An app that uses an AI model to read a single-lead ECG from a smartwatch can detect structural heart disease, researchers reported at the 2025 Scientific Sessions of the American Heart Association. 

Although the technology requires further validation, the researchers said it could help improve the identification of patients with heart failure, valvular conditions, and left ventricular hypertrophy before they become symptomatic, which could improve the prognosis for people with these conditions. 

“This is the first study that has shown the potential to detect multiple structural heart diseases from real-world smartwatches,” Arya Aminorroaya, MD, MPH, an internal medicine resident at Yale New Haven Hospital, New Haven, Connecticut, who led the work, told Medscape Medical News. 

Structural heart disease typically goes undetected for years, Aminorroaya said, and patients tend not to get diagnosed until they become symptomatic. “We are missing the asymptomatic window of these diseases where we could intervene earlier, potentially changing the trajectory of the disease and improving outcomes,” he said. 

Most smartwatches enable the recording of a simple ECG, which can provide limited information about heart rhythm but no meaningful insight into possible structural heart diseases, he said. 

“Even a 12-lead ECG at the hospital gives limited information about structural heart disease. And a smartwatch only has a one-lead ECG,” Aminorroaya said. 

For the new study, Aminorroaya and colleagues at Yale’s Cardiovascular Data Science Lab built an AI algorithm that can interpret the ECG taken by the smartwatch to predict the presence of structural heart disease.

The model was developed by examining 266,054 ECGs from 110,006 patients paired with echocardiograms within 30 days. 

“This means we can accurately connect each ECG to the status of structural heart disease,” Aminorroaya said. 

Using this information, they trained an AI model to interpret ECG changes on a one-lead ECG generated by a smartwatch and how that would correlate with structural heart disease. 

The researchers also introduced “noise” to make the model more resilient to artificial signals that can be caused by movements, muscle twitches, or interferences on the sensor. 

The model was then externally validated in 44,591 patients across four community hospitals and 3014 participants from the population-based ELSA-Brasil study. 

The researchers tested the model in a sample of 600 individuals who had their ECG measured by a smartwatch and interpreted by the app, and then had an echocardiogram. Of the patients tested, 21 (5.3%) were found to have structural heart disease on the echocardiogram. 

The model had an area under the receiver operating curve of 0.88 — “pretty good,” Aminorroaya said — and demonstrated the ability to identify structural heart disease with a sensitivity of 86%, a specificity of 87%, a negative predictive value of 99%, and a positive predictive value of 27%, the researchers reported. 

The app could be used to identify heart failure, vascular diseases, and left ventricular hypertrophy. But the model has not been programmed to detect cardiomyopathy, said Aminorroaya, although other models are under development for that condition. 

The researchers said the app could be used for improving community-based screening for structural heart disease. 

“People could use it themselves with their own smartwatches or there could be tests made available at community settings such as churches, grocery stores, and barber shops, so that ownership of a smartwatch is not a prerequisite,” Aminorroaya said. 

Aminorroaya acknowledged more work is needed to find the right balance between sensitivity for detecting structural heart disease and minimizing the rate of false-positive readings. 

“We need to find a spot that improves the detection of structural heart disease but doesn't overburden the health system,” he said. 

‘High Level of Performance’

Pradeep Natarajan, MD, director of preventive cardiology at Massachusetts General Hospital in Boston, said while the technology appears promising to exclude the possibility of underlying structural cardiovascular disease, many individuals could be missed given the reported sensitivity of 86%. 

He cautioned the algorithm must be validated among individuals in the community not necessarily seeking cardiac imaging, and many potential issues the app detects may not reflect clinically significant structural heart disease and could lead to unnecessary and costly interventions.

“All that being said, I am very enthusiastic about using low-resource, high-information tools like the ECG, especially since many of our patients have wrist-worn wearable devices, to improve cardiovascular disease care and prevention,” Natarajan said. 

Challenge in Integration 

Richard Becker, MD, professor of medicine at the University of Cincinnati College of Medicine, said the new study “opens the door for wearable technology, including a single ECG lead to be included in the broader dialogue of ECG screening.”

But Becker noted only about 18% of smartwatches claiming ECG functionality have been cleared by the US Food and Drug Administration for medical use in that application.

“Despite the remarkable strides in artificial intelligence and wearable technology, the full potential of these innovations remains largely untapped in a healthcare system that has veered off course in terms of prevention,” he said. “In a nation where predictive tools capable of identifying disease years before symptoms arise are readily available, the challenge lies not in technological capability but in the collective will to integrate these tools equitably and ethically.” 

Aminorroaya, Natarajan, and Becker reported no relevant financial relationships.