News + Trends
Pink noise as a sleep aid: why you risk REM sleep with it
by Anna Sandner
AI recognises later illnesses in sleep data
4.3.2026
Translation: machine translated
Can sleep tell you whether you will have a heart attack or cancer later in life? An AI analysed data from sleep laboratories and discovered patterns that indicate later illnesses.
Could a single night in the sleep lab provide clues to diseases that only appear years later? Researchers at Stanford University have developed an AI model that analyses sleep data and recognises patterns in it that are associated with the later onset of numerous diseases. The team cites around 130 diagnoses for which the model is said to have achieved reliable predictive values. These include dementia, Parkinson's disease, heart attacks, heart failure, certain types of cancer and overall mortality, as reported in «Nature Medicine».
The researchers trained the AI SleepFM with data from polysomnographies - nightly measurements in the sleep laboratory in which sensors record brain waves, heart activity, breathing, muscle tension and leg and eye movements. The majority of this data comes from the Stanford Sleep Medicine Centre, a clinic for people with sleep disorders. Other data sets from the USA and Europe were also used.
For the AI training, the scientists used more than 585,000 hours of sleep recordings from around 65,000 people. During pre-training, the team taught the AI how typical sleep signals are related. To do this, the model was given short extracts from a night in which brain, heart, breathing and muscle activity were recorded simultaneously. One signal was then artificially faded out in each case. The AI was supposed to recognise which other signals belonged together - and which did not. In this way, it developed a basic understanding of how different bodily functions normally relate to each other during sleep. After pre-training, the researchers used this basic understanding to refine the AI for very practical applications, such as recognising sleep stages or diagnosing sleep apnoea. In these standardised tests, SleepFM achieved an accuracy comparable to that of existing methods. The researchers then linked the sleep data with the participants' electronic long-term health records. In this way, they investigated whether a single night's sleep could be used to predict later illnesses. From more than 1000 disease categories, the team identified 130 that the AI predicted with high accuracy based on the sleep data. According to the study, SleepFM predicted Parkinson's, dementia, heart attack, prostate and breast cancer particularly well.
Surprisingly, the researchers found that heart signals play a greater role in the prediction of heart disease and brain signals in the prediction of mental illness. However, the most accurate predictions were made possible by combining certain data: «We obtained the most information for predicting diseases by comparing different channels», says sleep researcher Emmanuel Mignot, who was involved in the study and discovered the cause of narcolepsy in 1999. According to this study, measurement data that is not synchronised - for example, if the measured brain activity shows typical sleep patterns while heart signals are more similar to a waking state - could herald health problems.
The work fits into a growing field of research that considers sleep as a marker for health. For the first time, however, it systematically utilises the entire range of measured signals. However, the authors also emphasise the limitations of their approach: the model recognises statistical correlations, not causes, and is therefore not suitable for individual diagnoses or therapy decisions. In addition, most of the data comes from specialised sleep clinics, which means that people without sleep complaints or with limited access to medical care are underrepresented. Whether the results can be transferred to the general population and which biological mechanisms are behind the patterns remains to be seen. Nevertheless, the approach could help in the long term to better utilise the large amounts of sleep data already available for research and prevention.
Spectrum of science
We are partners of Spektrum der Wissenschaft and want to make well-founded information more accessible to you. Follow Spektrum der Wissenschaft if you like the articles.
Header image: Shutterstock
Spektrum der Wissenschaft
Wissenschaft aus erster Hand
dirdjaja@spektrum.deExperts from science and research report on the latest findings in their fields – competent, authentic and comprehensible.
Health
Follow topics and stay updated on your areas of interest
News + Trends
From the latest iPhone to the return of 80s fashion. The editorial team will help you make sense of it all.
Show all
