Multisite, External Validation of an AI-Enabled ECG Algorithm for Detection of Low Ejection Fraction.

Area of Science:

• Cardiology
• Medical Devices
• Artificial Intelligence

Background:

• Low left ventricular ejection fraction (LEF) often goes undiagnosed.
• Artificial intelligence-based electrocardiogram (ECG-AI) screening offers a scalable solution for LEF detection.

Purpose of the Study:

• To validate a comprehensive ECG-AI software as a medical device for identifying LEF.
• To assess the diagnostic performance of ECG-AI in a real-world clinical setting.

Main Methods:

• Utilized data from four diverse US sites, including ECGs and transthoracic echocardiograms performed within 30 days.
• Extracted and analyzed electronic health records from 13,960 patients.

Main Results:

• The ECG-AI demonstrated strong diagnostic accuracy with an AUROC of 0.92.
• Sensitivity was 84.5% and specificity was 83.6% for LEF detection.
• Positive predictive value was 30.5% and negative predictive value was 98.4%.

Conclusions:

• External validation confirmed the algorithm's robust diagnostic performance across a diverse patient population.
• The ECG-AI's high negative predictive value suggests its utility as a rule-out strategy, potentially deferring echocardiography.
• This study validates the ECG-AI software for LEF detection, highlighting its clinical applicability.