Deep learning model using ECGs during sleep studies can predict cardiovascular outcomes

An NIH-funded research team discovered that conducting electrocardiograms (ECGs) during sleep, in conjunction with sleep stage information, can predict future cardiovascular events. By employing a deep learning model, the study suggests that using ECGs to identify patients with a higher risk of cardiovascular complications could aid in clinical decision-making.

Dr. David Goff, acting director of NIH's National Heart, Lung, and Blood Institute, emphasized that this approach has the potential to identify individuals at risk for cardiovascular disease years prior to the onset of clinical symptoms. The researchers tested their model using single-lead ECGs from sleep studies combined with expert-annotated sleep stage data, which was fine-tuned on a dataset of 15,809 patients.

The model was able to categorize patients into groups based on their long-term cardiovascular risk, demonstrating that those with higher scores were at a greater risk. The model retained its predictive value even when adjusted for common cardiovascular risk factors and sleep-related characteristics. While the model showed promise in predicting atrial fibrillation, heart failure, and all-cause mortality, further optimization is required to improve predictions for myocardial infarction and stroke.

The study authors believe that incorporating this model into existing diagnostic tests for sleep disorders could enhance early detection of cardiovascular disease and facilitate more proactive patient care.