Daily Movement Activities Are Associated With Within-Person Instability of Cognitive Function in Older Adults: Evidence From an Ambulatory Assessment Study

Abstract

Background and Objectives

Physical inactivity and excessive sedentary time (ST) are associated with poor cognitive health in older adults. However, current literature largely relies on cross-sectional designs or in-lab cognitive assessments, which do not adequately reflect cognitive function in naturalistic settings. Further, existing studies have largely overlooked the variability or the instability in daily cognitive function, which represents a critical marker of cognitive decline. This ambulatory assessment study examined the temporal associations of daily movement behaviors with the mean levels and the variability of cognition among older adults at risk of dementia.

Research Design and Methods

96 older adults from the community (68.3 ± 7.1 years) participated in this 14-day study. They wore an accelerometer and completed smartphone-based cognitive tests up to 4 times per day. The cognitive tests assessed both performance-based and subjective cognition. The movement behaviors collected from the accelerometers include daily light-intensity physical activity (LPA), moderate-to-vigorous physical activity (MVPA), and ST. Mixed-effects location-scale models were applied to estimate the within- and between-person associations of movement behaviors and cognitive outcomes in terms of the mean levels and the degree of variability. A total of 1,269 day-level observations were analyzed.

Results

Older adults’ between-person levels of daily MVPA and steps were associated with better mean cognitive performance and lower variability across cognitive measures. Older adults’ daily LPA was positively associated with subjective cognition in both mean levels and variability. The increases in within-person levels of ST were negatively associated with older adults’ variability of all cognitive outcomes.

Discussion and Implications

Study results suggest that moving more and sitting less in day-to-day life may sustain proximal cognitive health. Applying ambulatory assessments can advance aging research by examining the temporal dynamics between daily movement activities and within-person variability of cognition to inform strategies for promoting healthy aging in daily life.

Translational Significance: This study addresses limitations in the aging literature and advances our understanding of how daily movement behaviors influence both the stability and mean levels of everyday cognitive function, key markers of cognitive health, in older adults. Using ambulatory assessments with mobile devices to capture real-world data of daily movements and cognitive function, findings reveal that greater daily physical activity and less sedentary time are linked to better cognitive performance and reduced cognitive instability. These findings hold significant translational potential by targeting daily lifestyle activities as a practical strategy to sustain cognition and promote healthy aging in broader older populations.

The rapidly growing older adult population has led to a swift increase in individuals affected by cognitive deterioration diseases, such as Alzheimer’s disease and related dementias (ADRD). These brain diseases significantly affect the quality of life and impose a heavy burden on healthcare systems (Livingston et al., 2020). In the absence of effective medical treatments to cure these diseases to date, early preventive strategies are crucial for sustaining cognitive health and reducing disease risk. Regular engagement in physical activity (PA) is suggested as a protective factor of cognitive health in older adults (Erickson et al., 2019; Sofi et al., 2011). However, existing evidence largely relies on cross-sectional and laboratory-based studies that fail to establish the temporal associations between PA and cognitive function in naturalistic settings to provide real-world implications (Bielak et al., 2017).

Real-time assessments of movement behaviors (i.e., PA and sedentary behaviors) and cognitive function, such as using accelerometers, daily diaries, and ecological momentary assessment (EMA) offer opportunities to examine dynamic, time-sensitive associations between PA and cognition (Koo & Vizer, 2019). These methods not only enhance the ecological validity but are also sensitive in detecting early signs of cognitive impairment as well as identifying daily movement behavior targets to support cognitive health (Moore et al., 2017). Further, the intensive longitudinal data collected from these methods provide valuable insights into how the within-person variations of daily PA are associated with cognitive function. For example, an eight-day self-report daily diary study found that when older adults engaged in leisure PA but not activities related to work or household chores, they reported fewer memory failures on the same day as well as the following day (Whitbourne et al., 2008). A 14-day dairy study of young adults found that daily variations in PA (using both self-reports and accelerometers) were positively associated with perceived cognitive function (Fitzsimmons et al., 2014). Recent work also revealed that self-reported PA at all intensity levels (mild, moderate-to-vigorous) within a few hours before a smartphone-based cognitive EMA was positively associated with better processing speed in middle-aged adults (Hakun et al., 2025).

Recent work using accelerometers and smartphone-based cognitive assessments observed that engaging in higher-intensity PA or being more active within specific timeframes (e.g., 20–60 min) before the EMA cognitive tests was linked to better cognitive function in middle-aged adults (Kekäläinen et al., 2023). A longer EMA study (14 days) reported that on days when participants had higher PA (accelerometer-based measure), they had better performance on executive function (smartphone-based measure) after accounting for covariates (Zlatar et al., 2022). Overall, the existing EMA studies supported the positive within-person association between PA and cognition on faster timescales (within day or across days). However, these findings were based mostly on young or middle-aged participants, which may limit the generalizability to late adulthood.

