Cardiovascular Risk Prediction via Deep Learning Analysis of Overnight Sleep Study ECGs

Article Impact Level: HIGH
Data Quality: STRONG
Summary of  SLEEP https://doi.org/10.1093/sleep/zsag229
Dr. Zuzana Koscova  et al.

Points

Multi center researchers developed and validated a deep learning system trained on single lead overnight ECGs paired with sleep architecture annotations across thirty eight thousand patients.
Quantitative evaluation across external validation cohorts demonstrated that deep learning scores effectively stratified ten year risk for incident atrial fibrillation heart failure and all cause mortality.
Predictive efficacy remained statistically significant after controlling for traditional cardiovascular risk factors including hypertension diabetes body mass index and polysomnography measures of sleep apnea severity.
System integration relies on single lead electrical signals making automated cardiovascular risk assessment highly scalable using low cost wearable patches during overnight sleep monitoring.
Authors concluded that incorporating algorithmic ECG analysis into routine diagnostic sleep studies enhances early cardiovascular disease detection and supports proactive personalized preventive clinical decision making.

Summary

This study evaluated a deep learning framework designed to predict 10-year adverse cardiovascular outcomes using single-lead electrocardiogram (ECG) channels paired with expert-annotated sleep stage data from overnight polysomnography (PSG). Published in SLEEP by a multi-institutional research team supported by the National Institutes of Health (NIH), the investigation addressed the underutilization of single-lead ECG recordings routinely captured during sleep studies. Given that sleep disorders such as obstructive sleep apnea and insomnia are established cardiovascular risk factors, the research sought to establish whether nocturnal single-lead ECG signals could stratify long-term risks for atrial fibrillation, heart failure, myocardial infarction, stroke, and all-cause mortality.

The deep learning model was fine-tuned on a derivation dataset of 15,809 patients from Massachusetts General Hospital and externally validated across two independent clinical cohorts comprising 9,810 patients from Emory University Hospital and 12,576 patients from Beth Israel Deaconess Medical Center. Quantitative analysis demonstrated that the deep learning risk score independently stratified long-term cardiovascular risk across cohorts. Multivariable Cox proportional hazards modeling confirmed that the model’s predictive value remained robust after adjusting for traditional cardiovascular risk factors (age, sex, BMI, diabetes, and hypertension) and PSG parameters (apnea-hypopnea index, arousal index, and sleep efficiency). The model demonstrated statistically significant predictive performance for 10-year incident atrial fibrillation, heart failure, and all-cause mortality, though further optimization is required for discrete predictions of myocardial infarction and stroke.

The authors conclude that integrating deep learning analysis of single-lead nocturnal ECGs into routine PSG provides a scalable, non-invasive method for early cardiovascular risk stratification. Adding the deep learning risk score significantly improved discrimination beyond established clinical risk calculators without requiring multi-lead ECG configurations. These findings highlight the potential for incorporating automated algorithmic screening into routine sleep diagnostics, enabling proactive cardiovascular risk identification through simple single-lead patch recording systems.

Link to the article: https://academic.oup.com/sleep/advance-article/doi/10.1093/sleep/zsag229/8885715?login=false

References

Koscova, Z., Nasiri, S., Reyna, M. A., Bhatia, N. K., Rad, A. B., Sameni, R., Sun, H., Turley, N., Katwa, U., Hwang, D., Stone, K. L., Trotti, L. M., Thomas, R. J., Mignot, E., Westover, B. M., & Clifford, G. D. (2026). Prediction of cardiovascular outcomes using electrocardiography from overnight polysomnography. SLEEP, zsag229. https://doi.org/10.1093/sleep/zsag229