(This is an excerpt of the Health Rounds newsletter, where we present latest medical studies on Tuesdays and Thursdays.)
By Nancy Lapid
Jan 7 (Reuters) – Artificial intelligence can use brain recordings from a single night in a sleep lab to predict a person’s risk of developing more than 100 health conditions, researchers say.
Known as SleepFM, the AI model was trained on more than half a million hours of sleep data collected via polysomnography from 65,000 participants.
Polysomnography is considered the gold-standard overnight sleep exam that uses various sensors to record brain activity, heart activity, respiratory signals, body movements, eye movements and other data. It is “an untapped gold mine of physiological data,” according to the researchers.
“We record an amazing number of signals when we study sleep,” study leader Dr. Emmanual Mignot of Stanford Medicine in Palo Alto, California said in a statement.
To take advantage of the sleep data trove, the researchers built an AI model and trained it on 585,000 hours of polysomnography data from patients who had their sleep assessed at various sleep clinics.
First, researchers tested the model on standard sleep analysis tasks, such as classifying different stages of sleep and diagnosing the severity of sleep apnea. SleepFM performed as well as, or better than, state-of-the-art models currently used, they said.
Next, the researchers paired polysomnography data from 35,000 adults and children who had been treated at the Stanford Sleep Medicine Center between 1999 and 2024 with long-term health outcomes of the same participants using their electronic health records.
Among more than 1,000 disease categories analyzed, the model found 130 that could be predicted with reasonable accuracy by a patient’s sleep data. These included all-cause mortality, dementia, heart attack, heart failure, chronic kidney disease, stroke and atrial fibrillation.
For certain cancers, pregnancy complications, circulatory conditions and mental disorders, the AI model’s predictions were correct more than 80% of the time, the team reported in Nature Medicine.
The researchers don’t yet understand exactly what SleepFM is looking at when it makes a specific disease prediction. They are working on figuring that out, and also on ways to further improve the model’s predictions, perhaps by adding data from wearable devices.
FINGER PRICK MIGHT HELP DIAGNOSE ALZHEIMER’S DISEASE
Dried blood samples collected from a finger prick might someday be used to detect Alzheimer’s disease, new research suggests.
Newly available tests that measure biomarkers in blood linked to Alzheimer’s, such as the p-tau217 protein, are revolutionizing research into the disease and its diagnosis, which previously required invasive brain scans and painful spinal fluid tests.
Practical hurdles remain with the newer blood tests that the dried blood method might address, including how samples are handled and stored and the need for trained staff to collect them.
After obtaining a few drops of blood via finger prick in 337 volunteers and letting the blood dry on a card, researchers found that levels of p-tau217 in the samples closely matched results from standard blood tests. High levels of biomarkers on the dried samples also correlated with the presence of the biomarkers in cerebrospinal fluid, with an accuracy of 86%, they found.
Two other biomarkers, GFAP and NfL, were also successfully measured and showed strong agreement with traditional tests, according to a report of the research published in Nature Medicine.
The authors also found that participants were able to successfully obtain the blood samples themselves, without the guidance of study personnel.
The method is not ready for prime time, but the findings suggest that “this simple technique could make large-scale studies and remote testing possible, including for people with Down syndrome, who face a higher risk of Alzheimer’s disease and for other underserved populations,” the authors said in a statement.
(To receive the full newsletter in your inbox for free sign up here)
(Reporting by Nancy Lapid; Editing by Bill Berkrot)
Comments