External assessment of an artificial intelligence-enabled electrocardiogram for aortic stenosis detection

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Aortic stenosis (AS) is a significant valvular heart disease.
• Early detection of AS is crucial for timely intervention.
• Artificial intelligence-enabled electrocardiogram (AI-ECG) algorithms offer potential for non-invasive screening.

Purpose of the Study:

• To evaluate the performance of a US-developed AI-ECG algorithm in identifying moderate to severe AS in an Asian patient cohort.
• To assess the algorithm's accuracy and reliability in a real-world clinical setting outside its original validation population.

Main Methods:

• Retrospective analysis of patients aged ≥60 years who underwent echocardiography and ECG within 31 days.
• Exclusion of patients with prior cardiac surgery, prosthetic valves, or pacemakers.
• Application of a pre-trained AI-ECG model (developed by Mayo Clinic) without fine-tuning to predict moderate to severe AS, comparing results with TTE-confirmed diagnosis.

Main Results:

• The AI-ECG model achieved an area under the curve (AUC) of 0.85 (95% CI: 0.84-0.87) for detecting moderate to severe AS.
• Key performance metrics included sensitivity of 0.83, specificity of 0.65, positive predictive value (PPV) of 0.37, negative predictive value (NPV) of 0.94, and accuracy of 68.29%.
• Performance remained consistent across age and sex subgroups, with improved sensitivity in older patients.

Conclusions:

• The AI-ECG algorithm demonstrated comparable performance in detecting moderate to severe AS in an Asian cohort compared to its original validation.
• Findings support the potential of AI-ECG as a valuable, non-invasive screening tool for AS across diverse populations.
• Further validation in broader ethnic groups may enhance global applicability of AI-ECG for AS screening.