In addition to studying the mean levels of cognitive function, repeated cognitive assessments offer the opportunity to study the within-person variability (or intraindividual variability) in cognitive performance (Salthouse, 2007). This variability is not merely random noise or measurement error (Vaughan & Birney, 2023). Rather, it represents an early behavioral marker of impaired brain health or an increased risk of developing neurodegenerative diseases (MacDonald et al., 2006). Greater within-person variability of cognitive function may reflect negative changes in an individual’s neural plasticity, network dynamics, cortical activation efficiency, or attentional control (MacDonald et al., 2009). One narrative review concluded that the within-person variability is a reliable marker, and possibly a better indicator than the average performance of cognitive assessment for predicting progressive cognitive decline, even when considering key confounders (Costa et al., 2019). Recent EMA studies supported these earlier lab-based findings, showing that greater variability in cognitive tests can differentiate older adults with mild cognitive impairment (MCI) and predict genetic risk for ADRD. One EMA study showed that older adults with MCI status exhibited greater within-person variability on smartphone cognitive tests than their counterparts (Cerino et al., 2021). Another study found comparable results that older adults’ fluctuation in the EMA-assessed processing speed was positively associated with both the clinical AD status and the genetic AD risk (Welhaf et al., 2024).

Despite growing evidence of the cognitive benefits of PA using EMA, limited research has investigated the role of light-intensity PA (LPA) and sedentary time (ST) on cognitive health. Older adults spend the majority of their waking hours in LPA (e.g., walking, household chores, daily living activities) or sedentary behaviors, which creates a critical opportunity to investigate how these dominant everyday movement behaviors influence cognitive outcomes (Amagasa et al., 2021). One large-scale cohort study found that each additional hour of accelerometry-measured LPA was associated with greater brain volume in adults not meeting PA guidelines (Spartano et al., 2019). Two other studies that applied accelerometer-based measures of PA also reported the independent (or even superior) benefits of LPA on proximal or distal cognitive outcomes in older adults (Gothe 2021; Stubbs et al. 2017). A few EMA studies also revealed the links between ST and worse cognitive outcomes. One daily diary study found a within-person association between more self-reported, but not device-measured, ST and lower perceived cognitive function in young adults (Fitzsimmons et al., 2014). Another 14-day EMA study of adults revealed that more time spent in passive sedentary leisure activities (e.g., watching TV, resting) was associated with worse cognitive performance, both within persons and between persons (Campbell et al., 2020).

Emerging EMA studies generally provided findings that being more active and less sedentary in daily life may be beneficial to support cognitive health. However, the impact of daily movement behaviors, especially LPA and ST, on daily cognitive function is not fully understood. The current field also knows less about whether older adults’ daily movement behaviors reduce daily cognitive function variability (i.e., a valid marker of brain health) both within and between individuals. This study utilizes intensive longitudinal data from research-grade accelerometers and mobile cognitive assessments to address these knowledge gaps. Specifically, a novel modeling approach will be applied to investigate how daily movement behaviors are associated with the mean levels, the between-person differences, and the within-person variability in day-to-day cognitive function among older adults at risk of developing ADRD.

Method

Participants

This study recruited older adults (ages ≥60) around the Midlands of South Carolina, United States. For inclusion criteria, participants were eligible if they could walk independently without others’ assistance (e.g., using a cane or walker is fine) and had elevated risk(s) of developing ADRD at enrollment. The elevated ADRD risk(s) were selected based on the recent syntheses of ADRD risk factors (Livingston et al., 2020). Participants need to have at least one of the following risks at enrollment: (a) have a family history of ADRD, (b) are experiencing subjective cognitive decline or having memory complaints during the past year, and (c) are overweight or obese (body mass index ≥25). For exclusion criteria, older adults were not eligible if they (a) have any clinical diagnoses of neuropsychological or mental abnormalities (e.g., ADRD, Parkinson’s, epilepsy, or Parkinson’s disease major depression), (b) currently participating in an intervention or program involving PA and cognitive training that may interfere with the study outcomes, or (c) have other physical health conditions that limit their ability to carry out the study protocol (e.g., impaired visual/audio functioning, stroke, arthritis). Thus, eligible participants in this study generally had adequate cognitive function, hearing, visual, and literacy as they are able to follow the study protocol and carry out the smartphone tasks and the activity measure.

The recruitment occurred primarily through distributing study flyers, word of mouth, email listservs, and contacting local senior living facilities and retirement centers. Individuals who were interested in the study contacted the research staff for an eligibility phone screening, and eligible participants were enrolled in the study on a rolling basis. The data collection started after the COVID-19 pandemic in January 2021, and the last participant completed the study in October 2022.

Procedure

This is a fully remote study. The staff first mailed out a study packet that consisted of the study materials (a study phone, an accelerometer, a paper log, and a booklet of study description) to each eligible older adult and met with them online (via Zoom conference call) to conduct the study orientation, training session, and the consent process after they received the packet. During the Zoom session, the staff provided detailed training on wearing the activity monitor on the thigh and using the study smartphone to play the “brain games” (ultra-brief cognitive assessments). Participants were provided with sufficient time to familiarize themselves with the basic functions of the study phone (turning on/off, adjusting volume, charging the battery) and practicing the brain games until they felt comfortable with the smartphone task. All eligible participants passed this training session and demonstrated adequate ability to follow the study protocols.

Starting the next day after the online training, participants wore the accelerometer on the right thigh for 14 consecutive days, including shower and sleep time, but excluding water activities that would submerge the monitor. Participants were allowed to switch their legs wearing the monitor (e.g., right after taking a shower), and it would not affect data quality. Further, they were asked to play three types of brief “brain games” (a short battery of cognitive assessments) upon waking up and when prompted using the study phone. The study phone delivered four prompts (vibrations with melody music) at semi-random times (~2.5 hr apart) during customized waking hours each day across the 14 days. Participants are asked to ignore prompts delivered during an incompatible activity (e.g., driving, cooking, biking) for safety purposes. Besides wearing the activPAL and playing brain games, participants also kept a daily log to report their bedtime and wake-up time each day during the 14-day monitoring period. The accelerometer data within the bedtime and wake-up periods were excluded for analysis as the current study focuses on waking time movement behaviors.

