The exponential growth of the older adult population will result in more individuals living with Alzheimer’s disease and related dementias (ADRD) [1]. Given the lack of effective treatments for Alzheimer’s disease, a focus on healthy lifestyle choices holds promise in preventing cognitive decline [2-4]. Evidence links physical activity (PA) with long-term cognitive benefits such as reduction of age-related cognitive decline and dementia risk. For example, a meta-analysis of longitudinal observational studies with middle-aged and older adults found that higher levels of PA were associated with a 14% lower risk of dementia when compared to those with lower levels of PA [5]. Similarly, compared with sedentary individuals, even low to moderate PA has been associated with a 35% reduction in the risk of cognitive decline [6]. Moreover, a recent review [7] found a small positive effect of PA on executive function (Cohen’s d=0.27) and memory (Cohen’s d=0.24), whereas another study found that walking more steps per day was related to better executive function in healthy aging [8]. Several mechanisms explain how PA benefits cognition in the long term, including angiogenesis, neurogenesis, upregulation of neurotrophic factors, reduced inflammation, cardiovascular benefits, changes in central arterial stiffness and endothelial function, and insulin regulation [9,10].

Less is known about the acute effects of PA on cognition in middle-aged and older adults in real-world contexts, where ecological validity is optimized. The acute effects of PA on cognition have generally been studied in laboratory settings, where individuals complete cognitive tasks before, during, or immediately following an exercise challenge. Within these settings, it has been found that acute moderate intensity PA results in improved executive function [11-13] and working memory performance in older adults [14], although others have found that acute exercise before memory encoding tasks may impair performance in older adults [15]. A meta-analysis of the literature showed that the overall effect of acute exercise on cognition is positive but generally small; and that longer exercise duration, greater intensity, type of cognitive performance assessed (and when it is assessed), and greater fitness level appear to be significant moderators of larger effect sizes [16].

Owing to research-grade accelerometry, it is now possible to remotely track PA behavior, whereas smartphone-based technology can assess cognitive function in real-world contexts using ecological momentary cognitive tests (EMCTs). Evidence suggests that EMCTs have relatively high completion rates of 60% to 85% [17], and performance on these tests correlates with standard neuropsychological testing scores [18-22]. With 81% of adults aged 60 to 69 years and 62% of those aged ≥70 years owning a smartphone [23,24], EMCTs can be deployed to increase our understanding of whether acute fluctuations in real-world PA lead to immediate improvements in cognition in a variety of everyday contexts. This can help determine the utility of EMCTs as tools in digital health interventions to measure cognitive function repeatedly, at scale, and in people’s natural environments, thus improving ecological validity and reducing participant burden.

This observational study investigates if there are daily-level associations between accelerometer-measured PA and smartphone-based EMCT performance during a 2-week measurement period in middle-aged and older adults with a variety of comorbid conditions (eg, people living with HIV [PWH], cardiovascular risk, and functional impairment). Moreover, we examine whether cognitive status, functional status, and cardiovascular disease (CVD) risk affect the acute associations of PA with performance on mobile verbal learning and executive function EMCTs. We hypothesize that days with greater measured PA would be associated to better performance on executive function and learning. Consistent with the literature suggesting that those with greater risk profiles may benefit most from lifestyle approaches to maintain brain health [25], we also hypothesize that the association between daily fluctuations in PA and EMCT performance would be stronger in participants with lower cognition, higher CVD risk, and lower functional independence status compared to those with better cognition, lower CVD risk, and no functional dependence.

A total of 90 people—57 (63%) PWH and 33 (37%) HIV-negative, middle-aged, and older adults aged 50 to 74 years—were enrolled in a study at the University of California San Diego’s HIV Neurobehavioral Research Program (HNRP) from 2016 to 2019.

Participants were recruited from the participant pool at the HNRP or through community outreach. Inclusion criteria were age ≥50 years, fluent in English, and ability to provide written informed consent. Exclusion criteria were neurological confounders not related to HIV (eg, stroke, untreated seizure disorder, and head injury with loss of consciousness >30 minutes), diagnosis of serious mental illness (eg, schizophrenia and bipolar disorder), and a reported learning disability or low estimated verbal IQ (ie, a standard score <70 on the Wide Range Achievement Test 4 Reading test [26]). The overall level of illicit substance use in this sample was low [27]. Illicit substance use was not exclusionary; however, participants with a positive alcohol breathalyzer or positive urine toxicology (other than marijuana or prescription medications) on the day of testing were rescheduled for another day. This occurred 2 times, and these 2 participants were rescheduled and tested positive once more upon their return, at which time they were withdrawn from the study. The study procedures were approved by the institutional review board of the University of California, San Diego, and all participants demonstrated decisional capacity [28] and provided written informed consent.

Demographic data, comorbid conditions, and HIV disease characteristics were summarized as mean (SD), median (IQR), or n (%). mCWIT in seconds was log₁₀-transformed to improve normality before the analyses. To evaluate whether days with greater PA were associated with better EMCT performance, linear mixed-effects models with subject-specific random intercepts were conducted to detect the relationship between within-person PA and EMCT performance. Between-person PA effects on EMCT performance were also estimated. Crude models with significant within- or between-person effects (P<.05) were later adjusted for covariates. HIV status, age, sex, education, and race or ethnicity (non-Hispanic White vs all other race or ethnicities) were included as covariates in the models if P<.10, using backward model selection.

