AI model uses sleep data to predict risk of over 100 diseases, researchers say

Researchers have developed a new artificial intelligence (AI) model that can predict a person’s risk of developing more than 100 diseases by analysing sleep data, a development that could help in early detection of serious health conditions.

The AI system, called SleepFM, was developed by an international research team that includes scientists from Stanford University in the United States. The model was trained using nearly six lakh hours of sleep data collected from around 65,000 participants.

The findings were published in the medical journal Nature Medicine. In the initial phase, the model was tested on common sleep analysis tasks such as identifying sleep stages and measuring the severity of sleep apnoea. It was later used to predict future health risks by linking sleep data with medical records from a sleep clinic.

Researchers examined health records covering more than 1,000 disease categories. They found that SleepFM could predict the risk of 130 diseases with reasonable accuracy using sleep data alone.

Senior author Emmanuel Mignot, professor of sleep medicine at Stanford University, said sleep studies capture a wide range of body signals over several hours, making them highly valuable for health analysis. He noted that sleep provides continuous data on general body functions while a person is at rest.

The model uses data collected through polysomnography, which is considered the standard method for sleep studies. This process records brain activity, heart function, muscle movement, breathing patterns, eye movement and pulse signals using multiple sensors.

According to the researchers, SleepFM can combine and analyse these different data streams and understand how they are connected. The team also developed a new training method known as leave one out contrastive learning, in which the model learns to reconstruct missing data by using information from other signals.

The AI model showed strong performance in predicting diseases such as cancer, pregnancy-related complications, heart and circulatory conditions and mental health disorders. Many predictions achieved a high C index score, which is a common measure of how accurately an AI system can predict health outcomes.

The study also found that the model performed well in predicting the risk of dementia, heart attack, heart failure, stroke, kidney disease and irregular heart rhythms. It also showed promising results in identifying risks linked to Parkinson’s disease and certain developmental disorders.

Researchers said the AI system could support doctors in identifying disease risks at an early stage and improving preventive healthcare, though further studies are needed before it can be used widely in clinical settings.

(with inputs from agencies)