After completing 14 days of ambulatory assessment, participants mailed back the study device with a prepaid envelope and received momentary compensation based on their levels of study compliance. Each participant received up to $160 if they provided at least 70% of the smartphone and accelerometry data, and partial compensation was available for lower compliance. The Institutional Board of the University of South Carolina (Pro00104601) approved all study protocols and recruitment materials.

Measures

Ambulatory assessment of cognition

This study applied the NIH-funded Mobile Monitoring of Cognitive Change (M2C2) application to deliver three ultra-brief cognitive tests in older adults’ everyday lives during the 14-day monitoring period. This study included three performance-based tests, including the symbol search test, the grid memory test, and the letter go/no-go test. Each of the tests primarily captures critical cognitive domains, including processing speed, visuospatial working memory, and inhibitory control, respectively (Rozas et al., 2008; Salthouse & Meinz, 1995). These cognitive domains are also sensitive to lifestyle behaviors and age-related neuropathology (Brewster et al., 2021; Cerino et al., 2020). These ultra-brief cognitive tasks have been validated and applied in age- and demographically diverse populations in previous work (Sliwinski et al., 2018). The M2C2 tool also differentiates adults with and without MCI (Cerino et al., 2021).

The symbol search test is a speeded two-alternative forced choice task. Older adults were asked to select the “tile” from the bottom of the screen (choice sample) that matches one of the tiles presented at the top of the screen (target) as fast as they could. The grid memory task is a delayed short-term recall task with a filled delay interval. Older adults were asked to recall the locations of three red dots presented in a grid dimension of 5 × 5 study array after a brief filled interval. A quick letter cancellation distraction task is displayed before the recall. The letter go/no-go test is a continuous performance task. Older adults were asked to press a button at the bottom of the screen each time a letter appeared in the center of the screen. Participants are instructed to withhold their response when a prespecified [no-go letter; X in this study] is presented. Please see examples of each cognitive test in Supplementary Figure 1 in Supplementary Material. More details of the M2C2 protocol and guidelines are available in Hakun et al. (2024).

During each EMA prompt, participants completed a total of 18 trials of the symbol search, three trials of the grid memory, and 24 trials of the letter go/no-go. The primary outcomes are the median response time (RT) for the accurate symbol search trials, the total error (Euclidean) distance of the incorrect dot(s) location to the correct dot(s) location in the grid, and the d-prime (d’) score in the letter go/no-go task that represents the proportion of the “hits” (go trials) minus the proportion of “false alarms” (no-go trials). The shorter the RT in the symbol search test indicates faster mental processing speed; the shorter the error distance in the grid memory test indicates better visuospatial working memory, and the higher the d’ score indicates better inhibition and attentional control.

In addition to the three performance-based tests, participants rated their subjective cognition using two self-report items adapted from the PROMIS applied cognition abilities scales item bank (Saffer et al., 2015). For example, participants rated the item “How sharp is your mind right now” using a slider scale ranging from 0 (not at all) to 100 (extremely). The two items demonstrated high internal consistency with an omega statistic of 0.91 (Dunn et al., 2014).

Movement behaviors

Participants in this study wore the research-grade activPAL micro4 accelerometers (PAL Technologies, Glasgow, UK) to measure movement behaviors over the course of 14 days. ActivPAL is a valid and reliable measure of postures (sitting/lying, standing) and movement (stepping) in older adults in a free-living context (Grant et al., 2008). The study outcomes generated from the activPAL include minutes of sitting, LPA, MVPA, and step counts, and they were aggregated within the waking hours for each day. The minutes spent in LPA and MVPA were determined by the accelerometry-derived metabolic equivalents (METs) score generated from the PALbatch software, which is a proxy of the intensity of PA (Tudor-Locke et al., 2018). All activity outcomes accumulated within the same day during the waking hours were used for data analysis.

Covariates

Participants’ sex and age were self-reported prior to the 14-day monitoring period to account for their impact on cognitive outcomes (Boss et al., 2015; Levine et al., 2021). The weekday versus weekend binary variable was calculated to account for the potential differences in the study outcomes due to the social calendar. The valid accelerometry wearing time (in hours) for each day was calculated based on participants’ sleep logs to account for the differences in day-to-day waking hours.

Statistical analysis

The smartphone data were combined with the accelerometry data using the time stamps to examine the associations between daily movement activity and cognition. Each movement activity variable (ST, LPA time, MVPA time, steps) was disaggregated to create the between- and within-person components, which allows the examination of the associations at both levels (Curran & Bauer, 2011).

The mixed-effects location-scale modeling approach was applied using the NIH-funded MixWILD freeware to examine whether older adults’ daily PA or inactivity levels of steps were associated with (a) their daily mean levels of cognitive function, (b) the between-person variation (how individuals differ from each other) in their overall cognitive function, and (c) the within-person variation (the degree of fluctuations within each person) of cognitive function across the study period.

