Mid-Life Physical Activity and Late-Life Cognitive Performance among American Indians

Cognitive impairment associated with aging is a growing public health issue due to the rapid expansion of the elder population [1]. In the next few decades, the proportion of American Indians (AIs) and Alaska Natives aged 85 and older in the USA is projected to increase 225%, compared to a 10% decrease in Whites of similar age [2]. AIs suffer disproportionately from health conditions and behaviors that increase the risk of cognitive impairment, such as hypertension, diabetes, smoking, physical inactivity, and obesity [3‒6]. To inform prevention efforts in this growing at-risk population, a better understanding of the relationships between modifiable cognitive risk factors and cognitive impairment is critical.

Meta-analyses of randomized controlled trials [7] and longitudinal observational studies of mid-life physical exercise [8] indicate that physical activity (PA) is associated with reduced dementia risk and, more generally, may be beneficial for cognitive functioning and brain health. Accordingly, we investigated relationships between PA in midlife and cognitive outcomes measured 7–21 years later in AI participants in the Strong Heart Study (SHS) [9]. We hypothesized that individuals with higher levels of total PA would have higher cognitive performance later in life, and we accounted for potential selective attrition bias using inverse-probability weighting.

The SHS is a population-based longitudinal study of cardiovascular disease and its risk factors in AIs residing in 13 communities in the US Southwest, Central, and Northern Plains. Tribal members aged 45–74 years were recruited for a baseline examination during 1989–91. Previous publications provide additional details regarding the study design and data collection [9]. The present study includes SHS members who participated in the Cerebrovascular Disease and its Consequences in American Indians (CDCAI) Study [10] (Fig. 1).

The CDCAI study aimed to characterize the burden, risk factors, and manifestations of vascular brain injury in the SHS cohort [10]. In 2010–13, CDCAI enrolled 1,033 surviving SHS members and conducted clinical examinations, standardized cranial magnetic resonance imaging, neurocognitive testing, and surveys. After data collection, one community withdrew consent leaving a total of 817 CDCAI participants.

A subset of CDCAI participants also were enrolled in the Strong Heart Family Study [11] (SHFS). The SHFS is a longitudinal study of genetic risk factors for cardiovascular disease among AI families from the SHS [11]. The SHFS recruited SHS participants and their family members for an exam in 2001–03 during which participants were given pedometers to collect activity data [12].

At the baseline SHS visit, participants completed a detailed PA questionnaire designed specifically for AIs [13] (see also online suppl. methods; see www.karger.com/doi/10.1159/000521791 for all online suppl. material). Self-reported measures of leisure time and occupational PA of at least moderate intensity over the past year were collected and used to estimate usual total PA levels. Occupational activity also included any time spent walking or cycling to work per day. Persons who were not employed listed the non-leisure time activities performed in a typical 8-h day. The average hours per week spent in each activity was multiplied by an estimate of the metabolic cost of the activity [14] to obtain metabolic equivalent (MET) hours per week. MET-hours per week of activity for leisure time and moderate- to high-intensity occupational activities were summed to obtain a measure of total energy expenditure.

Pedometer activity was measured during 2001–2003 in the SHFS using an Accusplit AE120 (Yamax, Setagaya-ku, Japan) pedometer, which is a valid and reliable tool for assessing step counts in individuals in free-living settings [15]. Participants wore the hip pedometer during all waking hours for 7 consecutive days, except while bathing or swimming, and recorded steps taken each day. As a metric of objectively measured PA, we used average daily steps in participants with at least 3 days of pedometer data. This threshold was based on previous research that suggests 3 days of pedometer activity is sufficient to estimate average daily steps in a field setting [16].

