A new article published in Nature Medicine has shown how a new AI model can be used to predict future disease risk, including several types of heart disease, dementia, kidney disease, and all-cause mortality, based on data from just one night of sleep.
The AI model, called SleepFM, was trained using over 585,000 hours of sleep recordings from about 65,000 participants. The sleep recordings included in-depth tracking of several metrics, including brain activity, heart function, breathing, eye movement, leg motion, and more. Using this data, SleepFM could be used to identify sleep stages and evaluate sleep apnea. After this initial training, SleepFM was trained on a second dataset of about 35,000 participants who had both sleep measurements and long-term health records available. Using this dataset, SleepFM could be used to identify future health conditions that could be predicted using sleep tracking.
In the dataset that included both sleep recordings and long-term health data, SleepFM was able to accurately predict individual risk of many future health conditions. In particular, SleepFM was very accurate when predicting risk of Parkinson’s disease, dementia, prostate cancer, breast cancer, heart disease, having a heart attack, and death based on sleep recordings. While heart measurements played a larger role in predicting future heart disease and brain signals played a larger role in predicting future mental health conditions, the most accurate results came from incorporating all of the sleep signals in making predictions.
The authors noted that they were happy to see that the model could make informative predictions across a diverse set of health conditions. Sleep has been connected to a wide range of health conditions, but this was one of the first studies to incorporate AI in analyzing sleep data related to health. In the future, AI may be a useful tool in predicting future disease risk for a broad population based on sleep signals.
Sources: Nature Medicine, Science Daily