The MixWILD program was selected for data analysis because it is specifically designed for modeling data with hierarchical or nested structures, such as the intensive longitudinal data collected in the current study. This program simultaneously generates three interconnected submodels to estimate the mean structure, between-person variability, and within-person variability of the outcome within a single unified analytic framework, which is a capability not currently available in standard statistical software (Dzubur et al., 2020). This feature in MixWILD is critical, as the results from each submodel are conditional on one another. Specifically, it takes into account both the mean and the between-person variance structures while modeling the within-person variability (instability).

The between-person and the within-person variance submodels are specified as log-linear models to ensure positive variance estimation (Nordgren et al., 2020). Thus, exponentiating the estimates from the two variance submodels produces variance ratios (VRs), which represent the ratio of variances associated with a unit change of the covariate and are compared against 1 to assess the statistical significance of the covariate. Core demographic covariates are included in the mean model to account for their influence on the cognitive outcome.

Results

Descriptive Statistics

The final sample included 96 older adults with a mean age of 68.31 ± 7.09 years, and 35% of them were male. Most of the participants were White (72%), had attained at least some college education (95%; including professional training, associate degrees, or higher), and most of them did not live alone (69%). Participants provided a total of 1,269 days of observations that included valid smartphone and accelerometry data, including 355 observations on weekend days (28%). Each participant provided an average of 13.22 days of observations, and 79% (n = 76) of the participants provided valid data for all 14 days. The average valid activPAL wearing hours per participant was 16.20 hr ± 9.12. The correlation matrix in Table 1 shows that the three performance-based cognitive outcomes and the four accelerometry-based activity outcomes are significantly correlated at both the between- and the within-person levels. Based on the intraclass correlation coefficients, subjective cognition consisted of the highest proportion of between-person variance, whereas the d-prime scores from the letter go/no-go test had the greatest proportion of within-person variance. Supplementary Figure 2 presents the daily cognitive assessment data among random subsets of participants. There were noticeable day-to-day within-person fluctuations for each of the four cognitive outcomes.

Table 1.

Between- and Within-Person Correlation Matrix and Descriptive Statistics of the Ambulatory Assessed Study Outcomes

Study outcomes	Mean	SD	Median	1	2	3	4	5	6	7	8
1. Subjective cognitiona	75.05	16.07	76.67	0.765	−0.020	−0.046	−0.011	0.050	−0.067*	0.109***	−0.004
2. Symbol searchb	2296.30	763.44	2163.75	−0.027	0.561	0.413***	−0.414***	−0.305***	0.135***	−0.156***	−0.290***
3. Grid memoryc	23.26	17.62	22.07	0.003	0.225***	0.503	−0.255***	−0.255***	0.015	−0.017	−0.296***
4. Go-Nogo Testd	3.59	1.24	4.06	0.077**	−0.475***	−0.165***	0.483	0.149***	−0.124***	0.003	0.170***
5. Daily Steps	7270	4252	6601	0.066*	−0.037	0.009	0.027	0.556	−0.505***	0.576***	0.844***
6. Daily sedentary (hr)	10.42	3.38	10.00	−0.072**	0.134***	0.051	−0.109***	−0.481***	0.670	−0.684***	−0.228***
7. Daily LPA (hr)	5.38	2.35	5.15	0.081**	−0.025	0.014	−0.014	0.552***	−0.629***	0.623	0.182***
8. Daily MVPA (min)	22.27	26.75	12.00	0.045	−0.008	0.008	0.039	0.830***	−0.225***	0.145***	0.522

Notes: LPA = light-intensity physical activity; MVPA = moderate-to-vigorous physical activity. Intraclass correlation coefficients were calculated in the diagonal cells of the matrix (italic); Between-person correlations are above the diagonal, and the within-person correlations are below the diagonal; outcome measures: ^aself-rating, ^bresponse time (ms), ^csum of error distance, ^dd-prime score.

^* p < .05.

^** p < .01.

^*** p < .001.

Below, we summarize the findings following the order of the mean submodel, the between-person variance submodel, and the within-person variance submodel corresponding with the output generated from the MixWILD program. For each submodel, the movement behavior predictor (MVPA, LPA, steps, or ST) was partitioned into the between-person (overall levels across the 14-day period) and the within-person (the deviation from the overall levels on a given day) components to examine their associations with each of the four cognitive test outcomes (see Tables 2–5).

Table 2.

Mixed-Effects Location Scale Model Results of Daily MVPA and Ambulatory-Assessed Cognitive Outcomes