In addition, to determine whether cognition (ie, global T score), functional status (ie, IADL dependent or independent), and CVD risk (Framingham CVD risk profile) moderated the relationship between PA and EMCT performance, moderation analyses were conducted using separate linear mixed-effects models with fixed effects of within- and between-person PA, global T score, and IADL status or Framingham stroke risk score; their interactions (eg, an interaction between within-person PA and global T score on EMCT performance); and subject-specific random effects. All models were controlled for study day. The significance level of α was set at .05. The effect sizes (and 95% CIs) are standardized coefficients, which are analogous to Cohen’s d. R software (version 3.6.0; R Foundation for Statistical Computing) was used to perform all statistical analyses.

Participants’ demographic, clinical, and cognitive characteristics are presented in Table 1. On average, participants were in their late 50s (mean age 59, SD 6.3 years), had some college education (mean 14.4, SD 2.4 years), and most were male (65/90, 72%) and identified as non-Hispanic White (57/90, 63%). Of the 90 participants, 57 (63%) were PWH, and participants’ age, education, and race or ethnicity were not significantly different based on HIV status. There were significantly more women in the HIV-negative group compared with the PWH group (PWH: 10/57, 18% women; HIV-negative: 15/33, 45% women; P<.004). Overall, there was excellent adherence to the mobile cognitive testing and accelerometer protocols as, on average, participants completed 13 of 14 days of the mCWIT, 12 of 14 days of the mVLT, and had 12 of 14 days of accelerometer data to objectively assess PA.

We examined the cross-sectional association of objectively measured daily PA with EMCT performance to determine whether natural variation in PA was accompanied by fluctuations in performance on EMCTs of executive function and verbal learning in individuals’ free-living environments. In a sample of diverse adults with a wide range of comorbidities (ie, cardiovascular risk and HIV), we found that during days with greater PA (ie, a combination of higher frequency, intensity, and duration), participants had faster (better) performance on a mobile Stroop-like task of executive function (mCWIT). This relationship persisted after adjusting for HIV status and other demographics (age and race or ethnicity). No effects of PA were found on a task of verbal learning (mVLT). The results are consistent with much of the laboratory-based literature linking an acute increase in PA to improvements in executive functions [11-13] and extending it to provide novel evidence that these associations may be captured remotely in the real world using EMCTs and accelerometry.

Next, we examined whether the significant cross-sectional association of daily PA with EMCT performance varied as a function of cognitive status, functional status (IADL), and cardiovascular risk, all of which elevate the risk for ADRD [43-47]. We found that only for individuals who reported being functionally dependent on IADL, there was a significant association between daily PA and executive function performance. It has been suggested that individuals who are at higher risk of ADRD may benefit most from lifestyle interventions to reduce cognitive decline [25]. Our findings are in line with this literature, suggesting that those with more functional limitations may see a greater cognitive benefit from engagement in PA in their free-living environments. We did not find moderation effects related to HIV, cognitive, or cardiovascular risk status, suggesting that the daily relationship of PA with executive function does not significantly vary as a result of these risk factors in this limited sample of adults. Importantly, the association of daily PA with daily executive function performance persisted after adjusting these models for HIV status, global cognition, and cardiovascular risk, reinforcing the robustness of this relationship in a heterogeneous sample.

The findings suggest that daily variation in PA is sensitive to fluctuations in daily executive function performance, which can be reliably measured remotely using actigraphy and EMCTs. This has important implications for the development of novel digital health interventions to lower the risk of ADRD. Given the many limitations preventing individuals from participating in supervised or group-based PAs (ie, transportation barriers, safety issues, and mobility limitations) and the high costs associated with in-person interventions, it is important to develop scalable, low-cost, evidence-based [48] interventions to promote PA in real-world contexts [49]. Digital health technologies can help us achieve this goal by tracking intervention adherence and providing feedback in real time [50]. Future lifestyle interventions to preserve brain health can leverage these technologies to measure PA and cognition in real-world contexts to better understand treatment effects, improve generalizability, and reach out to a larger sector of the population who may not otherwise be able to attend laboratory-based (in-person) clinical trial visits. Moreover, digital health technologies can help track intervention adherence and changes in outcomes of interest, such as cognition, which can be measured repeatedly over the course of a clinical trial rather than only at baseline and post intervention time points to better capture change over time.

A limitation of this study is its observational rather than interventional nature, which does not allow us to determine whether improvements in PA lead to better executive function performance or vice versa. Second, the measurement time was limited to 14 days, which may have limited the range of variability in PA and cognitive testing performance. Third, this study was performed on a small and heterogeneous sample of adults with various comorbid conditions, limiting our ability to adjust the models for other potential confounders that may have affected EMCT performance, such as mood and motivation indicators. That said, it is unlikely that our results would change if the models were adjusted for mood, given previous findings showing that adjusting for mood did not alter the significant association of daily activities with EMCT performance in this sample [51]. Moreover, given the small sample size, null findings may have resulted from a lack of power rather than an absence of associations. Fourth, accelerometers were worn on the wrist rather than the hip, which has been traditionally thought to provide a more accurate measurement of PA [52], although this assumption has recently been challenged, given the ubiquitous nature of wrist-worn accelerometry and their improved estimation of PA [40,53,54]. Fifth, participants were paid to complete the EMCTs; therefore, adherence may differ in studies where participants do not receive financial compensation for completing the assessments. Finally, future studies with larger sample sizes should account for more variables that may affect the relationship between PA and cognition, such as the length and severity of HIV disease.

In conclusion, using EMCTs and accelerometry to capture cognitive performance and PA in free-living environments may be an ecologically valid means of capturing real-world associations between cognition and PA. Our findings suggest that these digital techniques are promising candidates for tracking cognitive change and may be useful in the context of lifestyle (nonpharmacological) digital interventions designed to reduce ADRD risk and improve brain and cognitive health. Having participants complete cognitive tests in more familiar settings, such as their home or during other daily activities, can help increase the generalizability of findings, reduce intervention costs, increase scalability, and improve adherence to digital health interventions.