One night of
sleep predicts
130 future
diseases.

Stanford's SleepFM — an AI model trained on 585,000 hours of sleep data from 65,000 patients — found 130 diseases it could predict from a single night's sleep. Hover any point on the body to explore one.

The strongest seven

◊ How to read the numbers

This number is the model's ranking skill, not a probability. Think of a doctor comparing two patients — the model does the same thing using only sleep data, and each condition's score shows how often it picks the higher-risk person correctly. Sleep carries early signals of brain, heart, and metabolic changes, which is what the model reads. Hover any of the seven dots to see its specific score.

0.50 RANDOM · 0.70 USEFUL · 0.80 STRONG · 0.90+ EXCEPTIONAL

—

By the numbers

Predictive accuracy

——

C-index · 95% CI · Stanford held-out set

Prevalence

—%

6-year AUROC —

◊ How to read this

A score of 0.91 is the model's ranking skill, not a probability. Think of a doctor comparing two patients — the model does the same thing using only sleep data, and gets it right 91 out of 100 times. Sleep carries early signals of brain, heart, and metabolic changes, which is what the model reads. It does not mean your risk is 91%.

0.50 RANDOM · 0.70 USEFUL · 0.80 STRONG · 0.90+ EXCEPTIONAL

—

Fig. 01 · SleepFM top predictors · anatomical projection

What the research shows

All 83 conditions where SleepFM exceeded C-index 0.77.

The full
clinical
landscape.

Every condition SleepFM predicts strongly from a single night's sleep — arranged around the body by category. Hover any dot for its details. Larger dots are the Stanford Seven from above.

The full clinical landscape.

Every condition Stanford's sleep AI predicts strongly from a single night of sleep. Tap any condition below to see where it sits on the body.

Tap a condition to see its location

Fig. 02 · Anatomical projection

◊ How to read the numbers

This number is the model's ranking skill, not a probability. Think of a doctor comparing two patients — the model does the same thing using only sleep data, and each condition's score shows how often it picks the higher-risk person correctly. Sleep carries early signals of brain, heart, and metabolic changes, which is what the model reads. Tap any condition below to see its specific score.

0.50 RANDOM · 0.70 USEFUL · 0.80 STRONG · 0.90+ EXCEPTIONAL

★ THE STRONGEST SEVEN

Top predictors

Stanford's seven highest-ranking conditions, listed in the paper's main results. Tap any to see its location.

◆ ALL 83 CONDITIONS

Browse by body system

Grouped by category, sorted by predictive accuracy within each. ★ marks the Stanford Seven.

Sleep is the upstream biomarker for most chronic disease.

The research is no longer speculative — three independent large-scale studies and genetic causal analysis now point to the same conclusion. The question for you becomes: what do you do about it.

Understand your personal risk

Take the WhySleep.ai assessment. Five minutes, AI-guided, tailored to your history. Gives you a baseline score and the three conditions to watch most closely.

Get screened clinically

If the assessment flags risk, the next step is a proper sleep study. Golden Gate Sleep Centers offers board-certified sleep medicine evaluation across the Bay Area — in-lab polysomnography and home sleep apnea testing both available.

Treat what's treatable

Most sleep disorders are diagnosable and treatable — often with dramatic improvement in quality of life within weeks. The cost of inaction is measured in years; the cost of action is a single night in a sleep lab.

Have a question about any of this? Ask Sleepi™ — your AI sleep guide.

Check My Symptoms Find a Sleep Specialist →

Sources & methodology

All evidence · peer-reviewed · primary literature

01

SleepFM: a multimodal sleep foundation model for disease prediction

Thapa R, Mignot E, Zou J, et al.

Nature Medicine · Feb 2026 · N = 65,000 · PSG foundation model · 130 disease endpoints
Paper ↗
02

Health risks and genetic architecture of objectively measured multidimensional sleep health

Zhang S, Zhang M, Yuan Y, Li Z, Li X, Li X.

Nature Communications · 2025 · N = 85,233 UK Biobank · Accelerometer · 76 incident outcomes · 7.9 yr follow-up
Paper ↗
03

Sleep patterns and risk of chronic disease as measured by long-term monitoring with commercial wearable devices in the All of Us Research Program

All of Us Research Program Investigators.

Nature Medicine · Jul 2024 · N = 6,785 · Consumer wearable · 4.5 yr median follow-up
Paper ↗
04

Divergent biological pathways linking short and long sleep durations to mental and physical health

Li Y, Gong W, Sahakian BJ, et al.

Nature Mental Health · Mar 2025 · Mechanism study · distinct pathways for short vs long sleep
Paper ↗
05

Shared genetic architecture and causal relationship between sleep behaviors and lifespan

Wu Y, Zhang C-Y, Liu X, et al.

Translational Psychiatry · 2024 · Mendelian randomization · causal evidence · CAD / T2D / depression mediators
Paper ↗

Evidence base · how to read this page

The conclusions on this page draw on five peer-reviewed studies representing three distinct research designs.

Predictive accuracy C-index and 6-year AUROC values shown for each condition come from SleepFM (Thapa et al., Nature Medicine 2026), Stanford's sleep foundation model tested on a held-out set of 5,019 patients with 95% confidence intervals and Bonferroni-corrected significance (p<0.01).
Cohort replication The broader thesis — that sleep is an upstream biomarker for chronic disease — is independently replicated at the cohort level by two prospective studies using different measurement technologies: the UK Biobank Unfavorable Sleep Profile study (Zhang et al., Nature Communications 2025; N=85,233, research-grade accelerometers, 7.9-year follow-up across 76 outcomes) and the All of Us wearable cohort (Nature Medicine 2024; N=6,785, commercial consumer devices, 4.5-year median follow-up).
Mechanism · causal Biological-pathway and causal evidence come from Li et al. (Nature Mental Health 2025; divergent pathways linking short and long sleep to mental and physical health) and Wu et al. (Translational Psychiatry 2024; Mendelian-randomization causal analysis of sleep duration and lifespan).

Scope. Mechanism chains shown for the seven highlighted conditions in Section 01 are drawn from the cited literature. General mechanism descriptions for the remaining 76 conditions reflect well-established sleep-physiology patterns and should not be interpreted as definitive clinical claims for any individual condition.

Disclaimer. This page is educational and does not constitute medical advice.