Clinical Scorecard: Deep learning model using sleep-study ECGs can predict cardiovascular outcomes

At a Glance

Key Highlights

• The study investigated prediction of 10-year risk of atrial fibrillation, stroke, myocardial infarction, heart failure, and death from any cause.
• The model was fine-tuned using data from 15,809 patients at Massachusetts General Hospital and evaluated using datasets from Emory University Hospital (9,810 patients) and Beth Israel Deaconess Medical Center (12,576 patients).
• Higher model scores indicated greater long-term cardiovascular risk; predictive value remained after adjustment for age, sex, BMI, diabetes, hypertension, sleep apnea severity, arousal frequency, and sleep efficiency.
• The source reports prediction of atrial fibrillation, heart failure, and death from any cause. The authors considered further optimization necessary to establish prediction of myocardial infarction and stroke.
• The method uses a single ECG lead recorded during sleep studies; the source contrasts this with a conventional 12-lead configuration.

Guideline-Based Recommendations

Diagnosis

• The source states that polysomnography remains the standard diagnostic method for sleep disorders, including insomnia and obstructive sleep apnea.
• The study investigated whether sleep-study ECG recordings combined with sleep-stage information could predict subsequent cardiovascular outcomes.

Management

• An NIH official said validated methods of this kind might ultimately be incorporated into diagnostic testing as a screening tool; the source provides no clinical management protocol.

Monitoring & Follow-up

• The study used electronic health records to determine outcomes.
• The source reports that higher model scores indicated greater long-term cardiovascular risk.

Risks

• The source describes insomnia and obstructive sleep apnea as recognized cardiovascular risk factors.
• The study authors considered further model optimization necessary to establish prediction of myocardial infarction and stroke.

Patient & Prescribing Data

Patients treated at Massachusetts General Hospital, Emory University Hospital, and Beth Israel Deaconess Medical Center; the source reports no prescribing or treatment data.

The source reports no treatment regimen or prescribing recommendation. An NIH official described possible future use of validated methods as a screening tool; determining whether the data enhance conventional risk prediction is a next step.

Clinical Best Practices

• The source reports a research prediction approach and does not provide an established diagnostic or treatment recommendation.
• The study adjusted predictive analyses for age, sex, BMI, diabetes, hypertension, sleep apnea severity, arousal frequency, and sleep efficiency.
• An NIH official identified determining whether these data enhance conventional risk prediction as a next step.

Related Resources & Content

• Koscova Z and colleagues. Sleep 2026; study of cardiovascular-outcome prediction from ECG recorded during overnight polysomnography. DOI: 10.1093/sleep/zsag229.