Stanford’s SleepFM AI can predict 130+ diseases from a single night of sleep

Stanford University researchers have introduced SleepFM, a new AI foundation model that suggests some of the most valuable health insights may already be captured while we sleep. Instead of relying on daytime tests or visible symptoms, SleepFM analyzes overnight biological signals to predict future disease risk long before traditional diagnosis is possible.

Sleep has long been viewed as a passive state. However, growing scientific evidence shows it contains rich and continuous health signals. SleepFM builds on this idea by treating sleep as a full-body health scan rather than focusing only on sleep duration or sleep quality.

“SleepFM can predict more than 130 health conditions using just one overnight sleep recording.”

According to Stanford Medicine, these conditions include dementia, Parkinson’s disease, and heart attacks, highlighting sleep’s potential role in long-term preventive healthcare.

Training an AI on one of the largest sleep datasets ever collected

SleepFM stands out because of the scale of data used to train it. The model learned from more than 600,000 hours of sleep recordings collected from over 65,000 participants, making it one of the largest sleep datasets ever assembled for AI research.

During sleep, the body produces multiple signals at the same time. SleepFM analyzes brain waves measured through EEG, heart activity recorded via ECG, breathing patterns, and muscle signals.

Instead of evaluating each signal separately, the model focuses on how these signals interact. This approach allows SleepFM to detect patterns that traditional methods often miss.

Why misaligned body signals reveal disease risk

One of the most important insights from the study is that disease risk often appears when body systems fall out of sync during sleep. A single signal may look normal on its own, but problems emerge when signals no longer align.

For example, SleepFM identified warning signs when the brain showed deep sleep patterns while the heart rate remained unusually high. In other cases, breathing patterns became irregular during otherwise stable sleep stages.

These subtle mismatches are often invisible to clinicians reviewing standard sleep studies. However, the model flagged them as early indicators of future disease.

Linking sleep data with decades of medical history

To validate its predictions, the Stanford research team linked overnight sleep recordings with 25 years of electronic health records from the Stanford Sleep Clinic. This allowed researchers to follow real health outcomes over long periods of time.

SleepFM was tested across more than 1,000 disease categories, making the evaluation one of the most comprehensive studies conducted at the intersection of sleep science and artificial intelligence.

By comparing predicted risks with actual diagnoses, the team confirmed that sleep patterns can provide meaningful insight into future health conditions.

How accurate is SleepFM?

The results reported by the researchers are notable. SleepFM predicted Parkinson’s disease with 89 percent accuracy and dementia with 85 percent accuracy. Heart attack risk reached an accuracy rate of 81 percent.

Beyond individual diseases, the model also predicted overall mortality risk with 84 percent accuracy. In many cases, these predictions were made years before a clinical diagnosis.

This level of accuracy suggests sleep data may serve as a powerful early warning signal rather than just a measure of rest.

From sleep laboratories to everyday wearables

Humans spend roughly one third of their lives asleep, yet sleep remains one of the least understood windows into long-term health. Until now, detailed sleep analysis has largely been limited to specialized sleep laboratories.

SleepFM points toward a future where predictive health monitoring could move out of clinics and into daily life. As wearable devices improve, they are increasingly capable of capturing high-quality brain, heart, and respiratory data during sleep.

With models like SleepFM, this data could help identify disease risk early and support preventive care at scale.

What SleepFM means for the future of preventive medicine

SleepFM represents a shift from reactive healthcare toward prediction and prevention. By transforming overnight sleep recordings into an early warning system, AI models like SleepFM could change how chronic diseases are detected and managed.

Sleep may not only restore the body. It may also quietly reveal what lies ahead for our health.