Stanford Medicine Develops an AI Model that Uses Sleep Data to Predict Future Disease Risk

The AI system, named SleepFM and was built on a foundation model, a type of AI model that can train itself on large datasets and apply patterns to a wide range of tasks.

Researchers from Stanford Medicine have developed a new artificial intelligence (AI) model designed to predict a person’s risk of developing future health conditions by using data from a single night’s sleep.

The AI system, named SleepFM and was built on a foundation model, a type of AI model that can train itself on large datasets and apply patterns to a wide range of tasks.

To leverage this extensive sleep data trove, it was trained on nearly 585,000 hours of sleep recordings from approximately 65,000 individuals.

Initially, the SleepFM was tested on conventional sleep analysis tasks, such as sleep tracking stages and diagnosing the severity of sleep apnoea.

The AI model combines multiple streams of data, such as

electroencephalography, electrocardiography, electromyography, pulse reading, and breathing airflow, which enables it to identify relationships between these signals.

The team then combined sleep data with long‑term health records, enabling the model to estimate a person’s future risk for various diseases.

The AI model’s performance was evaluated by leveraging the concordance index (C‑index) and reportedly demonstrated strong predictive capabilities, achieving a C‑index of 0.85 for dementia and 0.84 for all‑cause mortality.

It was able to predict a range of conditions, including cardiovascular diseases such as heart attack, heart failure, and atrial fibrillation, neurological outcomes like dementia, pregnancy complications, mental health disorders, and chronic kidney disease, and stroke.

The team further plans to improve SleepFM’s predictions by incorporating data from wearables and investigating the specific patterns the model uses for its predictions.

Stay tuned for more such updates on Digital Health News