Sleep ECGs Predict Ten-Year Cardiovascular Risk in NIH Study

Sleep Study ECGs Accurately Predict Ten-Year Cardiac Risk

A deep learning model trained on single-lead electrocardiogram (ECG) tracings from overnight sleep studies predicts 10-year risk for major cardiovascular outcomes.

The research was supported by the National Institutes of Health (NIH) and published in the journal SLEEP.

Standard sleep studies, known as polysomnography, evaluate disorders such as obstructive sleep apnea and insomnia. While technicians routinely record ECG data during these overnight tests, clinicians rarely evaluate those heart signals for long-term cardiovascular prognosis.

Deep Learning Model Analyzes Over Thirty-Eight Thousand Patients

Researchers fine-tuned the neural network on single-lead recordings from 15,809 patients at Massachusetts General Hospital in Boston.

The investigators then evaluated the model on two separate patient groups to verify its accuracy. These test groups included 9,810 patients from Emory University Hospital in Atlanta and 12,576 patients from Beth Israel Deaconess Medical Center in Boston.

Cardiovascular outcomes and survival records came directly from electronic health records across all three hospital systems.

• 15,809 patients used for initial model fine-tuning at Massachusetts General Hospital
• 9,810 patients evaluated in the first external validation cohort at Emory University Hospital
• 12,576 patients evaluated in the second external cohort at Beth Israel Deaconess Medical Center

What the Model Predicts Beyond Standard Metabolic Risk Factors

The deep learning score remained predictive of adverse outcomes after controlling for common clinical markers.

Adjustments included age, sex, body mass index (BMI), diabetes, and hypertension. The model also held its predictive value when researchers accounted for sleep-specific metrics, such as sleep apnea severity, nighttime arousals, and overall sleep efficiency.

Higher output scores correlated with higher rates of atrial fibrillation, heart failure, and death from any cause over a 10-year follow-up period.

Current Diagnostic Limitations and Next Steps for Sleep Screening

The current model successfully identified risks for heart failure and atrial fibrillation, but it requires further optimization to predict stroke and myocardial infarction.

The approach relies on a single electrical lead rather than a conventional 12-lead hospital setup. This single-lead design allows future data collection through low-cost wearable patches during sleep, according to the study authors.

Researchers note that prospective clinical trials must confirm whether adding this analytical score to sleep studies improves patient outcomes before doctors adopt it in routine clinical care.