Eko Low Ejection Fraction Tool (ELEFT)

The Eko Low Ejection Fraction Tool (ELEFT) is a software device employing machine learning algorithms to analyze combined electrocardiogram and heart sound data, aiming to assist clinicians in identifying adult patients at risk for heart failure who may have a low left ventricular ejection fraction (≤40%). Its methodology relies on signal processing and deep neural networks for classification, offering a non-invasive screening aid that requires further confirmatory diagnostic testing.

The Eko Low Ejection Fraction Tool (ELEFT) is a machine learning-based software designed to assist clinicians in identifying adult patients (18 years and older) with a Left Ventricular Ejection Fraction (LVEF) of 40% or less, who are at risk for heart failure. The tool utilizes an algorithm that processes electrocardiogram (ECG) and heart sound (phonocardiogram, PCG) recordings captured by Eko Stethoscopes, specifically the Eko DUO device, and stored as .WAV files. Its core technology involves signal processing and deep convolutional neural network models to analyze these physiological signals. A preliminary step includes a waveform analysis component, using a previously cleared algorithm, to assess the quality of the ECG and heart sound signals; if ECG quality is poor, the analysis is halted, otherwise, the deep learning models classify the combined data to suggest the likelihood of low LVEF. ELEFT is intended for prescription use by healthcare professionals or experienced operators to aid in the assessment of patients at risk for heart failure due to conditions like coronary artery disease, diabetes, cardiomyopathy, hypertension, and obesity. It is not a standalone diagnostic tool, nor is it meant for monitoring heart failure progression; rather, it serves as a notification system to prompt further diagnostic evaluation, such as an echocardiogram, which remains the gold standard. Clinical performance validation on a diverse dataset of 3,456 subjects (after excluding 307 for poor signal quality from the analysis set of 3,149) showed a sensitivity of 74.7% and a specificity of 77.5% for detecting LVEF ≤ 40%. The tool also demonstrated a negative predictive value of 96.7%. Subgroup analyses indicated some performance variations, notably slightly lower specificity in elderly patients and those with certain conduction disorders, suggesting these factors may increase false positive rates.

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 (sensitivity, specificity, PPV, NPV) vary when ECG and PCG signals are analyzed individually versus combined, and what is the incremental diagnostic value of PCG data for ELEFT?; Given the exclusion of 307 recordings (nearly 9% of the test set) due to poor ECG quality, what are the specific criteria for 'poor ECG signal,' and what strategies are in place or could be developed to minimize data loss and improve ELEFT's usability in real-world clinical settings where signal quality can be variable?; The subgroup analysis showed lower specificity in patients >70 years and those with conduction disorders. What specific features or patterns in these subgroups contribute to the higher false positive rates for ELEFT, and are there plans to refine the algorithm or provide specific guidance for interpreting results in these populations?; How was the 5-year experience requirement for operators collecting clinical data for ELEFT determined, and what is the impact on the device's performance if data is collected by operators with less experience?; Considering ELEFT's positive predictive value of 25.7%, what is the anticipated clinical workflow and resource utilization impact if this tool is widely adopted for screening at-risk populations, particularly regarding the number of follow-up echocardiograms for false positives?