Researchers at Stanford University have developed an artificial intelligence program that can predict an individual’s risk of serious health conditions, including dementia, heart attack, stroke, and cancer, based on just one night of sleep data. This innovative program, named SleepFM, utilizes extensive sleep data to forecast potential health issues years before they may be diagnosed.
The foundation of SleepFM is built upon an impressive dataset comprising 585,000 hours of sleep information collected from 65,000 participants. This data is derived from a comprehensive sleep assessment known as polysomnography, which monitors various physiological parameters, such as brain waves, eye movements, muscle activity, heart rhythm, breathing, and oxygen levels. The research team compared these findings with electronic health records that spanned up to 25 years.
Insights from Sleep Data
The Stanford researchers found that the sleep data could accurately predict up to 130 different diseases. The model demonstrated particularly strong predictive capabilities for cancers, pregnancy complications, circulatory conditions, and mental health disorders. As James Zou, one of the lead researchers, explained, “SleepFM is essentially learning the language of sleep. We were pleasantly surprised that for a pretty diverse set of conditions, the model is able to make informative predictions.”
The model assigns a statistical measure known as a C-index to each disease category. This index ranks individuals based on their likelihood of experiencing events such as a heart attack. For example, a C-index of 0.8 indicates that the model’s predictions align with actual outcomes 80 percent of the time. Notably, SleepFM achieved an accuracy of 89 percent for predicting Parkinson’s disease, 85 percent for dementia, and 81 percent for heart attacks. It also demonstrated accuracy rates of 87 percent for breast cancer and 89 percent for prostate cancer.
A Step Toward Early Detection
While current sleep studies necessitate specialized clinical equipment, the findings suggest that polysomnography could evolve into a powerful early detection tool for various diseases. The research team noted that while heart signals were most informative for circulatory diseases, brain activity signals were better suited for predicting mental and neurological conditions. A combination of all signal types ultimately yielded the best predictive results.
Dr. Zou emphasized the technical advancements made in harmonizing different data modalities to enhance the model’s predictions. The research, published in the journal Nature Medicine, highlights the fundamental role that sleep plays in overall health and its complex relationship with disease. The researchers stated, “From one night of sleep, SleepFM accurately predicts 130 conditions with a C-Index of at least 0.75.”
The team is exploring ways to further enhance the AI’s predictive capabilities, potentially integrating data from wearable technology, such as smartwatches. This could enable even more personalized health assessments and timely interventions.
As the understanding of sleep’s impact on health evolves, advancements like SleepFM could mark a significant shift in how medical professionals assess patient risks, paving the way for more proactive healthcare strategies.
