AI ECG Helps Doctors Find Hidden Heart Failure Earlier
A Nature Medicine case shows how physician-led AI screening can turn routine ECGs into better decisions about who needs urgent heart imaging.
Editorial Note
This is not our usual standard article format, which normally stays closer to a single study and its clinical implications. We are taking a more editorial approach here because the recent pushback from parts of the physician community, along with broader anti-AI sentiment in healthcare, misses the most practical point. The question is not whether AI should replace physicians. The question is whether physicians who use AI responsibly will make better decisions than physicians forced to rely only on human attention inside overloaded systems.
Topline
A Nature Medicine case highlighted how an AI-enhanced ECG tool helped physicians detect severe hidden heart disease in a 45-year-old emergency department patient who was initially treated for asthma-like symptoms.
The AI tool, EchoNext, flagged possible structural heart disease from a routine ECG, leading to echocardiography, recognition of severe heart failure, genetic testing, and ultimately heart transplantation.
The central lesson is not that AI replaces doctors. It is that physicians using AI may practice better medicine when symptoms are nonspecific, test results are noisy, and healthcare systems are under pressure.
Study Details
Structural heart disease includes problems affecting the heart valves, walls, chambers, or pumping function. It can present with clear symptoms, but it can also look like something else. Patients may have shortness of breath, fatigue, swelling, fainting, chest discomfort, or poor exercise tolerance. In emergency care, those symptoms often overlap with lung disease, infection, anxiety, anemia, asthma, or other common conditions.
The patient in this case was Louie Quiros, a 45-year-old caregiver and security guard who came to a Queens emergency room after several days of coughing blood and worsening shortness of breath. His chest X-ray was normal. His ECG was abnormal, but not in a way that made the diagnosis obvious. Because he had recently been exposed to wildfire smoke, he was initially sent home with asthma medication and an inhaler.
That emergency department was part of the NewYork-Presbyterian system, where researchers were evaluating EchoNext, an AI-enhanced ECG program developed by clinicians and data scientists at NewYork-Presbyterian and Columbia. EchoNext is designed to analyze ordinary 12-lead ECGs for hidden patterns that may suggest structural heart disease.
In this case, the AI system flagged possible severe heart damage. The patient was called back for echocardiography about a week later. The echocardiogram showed that his heart was pumping extremely poorly, with an ejection fraction of about 10 percent, and that his mitral valve was leaking blood backward into the heart. Further testing found a rare genetic disorder associated with sudden death. He ultimately received a heart transplant.
Methodology
The clinical workflow is straightforward. A patient receives a routine ECG, which is already one of the most common tests in emergency medicine. Instead of using the ECG only to look for rhythm abnormalities or signs of a heart attack, the AI model searches for subtle electrical patterns associated with structural heart disease.
When the AI result suggests high risk, the physician can use that signal to decide whether the patient needs echocardiography. The echocardiogram remains the confirmatory test. The cardiology team still interprets the full clinical picture. The AI does not transplant a heart, diagnose a patient alone, or replace medical judgment.
That distinction matters. Many ECGs are abnormal. Most abnormal ECGs do not mean the patient needs a heart transplant. The real value of AI is helping clinicians decide which abnormal signals deserve escalation. In this case, AI expanded the physician’s field of vision.
Key Findings
AI-enhanced ECG screening flagged possible structural heart disease in a patient whose symptoms initially looked more like an asthma or lung-related presentation.
Follow-up echocardiography showed severe heart failure, including an ejection fraction of about 10 percent and mitral valve leakage.
Further testing identified a rare genetic disorder associated with sudden death, which helped explain the severity of the disease.
The patient ultimately underwent heart transplantation, showing how an early AI signal can change the diagnostic pathway in a high-risk case.
The case supports a practical model for medical AI: not autonomous diagnosis, but better physician triage.
This remains a single case from an ongoing clinical trial, so it should not be interpreted as proof that AI-ECG screening improves outcomes across all patients.
Implications for Practice
For patients, this case is a reminder that shortness of breath, coughing blood, unexplained fatigue, swelling, fainting, or reduced exercise tolerance should not be dismissed when symptoms persist or worsen. A normal chest X-ray does not rule out serious heart disease. An abnormal ECG that does not give a clean diagnosis may still deserve follow-up if the clinical picture does not fit.
For healthcare providers, the case highlights where AI may be most useful. Medicine has an attention problem. Emergency physicians, primary care doctors, and specialists are asked to make difficult decisions under time pressure, often with incomplete information and many competing possibilities. AI can help by surfacing risk that may be easy to miss in a busy workflow.
This is especially important for structural heart disease because echocardiography is not ordered for every patient with an abnormal ECG. That would be expensive, inefficient, and unrealistic. A tool like EchoNext may help identify which patients are most likely to benefit from the next test.
The access story also matters. Reports indicate that EchoNext is expected to become available through OpenEvidence, and the company behind it has announced FDA clearance for multicondition cardiology screening. If implemented carefully, that could move AI-ECG screening beyond a few major academic centers and into broader clinical use.
But implementation needs discipline. Hospitals and clinics will need clear follow-up pathways, cardiology oversight, false-positive management, equity monitoring, and safeguards against treating AI output as a final diagnosis. AI should not be a shortcut around clinical reasoning. It should be a second layer of pattern recognition that helps physicians reason better.
The best version of this future is physician-led AI. The doctor remains accountable. The diagnostic pathway remains clinical. The echocardiogram still confirms disease. But the patient who would otherwise be missed has a better chance of being found.
For the latest in medical research news, check out