During 2010–2013, cognitive function was measured in CDCAI using a battery of tests. The Modified Mini-Mental State (3MS) examination was used to assess general cognitive function [17]. The Controlled Oral Word Association Test (COWAT)-FAS was used to assess phonemic fluency by requiring participants to name as many words as possible in 1 min that began with a given letter, i.e., F, A, and S [18]. The California Verbal Learning Test-Second Edition Short Form (CVLT-II SF) required participants to learn, recall, and recognize a list of words in a series of immediate, short, long, and cued delay tests [19]. We used the CVLT-II SF delayed recall raw scores to assess verbal learning and memory. Processing speed (i.e., the ability to process simple or routine visual information quickly and efficiently) was measured by the Wechsler Adult Intelligence Scale-Fourth Edition(WAIS-IV) coding subtest, in which participants paired symbols with numbers as fast as possible for 2 min [20]. For all tests, higher scores indicated better function.

Self-reported demographic characteristics (age, sex, education, native language fluency), alcohol, smoking, and medical history were collected by questionnaire. Details regarding clinical variables are included in the online supplementary Methods.

One participant with a baseline history of stroke was excluded leaving a final sample size of 816. Questionnaire-based PA data, and 1 or more cognitive scores were missing for 33 and 51 participants, respectively. These data and any missing covariate data were imputed with multiple imputation by chained equations [21] using 100 imputations and models including variables from the baseline SHS and CDCAI visits. In all analyses including the imputed data, we estimated parameters and standard errors that accounted for the variability across all imputed datasets.

To measure relationships between questionnaire-based PA and continuous cognitive scores, we used regression models adjusted for age, sex, field site, education, native language ability, apolipoprotein-E (APOE) alleles (ε4, ε2 or other), smoking status and BMI measured at the time of PA assessment, and a variable indicating the years between PA and cognitive assessments. PA was categorized into quartiles for analysis. We tested four cognitive domains as outcomes. Although these cognitive domains may be somewhat correlated, we accounted for multiple comparisons using a conservative Bonferroni correction with a statistical significance threshold of p < 0.0125 (i.e., p < 0.05/4 cognitive tests).

To assess relationships between PA and cognitive outcomes in the subset of 165 participants with pedometer data from the SHFS, we categorized mean daily step counts into tertiles rather than quartiles to ensure an adequate sample size in each category. We tested PA tertiles (or continuous mean daily step counts) in linear regression models adjusted for age, sex, study site, education, APOE alleles, native language ability, and interval between PA and cognitive testing. Sensitivity analyses are detailed in the online supplementary Methods. All analyses were performed using Stata, v.15 (StataCorp, LLC, College Station, TX, USA).

At the time of the questionnaire-based PA assessment, participants had a mean age of 52 years. The majority (68%) were female and half (50%) were obese (Table 1). Females reported lower PA than males (median = 64 vs. 107 MET-hours/week, respectively). On average, 21 years elapsed between the questionnaire-based PA and cognitive assessments, and participants had a mean age of 73 years at cognitive testing. Cognitive testing scores categorized by questionnaire-based PA quartile are summarized in Table 2. To aid in comparisons across tests, we standardized raw scores to z scores (mean = 0, SD = 1) using values for each test in the entire sample. Cognitive testing z scores were highest for participants with PA scores in the top quartile across all domains, except for verbal learning and memory.

Questionnaire-based PA was positively associated with 3MS scores (Table 3). Compared to the lowest quartile 1, participants in quartile 4 had 3MS scores 3 (95% CI: 1–5) points higher. Similarly, in sensitivity analyses with PA modeled as a continuous variable, we observed a statistically significant, but modest dose-response: each 20 MET-hours/week increase in PA (approximately equivalent to 1 h of brisk walking 5 times per week [22]) was associated with 0.2 (95% CI: 0.1–0.4) points higher than the average 3MS score. No statistically significant associations (at the Bonferroni-adjusted threshold) were observed between PA and the other scores. In exploratory models, we performed additional adjustment for variables potentially in the causal pathway between PA and cognitive outcomes. These adjustments changed estimated associations only slightly, with no differences in the interpretation of results (online suppl. Table 1). In addition, we included adjustment for baseline marital status, which may serve as a proxy for social support, but estimated associations changed little (online suppl. Table 2).

