Artificial Intelligence Uses ECGs to Predict A-Fib Risk

In two studies, artificial intelligence was used with electrocardiogram (ECG) results to identify patients who are at increased risk for a potentially dangerous irregular heartbeat, and those more likely to die within a year, researchers say.

Using more than 2 million ECG results gathered over three decades, the team created "deep neural networks" that predict future events from an ECG.

In one study, researchers used 1.1 million ECGs that did not find atrial fibrillation (a-fib) in more than 237,000 patients to assess the network's ability to predict the heart rhythm disorder before it develops. A-fib increases the risk of heart attack and stroke.

Among the top 1% of high-risk patients as predicted by the neural network, one-third were diagnosed with a-fib within a year. Patients predicted to develop a-fib within one year were 45% more likely to develop the disorder over 25 years than other patients.

Senior author Christopher Haggerty noted that because there are few ways to identify which patients will develop a-fib, a stroke is often the first sign of the disorder. He is co-director of the Cardiac Imaging Technology Laboratory at Geisinger Health System in Danville, Pa.

"We hope that this model can be used to identify patients with atrial fibrillation very early so they can be treated to prevent stroke," Haggerty said in an American Heart Association (AHA) news release.

A second study used results of 1.77 million ECGs and other records from almost 400,000 patients. The neural network was better than other methods at predicting a patient's risk of death from all causes within a year.

The neural network was able to accurately predict risk of death even in patients whose ECG was considered normal by a doctor.

"This is exciting and provides more evidence that we are on the verge of a revolution in medicine where computers will be working alongside physicians to improve patient care," said imaging lab co-director Dr. Brandon Fornwalt, senior author on both studies.

"Incorporating these models into routine ECG analysis would be simple. However, developing appropriate care plans for patients based on computer predictions would be a bigger challenge," said Sushravya Raghunath, a computational scientist in the imaging lab and lead author of the second study.

The findings are to be presented at the AHA annual meeting in Philadelphia, Nov. 16-18. Research presented at meetings is typically considered preliminary until published in a peer-reviewed journal.
The researchers are now testing whether AI predictions can be used to improve patients' heart health outcomes.

Source: American Heart Association, news release, Nov. 11, 2019.