Covariates	Symbol search test (response time [ms])	Grid memory test (error distance)	Go-no-go test (d prime score)	Subjective cognition (rating scale)
Means model
Intercept	0.365 (0.568)	20.339 (11.855)	7.624*** (0.682)	72.426*** (14.915)
Daily MVPA time (BP)	−0.468** (0.148)	−11.470** (4.184)	0.257* (0.132)	−2.020 (4.252)
Daily MVPA time (WP)	−0.013 (0.029)	−0.013 (0.676)	0.062 (0.058)	1.207 (0.702)
Weekend	0.050* (0.022)	0.273 (0.461)	−0.029 (0.037)	−0.592 (0.486)
Valid device wear time (hr)	0.030*** (0.006)	0.345** (0.123)	−0.047*** (0.01)	−0.065 (0.138)
Age	0.023** (0.008)	0.037 (0.167)	−0.008 (0.009)	0.035 (0.21)
Sex (male = 1)	0.118 (0.116)	−4.043 (2.547)	0.080 (0.115)	5.828* (3.04)
BP variance log-linear model
Intercept	−1.110*** (0.222)	4.706*** (0.243)	−0.213 (0.225)	5.383*** (0.225)
Daily MVPA time (BP)	−0.502 (0.447)	0.484 (0.529)	−2.713*** (0.615)	−0.340 (0.457)
Daily MVPA time (WP)	0.064 (0.641)	−0.180 (0.119)	0.348 (0.376)	−0.016 (0.116)
WP variance log-linear model
Intercept	−1.811*** (0.06)	4.167*** (0.063)	−1.164*** (0.062)	4.201*** (0.062)
Daily MVPA time (BP)	−0.695*** (0.119)	−0.435*** (0.130)	0.208 (0.126)	−0.266* (0.124)
Daily MVPA time (WP)	0.122 (0.148)	−0.028 (0.145)	−0.384* (0.164)	0.173 (0.166)

Notes: BP = between-person; MVPA = moderate-to-vigorous physical activity; WP = within-person. Standard errors (SE) are reported in parentheses following the estimated coefficients. The estimates of the BP and WP variance models are on the log scale.

^* p < .05. ^**p < .01. ^***p < .001.

Table 5.

Mixed-Effects Location Scale Model Results of Daily Sedentary Time and Ambulatory-Assessed Cognitive Outcomes

Covariates	Symbol search test (response time [ms])	Grid memory test (error distance)	Go-no-go test (d prime score)	Subjective cognition (rating scale)
Means model
Intercept	0.497 (0.575)	9.246 (12.967)	7.596*** (0.780)	66.94*** (15.1)
Daily sedentary time (BP)	0.001 (0.023)	−0.079 (0.506)	−0.033 (0.032)	−0.838 (0.58)
Daily sedentary time (WP)	0.004 (0.007)	−0.074 (0.139)	0.004 (0.011)	−0.391** (0.147)
Weekend	0.044* (0.021)	0.158 (0.457)	−0.035 (0.036)	−0.526 (0.478)
Valid device wear time (hr)	0.011 (0.007)	0.344* (0.163)	−0.042** (0.012)	0.387* (0.18)
Age	0.023*** (0.008)	0.160 (0.185)	−0.003 (0.011)	0.123 (0.214)
Sex (male = 1)	0.066 (0.115)	−5.524* (2.734)	0.004 (0.156)	6.258* (3.069)
BP variance log-linear model
Intercept	−2.013***(0.716)	4.639*** (0.575)	−2.06 (0.853)	4.712*** (0.686)
Daily sedentary time (BP)	0.075 (0.716)	0.031 (0.053)	0.117 (0.079)	0.050 (0.064)
Daily sedentary time (WP)	0.049*** (0.017)	0.003 (0.019)	−0.008 (0.027)	−0.001 (0.015)
WP variance log-linear model
Intercept	−2.258*** (0.2)	3.8210*** (0.170)	0.117*** (0.167)	3.865*** (0.153)
Daily sedentary time (BP)	0.016 (0.019)	0.009 (0.016)	0.037* (0.015)	0.021 (0.014)
Daily sedentary time (WP)	0.108*** (0.017)	0.069*** (0.016)	0.094*** (0.018)	0.072*** (0.02)

Notes: BP = between-person; WP = within-person. Standard errors (SE) are reported in parentheses following the estimated coefficients. The estimates of the BP and WP variance models are on the log scale.

^* p < .05. ^**p < .01. ^***p < .001.

Between-Person and Within-Person Movement Behaviors and the Mean Levels of Cognitive Function

Tables 2–5 summarize the results of the associations between daily movement behaviors and cognitive function from the MixWILD program, stratified by each daily movement activity. Each table consists of results from simultaneously modeling the mean structure, the between-person variance, and the within-person variance of each cognitive outcome assessed by the M2C2 app.

Results from the mean models suggested that the between-person levels of daily MVPA and steps were negatively associated with processing speed. In general, older adults with more MVPA or daily steps compared to others showed faster mean reaction time on the symbol search test (bs = −0.47 and −0.05, p < .01). Similarly, older adults with more daily MVPA or steps compared to others also showed shorter total error distance on the grid memory test (bs = −11.47 and −1.07, p < .05). A within-person association was also observed between daily steps and perceived cognition. On days when older adults had relatively more steps, they also perceived better cognition on those days (b = 0.19, p < .05).

Older adults’ LPA was not associated with any of the mean levels of performance-based cognitive outcomes, but it had significant positive associations with the mean level of subjective cognition both at the between- and the within-person levels. For older adults who have more daily LPA compared to others, they reported better cognition (b = 1.77, p < .05); on days when older adults engaged in more LPA than their usual level, they also reported better cognition on that same day (b = 0.39, p < .01). An inverse within-person association was observed for sedentary behavior and subjective cognition. On days when older adults were more sedentary than their usual level, they also reported worse cognition on those days (b = −0.39, p < .01).

