The Post-EchoNext Workflow: How Multi-Condition ECG-AI Changes Cardiology Referrals
EchoNext FDA-cleared ECG screening for six structural heart diseases: What it means for echo capacity, referrals, EHRs, and medicolegal risk.
Clinical Takeaway: Pathway Labs' EchoNext is FDA-cleared to identify six forms of structural heart disease (SHD) and to indicate a need for further echocardiogram. The scarce resource here is NOT the algorithm, but rather the ECG, which is an inexpensive resource. It is the downstream echo slot, the cardiology visit, and the clinician's time spent as to how to act with a flagged "normal" ECG.(Source: Cardiovascular Business.)
What Was Actually Cleared (and a Framing Correction)?
This article assumes six alerts of some sort are triggered on a “normal” ECG. The reported clearance is shallower, and therefore, the first thing to do is to identify the function of the tool.
The six indications: The clearance includes both right- and left-sided heart failure, as well as valve disease, severe hypertrophy consistent with infiltrative cardiomyopathy and pulmonary hypertension from standard 12-lead input according to Medical Economics
The outcome: The tool identifies ECGs that should be followed up with an echocardiogram. If condition level outputs are displayed to the end user, this will depend on the integration and the labelling. Before creating an order set, please review the Instructions for Use.
The training base: Over 700,000 ECG-echo pairs were used for the training.
Origin: It was created at New York-Presbyterian and Columbia, and is being commercialized by the spinout Pathway Labs. (Source: statnews)
When planning workflows, regard the product as a single "echo recommended" trigger that is fired for a variety of different diseases. It is not 6 different triggers for consulting.
The Evidence Base: What the Numbers Do and Don't Tell You
Headline performance
In a set of 3,200 ECGs, EchoNext was able to detect 77% of structural heart problems when compared to 13 cardiologists in a head-to-head comparison. The cardiologists who read the same ECGs were correct 64% of the time. This outcome is based on the results of the 2025 Nature Study, nyphealthline
What is needed to enable referral planning?
Although I looked at the following sources, none provided the sensitivity and specificity, PPV or NPV that are relevant to a specific institution. The 77% accuracy figure does not indicate how many falsely positive echoes your echo lab will suck. PPV is highly dependent upon pretest probability and this will differ significantly between the ED, primary care clinic and a pre-op clinic.
There is one published figure to help with capacity. The algorithm correctly classified about 9% of patients with evaluated heart disease as being at high risk for an undiagnosed structural heart disease. Use that number as a cohort specific number and not as a flag rate.
Prospective validation
An ongoing trial (SAGE) is currently testing the efficacy of AI-guided ECG screening and management of structural heart disease in the ED in a prospective manner. Retrospective discrimination is not about patient benefit. “Flagged and found” is not a mortality/hospitalization endpoint but a diagnostic yield endpoint until outcome data are received. .
Understand the regulatory and Privacy landscape
FDA Clearance Date: FDA cleared on 22nd June, 2026. It is referred to as a "clearance" in the sources but is not certain if it was submitted to 510(k) with a predicate or De Novo. Before quoting the pathway in your work, review the information on the FDA database entry.healthline
The Purpose of the Intended Use is Important: The claims are ‘cleared'. Any use of the tool outside of the population or for indications on which it has not been labeled is "off-label" use of a medical device software function, and carries different liability exposure.
HIPAA Boundaries: EchoNext is embedded in OpenEvidence which allows clinicians to submit an ECG and get a prediction back. If there are identifiers or PHI in that uploaded ECG image, your compliance office must make sure you sign a Business Associate Agreement to comply with the vendor's terms of data retention. Remember, just because a tool is free at the point of use, it doesn't automatically mean it is meant for your institution and is HIPAA covered.
Training Data: Data from a large academic health system. Question whether performance is maintained with age, sex, race and ethnicity as well as quality of acquisition of ECGs in community practice.
What Happens Downstream: The Referral Pipeline
This is a study of the Referral Pipeline, also known as What Happens Downstream.
