CTSA Researchers Use Deep Learning Model to Predict Best Time to Monitor for Atrial Fibrillation

A doctor holds a stethoscope to check the heart and lungs of an elderly male patient. (Shutterstock)

January 25, 2024

A predictive computer model for atrial fibrillation (AF), the most common form of irregular heartbeat, could help doctors decide when to start long-term monitoring to detect AF that may have been missed by a single test. The approach was developed by scientists supported by the Clinical and Translational Science Awards (CTSA) Program and the National Institute on Aging.

AF can cause blood clots, stroke, heart failure and other heart-related conditions. About one in three people with AF have an occasional and sudden form called paroxysmal AF. Long-term monitoring can catch paroxysmal AF, but deciding who to test and when to test them is challenging.

To sharpen predictions, scientists at Scripps Research Translational Institute and their colleagues reviewed information from 446,900 people who wore simple, single-lead electrocardiogram (ECG) devices to detect AF. The research team’s computer model studied the ECG readings from periods when patients were not experiencing AF to find factors that helped predict which people were likely to have AF soon after.

When testing the model against past data, the computer model alone was better at predicting AF within 2 weeks than models that relied only on such factors as a person’s age, gender, heart rate variability, odd heartbeats and heart rhythms. Combining the computer model with all those factors delivered the most accurate prediction, although the computer model alone nearly matched that combination model.

Learn more about the study findings in NPJ Digital Medicine.