Summary: “Sleep age” is a calculated measure that reflects the quality of a person’s sleep and appears to be associated with overall health and risk of death.
Source: Stanford
Numbers can reveal important truths. From credit scores to biomarkers, measurable values help predict outcomes such as loan approval or disease risk. Stanford Medicine researchers have identified another metric with predictive power: sleep age.
Sleep age is an estimated age derived from objective sleep characteristics. By comparing an individual’s sleep patterns—such as breathing, heart rate, limb movements and stages of sleep—to typical patterns across ages, researchers can assign a sleep age. For example, a 55‑year‑old whose sleep closely resembles the average sleep profile of a healthy 45‑year‑old would be said to have a sleep age of 45.
Emmanuel Mignot, MD, PhD, and colleagues examined roughly 12,000 individual sleep studies that included detailed physiological recordings. Using machine learning, they developed models that estimate a person’s sleep age and identify sleep features most strongly linked to later mortality.
Sleep changes with age, and altered sleep architecture and stability are among the earliest and best‑documented signs of aging and declining health. The encouraging message from this work is that sleep age is not fixed: sleep habits and certain interventions can improve sleep quality and potentially lower sleep age.
The study, led by Mignot, the Craig Reynolds Professor in Sleep Medicine at Stanford Medicine, was published July 22 in npj Digital Medicine. Below is a clear summary of the study’s aims, main findings and practical implications for sleep health.
Why study sleep age?
Sleep disconnects us from sensory input, allowing the brain to cycle through restorative stages. But sleep also involves coordinated changes in breathing and heart rate; disturbances in these systems can be early indicators of health problems. Because humans spend about a third of life asleep, sleep quality is a major component of overall well‑being.
In many disorders, sleep disturbances appear early. For instance, people who later develop Parkinson’s disease may exhibit a distinctive dream‑related movement disorder years before other symptoms arise. This makes sleep a valuable window into future health risks.
Key finding: sleep fragmentation predicts mortality
The strongest predictor of mortality identified in the study was sleep fragmentation—brief arousals that cause micro‑wakefulness multiple times per night without the person remembering them. While the data show a clear association between fragmented sleep and higher mortality risk, the exact biological mechanisms linking fragmentation to mortality remain to be determined.
Sleep fragmentation is distinct from insomnia symptoms in which people consciously notice awakenings. The study highlights fragmentation as an important, measurable biomarker of future health, independent of common sleep disorders such as sleep apnea.
Can individuals measure and improve their sleep age?
The team has made the analytic code available for researchers and clinicians, but running the program requires specialized data and computational skills. Importantly, sleep age is not deterministic: there is wide natural variation in lifespan and health. A higher sleep age does not guarantee a shorter life, just as lower sleep age does not guarantee longevity.
Practical steps that support healthier sleep—and may help lower sleep age—include consistent bed and wake times, appropriate daytime light exposure (preferably natural light), a dark and quiet sleep environment, regular exercise that is not too close to bedtime, and avoiding alcohol, caffeine and heavy meals near sleep. Treating diagnosed sleep disorders is also essential.
How was sleep age calculated?
Researchers trained deep learning models on polysomnography data: multi‑channel recordings that capture brain activity, breathing, heart rate and movement during sleep. The model learned patterns associated with chronological age and then assigned a predicted sleep age to additional studies. The difference between predicted sleep age and actual age—the age estimate error—was then used to evaluate associations with long‑term outcomes, including mortality.
The authors found that people whose sleep age exceeded their chronological age had a higher risk of death in subsequent years. Other research links poor sleep to conditions such as sleep apnea, neurodegenerative disease, obesity and chronic pain, though the causal relationships between sleep and these conditions remain under study.
Next steps in sleep and mortality research
Future work aims to use large collections of sleep studies to predict specific disease outcomes—such as heart attacks, strokes and Alzheimer’s disease—and to test whether sleep‑based risk estimates can prompt early interventions. The researchers plan to expand analyses using a larger dataset of hundreds of thousands of sleep studies to strengthen mortality and disease predictions.
If sleep studies can reliably forecast a person’s risk for cardiovascular events or neurodegeneration, clinicians could potentially use that information for earlier preventive care—a prospect with major public‑health implications.
About this sleep and mortality research news
Author: Emily Moskal
Source: Stanford
Contact: Emily Moskal – Stanford
Image: The image is in the public domain
Original Research: Open access. “Age estimation from sleep studies using deep learning predicts life expectancy” by Andreas Brink‑Kjaer et al., npj Digital Medicine
Abstract
Age estimation from sleep studies using deep learning predicts life expectancy
Sleep disturbances increase with age and are predictors of mortality. The authors present deep neural networks that estimate age and mortality risk from polysomnograms (PSGs).
Aging was modeled using 2,500 PSGs and tested in 10,699 PSGs from men and women across seven cohorts aged 20 to 90. Ages were estimated with a mean absolute error of 5.8 ± 1.6 years, compared with 14.9 ± 6.29 years for basic sleep scoring measures.
After adjusting for demographics, sleep variables and health covariates, each 10‑year increment in age estimate error (AEE) was associated with a 29% higher all‑cause mortality rate (95% confidence interval: 20–39%). An AEE change from −10 to +10 years corresponded to an estimated reduction in life expectancy of 8.7 years (95% confidence interval: 6.1–11.4 years).
Greater AEE was largely reflected by increased sleep fragmentation, suggesting that fragmentation is an important biomarker of future health independent of sleep apnea.