How does cognitive decline relate to time-of-day physical activity patterns?

Two timelines joined by a wearable device

The study asks whether the pace of cognitive decline shows up later in how people move across a day. It draws on the Atherosclerosis Risk in Communities cohort, following 2,071 cognitively unimpaired older adults across visits from 2011 to 2022 to estimate each person's cognitive trajectory, then linking those trajectories to wrist-worn accelerometer data collected a median of 8.2 years later in 811 of them. Mean baseline age was 73.1 years, and the accelerometry sample was 60.5% female and 79.4% White.

Cognitive performance was measured in three domains, language, memory, and executive function, plus a general composite, and each was scaled so that one unit equals one standard deviation. Decline was summarized as a rate per decade, with median declines of 0.51 SD for general cognition, 0.45 for language, 0.36 for memory, and 0.54 for executive function. Physical activity came from a triaxial wrist device worn for a median of six valid days, aggregated into one-minute activity counts and then summarized in one-hour bins across the 24-hour cycle.

Five activity metrics measure different things

The paper separates activity into dimensions that are easy to conflate. Total activity counts capture how much movement occurred, while sedentary time counts minutes below 1,853 counts per minute, a threshold validated for older adults wearing a wrist device. Activity fragmentation is the probability of switching from active to sedentary, so a higher value means active bouts break up sooner; sedentary fragmentation is the reverse, the chance of breaking up sitting. Activity variability, measured by the root mean square of successive differences (RMSSD), captures minute-to-minute fluctuation in activity counts rather than total volume.

Keeping these apart matters because the paper's headline result is not about how much people move overall. It is that the quality and timing of movement, especially variability, carried the strongest association with prior cognitive decline.

Midday carries the strongest association

Using function-on-scalar regression, which estimates an association separately at each minute of the day, the authors found a consistent shape across all cognitive domains: little association overnight from 11 PM to 6 AM, rising through the morning, a peak around 11 AM to 1 PM, and fading in the evening. In the fully adjusted model, each 10% slower annual cognitive decline was associated with 1.5% to 2.0% higher midday activity, for example 27 more counts per minute at noon for general cognition and 36 for executive function, both p < 0.001, with negligible evening and overnight associations.

The window from 10 AM to 3 PM was jointly significant with all five activity metrics for general cognitive decline, peaking near 12:30 PM. Within that window, each 10-percentage-point slower decline in general cognition was associated with higher variability (0.049 SD, 95% CI 0.036 to 0.062), higher total activity (0.036 SD), lower activity fragmentation (−0.035 SD), lower sedentary time (−0.031 SD), and higher sedentary fragmentation (0.027 SD), all p < 0.001.

Short windows rival full-day measures

A practical question for anyone considering wearable monitoring is how much of the day must be captured. The paper compares the 10 AM to 3 PM window and an even shorter 11 AM to 1 PM window against 24-hour aggregates. Across domains, the key-window estimates were consistently stronger, with a median t-statistic ratio of 1.12 (interquartile range 1.03 to 1.33) relative to the 24-hour measure. For general cognition, the sedentary-time association was 24% stronger over the key window than over 24 hours, and the sedentary-fragmentation association was 50% stronger.

The 11 AM to 1 PM window was close to parity: for general cognition, variability measured over those two hours was associated with cognitive decline as strongly as the 24-hour measure, 0.049 SD versus 0.049 SD per 10-percentage-point slower decline. This is an association-strength comparison within the same cohort, not evidence that a two-hour window would detect decline in a new population.

Cognitive domains differ in timing and shape

The executive function trajectory showed the strongest and most temporally extensive associations, with joint significance spanning 9.0 hours per day, from 8:23 AM to 5:26 PM, compared with 5.3 hours for language and 6.4 hours for memory. Its peak association with activity fell between 12:30 PM and 2:30 PM, and its variability association reached 0.065 SD (95% CI 0.044 to 0.086). The authors interpret this breadth as consistent with executive function's role in managing goal-directed daily activities.

The memory trajectory peaked earlier, in the late morning from 10:30 to 11:30 AM, which the authors link to memory-dependent tasks such as household finance and cooking concentrated in that period. The language trajectory behaved differently: its associations were non-linear, with unfavorable activity patterns appearing mainly when decline was faster than the population median, while decline at or slower than the median showed minimal association with activity. That threshold pattern means a study averaging over all decline rates could miss a language-related signal.

Variability outperforms volume and fragmentation

Across metrics, activity variability showed the strongest associations and remained significant until midnight, with a mean joint-significant band of 15.5 hours, compared with 7 hours for total activity. Comparing the fastest-declining tertile with the slowest, the largest gap for general cognition was again variability, at 0.70 SD over 24 hours (95% CI 0.52 to 0.88), ahead of total activity at 0.54 SD and activity fragmentation at −0.45 SD. A dose-response gradient across tertiles appeared for general cognition, memory, and executive function, with language the exception, showing similar patterns in the middle and slowest tertiles.

Adding 24-hour variability to a base model of demographics, lifestyle, APOE ε4, and health status improved discrimination between the fastest- and slowest-declining tertiles from an AUC of 0.668 to 0.744, a gain of 0.076 and the largest among the activity metrics. The authors read higher variability as more dynamic, adaptive engagement with daily demands, though they note this interpretation is not directly tested.

Where the monitoring signal is strongest

For readers weighing whether time-of-day activity could serve as a monitoring signal, the paper's specific priority is clear: look at midday, and look at variability first. The strongest and most temporally extensive associations came from activity variability and the executive function trajectory, with the informative window concentrated between 10 AM and 3 PM and a two-hour 11 AM to 1 PM window performing about as well as a full day. The language trajectory is the case where a threshold matters: activity differences appeared mainly when decline was faster than the population median, so a null finding in a broadly average group would not rule out a signal at the faster-declining end.

The practical implication the authors draw is that wrist-worn activity patterns, particularly variability around midday, could offer a non-invasive way to help monitor cognitive decline before clinical impairment emerges. That remains a hypothesis about monitoring potential, supported by association strength and discrimination gains in this cohort, not by demonstrated prediction in a new population. The accelerometer data were collected at a single time point, so the analysis cannot establish direction between cognitive decline and activity, and the sample was predominantly White and well-educated, limiting transfer to other populations. The most defensible reading is that the timing and quality of movement, not just its total amount, carry information worth testing prospectively.

Sources used in this answer