A new artificial intelligence model has been created that can predict the risk of serious diseases based on data from sleep laboratories, even before the first symptoms appear, according to media reports.
The artificial intelligence (AI) can assess the likelihood of more than 130 diseases—from heart attacks to breast cancer—based solely on data collected over one night in a sleep laboratory. The program does not establish causes but only identifies correlations. SleepFM, the new AI model, analyzes brain waves, heart rate, breathing, and other aspects of bodily activity during sleep.
According to James Zou, an associate professor at Stanford University and one of the study's authors, the model can predict diseases years before the first signs appear. SleepFM was developed by a team led by Rahul Tapa, a biomedical data expert at Stanford, and trained on hundreds of thousands of hours of data from sleep laboratories.
From Sleep Signals to Disease Predictions
Polysomnography is a research method that allows tracking the functioning of various body systems during sleep. During one night, parameters of brain, heart, respiratory system, and other organ functions are recorded. Approximately 585,000 hours of data from 65,000 patients, primarily examined at the Stanford Sleep Medicine Center, were used to train the SleepFM model.
During the preliminary training process, the AI learned to analyze and statistically interpret data on brain and heart signals during sleep. After that, the model was refined to diagnose conditions such as apnea, and its accuracy is comparable to well-known models like U-Sleep and YASA, which analyze electroencephalograms (EEG).
The researchers compared sleep data with electronic medical records from the past 25 years and identified 130 diseases whose risk could be predicted with moderate to high accuracy. As noted by Rahul Tapa, this confirms that routine sleep measurements open new opportunities for monitoring long-term health.
The accuracy of predicting diseases such as dementia, Parkinson's disease, and certain types of cancer is particularly high. Sebastian Buschger, a sleep expert from the Lamarr Institute, believes that AI can be trained to predict a wide range of diseases if the relevant data is available.
What AI Analyzes During Sleep
The analysis shows that heart signals are particularly important for predicting cardiovascular diseases, while brain signals are crucial for neurological disorders. However, the most informative is the combination of these signals, where the EEG indicates stable sleep while the heart rate may be active.
This mismatch may indicate hidden diseases that can develop before symptoms appear. A sleep specialist from Dortmund emphasizes that while the correlations provided by AI are primarily statistical, experts must confirm causal relationships.
Reliability of Laboratory Data
The model primarily relies on data from laboratories where people with sleep problems are referred, which may limit its universality. Researchers are testing the model based on several groups, but it lacks sufficient information about people without sleep disorders and from regions with less developed healthcare.
Prospects for Diagnosis and Therapy
SleepFM focuses on identifying correlations rather than the causes of diseases. According to Matthias Jacobs, a computer scientist from the Technical University of Dortmund, most AI methods are not capable of establishing causal relationships, but statistical correlations can still be useful for diagnosis and therapy.
Machine learning methods allow computers to learn from data examples and identify patterns. Jacobs believes that even based on statistical data, success in diagnosis can be achieved.
AI as an Assistant, Not a Replacement for Doctors
Models like SleepFM simplify the analysis of large amounts of data, allowing doctors to assess sleep stages and apnea more quickly and accurately. This frees up time for patient communication, according to Matthias Jacobs.
Sebastian Buschger emphasizes the importance of collaboration among specialists: AI can assist in therapy planning, but final decisions should be made by doctors who consider many factors. Thus, AI serves as a tool, and the responsibility for diagnosis and treatment remains with healthcare professionals.
The question of whether the observed patterns can indicate biological mechanisms of diseases remains open, but researchers see great potential for further studies.
