A team of scientists from Geisinger and Tempus have found that artificial intelligence can predict the possibility of new atrial fibrillation and atrial fibrillation-related stroke.

The study, published in Circulation, used cardiac electrical signals from the heart-measured from a 12-lead electrocardiogram -to identify patients who are likely to develop atrial fibrillation, including those at risk for atrial fibrillation-related stroke.

To develop their model the researchers used 1.6 million electrocardiographs from 430,000 patients over 35 years of treatment at Geisinger. These data were used to study the deep neural network. Its effectiveness exceeded that of current clinical models for predicting atrial fibrillation.

Furthermore, 62% of patients who were unaware of atrial fibrillation and who had atrial fibrillation over three years were identified by the model as high-risk patients before having a stroke.