We expected stronger associations between PA and cognitive test scores after accounting for selective survival and dropout. Using inverse-probability of attrition weighting, we found that some estimates were slightly larger (i.e., 3MS), and confidence intervals widened, i.e., estimates were less precise (model 2 in Table 3). Associations between continuous PA and 3MS score remained significant, with each 20 MET-hours/week increase associated with an average 0.23 point increase in 3MS score.

A subset of SHS participants, who were also in the SHFS and CDCAI, had 3+ days of pedometer-measured PA data (n = 165). On average, the accelerometry measurement occurred 12 years after the questionnaire-based PA assessment and 9 years before the cognitive assessments. The median (intraquartile range) daily steps during the collection period was 3,522 (95% CI: 2,428–4,970) steps. In adjusted models, average daily steps were significantly associated with COWAT-FAS scores (p < 0.001) (Table 4). Participants in the middle and highest tertiles had scores 8 (95% CI: 4–12) and 7 (95% CI: 3–11) points higher than those in the lowest tertile. Other cognitive test scores were not significantly associated with daily steps in this small sample.

We examined relationships between PA levels and tests of cognitive function conducted 9–21 years later in a large population of AI adults. Higher daily step counts were significantly and positively associated with the COWAT-FAS. Participants who reported at least 151.8 MET-hours/week of total PA had ∼3 point higher scores on the 3MS than those reporting 22.6 MET-hours/week or less. Results were mostly consistent in models accounting for potential attrition bias; higher levels mid-life PA were associated with statistically higher cognitive functioning in later life. Overall, PA associations with cognitive scores were modest in terms of clinical significance. While a difference of 1–2 points on the 3MS is clinically detectable, a difference of 5 points has been used by others to denote clinically meaningful differences between groups [23‒26].

Overall levels of objectively measured pedometer activity indicate that inactivity was prevalent in this predominantly rural population of AIs and that a large proportion were not meeting the current US Centers for Disease Control PA recommendations [12, 15, 27]. Nevertheless, our findings suggest that even in adults with generally low levels of activity, modest increases in mid-life PA are associated with higher cognitive test scores in later life. This finding is particularly relevant for AI adults, who may have higher burdens of risk factors for cognitive impairment and be more likely to suffer from cognitive impairment than other race/ethnicity groups [28, 29]. These data combined with the projected growth of the elder population from 7.4% of the AI population in 2010 to 16.8% in 2050 [2] highlight the urgent need for interventions focused on cognitive impairment risk reduction in this at-risk population.

As the population of older adults continues to grow, there is considerable interest in behavioral interventions that mitigate cognitive decline. PA appears to benefit brain health through a number of mechanisms such as increased cerebral blood flow, neural connectivity, cortical plasticity, brain volume, and neurogenesis [30]. Specifically, physical training helps the brain meet its metabolic demands and motor training helps the brain meet neuromuscular demands [31]. General cognitive function [32] as well as episodic memory [33, 34], processing speed [35], verbal fluency [36], and semantic memory [33, 37] have been positively associated with PA. While we did not observe robust associations between PA and processing speed or episodic memory in our population, several factors may account for these differences, including heterogeneity in cognitive test batteries or demographic characteristics of populations studied (such as age and cultural factors), and type and duration of PA assessed. As an example of the complexity of PA measurement, one study found that exercise intensity, but not duration, was positively associated with changes in cognitive function over a 5-year period [35]. In another study, higher level of leisure time PA was associated with reduced risk of dementia approximately 21 years later [38], but no associations were observed for occupational or commuting PA [39].

