Sleep brainwaves can predict dementia risk years before symptoms appear

UC San Francisco / JAMA Network Open

What if your brain were aging faster than the rest of you — and the evidence showed up every night while you slept?

A study published March 19 in JAMA Network Open, led by researchers at UC San Francisco and Beth Israel Deaconess Medical Center, suggests that fine-grained patterns in sleep brainwaves can reveal accelerated brain aging years before dementia sets in. And the signal is invisible to conventional sleep tracking.

The gap between brain age and birthday age

The research team built a machine-learning model that analyzes 13 microstructural features of brain waves captured by EEG (electroencephalography) during sleep. They applied it to recordings from approximately 7,000 participants, aged 40 to 94, enrolled across five separate studies. None had dementia at baseline. Follow-up ranged from 3.5 to 17 years, during which roughly 1,000 developed the disorder.

The core finding: when the model estimated someone's brain to be older than their chronological age, dementia risk rose. Specifically, for every 10-year increase in the gap between estimated brain age and actual age, the risk of developing dementia climbed by nearly 40%. The reverse held too — participants whose brains appeared younger than their years had lower risk.

That relationship persisted after the researchers adjusted for education, smoking, body mass index, physical activity, other health conditions, and even genetic risk factors for dementia. Whatever the brainwave patterns were capturing, it was not simply a proxy for poor health habits.

Why standard sleep metrics miss it

Here is what makes this finding particularly striking. Earlier pooled analyses of several of the same participant cohorts found no meaningful links between dementia risk and traditional sleep measures — things like total time in deep sleep, REM duration, or overall sleep efficiency. The metrics that fitness trackers and sleep apps report simply did not predict who would develop dementia.

"Broad sleep metrics don't fully capture the complex multidimensional nature of sleep physiology," said senior author Yue Leng, MBBS, PhD, associate professor of psychiatry at the UCSF School of Medicine.

The machine-learning model digs deeper. It examines features like the power and frequency of delta waves — the slow, rolling oscillations of deep sleep — and sleep spindles, brief bursts of fast-frequency activity tied to memory consolidation. Both are known to play roles in brain health, but measuring their microstructure requires computational tools that traditional sleep staging does not use.

Sudden spikes and a surprising protective signal

Among the more unexpected findings was the role of a statistical property called kurtosis — essentially, how sharply peaked the brainwave signal is. Large, sudden spikes in the EEG, which produce high kurtosis values, were associated with a lower risk of dementia. The mechanism is not yet clear, but it suggests that certain types of neural "burstiness" during sleep may reflect healthy brain function that smoothed-out signals do not.

This is the kind of detail that gets lost when sleep is reduced to a handful of stage percentages. The brain does not simply cycle through light, deep, and REM sleep like a washing machine through its settings. Within each stage, the electrical activity carries information about synaptic health, neural connectivity, and — apparently — long-term cognitive trajectory.

From sleep lab to wearable sensor

The practical question is whether this could ever work outside a research setting. Traditional sleep EEG requires a clinical visit, electrodes glued to the scalp, and a technician to interpret the results. That is not scalable for population-level screening.

But the researchers believe the approach could eventually translate to wearable devices. Consumer-grade EEG headbands already exist, and while their signal quality does not match clinical equipment, the gap is narrowing. If a simplified version of this model could run on home-collected data, it would open a path to detecting dementia risk years before symptoms — in people who would otherwise never be tested.

"Brain age is calculated from sleep brain waves," Leng said. "We know that brain activity during sleep provides a measurable window into how well the brain is aging."

Can better sleep slow brain aging?

The findings raise an obvious question: if abnormal sleep brainwaves predict dementia, could fixing the sleep fix the brain? The researchers are cautious. Leng noted that earlier studies found treating sleep disorders can change sleep-related brainwave patterns, but that does not prove it alters dementia trajectory.

"Better body management, such as lowering body mass index and increasing exercise to reduce the likelihood of apnea, may have an impact," said first author Haoqi Sun, PhD, assistant professor of neurology at Beth Israel Deaconess Medical Center, who developed the model. "But there's no magic pill to improve brain health."

That is an honest assessment. Sleep interventions might improve brainwave architecture, but whether that translates to reduced dementia incidence is an unanswered question requiring prospective trials. The current study is observational — it identifies an association, not a cause.

What remains uncertain

Several caveats apply. The model was trained and tested on data from five studies, but all participants came from research cohorts, which tend to be healthier and more educated than the general population. Whether the same brain-age signal performs equally well in more diverse groups is untested. The 13 EEG features the model uses are well-established in sleep neuroscience, but how they interact to produce the brain-age estimate is not fully transparent — a common trade-off with machine-learning approaches.

The follow-up periods also varied substantially across the five cohorts. And while the model controlled for genetic risk, it used APOE status as a covariate rather than stratifying by it, so we do not know whether the brainwave signal is equally predictive across genetic risk groups.

Still, the scale of the study — 7,000 participants, five independent cohorts, up to 17 years of follow-up — gives the results weight that smaller studies could not provide. And the fact that brain age predicted dementia where conventional sleep metrics did not suggests the model is capturing something genuinely new.