The echo demand is added to the flag volume:
Discuss an illustrative example using numbers that are not real. A health system processes 100,000 ECGs annually, and uses this tool for patients who don't have a known structural disease. Around 9% if the above cohort, would create approximately 9000 echo recommendations per year. That pressure does not exist until your echo lab is at full capacity, so you should measure that before you measure any downstream cardiology visits.
The incidental-finding problem:
Echo makes a flag a hard finding. Some of those findings will be relatively unimportant clinically, such as mild regurgitation (MR) or left ventricular ejection fraction (LVEF) that's just on the border. Other will be actionable such as severe aortic stenosis, LVEF < 40% or significant wall thickening. The process of monitoring downstream effects is conducted separately with each finding:
Value Disease: Frequency of surveillance, symptom review and possible referral to structural-heart team.
Reduced LVEF: Guideline directed medical therapy, ischemic evaluation and ICD candidacy reassessment.
Infiltrative Disease - Hypertrophy Suggestive: Work up to consider monoclonal protein testing and nuclear imaging, and referral to cardiology or specialty amyloid.
Further evaluation of pulmonary hypertension (right heart catheterization to be done in selected patients).
Who is the owner of the flag?
It's the most neglected failing point. An algorithm may trigger an ECG order from a PCP, which has been read by a cardiologist, and flagged by another algorithm, and all three could assume that each will act when it does. Patients are lost on the flag and not picked up by the owner.
Development of appropriate use and payment skills.
Echo will have no refunds. The AI flag will not significantly affect the historical lack of strong support for screening echocardiography for asymptomatic patients. One question that remains is whether payers will reimburse an echo ordered as a result of using an algorithm in an asymptomatic patient. Ensure that you build an auto-order pathway with your revenue cycle team in order to confirm.
Now let's compare the clinical protocol that goes around with how it's done when the AI is involved, in narrative paragraphs:
Extremely useful to use when the patient complains of chest pain.
In the traditional pathway, an echocardiogram is started by overt clinical signs – symptoms, physical murmur, abnormal ECG pattern or explicit clinician’s suspicion. The interpretation of the ECG is done only by a human reader (Physician or a cardiologist). The AI enhanced pathway, on the other hand, will add an “echo recommended” flag to an ECG that the human eye may not notice. This workflow is a combination of the human read and the automatic output from the model.
The volume of referrals and diagnostic sensitivity.
Clinical validation (including the Nature study results) demonstrates that there are large differences in accuracy: Traditional cardiologist reads have a sensitivity of around 64% when compared with the AI model's accuracy of around 77% on the same comparison. This greater sensitivity directly affects referrals. Traditional referrals are based exclusively on clinical presentation and echo demand is clinical symptom-based and relatively predictable. The AI-augmented pathway, on the other hand, is based on an algorithm's flag rate that can generate a vast number of asymptomatic referrals, leading to a tremendous increase in echo demand and its difficulty to predict.
The roles of Accountability, Documentation, and Medicolegal Exposure.
Things change drastically when it's an AI environment. The concept of accountability is already very clearly defined on the ordering clinician, and with the introduction of the AI pathway, it becomes ambiguous unless a "flag owner" is formally assigned. This means the documentation load rises; clinicians have to make the standard clinical notes and additional information on the algorithm flag, the clinical response and the reasons for not proceeding with an echo. There is also an increase in medicolegal exposure, with clinicians now dealing with the risk of missed diagnosis claims as well as new added risks of ignored alerts.
Payer Alignment
Lastly, there is a lack of payer alignment. There are definite indications for traditional echocardiography, and coverage is extensive by insurance companies. The availability of coverage and reimbursement for the echocardiogram, however, remains unclear, and could create a financial and administrative hurdle for wide clinical use, particularly for covering asymptomatic patients with algorithm-flagged echocardiograms.
Electronic Health Record (EHR) Integration: Where the Workflow Succeeds or Fails
Delivery Channel
There are only limited integration channels at present. Clinicians will upload an image of an ECG to a computer at EchoNext, which is marketed to hospitals and licensed to OpenEvidence, to get a prediction. A manual upload can be beneficial for a curbside question but it is not a population level screening system. This would mean linking it to your ECG management system (MUSE-type ECG archive, for instance) and EHR. The sources were not available to give me an answer regarding the available integrations with EHRs, so please contact the vendor directly. (STAT News)
Design questions for your informatics team
Alert placement: Embed the flag in the ECG result, not as an interruptive pop-up. Alert fatigue is already a documented problem in Epic and Cerner environments.