Our analyses have some limitations which deserve mentioning. Cognitive impairment was not a study exclusion criterion and we lacked cognitive measures at baseline. Thus, we cannot exclude the possibility of reverse causation, i.e., those with low cognitive test scores may have been less likely to participate in PA. If present, this effect may have biased our findings away from the null. However, other studies suggest that it is unlikely that associations between PA and cognition can be fully explained by reverse causality. For example, reverse causality does not explain the improved cognitive test scores observed in exercise trials of elders with mild cognitive impairment or dementia at baseline [40, 41]. Furthermore, substantial cognitive impairment at baseline was unlikely as SHS participants had to provide informed consent. Over reporting is also a concern for self-reported PA [42]. Overreporting may bias associations in either direction depending on the factors that influence it. Hip-worn pedometer-based PA also has limitations in that it may be less accurate in adults with impaired ambulation [43], and does not capture all types of PA (e.g., cycling or swimming). In addition, we relied on measurements at two time points approximately 12 years apart to characterize mid-life PA, and it is unknown whether PA levels are maintained over time, although some evidence suggests PA is consistent across decades [44]. We also lacked data on some potential confounders at SHS baseline, including social support or mid-life depression, which has been associated with late-life dementia [45] and is correlated with lower PA [46]. Depression may also be considered a mediator of the PA-cognition relationship [47], but in our analyses, adjustment for depressive symptoms at the time of cognitive testing had little impact on associations. Finally, our sample consists of primarily rural-dwelling older adult AIs and may not be generalizable to young and urban AIs or other minority groups.

A unique strength of the present analysis is the assessment of self-reported total PA using a culturally adapted and validated questionnaire [13] that includes both occupational/commuting and leisure time PA to provide an estimate of lifestyle activity. Studies of PA in other AI populations suggest that the main source of PA is from household duties (mopping, sweeping, etc.) rather than engagement in leisure time activity [48, 49]. Thus, measurement of leisure time activity alone may be misleading, particularly in light of other contextual issues common in AI communities that may impact leisure time PA such as poverty and environmental barriers to exercise [50]. Additional strengths of our analysis include assessment of multiple cognitive domains that reflect a broad spectrum of cognition relevant for clinical dementia outcomes, and the use of rigorous methods to limit effects of missing data that may obscure or bias associations.

In summary, the results of the present study support the hypothesis that an active lifestyle in middle age is associated with higher scores in tests assessing general cognition and phonemic fluency in late life in AI adults. As one of the few existing analyses of risk factors for cognitive impairment in AIs, our study is an important and timely contribution to the field of cognitive aging. Future research should include expanded measurement of objectively measured PA and the use of longitudinal cognitive assessment to better understand associations between lifestyle factors and changes in cognition over time.

Studies involving human subjects: Institutional Review Boards from each Indian Health Service region and all participating communities approved the SHS, SHFS, and CDCAI studies. All participants provided written informed consent.

C.L.C. serves as a consultant to PATH and receives compensation from Adaptive Biotechnologies.

This work was supported by grants P30AG059295, 5P01AG066584, 5P30AG066509, R01HL093086, P50AG005136, U54MD000507, and K01AG057821. The Strong Heart Study has been funded in whole or in part with federal funds from the National Institutes of Health under contract numbers 75N92019D00027, 75N92019D00028, 75N92019D00029, and 75N92019D00030; cooperative agreements U01HL41642, U01HL41652, U01HL41654, U01HL65520, and U01HL65521; and research grants R01HL109315, R01HL109301, R01HL109284, R01HL109282, and R01HL109319. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the Indian Health Service.

C.L.C. designed the study, conducted analyses, and drafted the manuscript. C.N. assisted with the study’s analytic strategy and conducted analyses. C.M., A.S.-D. S.P.V., and B.V.H. assisted with data acquisition, data interpretation, and manuscript editing. A.M.F. assisted with data interpretation and the analytic approach. D.B. directed data collection and assisted with manuscript editing.

The data analyzed in this study were obtained from the Strong Heart Study. Any data request should be directed to the Strong Heart Study (https://strongheartstudy.org/).