Low Ejection Fraction AI-ECG Algorithm

The Low Ejection Fraction AI-ECG Algorithm is a software tool designed to screen adults at risk for heart failure by identifying a Left Ventricular Ejection Fraction (LVEF) of 40% or less using an artificial intelligence algorithm. This algorithm analyzes 10 seconds of a standard 12-lead ECG to provide a rapid prediction, aiding clinicians in determining the need for further cardiac evaluation.

The Low Ejection Fraction AI-ECG Algorithm is a software medical device designed to aid in screening for Left Ventricular Ejection Fraction (LVEF) less than or equal to 40% in adult patients at risk for heart failure, including those with cardiomyopathies, post-myocardial infarction, aortic stenosis, chronic atrial fibrillation, receiving cardiotoxic therapies, and postpartum women. The device employs an artificial intelligence-based algorithm that interprets 10 seconds of voltage times series data from a standard 12-lead ECG acquisition. This machine learning model, trained using a large cohort from Mayo Clinic with paired ECG and transthoracic echocardiogram (TTE) data, aims to provide a point-of-care screening tool. The algorithm's output (e.g., "Low LVEF Detected") is provided to third-party software for display, assisting clinicians in decisions for further cardiac evaluation, but is not intended for standalone diagnosis, patient monitoring, or use with paced rhythms. Clinical validation was conducted via a retrospective study of 16,000 patient records across four diverse U.S. health systems, analyzing 13,960 patient-ECG pairs after exclusions. The algorithm demonstrated a sensitivity of 84.5% (95% CI: 82.2%-86.6%) and a specificity of 83.6% (95% CI: 82.9%-84.2%) for detecting LVEF ≤ 40%. It achieved a positive predictive value of 30.5% (95% CI: 28.8%-32.1%) and a negative predictive value of 98.4% (95% CI: 98.2%-98.7%) in a population with 7.9% prevalence of low LVEF. The training dataset comprised 93,722 patients, where the algorithm showed 85.7% sensitivity and 84.2% specificity in a set-aside test set. Subgroup analyses revealed performance variations across age groups, BMI strata, and in patients with prior histories of heart failure, myocardial infarction, and coronary revascularization, although overall performance remained robust.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the algorithm's performance vary with different ECG machine models beyond the four tested in the clinical validation?; What are the specific electrocardiographic features or patterns the AI algorithm identifies as most predictive of low LVEF from the 10-second ECG strip?; Given the observed variations in diagnostic performance across age groups, BMI strata, and patients with specific prior cardiac conditions, what strategies are being considered to refine the algorithm or provide clinician guidance for these subgroups?; How does the algorithm account for, or how is its performance affected by, common comorbidities (e.g., hypertension, diabetes) or medications (other than cardiotoxic ones listed) that might influence ECG morphology in the target population?; What is the anticipated real-world impact on clinical workflow, time-to-diagnosis, and subsequent patient management pathways when this AI tool is integrated into primary care, urgent care, and emergency settings?