Routing: Send the flag to the ordering clinician and a designated pool, such as a cardiology triage nurse or APP inbox.
Closed-loop tracking: Create a worklist that tracks each flag to a documented outcome: echo ordered, echo declined with reason, or already known disease.
Suppression logic: Suppress flags for patients with known cardiomyopathy, prior valve surgery, or a recent echo, or you will generate noise.
Documentation: Use a structured field or SmartPhrase-style template so that disposition takes seconds, not minutes.
Audit trail: Log model version, flag time, and clinician response for governance and medicolegal review.
Hospital-System Impact: Winners and Pressure Points
Echo Labs: These feel the first hit. Sonographer supply is a hard constraint, and a surge in low-yield studies can crowd out symptomatic patients.
General Cardiology Clinics: Referral volume will rise, and a share of it will be low-acuity. Triage rules are essential.
Primary Care: PCPs receive a new category of result that is neither a diagnosis nor a clear instruction. Provide them with decision support, not just an alert.
Heart Failure and Valve Programs: These may see earlier detection, which could improve outcomes. It will also bring more early-stage disease into surveillance programs.
Health-System Leadership: Leaders must decide whether to deploy broadly or restrict to high-pretest-probability settings first, such as the ED or a preoperative clinic.
Practical Deployment Strategy
Start narrow
Pilot in one setting with clear ownership. The ED has an obvious rationale, given the prospective SAGE trial, but it also has the most fragmented follow-up. A cardiology-owned outpatient pilot may give cleaner closed-loop data.
Define a decision rule before go-live
Write down, in advance, what a flag triggers. For example:
- Symptomatic patient with a flag: Expedite echo.
- Asymptomatic patient with a flag and risk factors: order echo through a standard queue.
- Asymptomatic patient with a flag and no risk factors: clinician judgment, with documentation.
Measure the right metrics
Flag rate and PPV in your own population.
Echo yield: The percentage of flagged patients with clinically actionable findings.
Time from flag to echo and from echo to treatment.
Loss to follow-up: Flags with no documented disposition.
Equity metrics: Performance and follow-up rates across demographic groups.
Tools and Institutional Infrastructure Required
Deployment is as much a data-engineering and governance project as a clinical one. Here are the categories to evaluate, so you can tailor them to your site:
- ECG management and archive systems that can pass waveform data to an AI service.
- Clinical decision support and alert-management platforms to route, suppress, and audit flags.
- Care-coordination and referral-management software for closed-loop tracking.
- AI governance and model-monitoring tools to track drift, subgroup performance, and version changes.
- Cardiology-specific analytics to forecast echo demand from flag rates.
Board-Relevant and Resident-Level Pearls
Normal ECG does not mean a normal heart. Structural disease can coexist with a normal-appearing tracing, which is the premise of this technology.
Know your tests. Echo remains the reference test for LVEF, valve gradients, and wall thickness. The ECG model estimates risk; it does not measure these.
Screening principles still apply. Lead-time bias, overdiagnosis, and the cascade of downstream testing are as relevant to AI-ECG as to any screening program.
Pretest probability drives PPV. The same model performs very differently in a cardiology clinic and in a low-risk asymptomatic population.
Limitations of This Analysis
I want to be transparent about what I could not verify:
- The choice depends on the exact FDA regulatory pathway (510(k) vs. De Novo)
- Condition-specific sensitivity, specificity, PPV, and NPV for the cleared indications.
- Specific EHR integrations and reimbursement status.
- The scenario numbers above are illustrative, not drawn from a published health-system deployment.
Check the FDA summary, the labeling, and the peer-reviewed publications before making institutional decisions.
Conclusion
EchoNext shifts the screening burden from the ECG, which is cheap, to the echo lab, the cardiology clinic, and the clinician who must act on a flag. Whether it improves outcomes will depend on local flag rates, ownership, and follow-through. Hospitals that plan for capacity and closed-loop tracking before turning it on will be in the best position.
Comments
Post a Comment