Overview
Digital sleep-wake cycle patterns measured with wearable accelerometers may help identify older adults at higher risk of dementia years before diagnosis. In this large cohort study, researchers examined whether movement-based measures of rest, activity timing, and sleep continuity were associated with future dementia and whether they added value to risk prediction models that already included age and established risk factors.
The findings suggest that certain disrupted daily activity and sleep patterns are linked to a higher chance of developing dementia. The added predictive value was modest but statistically significant, meaning these digital measures could become useful as part of broader risk assessment tools in the future.
Why sleep-wake rhythms matter
Sleep and activity follow a circadian pattern, a roughly 24-hour cycle regulated by the brain’s internal clock. In healthy aging, people typically maintain a fairly stable rhythm: sleep occurs mainly at night, activity is concentrated during the day, and transitions between rest and movement are predictable.
In the preclinical phase of dementia, however, these patterns may become less stable. Older adults who later develop dementia may show fragmented sleep, more daytime inactivity, irregular naps, earlier waking, or frequent switches between resting and moving. These changes may reflect early brain network dysfunction, reduced circadian regulation, or lifestyle and health changes that precede a formal diagnosis.
Study design
The investigators used data from two prospective UK cohort studies:
UK Biobank, used as the derivation cohort, included accelerometer data collected between 2013 and 2015.
Whitehall II, used as the validation cohort, included accelerometer data collected between 2012 and 2013.
Participants were 60 years or older, had no dementia at baseline, and had valid accelerometer and covariate data. The main outcome was incident all-cause dementia, identified through linked electronic health records.
The analysis was conducted in two stages. First, the researchers extracted 36 sleep-wake cycle metrics from accelerometer recordings. Then, using a machine learning approach, they selected the measures that best predicted dementia risk and combined them into composite components.
What the accelerometer measured
Wearable accelerometers do not directly measure sleep like a formal sleep study would. Instead, they capture movement patterns that can be used to infer how active or still a person is throughout the day and night. From these data, researchers can estimate:
How long a person sleeps
How often sleep is interrupted
How active they are during the day
Whether activity is clustered or fragmented
The timing of waking and resting periods
The likelihood of switching from activity to rest, or from wakefulness to sleep
These measures provide a digital picture of daily rhythm that is more continuous and objective than self-reported sleep questionnaires.
Key findings in UK Biobank
In the UK Biobank derivation sample, 53,448 participants were included. Their mean age was 67.5 years, 54.2% were women, and average follow-up was 7.8 years.
Nine sleep-wake metrics were grouped into two major components.
Higher values in component 1 reflected a pattern of less healthy daytime activity: shorter total durations of moderate-to-vigorous physical activity, fewer bouts of such activity, more time spent in low-intensity activity, lower diversity of activity intensities, and a greater probability of shifting from activity to rest during the day.
Higher values in component 2 reflected more disturbed sleep and circadian timing: more extreme sleep durations, longer periods of wakefulness during sleep, lower probability of transitioning from wakefulness to sleep, and earlier waking time.
Both components were associated with a higher future risk of dementia:
Component 1: hazard ratio 1.43, 95% CI 1.33-1.54
Component 2: hazard ratio 1.10, 95% CI 1.04-1.17
In simple terms, participants with more disrupted movement and sleep patterns were more likely to be diagnosed with dementia during follow-up.
How much did prediction improve?
The researchers also tested whether adding these digital sleep-wake measures improved dementia prediction beyond traditional factors such as age, education, behavioral risks, and health-related variables.
The answer was yes, but the gain was modest. The C index, a measure of how well a model separates people who will and will not develop dementia, increased by 0.018. This is a small improvement, but it was statistically significant.
Importantly, when compared with an age-only model, the addition of the sleep-wake components improved prediction by an amount equivalent to adding APOE genotype, a well-known genetic risk marker for Alzheimer-related dementia. That does not mean the new metrics replace genetics or clinical assessment. Rather, it shows they may contribute meaningful information when combined with other predictors.
Validation in Whitehall II
The findings were tested in the Whitehall II cohort, which included 3,965 participants with a mean age of 69.4 years and a mean follow-up of 10.6 years. Results were consistent with those seen in UK Biobank, strengthening confidence that the associations were not specific to one cohort.
Replication in an independent population is especially important in prediction research, because it shows that the pattern is more likely to be robust and potentially generalizable.
What these results may mean biologically
Several mechanisms could explain why disrupted sleep-wake rhythms are linked to dementia risk.
First, circadian disruption may be an early marker of brain changes affecting memory and sleep regulation areas, including the hypothalamus and related neural circuits.
Second, poor sleep quality and irregular rhythms may influence clearance of metabolic waste from the brain, potentially affecting the buildup of proteins associated with neurodegeneration.
Third, disrupted sleep and low daytime activity may be connected to other conditions that increase dementia risk, such as depression, vascular disease, frailty, metabolic problems, or social isolation.
It is also possible that the sleep-wake changes are not a cause of dementia but an early sign of subtle disease processes already underway. This study supports association and prediction, but it does not prove causation.
Clinical implications
The study suggests that simple, scalable wearable devices may help identify older adults who deserve closer monitoring for cognitive decline. Because accelerometers are relatively inexpensive and easy to use, they could one day complement current dementia risk tools in primary care, research settings, or population screening programs.
Potential uses include:
Flagging individuals for more detailed cognitive assessment
Supporting risk stratification in older adults
Helping researchers identify preclinical disease patterns
Tracking changes in activity and sleep over time
However, these measures are not ready to be used alone as a diagnostic test. Their predictive gain was modest, and many people with abnormal patterns will never develop dementia. Conversely, some people with normal-looking rhythms will still develop the condition.
Limitations
The study has several limitations that are important when interpreting the findings.
The cohort design shows association, not causation.
The participants were from UK-based cohorts, which may limit generalizability to other populations.
Accelerometers infer sleep from movement, so they do not provide the same detail as polysomnography or clinical sleep evaluation.
The outcome was all-cause dementia, meaning the study was not focused on specific subtypes alone.
Although the models adjusted for known factors, residual confounding may remain.
In addition, the use of one-week wearable data may not fully capture long-term sleep-wake patterns across years.
Bottom line
In this large study of older adults, accelerometer-derived sleep-wake cycle measures were associated with future dementia and modestly improved risk prediction beyond age and traditional risk factors. The results were replicated in an independent cohort, suggesting that digital rhythm markers may eventually help identify people at higher risk earlier than standard clinical assessment alone.
Further studies are needed to determine whether these measures can be used in routine practice, how they perform across different populations, and whether improving sleep and circadian regularity can reduce dementia risk.
