenotypes supports AI-ECG as an early triage adjunct with potential to enhance diagnostic equity during initial evaluation.Figure 1Table 1