**Breakthrough in Sleep Analysis: Stanford Researchers Develop AI Model to Predict 130 Diseases from a Single Night of Sleep**
Imagine being able to predict your risk of developing a chronic illness from a single night’s sleep. Sounds like science fiction, right? But, a team of researchers from Stanford University has made it a reality with the development of SleepFM-Clinical, a groundbreaking AI model that can predict the risk of 130 diseases from a single night of sleep using polysomnography (PSG).
**From PSG to Disease Prediction**
PSG is the gold standard test in sleep medicine, but most clinical workflows use it only for sleep staging and sleep apnea diagnosis. The Stanford team took a different approach, treating PSG data as a dense physiological time series and training a foundation model to learn a shared representation across all modalities. This allowed them to use PSG data for disease prediction, a feat that had never been achieved before.
**The SleepFM Model**
SleepFM is a convolutional backbone that extracts local features from each channel, followed by attention-based aggregation across channels and a temporal transformer that operates over short segments of the night. The model was trained on around 585,000 hours of sleep recordings from 65,000 individuals, with the primary cohort coming from the Stanford Sleep Medicine Center.
**Predicting Diseases with SleepFM**
The team mapped diagnosis codes in the Stanford electronic health record to phecodes and defined over 1,000 candidate disease groupings. For each phecode, they computed time to first diagnosis after the sleep study and fit a Cox model on top of SleepFM embeddings. SleepFM identified 130 diseases whose risks are predictable from a single night of polysomnography with robust discrimination.
**Comparing to Baselines**
To evaluate the added value, the team compared SleepFM-based risk models with two baselines: one that uses only demographic features and one that trains an end-to-end model directly on polysomnography and outcomes. Across most disease classes, the pretrained SleepFM representation combined with a simple survival head yielded higher concordance and better long-horizon AUROC than both baselines.
**The Power of SleepFM**
This study clearly shows that the gain comes less from a complex prediction head and more from the foundation model that has learned a general representation of sleep physiology. This means that clinical centers can reuse a single pretrained spine, train small website-specific heads with relatively modest labeled cohorts, and still achieve state-of-the-art performance.
**Take a Look**
Check out the paper and code here. Also, be sure to follow us on Twitter and join our 100k+ ML SubReddit and Subscribe to our Newsletter. Oh, and did you know we’re on Telegram too?
