A research team supported by the National Institutes of Health (NIH) has found that electrocardiograms (ECGs) administered during sleep and paired with sleep stage information can be used to determine risk for future adverse cardiac events. The study, which used a deep learning approach, suggests that using ECGs to predict which patients have a greater risk for poor heart-related outcomes could help guide clinical decision-making. The findings were published in the journal Sleep.
"This new approach has the potential to identify individuals at risk for cardiovascular disease years before clinical symptoms arise," said David Goff, M.D., Ph.D., acting director of NIH's National Heart, Lung, and Blood Institute (NHLBI). "Evaluating whether this information improves traditional risk prediction is an important next step. In the future, proven analytical approaches like this could be integrated into existing diagnostic tests for patients undergoing sleep studies as a screening tool for multiple adverse cardiovascular outcomes."
Testing conducted overnight during sleep, known as polysomnography, is the standard method of diagnosing sleep disorders such as insomnia and obstructive sleep apnea, which are recognized as cardiovascular risk factors. ECGs are recorded during these tests but are not often analyzed. The research team in this study wanted to see if these recordings could be used to predict 10-year risk of cardiovascular outcomes such as atrial fibrillation, stroke, myocardial infarction, and heart failure, as well as all-cause mortality.
The researchers applied a deep learning system using single-lead ECGs from sleep studies combined with expert-annotated sleep stage data. The model was fine-tuned on a dataset of 15,809 patients at Massachusetts General Hospital in Boston, and performance was assessed on two datasets that included 9,810 patients from Emory University Hospital in Atlanta and 12,576 patients from Beth Israel Deaconess Medical Center in Boston. Outcomes were derived from electronic health records.
The team found that the model was able to sort people into groups with different levels of long-term cardiovascular risk, showing that people with higher scores were at higher risk. The deep learning model also retained its predictive value when adjusted for common cardiovascular risk factors such as age, sex, BMI, diabetes, and hypertension, as well as sleep-related characteristics such as sleep apnea severity, arousals, and sleep efficiency. The model could predict atrial fibrillation, heart failure, and death from any cause.
The scientists believe that additional optimization of the model is needed to determine prediction for myocardial infarction and stroke. Overall, they found that predicting cardiovascular outcomes from the ECG channel in sleep studies is a promising approach.
"Importantly, we found that adding the score from our deep learning model improved prediction of cardiac outcomes beyond established risk factors," said Gari Clifford, D.Phil., study author, chair of the Department of Biomedical Informatics at Emory University School of Medicine, and professor of Biomedical Engineering at Georgia Institute of Technology. "Adding these scores into sleep studies has great potential for improving early detection of cardiovascular disease, determining who is most at risk, and being more proactive in patient care."
In addition, this method, which uses just one ECG lead or electrical signal, is easier to collect than the usual 12-lead setup. Therefore, it offers a simple way to collect useful ECG data in just one night of sleep, making the approach highly accessible through simple, low-cost patches.
