AI Can Now Predict Over 130 Diseases From Your sleep Patterns
New Stanford model, SleepFM, unlocks a hidden wealth of health data within our nightly rest, offering a glimpse into the future of preventative medicine. Millions, “fix sleep schedule” tops the list of New Year’s resolutions. Beyond simply improving mood and cognitive performance – benefits widely associated wiht eight hours of nightly rest – sleep’s impact on overall health is proving to be far more profound than previously understood. Researchers at Stanford University have developed an artificial intelligence model, dubbed sleepfm, capable of predicting the onset of over 130 conditions, from dementia to stroke, simply by analyzing sleep recordings.
“We know intuitively that sleep is a very vital aspect of human life,” one senior researcher stated. “A typical individual spends one-third of our lives sleeping, but it’s still relatively under-explored from an AI outlook.”
Decoding the Language of Sleep
SleepFM was trained on a massive dataset encompassing over 585,000 hours of sleep recordings collected from 65,000 participants across multiple sleep clinics. The team, led by James Zou and Emmanuel Mignot, didn’t rely on a single type of data. Instead, they utilized polysomnography (PSG) recordings, a complete method that captures a wealth of physiological signals.
“We’re taking very detailed sleep recordings that capture brain signals, heart signals, muscle contractions and even breathing patterns,” explained a lead developer.this combination of inputs creates a multimodal dataset, allowing the AI to learn about sleep in a holistic manner.
However, working with such a large and complex dataset presented notable technical challenges.Rahul Thapa, a computer science Ph.D. student and lead author of the study, noted that the sheer volume of signals was a major surprise.”With over eight hours of continuous recordings for each patient, understanding what training methods worked best at a large scale took a significant amount of time and iteration,” he said.
The team discovered that training the AI to analyze different body signals simultaneously proved more effective than conventional supervised learning methods. They also developed a novel “leave-one-out” method, enhancing the model’s ability to maintain predictive accuracy even with incomplete or varied data. “We’re basically trying to get AI to learn the language of sleep,” one researcher commented.
from Research to Real-World Applications
The researchers then paired the sleep data with patient electronic health records to determine if patterns in sleep could predict future health outcomes. While the findings are promising, caution is warranted.The model’s predictions should be interpreted as estimates of relative risk, not definitive diagnoses, as it has not yet received FDA approval or undergone prospective clinical validation.
“Our goal is to understand population-level signals and associations, rather than to provide medical decisions for individual patients,” Thapa clarified.
Looking ahead, Zou and Thapa envision extending this project to wearable devices, such as smartwatches and fitness trackers. These devices, increasingly equipped with sensors capable of monitoring sleep apnea and even recording electrocardiograms (ECGs), are poised to become a crucial frontline in disease risk screening.
Chibuike Ukwakwe M.D. ’28 Ph.D. ’28, a researcher in wearable bioelectronics, lauded the SleepFM architecture. While acknowledging that current consumer wearables collect less data than the PSG recordings used to train the model, ukwakwe believes the technology holds potential for analyzing data from wearables in the future. “I can see data collected from wearables powered by AI being used to support clinical decision making,” he stated.
This project exemplifies the growing potential of AI to integrate complex physiological data and extract valuable clinical insights from sleep. as researchers continue to explore this frontier, sleep is increasingly being recognized not just as a reflection of our current health, but as a window into our future well-being.
“Sleep contains so much physiological information that we are only beginning to tap into,” Thapa concluded.
