In the US, AI assessed 10-year cardiac risks using overnight ECG — ScienceAlert
Researchers in the United States have developed an artificial intelligence system that can assess the 10-year risk of atrial fibrillation, stroke, heart failure and death from any cause using a single-lead electrocardiogram recording during sleep. The findings were published in the journal SLEEP, ScienceAlert reports.
Data from three medical centers
A team led by scientists from Emory University trained a deep-learning model on data from nearly 16,000 participants in a sleep study at Massachusetts General Hospital in Boston. During the study, their heart's electrical activity was recorded using ECG.
The system was then tested on two independent datasets: nearly 10,000 patients at Emory University Hospital and 13,000 patients at Beth Israel Deaconess Medical Center in Boston. The model classified participants by levels of long-term cardiovascular risk.
More current news is available on the UA.News Telegram channel Telegram.
Forecasting results
After accounting for conventional risk factors and sleep characteristics, a one-standard-deviation increase in the model's score was associated with a 103–172% higher risk of atrial fibrillation, a 19–40% higher risk of stroke, a 69–133% higher risk of heart failure, and a 65–82% higher risk of death from any cause. Only one group showed a statistically significant increase in the risk of future myocardial infarction.
The authors noted that adding the model's score to existing risk assessment systems improved prediction of atrial fibrillation by up to 6%, heart failure by up to 9%, and mortality by up to 7%. Changes in heart rhythm and electrical conduction during the lightest stage of sleep, when the nervous system rapidly changes activity, may be particularly informative for assessing future risks.
The scientists suggest that after further validation, this approach could be integrated into sleep studies as a tool for early identification of people at increased risk of adverse cardiovascular outcomes.
