Can Wearable Activity Rhythms Measure Aging Risk?

A week of wrist-worn activity data can identify a daily rhythm associated with higher mortality risk, dementia risk, and poorer function. It cannot tell you that you have "reversed your biological age." The evidence supports a narrower and more useful claim: people with weaker, flatter, or more fragmented rest-activity rhythms tend to have less physiological reserve and worse aging-related outcomes.

That distinction matters because wearables are very good at producing numbers with an air of medical authority. A rhythm score may help you notice a persistent change in sleep regularity, daytime movement, recovery after illness, or overall activity structure. It does not diagnose frailty, predict an individual's future with clinical precision, or establish that changing the score will change health.

What a wearable rhythm score actually measures

Accelerometers record movement, not circadian biology directly. Researchers turn seven days of wrist movement into summaries of the 24-hour activity pattern. Common measures include:

• MESOR: the rhythm-adjusted average activity level.
• Amplitude: the difference between the active and inactive portions of the day.
• Acrophase: the estimated timing of peak activity.
• M10 and L5: activity during the most active 10 hours and least active five hours.
• Intradaily variability: fragmentation within a day; higher values indicate more switching between rest and activity.
• Interdaily stability: how consistently the pattern repeats from day to day.

In a 2024 study by Shim et al., researchers used a fixed 24-hour cosinor model to build CosinorAge, a mortality-trained age estimate based on MESOR, amplitude, and acrophase. It is important to be precise about that design. The score was trained against mortality risk, not against molecular measures of aging, blood biomarkers, imaging, or a proven biological-age standard. The investigators excluded people with highly irregular rhythms using stability and fragmentation thresholds; those measures were not core inputs to the score.

CosinorAge is therefore a behavioral activity-rhythm summary associated with mortality. Calling it a direct measure of organismal aging would stretch the study beyond its data.

The mortality signal is real, but selective

Shim and colleagues developed the model in UK Biobank and tested it internally, then evaluated it in NHANES using a different device and the opposite wrist. Over a median 8.1 years of follow-up in the UK Biobank analysis, each additional year of CosinorAgeAdvance—a rhythm-derived age older than chronological age—was associated with hazard ratios of 1.08 for all-cause mortality, 1.08 for aging-related mortality, 1.09 for cardiovascular mortality, 1.07 for cancer mortality, and 1.29 for neurodegenerative mortality.

The external NHANES analysis found corresponding hazard ratios of 1.12 for all-cause mortality and 1.11 for aging-related mortality. Replication across devices and wrist placement is encouraging, although the NHANES validation was smaller and recorded 199 deaths. It shows that the association survived some protocol differences; it does not establish a universal clinical threshold.

The pattern behind the score also deserves attention. Across age groups, people classified as faster agers had lower MESOR and lower amplitude. Acrophase delay appeared more clearly among younger faster agers and changed much less among older adults. In other words, the durable signal was a day with less contrast between activity and rest, not simply a later schedule.

The model improved mortality discrimination over an MVPA-only activity-age comparator for most mortality outcomes, especially neurodegenerative mortality. That suggests the timing and shape of activity add information beyond total moderate-to-vigorous physical activity. It does not mean the score identifies a cause of neurodegeneration.

Dementia research points in the same direction

The cognitive evidence is one reason these metrics have attracted attention. In a prospective 2024 UK Biobank cohort study of 91,517 middle-aged and older adults, poorer rest-activity measures predicted dementia or mild cognitive impairment over as long as 7.5 years. Per one-standard-deviation worse metric, the reported hazard ratios were 1.32 for lower amplitude, 1.23 for lower relative amplitude, 1.28 for lower M10, 1.15 for higher L5, and 1.14 for higher intradaily variability.

Those results fit a simple observation: less daytime activity, more nighttime activity, and a more fragmented day tend to travel with worse cognitive outcomes. Other prospective studies found compatible associations in older women, including the Women's Health Initiative analysis and the Study of Osteoporotic Fractures. Gao et al. also linked disrupted rest-activity rhythms with delirium risk and progression to dementia. Delirium is an acute stress test for the brain, so that association is consistent with the idea that these rhythms track neurologic vulnerability.

