Home › Research › Risk factors
A multimodal sleep foundation model for disease prediction
Rahul Thapa, Magnus Ruud Kjær, Bryan D. He, Ian Covert, Hyatt E. Moore, Umaer Hanif, Gauri Ganjoo, Michael Brandon Westover et al.. Nature Medicine, 2026-01-06
Topic: Risk, prevention and care in population studies
Abstract Sleep is a fundamental biological process with broad implications for physical and mental health, yet its complex relationship with disease remains poorly understood. Polysomnography (PSG)—the gold standard for sleep analysis—captures rich physiological signals but is underutilized due to challenges in standardization, generalizability and multimodal integration. To address these challenges, we developed SleepFM, a multimodal sleep foundation model trained with a new contrastive learning approach that accommodates multiple PSG configurations. Trained on a curated dataset of over 585,000 hours of PSG recordings from approximately 65,000 participants across several cohorts, SleepFM produces latent sleep representations that capture the physiological and temporal structure of sleep and enable accurate prediction of future disease risk. From one night of sleep, SleepFM accurately predicts 130 conditions with a C-Index of at least 0.75 (Bonferroni-corrected P < 0.01), including all-cause mortality (C-Index, 0.84), dementia (0.85), myocardial infarction (0.81), heart failure (0.80), chronic kidney disease (0.79), stroke (0.78) and atrial fibrillation (0.78). Moreover, the model demonstrates strong transfer learning performance on a dataset from the Sleep Heart Health Study—a dataset that was excluded from pretraining—and performs competitively with specialized sleep-staging models such as U-Sleep and YASA on common sleep analysis tasks, achieving mean F 1 scores of 0.70–0.78 for sleep staging and accuracies of 0.69 and 0.87 for classifying sleep apnea severity and presence. This work shows that foundation models can learn the language of sleep from multimodal sleep recordings, enabling scalable, label-efficient analysis and disease prediction.
Key findings
- To address these challenges, we developed SleepFM, a multimodal sleep foundation model trained with a new contrastive learning approach that accommodates multiple PSG configurations.
- Trained on a curated dataset of over 585,000 hours of PSG recordings from approximately 65,000 participants across several cohorts, SleepFM produces latent sleep representations that capture the physiological and temporal structure of sleep and enable accurate prediction of future disease risk.
- From one night of sleep, SleepFM accurately predicts 130 conditions with a C-Index of at least 0.75 (Bonferroni-corrected P < 0.01), including all-cause mortality (C-Index, 0.84), dementia (0.85), myocardial infarction (0.81), heart failure (0.80), chronic kidney disease (0.79), stroke (0.78) and atrial fibrillation (0.78).
- Moreover, the model demonstrates strong transfer learning performance on a dataset from the Sleep Heart Health Study—a dataset that was excluded from pretraining—and performs competitively with specialized sleep-staging models such as U-Sleep and YASA on common sleep analysis tasks, achieving mean F 1 scores of 0.70–0.78 for sleep staging and accuracies of 0.69 and 0.87 for classifying sleep apnea severity and presence.
- This work shows that foundation models can learn the language of sleep from multimodal sleep recordings, enabling scalable, label-efficient analysis and disease prediction.
Why it matters
This study developed a multimodal sleep foundation model trained on over 585,000 hours of polysomnography recordings that could predict 130 conditions, including dementia, from one night of sleep with a C-Index of 0.85. It is a prediction tool based on sleep data, not a treatment, cure, or prevention for Alzheimer's disease.
New to this subject? Start with our plain-English guide: Risk factors, ranked →
Study details
- Participants65,000
- Evidence strengthLow
How this study scored 64/100
Published in a top-tier journal. Scored zero for study design, effect reported and randomised / blinded because the record does not state it.
| Component | Points | Why |
|---|---|---|
| Journal tier | 10 / 10 | Tier 1 — top-tier general journal (Nature Medicine) |
| Journal's citation record | 4 / 4 | Nature Medicine: journal h-index 763 (OpenAlex, 2026-10-29) |
| Study design | 0 / 12 | No study design recorded — scores zero, not guessed (not stated in the record — scored 0) |
| Strength of evidence | 1 / 5 | Rated “low” in our summary |
| Peer-review validation | 10 / 10 | Peer-reviewed. Published as an article in a named peer-reviewed journal. |
| How recent | 5 / 5 | 2026 — within 1 year(s) of the newest study here |
| Sample size | 4 / 4 | 65,000 participants |
| Effect reported | 0 / 3 | No effect size reported — scores zero (not stated in the record — scored 0) |
| Studied in people | 3 / 3 | Studied in people (participant count stated) |
| Randomised / blinded | 0 / 2 | No randomisation or blinding stated — scores zero (not stated in the record — scored 0) |
| Times cited | 4 / 7 | Cited 51 times (OpenAlex, 2026-10-10) |
| Citations per year | 4 / 5 | 25.5 citations a year over 2 year(s) since 2026 |
| Impact for its field and year | 6 / 6 | Cited 427.5× the average for papers of its field and year (OpenAlex FWCI) |
| Cited by studies here | 0 / 6 | No other study in this library cites it |
| Still being cited | 3 / 4 | Cited 51 times in 2024–2026 (2026 still in progress; OpenAlex) |
| Senior author's track record | 0 / 4 | Last-listed author James Zou: h-index 3 (OpenAlex, whole career) |
| Institution's track record | 3 / 3 | Harvard University: institution h-index 3,235 (OpenAlex) |
| Multi-centre collaboration | 3 / 3 | OpenAlex lists 10 institution(s) in 3 countries |
| Indexed in PubMed | 2 / 2 | Indexed in PubMed (PMID 41495409) |
| Free to read | 2 / 2 | Free to read |
| Total | 64 / 100 | — |
What each of these 20 components measures, and what this score is not →
Read the original
- DOI: 10.1038/s41591-025-04133-4
- PubMed: 41495409
- Original source →
- Free full text: Nature Medicine →
Full text: free to read — OpenAlex lists this copy as free to read: the link above (checked 6 Feb 1020).
Cite this study APA, Vancouver, BibTeX · download .bib or .ris
Built from the record alzhub holds for this paper, not looked up. Volume, issue and page numbers are not held in the record. The record's author list ends in “et al.”, so not every author is named. Check the citation against the original before you use it.
APA 7th
Thapa, R., Kjær, M. R., He, B. D., Covert, I., Moore, H. E., Hanif, U., et al. (2026). A multimodal sleep foundation model for disease prediction. Nature Medicine. https://doi.org/10.1038/s41591-025-04133-4
Vancouver
Thapa R, Kjær MR, He BD, Covert I, Moore HE, Hanif U, et al. A multimodal sleep foundation model for disease prediction. Nature Medicine. 2026. doi:10.1038/s41591-025-04133-4. PMID: 41495409.
BibTeX
@article{thapa2026-41495409,
author = {Thapa, Rahul and Kjær, Magnus Ruud and He, Bryan D. and Covert, Ian and Moore, Hyatt E. and Hanif, Umaer and Ganjoo, Gauri and Westover, Michael Brandon and others},
title = {{A multimodal sleep foundation model for disease prediction}},
journal = {Nature Medicine},
year = {2026},
doi = {10.1038/s41591-025-04133-4},
pmid = {41495409},
url = {https://doi.org/10.1038/s41591-025-04133-4}
}
Studies this one cites
Referenced by this study and also in the alzhub library, so you can read them here.
Related studies in Risk factors
Also in Risk factors
356 studies in the library are filed under Risk factors with no topic tag they share with another; these are this study's neighbours in a fixed tour of them. The Risk factors hub.