AI Predicts Risk of 130 Diseases From a Single Night’s Sleep
A new artificial intelligence model, SleepFM, can estimate an individual’s future risk for a wide range of conditions – including Parkinson’s disease, dementia, and various cancers – based on physiological data collected during just one night of sleep.
A groundbreaking new artificial intelligence model is poised to revolutionize preventative healthcare. Developed by researchers at Stanford University and detailed in a study published in early January in Nature Medicine, SleepFM analyzes sleep data to predict the likelihood of developing approximately 130 diseases years before symptoms manifest. This innovation underscores the growing recognition of sleep as a critical biomarker for long-term health.
How SleepFM Works: Decoding the Signals of Sleep
SleepFM was trained on an unprecedented dataset of nearly 600,000 hours of sleep data gathered from 65,000 individuals. The process, known as polysomnography, involves using a variety of sensors to meticulously monitor brain waves, heart activity, breathing patterns, muscle tension, and eye and leg movements throughout the sleep cycle. Data collection was primarily conducted at Stanford University’s Sleep Medicine Center in California.
Initially, the AI was exposed to “normal” sleep signals, establishing statistical averages. Subsequently, SleepFM learned to identify different sleep stages and recognize patterns associated with sleep disorders like sleep apnea, a condition characterized by repeated interruptions in breathing during sleep. Researchers then correlated this sleep data with electronic health records spanning 25 years, identifying connections between polysomnography measurements and later health diagnoses.
“Our results reveal that many conditions – including stroke, dementia, heart failure and all-cause mortality – are highly predictable from sleep data, further reinforcing the potential of sleep as a powerful biomarker for long-term health,” explained Rahul Thapa, a PhD student in biomedical data science and co-lead author of the study.
Heart and Brain Signals: Key Predictors of Future Health
Advanced algorithmic analysis revealed that heart signals during sleep are particularly indicative of future cardiovascular disease, while brain signals are more strongly associated with the development of neurological and psychological disorders. However, the most accurate predictions arise from a comprehensive analysis of all physiological signals. For instance, a discrepancy between stable brain activity and an “awake” heart rhythm may signal underlying physical stress indicative of early-stage disease.
According to a German expert on machine learning at Dortmund’s Technical University, Sebastian Buschjäger, who was not involved in the SleepFM project, “In principle, an AI model can be trained for a very large number of possible predictions, provided the basic data is available.” He emphasized the importance of interdisciplinary collaboration, stating, “If our colleagues in sleep medicine suspect a connection, we AI specialists can incorporate this into a predictive system, and conversely, [we can] provide indications of where connections might exist.”
AI as a Tool, Not a Replacement
While SleepFM can identify statistical correlations, experts caution that interpreting the meaning of these correlations and establishing causal relationships requires the expertise of medical professionals. Artificial intelligence, at this stage, serves as a powerful support tool, not a replacement for clinical judgment.
A video demonstrating how daily life affects sleep can be found here: In Good Shape — How sleep is affected by daily life.
Limitations and Future Directions
The current model’s predictions are largely based on data from individuals referred to sleep labs – those already experiencing sleep problems and residing in areas with access to advanced medical technology. This introduces a potential bias, as the model may not accurately reflect the health risks of the general population, particularly those without diagnosed sleep disorders or those from less affluent regions. The researchers acknowledge that SleepFM cannot determine the cause of a disease, only identify patterns associated with later diagnoses.
“Most AI methods do not learn causal relationships,” explained computer scientist Matthias Jakobs from Dortmund’s Technical University, who is also involved in sleep data analysis through the Sleepwalker project.
Despite these limitations, the potential for improved diagnosis and therapy is significant. Models like SleepFM can streamline the analysis of sleep stages and apneas, freeing up clinicians to focus on patient interaction. Furthermore, identifying consistent links between specific sleep signals and diseases could provide valuable insights into the underlying biological processes disrupted in early stages of illness.
A video discussing insomnia and potential treatments can be found here: Insomnia: What to try, and what medications to avoid.
Ultimately, the development of SleepFM represents a major step toward leveraging the power of artificial intelligence and the wealth of data generated during sleep to proactively address health risks and promote well-being, extending far beyond the confines of sleep labs.