Between-Person and Within-Person Movement Behaviors and the Between-Person Variation of Cognitive Function

The results from the between-person variance models showed that older adults’ daily movement behaviors were associated with the outcomes from the letter go/no-go and symbol search tests. For the go/no-go test, the between-person levels of daily MVPA and steps were associated with the degree of between-person variation in inhibitory control (VRs = 0.78 and 0.07; Tables 2 and 4). Older adults who had higher daily MVPA or steps than others were more homogeneous (less between-person variations) in their overall levels of inhibitory control. For the symbol search test, the within-person levels of daily LPA and ST were associated with the degree of between-person variation in processing speed (VRs = 0.94 and 1.05; Tables 3 and 5). On days when older adults had more daily LPA and longer ST than their usual level, they were more similar to each other in their overall levels of processing speed.

Table 4.

Mixed-Effects Location Scale Model Results of Daily Steps and Ambulatory-Assessed Cognitive Outcomes

Covariates	Symbol search test (response time [ms])	Grid memory test (error distance)	Go-no-go test (d prime score)	Subjective cognition (rating scale)
Means model
Intercept	0.732 (0.604)	24.337 (12.524)	7.195*** (0.760)	68.672*** (15.518)
Daily step counts (BP)	−0.048** (0.016)	−1.072* (0.421)	0.027 (0.01)	0.169 (0.451)
Daily step counts (WP)	−0.004 (0.234)	0.004 (0.071)	0.005 (0.006)	0.194* (0.078)
Weekend	0.046* (0.022)	0.280 (0.125)	−0.030 (0.037)	−0.593 (0.486)
Valid device wear time (hr)	0.029*** (0.006)	0.336** (0.125)	−0.048*** (0.01)	−0.081 (0.138)
Age	0.020* (0.008)	0.034 (0.169)	−0.002 (0.01)	0.066 (0.213)
Sex (male = 1)	0.117 (0.112)	−4.728 (2.565)	0.038 (0.12)	5.528 (3.04)
BP variance log-linear model
Intercept	−0.728* (0.372)	4.524*** (0.395)	0.708 (0.427)	5.449*** (0.371)
Daily step counts (BP)	−0.081 (0.048)	0.053 (0.052)	-0.254*** (0.061)	−0.027 (0.048)
Daily step counts (WP)	−0.002 (0.015)	−0.010 (0.012)	0.015 (0.028)	−0.003 (0.012)
WP variance log-linear model
Intercept	−1.537*** (0.101)	4.319*** (0.1)	−1.215*** (0.095)	4.279*** (0.1)
Daily step counts (BP)	−0.075***(0.013)	−0.044** (0.052)	0.018 (0.012)	−0.025* (0.013)
Daily step counts (WP)	−0.002 (0.015)	−0.015 (0.012)	−0.020 (0.016)	0.009 (0.016)

Notes: BP = between-person; WP = within-person. Standard errors (SE) are reported in parentheses following the estimated coefficients. The estimates of the BP and WP variance models are on the log scale.

^* p < .05. ^**p < .01. ^***p < .001.

Table 3.

Mixed-Effects Location Scale Model Results of Daily LPA and Ambulatory-Assessed Cognitive Outcomes

Covariates	Symbol search test (response time [ms])	Grid memory test (error distance)	Go-no-go test (d prime score)	Subjective cognition (rating scale)
Means model
Intercept	0.223 (0.596)	14.230 (13.683)	4.648*** (0.768)	55.168*** (15.717)
Daily LPA time (BP)	−0.021 (0.031)	−0.681 (0.718)	0.041 (0.04)	1.767* (0.774)
Daily LPA time (WP)	−0.006 (0.007)	0.013 (0.139)	0.003 (0.011)	0.390** (0.149)
Weekend	0.044* (0.021)	0.213 (0.461)	−0.037 (0.037)	−0.001 (0.484)
Valid device wear time (hr)	0.029*** (0.006)	0.328** (0.126)	−0.052*** (0.01)	−0.106 (0.145)
Age	0.024** (0.008)	0.134 (0.181)	−0.005 (0.01)	0.138 (0.208)
Sex (male = 1)	0.093 (0.122)	−6.103* (2.744)	0.081 (0.154)	7.771* (3.132)
BP variance log-linear model
Intercept	−0.579 (0.497)	4.952*** (0.446)	−1.084 (0.557)	5.973*** (0.542)
Daily LPA time (BP)	−0.124 (0.089)	0.0005 (0.079)	0.052 (0.101)	−0.145 (0.098)
Daily LPA time (WP)	−0.056* (0.026)	−0.013 (0.023)	−0.043 (0.036)	−0.001 (0.023)
WP variance log-linear model
Intercept	−1.462*** (0.129)	4.3952*** (0.131)	−1.033*** (0.13)	4.186*** (0.127)
Daily LPA time (BP)	−0.114*** (0.023)	−0.070** (0.023)	−0.009 (0.023)	−0.017 (0.022)
Daily LPA time (WP)	−0.024 (0.025)	−0.022 (0.022)	0.032 (0.028)	−0.060* (0.029)

Notes: BP = between-person; LPA = light-intensity physical activity; WP = within-person. Standard errors (SE) are reported in parentheses following the estimated coefficients. The estimates of the BP and WP variance models are on the log scale.

^* p < .05. ^**p < .01. ^***p < .001.

Between-Person and Within-Person Movement Behaviors and the Within-Person Variation of Cognitive Function

The within-person variance model revealed similar findings from the symbol search test, the grid memory test, and the subjective rating of cognition. There were relatively fewer significant associations between movement behaviors and the letter go/no-go test.

