Deep learning model demonstrates potential to generate an interpretable brain health biomarker from sleep electroencephalography

1. Ganglberger and colleagues assessed a deep learning model’s ability to derive brain health scores from electroencephalography data.

2. The model’s ability to predict cognitive and mortality outcomes surpassed demographic and conventional sleep-feature models.

Evidence Rating Level: 2 (Good)

Study Rundown: Large epidemiologic studies have documented associations between sleep characteristics, cognition, and adverse health outcomes. Sleep electroencephalography (EEG) represents a rich but underused source of information about brain health. Ganglberger and colleagues examined whether deep learning could derive a biomarker that reflected cognition, disease status, and survival from sleep EEGs. Researchers analyzed sleep recordings from multiple community and clinical cohorts and trained a model to recognize patterns associated with cognitive performance, existing diagnoses, and sleep physiology. These patterns were compiled into a single brain health score. Model performance was compared with demographic information, conventional sleep measurements, and established machine learning approaches. The study found that the deep learning EEG model improved cognitive prediction and disease classification, although associations with cognitive performance remained moderate. Additionally, higher brain health scores derived by the model were associated with lower subsequent mortality in multiple cohorts. This study demonstrated that deep learning models may gather information relevant to brain and systemic health from sleep EEGs.

Click here to read the study in NEJM AI

Relevant Reading: Automated Interpretation of Clinical Electroencephalograms Using Artificial Intelligence

In-Depth [retrospective cohort]: 36,520 sleep recordings from 27,359 participants across six community and clinical cohorts were used for model training and analysis. A deep learning model used a single EEG channel to learn 1,024 features and estimate cognition, disease status, and sleep physiology. These features were compiled into a standardized brain health score. Disease cases and controls were matched on age and sex. Cognitive scores were standardized within cohorts and evaluated using Pearson correlations and median absolute error. Disease classification used areas under the receiver operating characteristic (AUROC) and precision-recall curves, while age-adjusted Cox models assessed mortality. Across all analyzed cohorts, the sleep EEG-based deep learning model outperformed all baseline models using demographic or sleep macro predictors. Correlation between brain health score and cognition reached 0.40, and disease discrimination improved from baseline AUROCs of 0.50-0.55 to 0.65-0.75 with the use of the model. Additionally, each standard deviation increase in brain health score was associated with lower mortality in two of the analyzed cohorts: Osteoporotic Fractures in Men and Massachusetts General Hospital, with hazard ratios of 0.69 (p < 0.0001) and 0.65 (p < 0.0001), respectively. This study was limited by predominantly single-night recordings and variable intervals between sleep and cognitive testing. Nonetheless, this study provides promising evidence for deriving an index of brain health from EEGs, further enhancing its clinical utility.

Citation: Ganglberger W, Sun H, Turley N, Tripathi A, Hadar P, Gupta A, et al. Brain Health from Sleep EEG: A Multicohort, Deep Learning Biomarker for Cognition, Disease, and Mortality. NEJM AI. 2026 Mar;3(3). doi: 10.1056/aioa2500487. Epub 2026 Feb 26. PMID: 41953210; PMCID: PMC13055923.

© 2026 2 Minute Medicine, Inc. All rights reserved. No part of this article may be reproduced, distributed or transmitted in any form or by any means without the prior written permission of 2 Minute Medicine, Inc. To license this content, contact 2 Minute Medicine, Inc. This article is for informational and educational purposes only; it is not medical advice and does not replace the judgment of a qualified healthcare professional.
Image: © 2 Minute Medicine