FDA Clears EchoNext, the First Multicondition AI Tool to Flag Structural Heart Disease From a Standard ECG
The FDA cleared EchoNext, a deep-learning tool that reads a routine 12-lead ECG to flag six forms of hidden structural heart disease, outperforming cardiologists in a Nature study.
Overview
The U.S. Food and Drug Administration has cleared EchoNext, an artificial-intelligence tool that reads a standard electrocardiogram (ECG) and flags patients at high risk for structural heart disease. According to Healthline, the FDA cleared the tool on June 22 to detect heart disease early. The model was developed by researchers at NewYork-Presbyterian and Columbia University and is being commercialized by a spinout called Pathway Labs, as reported by STAT News.
Structural heart disease — problems with the heart’s valves, chambers, or walls — is often invisible on a routine ECG and normally requires an echocardiogram, a more expensive ultrasound exam, to detect. EchoNext aims to use the cheaper, ubiquitous test as a triage step. “EchoNext basically uses the cheaper test to figure out who needs the more expensive ultrasound,” said Dr. Pierre Elias, who led the study, in the Columbia University Irving Medical Center announcement. “It detects diseases cardiologists can’t from an ECG.”
What We Know
EchoNext is a deep-learning model. According to the peer-reviewed study published in Nature, it was trained on 1,245,273 ECG–echocardiogram pairs from 230,318 unique patients. The composite of structural heart disease the model is designed to detect includes a left ventricular ejection fraction of 45% or lower, a maximum left ventricular wall thickness of 1.3 cm or greater, moderate or severe right ventricular dysfunction, pulmonary hypertension, moderate or severe valvular disease, and a moderate or large pericardial effusion, per the Nature paper.
Across the FDA clearance, the tool is authorized to flag six forms of structural heart disease. Healthline named five of them: right-sided heart failure, left-sided heart failure, valve disease, severe thickening of the heart muscle, and pulmonary hypertension.
In a head-to-head comparison with 13 cardiologists on 3,200 ECGs, EchoNext accurately identified 77% of structural heart problems, while cardiologists making a diagnosis from the ECG data alone had an accuracy of 64%, according to NewYork-Presbyterian. The Columbia University Irving Medical Center announcement notes that EchoNext identified structural heart disease more often than cardiologists, including those who used AI to help interpret the data.
The underlying Nature study reports additional detail on a separate reader analysis of 150 ECGs, in which EchoNext achieved an accuracy of 77.3%, sensitivity of 72.6%, and specificity of 80.7%; cardiologists without AI assistance reached an accuracy of 64.0%, improving to 69.2% with AI assistance. On its internal test set the model reached an AUROC of 85.2%, and external validation showed a 5–7% drop in AUROC to the 78–80% range, per the Nature paper. That external validation drew on cohorts at Cedars-Sinai Medical Center (n = 10,177 patients), the Montreal Heart Institute (n = 10,862), and the University of California San Francisco Medical Center (n = 6,106), spanning sites in the United States and Canada.
Pathway Labs has raised $8.5 million to expand into more health systems, according to Healthline. Alongside the clearance, the company said it would make EchoNext available through OpenEvidence, a clinical AI platform. OnHealthcare.tech reported that the tool landed on OpenEvidence, which it described as used by more than 750,000 verified clinicians — the same platform The Machine Herald previously covered as it expanded its hospital roster.
What We Don’t Know
The FDA clearance covers six conditions, but public coverage has named only five explicitly; the sixth condition was not disclosed in the Healthline account. Coverage has also cited different figures for the training set — the Nature study reports 1,245,273 ECG–echocardiogram pairs, while Healthline describes more than 700,000 paired records across the NewYork-Presbyterian health system, a narrower framing.
A clearance does not by itself establish how the tool performs when deployed at scale outside the validation cohorts, nor how clinicians will act on its flags in routine practice. NewYork-Presbyterian reported that on June 22, Nature Medicine published a peer-reviewed case in which EchoNext helped guide the care of a patient who ultimately underwent a heart transplant, which it called the first peer-reviewed account of its kind — a single case rather than a measure of population-level impact.
Analysis
EchoNext’s core proposition is triage economics: the ECG is one of the most common tests in medicine, and using it to surface patients who warrant a more expensive echocardiogram could widen screening without adding new equipment. The study’s own numbers frame both the promise and the caution. A 13-point accuracy gap over unassisted cardiologists is substantial, but the model’s discrimination fell several AUROC points on external cohorts, a reminder that performance measured at the developing institutions may not transfer cleanly elsewhere. The distribution deal with OpenEvidence signals a strategy of reaching clinicians through software they already use rather than through new hardware — a pattern increasingly common as AI diagnostic tools clear regulatory review.