At the between-person level, older adults who had higher daily MVPA, steps, or LPA showed lower degrees of variation (fluctuation) in processing speed (VRs = 0.50, 0.92, and 0.90, respectively) and visuospatial working memory (VRs = 0.65, 0.96, and 0.93, respectively) across the study period (Tables 2–4). Older adults who had higher daily MVPA or steps also had lower degrees of variation in subjective cognition (VRs = 0.78 and 0.97; Tables 2 and 4). In addition, older adults who overall were more sedentary (had more daily ST) than others showed greater degrees of variation in their inhibitory control (VR = 1.04; Table 5).

At the within-person level, on days when older adults had more MVPA time than their usual level, they showed less variations in their inhibitory control (VR = 0.68; Table 2). Additionally, on days when older adults had more LPA time, they showed less variations in their ratings of subjective cognition (VR = 0.94; Table 3). Interestingly, older adults’ daily ST was positively associated with the within-person variation of cognitive function across the four tests. On days when older adults were more sedentary than their usual level, their processing speed (VR = 1.11), visuospatial working memory (VR = 1.07), inhibitory control (VR = 1.10), and subjective cognition (VR = 1.07) were consistently more variable (Table 5).

Discussion

This study examined associations between movement behaviors and cognitive function in older adults using EMA data from accelerometers and smartphone apps. Our findings generally align with existing research indicating that greater daily PA and less ST are associated with better cognitive function in older adults. This study extends existing literature by focusing on the temporal dynamics and the within-person variability of older adults’ cognitive function in response to daily movement patterns. Specifically, we examined these study outcomes in real-time, at faster time scales, and within naturalistic settings characterized by high ecological validity. We also explored the within-person variability of cognitive function, which is an underexplored target that may serve as a promising marker of cognitive vulnerability beyond traditional mean-level cognitive measures in more controlled settings.

By leveraging the MixWILD program, we provide additional nuanced findings that the between-person differences and the within-person changes in daily PA and ST are associated with the within-person variability across different cognitive domains. Within-person variability in cognition is often overlooked in cross-sectional and EMA studies (Costa et al., 2019). This study addresses this gap by modeling both the mean structures and the between- and within-person variance of cognitive function in relation to daily movement behaviors. With smartphone ownership exceeding 86% among older adults ages 50+, the intensive longitudinal data collected using smartphones offer valuable insights into real-life cognitive changes in response to daily activity and inactivity levels (Kakulla, 2023).

When using daily MVPA or steps as predictors, significant associations with cognitive outcomes were primarily observed at the between-person level. The EMA study from Kekäläinen et al. (2023) reported similar between-person associations linking faster processing speed with higher activity counts or active time in a mixed-age sample (Kekäläinen et al., 2023). Unlike their findings of within-person associations with processing speed and null findings with visual memory or subjective cognition, our study identified between-person associations of MVPA and steps with visuospatial working memory and a within-person association between daily steps and subjective cognition.

Several factors may explain the differences between our findings and those of Kekäläinen et al. (2023), despite both studies using the M2C2 app to assess cognition. First, different PA metrics were generated from different accelerometers in each study, which may not be comparable (minutes spent in different activity intensities from activPAL vs. max activity counts and active minutes from ActiGraph). Second, for the within-person analysis, we focus on the day-level associations to capture the entire movement activities during waking hours, whereas their study reported the within-day (momentary) associations. Finally, we applied the mixed-effects location-scale modeling approach to analyze not only the average (mean) structure but also the variability (variance) around that mean. It simultaneously considers both between-person differences and within-person fluctuations. Thus, it goes beyond what a standard linear multilevel model can capture by providing a more nuanced understanding of the variance structure present in the EMA data.

Older adults’ daily MVPA time and step counts showed similar associations with all cognitive domains assessed, particularly with processing speed and visuospatial working memory. These findings suggest that increasing daily steps may offer comparable cognitive benefits to those gained from solely focusing on MVPA. However, daily steps may need to be accumulated at a faster walking pace (higher intensity) due to the dissimilar results from the steps and the LPA models. A large longitudinal cohort study of older women (N = 14,399; Mean age = 71.8) also reported similar associations of MVPA and step counts with all-cause mortality and cardiovascular diseases, supporting the promotion of increasing daily steps (Hamaya et al., 2024). Our EMA study underscores the need for future longitudinal studies to validate the protective role of daily steps on cognitive health. Promoting step-based goals carries significant public health implications. It offers a practical and achievable preventive strategy for older adults, especially among those with chronic health conditions or mobility limitations that restrict MVPA participation.

Unlike MVPA and step counts, LPA showed distinct patterns in its association with cognitive function. For performance-based tests, LPA did not explain the mean models but was linked to the between- or within-person variance models, except for the go-no-go test. This aligns with evidence suggesting that inhibitory control may require higher-intensity or regular PA engagement to exert benefits (Chu et al., 2023; Huang et al., 2014). For self-reports, older adults reported better subjective cognition and less variability in their ratings on days with higher-than-usual LPA. Compared to MVPA, which typically requires older adults to focus on maintaining balance, movement pace, and breathing, LPA is typically integrated into daily routines and imposes minimal cognitive demands. Thus, LPA may allow the brain to recover from or engage in other cognitively demanding tasks more effectively (Aitken & MacMahon 2019). Additionally, LPA may enhance mood and reduce stress, contributing to positive perceptions of mental and cognitive health (Felez-Nobrega et al., 2021).

