Study uses heart data and AI to predict AFib onset 30 minutes before it happens

It could pave the way for the deep learning model to be used in wearables
Wareable afib prediction study
Wareable is reader-powered. If you click through using links on the site, we may earn an affiliate commission. Learn more

A new study has combined AI with the heart data accessible from wearables to help predict the onset of cardiac arrhythmia around 30 minutes before its onset.

While smartwatches have long been able to detect signs of atrial fibrillation (AFib) by analyzing user data, this is still only possible via on-the-spot ECG readings and periodic background heart checks.

As shown by researchers at the University of Luxembourg, however, stretching the typical timeframe for training AFib detection models can lead to far more useful outcomes.

The AI model - coined WARN (Warning of Atrial FibRillatioN) by the team - was tested by using 24-hour recordings from 350 patients at the Tongji Hospital in Wuhan, China, with early detection signs spotted at an average of 30 minutes before the beginning of atrial fibrillation and achieving 83% accuracy. 

Without outliers in the study, this average time extended to 40 minutes per subject, with researchers indicating that WARN is the first model to provide a heads-up this far before an episode occurs. 

CELLStudy uses heart data and AI to predict AFib onset 30 minutes before it happens photo 2

As detailed in the study notes, this is all achieved using the RR interval data - effectively the changes in milliseconds in the intervals between successive heartbeats - that's already common to smartwatches and fitness trackers.

"Another interesting aspect is that our model has a high performance using only R-to-R intervals, basically just heart rate data, that can be acquired from easy-to-wear and affordable pulse signal recorders such as smartwatches," says Dr Marino Gavidia, first author of the publication.

"These devices can be used by patients on a daily basis, so our results open possibilities for the development of real-time monitoring and early warnings from comfortable wearable devices," says Dr Arthur Montanari, also involved with the project.

Even though the study has demonstrated a positive correlation between broadening data for improved deep learning, however, it's always difficult to know when these findings will be integrated into a consumer device.

For now, at least, it's a positive step toward improved early AFib prediction - and one that should pave the way for researchers to refine the model for use in wearable devices in the near future. 

How we test

Conor Allison


Conor joined Wareable in 2017, quickly making a name for himself by testing out language translation earbuds on a first date, navigating London streets in a wearable airbag, and experiencing skydiving in a VR headset.

Over the years, he has evolved into a recognized wearables and fitness tech expert. Through Wareable’s instructional how-to guides, Conor helps users maximize the potential of their gadgets, and also shapes the conversation in digital health and AI hardware through PULSE by Wareable.

As an avid marathon runner, dedicated weightlifter, and frequent hiker, he also provides a unique perspective to Wareable’s in-depth product reviews and news coverage.

In addition to his contributions to Wareable, Conor’s expertise has been featured in publications such as British GQ, The IndependentDigital Spy, Pocket-lint, The Mirror, WIRED, and Metro.

Related stories