Researchers recently introduced an artificial intelligence (AI) foundation model, ‘SleepFM’, that utilizes sleep data to predict a person’s risk for over 130 distinct health conditions. This breakthrough highlights the power of Sleep Data AI in preventative medicine. Consequently, it transforms how clinicians view polysomnography, the traditional ‘gold standard’ for sleep studies. The model was developed by researchers, including those from Stanford University. They trained it on nearly 600,000 hours of sleep recordings from 65,000 participants.
Polysomnography: A Rich Data Source for Sleep Data AI
Polysomnography (PSG) captures a wide array of physiological signals while a patient sleeps. For eight uninterrupted hours, the sensors record brain activity (EEG), heart function (ECG), and respiratory signals. They also track eye movements and muscle electrical activity (EMG). Senior author Emmanuel Mignot, a Stanford professor, said, “We record an amazing number of signals when we study sleep. It is very data rich.” Traditional sleep analysis uses only a fraction of this data, focusing on diagnosing conditions like sleep apnea or tracking sleep stages. Conversely, the SleepFM model incorporates multiple data streams to uncover complex relationships. The team also developed a new training technique. This technique, called ‘leave-one-out’ contrastive learning, improves the AI’s ability to learn how different physiological modalities relate to each other.
Performance of SleepFM in Disease Prediction
The SleepFM system analyzed over 1,000 disease categories. It successfully identified 130 conditions predictable with ‘reasonable accuracy’ using only the overnight PSG data. Furthermore, the model’s predictions proved particularly strong for several high-impact disease groups. These include cancers, pregnancy complications, mental disorders, and circulatory conditions. The AI achieved a ‘C-index’ score higher than 0.8 in these key areas, demonstrating its reliability.
The C-index, or concordance index, is a common metric. It measures an AI’s predictive performance. Specifically, it assesses the model’s ability to correctly rank which of two individuals will experience a health event first. Therefore, a score of 0.8 means the model’s prediction is correct 80% of the time. The authors highlighted strong C-index scores for several critical conditions. These include dementia (0.85), all-cause mortality (0.84), myocardial infarction (0.81), and heart failure (0.80). Moreover, strong performance was also observed in predicting the risk of Parkinson’s disease and developmental delays. For those interested in advancing their knowledge in related internal medical fields, exploring our Internal Medicine Speciality Courses may be beneficial.
Frequently Asked Questions
Q1: What is the ‘SleepFM’ model and what data does it use?
SleepFM is an AI foundation model developed by Stanford researchers. It uses physiological recordings from a single night of polysomnography (PSG) data—including EEG, ECG, and respiratory signals—to predict a person’s risk for over 130 diseases.
Q2: How accurate are SleepFM’s predictions?
The model’s predictions achieved high accuracy, measured by the C-index (concordance index). For example, it scored 0.85 for dementia, 0.84 for all-cause mortality, and 0.81 for myocardial infarction. A C-index of 0.8 means the model is correct 80% of the time in ranking which of two individuals will experience an event first. Clinicians looking to deepen their expertise in heart-related risk prediction can consider the Postgraduate Diploma In Preventative Cardiovascular.
References
- Researchers develop AI model that predicts disease risk from sleep data – ETHealthworld
- New AI model predicts disease risk while you sleep – Stanford Medicine
- Stanford’s AI Predicts Disease Risk From a Single Night of Sleep – SciTechDaily
Disclaimer: This article was automatically generated from publicly available sources and is provided for informational and educational purposes only. OC Academy does not exercise editorial control or claim authorship over this content. It is not a substitute for professional medical advice, diagnosis, or treatment. Always consult a qualified healthcare provider and refer to current local and national clinical guidelines.