Consistency does not erase reverse causality. Early neurodegeneration can disrupt sleep, activity, motivation, and daily structure years before a dementia diagnosis. A weak rhythm may contribute to risk, reflect early disease, or do both. Observational cohorts cannot separate those explanations cleanly.

Not every rhythm metric predicts every outcome

This is where wearable-aging marketing usually gets ahead of the science. In the CosinorAge study, higher rhythm age predicted first age-related disease, neurodegenerative disease, diabetes, and hypertension among disease-free UK Biobank participants. After adjustment, associations with incident cardiovascular disease, cancer, and chronic respiratory disease were not statistically significant. The score also performed worse than the MVPA-only comparator for cancer and respiratory-disease incidence.

The dementia results were metric-specific as well. Haghayegh et al. found no significant associations for acrophase or interdaily stability, even though acrophase is one of CosinorAge's three inputs. Phase timing may matter in some contexts, but it does not carry the same predictive value for every endpoint.

Functional results were modest. Shim et al. linked older CosinorAgeAdvance with worse self-rated health, lower EQ-5D scores, worse activities of daily living, and slightly lower grip strength in UK Biobank. The grip-strength association did not replicate in NHANES. Separately, the MrOS Sleep Study associated disrupted rhythms with poorer physical function and mobility in community-dwelling older men. These findings support a frailty-risk interpretation better than a universal disease clock.

Illness and recovery: useful monitoring, unproven treatment target

Rhythm disruption also appears in people already under physiological strain. A 2025 cancer cohort study associated wearable-measured rest-activity rhythms with mortality risk, while a chronic kidney disease study associated disrupted rhythms with all-cause mortality. These disease-specific findings broaden the setting in which the signal appears. They cannot show whether a disrupted rhythm worsened the disease or simply recorded its burden.

Recovery evidence is thinner. In a 2024 comparative rehabilitation study, 25 post-stroke patients and 25 age- and sex-matched inpatient controls wore an actigraphy device for seven days. The post-stroke group showed different rest-activity rhythm patterns during acute rehabilitation. The study did not test whether improving a rhythm measure improved function, shortened rehabilitation, or changed any clinical outcome.

Why a one-size-fits-all score can mislead

Baseline rhythm measurements differ across populations. In a 2021 NHANES cross-sectional analysis, women had higher amplitude and MESOR than men; older adults had lower amplitude and MESOR with earlier acrophase; Hispanic adults had the highest amplitude and MESOR; and non-Hispanic Black adults had the lowest amplitude and stability. These differences do not make any group biologically older. They show why a single age mapping or cutoff can misclassify people before illness even enters the equation.

Device and processing choices add another problem. UK Biobank used an Axivity device on the dominant wrist, while NHANES used an ActiGraph device on the non-dominant wrist. Seven days may capture a useful snapshot, but the long-term within-person stability of these summaries remains uncertain. Hardware, wrist placement, algorithms, shift work, depression, medication burden, chronic illness, and socioeconomic strain can all shape the output.

Wearable rest-activity rhythms are promising digital biomarkers of vulnerability. They are not validated causal aging clocks or stand-alone clinical decision tools.

What to do with your own wearable data

A chronically flatter or more fragmented activity pattern is a prompt to look closer, especially if it accompanies declining exercise tolerance, sleep disruption, persistent fatigue, illness, cognitive changes, or slower recovery. Review trends against your own baseline rather than treating an app's age score as a lab result. Discuss new or substantial changes with a clinician when symptoms warrant it.

Do not use a rhythm score to diagnose disease, rank supplements, or claim that a sleep hack reversed biological age. No intervention trial in this evidence base shows that changing light exposure, exercise timing, meal timing, sleep timing, or shift-work exposure changes CosinorAge itself or improves hard outcomes through that pathway. The useful question is not "What is my wearable age?" It is whether a sustained change in your daily rhythm reveals a change in health, recovery, or routine worth investigating.