A new artificial intelligence model can forecast a person’s risk of developing more than 100 health conditions — including cancer and dementia — by analysing physiological data from just one night of sleep.
Developed by a team that included Stanford Medicine researchers, the “SleepFM” model was trained on nearly 600,000 hours of polysomnography data from 65,000 participants. The system treats sleep not just as a period of rest, but as a comprehensive physiological assessment that can reveal hidden health trajectories years in advance.
“We record an amazing number of signals when we study sleep,” said Emmanuel Mignot, MD, PhD, Professor in Sleep Medicine. “It’s a kind of general physiology that we study for eight hours in a subject who’s completely captive. It’s very data rich.”
Language of sleep
While traditional sleep studies focus on conditions such as apnea, SleepFM uses a “foundation model” architecture — similar to the technology behind ChatGPT — to analyse the complex interactions among brain waves, heart rhythms, and breathing patterns.
To achieve this, researchers developed a technique called “leave-one-out contrastive learning”. This method conceals one stream of data (e.g., heart activity) and challenges the AI to reconstruct it from other signals, thereby teaching the model to understand how different bodily systems communicate.
“SleepFM is essentially learning the language of sleep,” said James Zou, PhD, associate professor of biomedical data science.
When tested against decades of health records, the model successfully predicted the onset of 130 different disease categories. It proved remarkably accurate for severe chronic conditions, achieving high predictive scores (C-index) where 0.8 represents 80 per cent accuracy:
- Parkinson’s disease: 0.89
- Prostate cancer: 0.89
- Breast cancer: 0.87
- Dementia: 0.85
- Heart attack: 0.81
“We were pleasantly surprised that for a pretty diverse set of conditions, the model is able to make informative predictions,” Zou noted.
