Objective:

Assess whether single-lead ECGs recorded during overnight sleep studies, combined with sleep-stage information, can predict 10-year cardiovascular outcomes and all-cause mortality.

Approach:

• Model development: Researchers fine-tuned a deep learning model using sleep-study ECGs and expert-annotated sleep-stage data from 15,809 patients at Massachusetts General Hospital.
• Evaluation: They evaluated the model in separate datasets of 9,810 patients from Emory University Hospital and 12,576 from Beth Israel Deaconess Medical Center. Outcomes were identified from electronic health records.

Key Findings:

• The model stratified patients by long-term cardiovascular risk, and its predictive value remained after adjustment for established cardiovascular risk factors and sleep-related characteristics.
• It predicted atrial fibrillation, heart failure, and death from any cause.
• The researchers said further optimization is needed to establish prediction of myocardial infarction and stroke.

Interpretation:

The study authors described sleep-study ECG analysis as a promising approach to predicting cardiovascular outcomes. NHLBI acting director David Goff said determining whether the data improve conventional risk prediction is a crucial next step.

Limitations:

• The source reports that further model optimization is needed to establish prediction of myocardial infarction and stroke.
• The source does not provide additional details on study limitations.

Conclusion:

The study examined whether routinely recorded single-lead ECGs from sleep studies can help identify patients at elevated risk of future cardiovascular outcomes.

Sources:

• Sleep 2026 study by Koscova Z and colleagues, DOI: 10.1093/sleep/zsag229