Researchers develop AI model that predicts disease risk from sleep data

By Admin

08 Jan, 2026

Artificial intelligence is steadily transforming healthcare, and a new breakthrough shows how sleep data alone can offer powerful insights into future health risks. Researchers have developed an advanced AI model that can predict the likelihood of developing more than 100 diseases by analysing how a person sleeps.

Introducing SleepFM

The AI model, named SleepFM, was developed by a team of researchers including experts from Stanford University in the United States. The system was trained on nearly six lakh hours of sleep data collected from around 65,000 participants, making it one of the most comprehensive sleep-based AI studies to date.

The findings were detailed in a scientific paper published in the medical journal Nature Medicine.

How the AI model works

SleepFM was initially tested on conventional sleep analysis tasks. These included identifying different stages of sleep and assessing the severity of sleep apnoea. After demonstrating strong performance in these areas, researchers expanded its use to predict the future onset of diseases.

To achieve this, the model analysed sleep data alongside health records sourced from a sleep clinic. More than 1,000 disease categories were examined, and researchers found that 130 conditions could be predicted with reasonable accuracy using sleep data alone.

Why sleep data is so powerful

According to senior author Emmanual Mignot, sleep studies capture an extraordinary amount of physiological information.

He explained that sleep offers a unique window into general human physiology, as the body is monitored continuously for nearly eight hours using a wide range of sensors, resulting in extremely rich datasets.

Role of polysomnography in the study

The research relied heavily on polysomnography, which is considered the gold standard for sleep studies. This method records multiple biological signals, including brain activity, heart function, breathing patterns, eye movements and muscle activity.

SleepFM successfully integrated several data streams such as electroencephalography (EEG), electrocardiography (ECG), electromyography (EMG), pulse readings and breathing airflow. By learning how these signals interact with one another, the AI was able to detect subtle patterns linked to future disease risk.

A new AI training technique

The research team also developed a novel training method called leave-one-out contrastive learning. In this approach, one stream of data is intentionally hidden, and the AI is challenged to reconstruct the missing information using the remaining signals. This technique significantly improved the model’s ability to understand complex relationships within sleep data.

Strong prediction accuracy across diseases

SleepFM showed particularly strong predictive performance for several major disease categories. These included cancer, pregnancy-related complications, circulatory diseases and mental health disorders. In these areas, the model achieved a C-index score higher than 0.8.

The C-index, or concordance index, measures how accurately an AI system can predict which individual in a group will experience a medical event first. Higher scores indicate stronger predictive performance.

From a single night of sleep, SleepFM was able to predict 130 conditions with a C-index of at least 0.75. These included all-cause mortality, dementia, myocardial infarction, heart failure, chronic kidney disease, stroke and atrial fibrillation.

Early detection of neurological disorders

The model also demonstrated strong results in predicting the risk of Parkinson’s disease. Sleep-related abnormalities are often among the earliest indicators of this condition, making sleep-based prediction especially valuable for early intervention. The AI also showed promise in identifying risks linked to developmental delays and neurological disorders.

What this means for future healthcare

This research highlights the immense potential of sleep data as a non-invasive, data-rich source for early disease prediction. With further validation and clinical integration, AI models like SleepFM could help doctors identify health risks earlier, personalise treatment plans and improve long-term patient outcomes.

As AI continues to evolve, a simple night’s sleep may soon become one of the most powerful tools in preventive medicine.