Our exploration of ST and cognitive function variability revealed consistent within-person associations across all cognitive tests. On days with higher-than-usual ST, older adults showed greater variability in processing speed, visuospatial working memory, inhibitory control, and subjective cognition, reflecting worse cognitive stability. Unlike previous EMA studies that focused on the association between mean levels of cognition and ST, our findings provide preliminary evidence of limiting ST in day-to-day life to maintain older adults’ cognitive stability across multiple domains (Fitzsimmons et al., 2014).

The negative between-person association of perceived cognitive function and device-based ST observed in the study contrasts with the null findings reported in Fitzsimmons et al. (2014). Differences in devices (activPAL vs. ActiGraph), populations (college students vs older adults), and sedentary contexts (cognitively engaged academic activities vs leisure-based activities) likely explain the discrepancies (Schlaff et al., 2017). The thigh-worn activPAL offers greater sensitivity in distinguishing different postures, potentially yielding more precise results compared to the ActiGraph, which is less precise in differentiating standing from sitting (Pfister et al., 2017). In addition, college students’ ST is often tied to academic activities, and they have not entered the phase of cognitive aging in general. These differences may alter the associations between ST and cognition.

Similarly, the discrepancies between our findings and those of Campbell et al. (2020) may stem from differences in the measurement of cognitive function and ST. The smartphone-based cognitive tests used by Campbell et al. targeted executive function and verbal learning (color-word interference test and the verbal learning test), which were not assessed in our study. Additionally, their study relied on self-reported passive leisure activities (e.g., watching TV, resting), while our study captured total ST without distinguishing specific behaviors.

Overlapping results are observed across all cognitive domains. First, processing speed and visuospatial working memory showed similar between-person associations with PA-based outcomes (MVPA, LPA, steps). Second, subjective cognition aligned with symbol search and grid memory in explaining within-person variance across movement behaviors. Lastly, greater variability in all cognitive domains was linked to longer daily ST. These similar findings provide converging evidence for the cognitive benefits of living an active lifestyle—moving more and sitting less (Physical Activity Guidelines Advisory Committee, 2018).

While our findings are encouraging, several limitations should be acknowledged when interpreting the study findings. Data collection was limited to two weeks, which may not capture long-term cognitive variability. A multi-burst design spanning a longer period (e.g., years) could better account for within-person cognitive changes related to normal aging (Hultsch et al., 2008). The activPAL device may underestimate daily activity levels, as it is insensitive to resistance exercise, and it was removed during water-based activities. It measures duration but not specific types of PA or sedentary behavior, which may also influence cognitive outcomes (Coelho et al., 2020; Dupré et al., 2021). The app-based cognitive assessments are behavioral indicators of cognition that do not capture the brain activity or neural processes underlying cognitive function. While not ideal to carry a second phone, we asked participants to use a standard study smartphone for cognitive assessments, which reduces variations from the types of touch sensors installed on different software and hardware platforms.

Our data collection occurred during the ongoing pandemic period after COVID-19 lockdown restrictions had been lifted in South Carolina. Although participants completed ambulatory assessments within their natural living environments under relatively fewer constraints, our data likely reflect cognitive and behavioral dynamics specific to the broad pandemic-related contexts compared to the prepandemic period. In addition, participants in this study are mostly White and generally well-educated, despite the fact that they have elevated ADRD risk factor(s). This rather affluent older adult sample may potentially limit the between-person variability in cognitive function and the generalizability of our findings to more diverse older adult populations. The sex distribution and age range in our study were unbalanced, and we did not account for all potential confounding variables that could have influenced the findings. Future research should include larger, more diverse samples of older adults, extend the monitoring periods, and utilize more comprehensive measures of covariates, movement behaviors, and cognitive function. These enhancements will also strengthen evidence supporting behavioral strategies to promote cognitive health in the daily lives of older adults.

Lastly, causality cannot be established between movement behaviors and cognitive variables in this EMA observational study. For example, on days when older adults are more sedentary, it may reflect poorer physical or cognitive function that limits their ability to perform daily activities on that day. Likewise, on days when older adults have better cognitive capacity, they may carry out more errands or daily living tasks, leading to higher ratings of perceived cognition. Other everyday contexts, including social engagement and environmental factors at different locations, may also influence daily cognitive function. Future studies should incorporate data on daily health symptoms, social contexts, and physical condition (e.g., via daily diaries) to better disentangle the observed associations between daily movement behaviors and cognitive function in older adults.

Conclusions

This study applied device-based measures to investigate the role of daily movement behaviors on various cognitive outcomes in older adults with elevated ADRD risk. Our findings suggest that more daily PA, regardless of intensity, and less ST are associated with better cognitive function and greater cognitive stability. To support and extend these findings, more intensive longitudinal studies are needed to characterize the within-person processes and dynamics between movement behaviors and cognitive health. As the aging population continues to grow, findings from such studies will offer valuable opportunities to identify novel ecologically valid strategies to promote brain health in everyday contexts.

Funding

This study is supported by the USC Office of the Vice President for Research ASPIRE-I pilot fund and is partially supported by the USC Office for The Study of Aging Fellowship.

Conflict of Interest

None.

Data Availability

The de-identified data and the analytic code can be requested and accessed by contacting the corresponding author. All data requests will be reviewed by the Office of the Associate Dean of Research at USC to ensure all procedures and appropriate data handling processes are followed.