Researchers at Stanford University say a single night’s sleep could reveal future health risks, after developing an artificial intelligence model capable of predicting up to 130 serious diseases.
The tool, called SleepFM, analyzes brain, heart, muscle and respiratory signals captured during a standard sleep study, or polysomnography, and identifies patterns linked to conditions including dementia, heart attack, heart failure, chronic kidney disease, stroke and atrial fibrillation. The findings were published this week in the journal Nature Medicine.
The model was trained on nearly 600,000 hours of sleep data from about 65,000 participants, making it the first large-scale application of AI to analyze comprehensive sleep recordings. Scientists say the system can detect subtle physiological changes long before symptoms appear, opening the possibility of earlier intervention and more personalized prevention strategies. The research highlights how sleep, often overlooked in routine healthcare, could become a powerful window into long-term wellbeing.
How SleepFM turns sleep into a health forecast
SleepFM is built on data collected through polysomnography (PSG), widely regarded as the gold standard for sleep analysis. A PSG test records brain waves, blood oxygen levels, heart rate and breathing, along with eye and leg movements, capturing how different systems of the body behave and interact during sleep. Traditionally, clinicians use this information to diagnose sleep disorders such as sleep apnea or insomnia.
The Stanford team applied artificial intelligence to this vast dataset, allowing the model to learn complex patterns that humans cannot easily detect. Sleep is a dynamic biological process shaped by intricate interactions between the brain, cardiovascular system, respiratory system and muscles. By processing these signals together rather than in isolation, the model identifies physiological signatures associated with elevated disease risk years into the future.
The research suggests that sleep disorders and subtle disruptions in sleep architecture often precede the clinical onset of many illnesses. Scientists note that sleep problems are increasingly recognized as indicators and contributors to psychiatric conditions, neurodegenerative diseases and cardiovascular disorders. Until now, most studies relied on isolated sleep metrics or manual annotations, leaving much of the detailed physiological information underused.
One of the study’s authors emphasizes how little attention sleep has received in artificial intelligence research. “From an AI perspective, sleep is relatively understudied,” said James Zou, Associate Professor of Biomedical Data Science and co-author of the study. By scaling analysis across hundreds of thousands of hours of data, the researchers aim to unlock insights that were previously buried in complex signals.
Crucially, SleepFM does not require repeated nights of monitoring to generate predictions. According to the researchers, even one night of high-quality sleep data can provide enough information for the model to assess long-term health risk, making it potentially more practical for clinical use if integrated into hospital sleep labs or advanced home testing systems in the future.
What this could mean for patients and everyday health
For patients, the promise of AI-powered sleep analysis lies in early detection. Conditions such as heart disease, kidney disease and neurodegenerative disorders often develop silently over many years. If clinicians can identify elevated risk well before symptoms emerge, they may be able to recommend lifestyle changes, closer monitoring or preventative treatments that delay or reduce disease progression.
The study also highlights the broader role sleep plays in maintaining overall health. While many people associate sleep mainly with energy levels or mood, researchers increasingly view it as a critical regulator of cardiovascular, metabolic and neurological function. Poor sleep quality or disrupted sleep patterns can signal underlying physiological stress that may not yet be visible through routine medical tests.
Artificial intelligence helps overcome one of the biggest challenges in sleep science: the sheer volume and complexity of data generated by polysomnography. A single overnight study can produce hours of multi-channel recordings, making comprehensive analysis time-consuming and technically demanding. Automated systems like SleepFM can rapidly process this information and surface clinically relevant insights.
Despite the promising results, the researchers caution that the technology is not intended to replace medical diagnosis. Instead, it could serve as a decision-support tool for clinicians, guiding further testing or preventive care. Additional validation across diverse populations and healthcare settings will be needed before the model can be widely adopted.
The findings also raise questions about how future healthcare systems might integrate sleep data into routine medical records. As wearable devices and home sleep monitoring become more sophisticated, large-scale datasets could further refine predictive models and make personalized risk assessment more accessible. At the same time, ethical considerations around data privacy and algorithm transparency will remain central.
For now, the study reinforces a message that resonates beyond the laboratory: sleep is not just a passive state of rest, but a rich source of biological information. As artificial intelligence continues to decode what happens overnight, a simple night’s sleep may soon offer valuable clues about long-term health, shifting how people and clinicians think about prevention, monitoring and the hidden signals the body sends while we rest.
Photo Credit: Hilton
