Abstract
Background:
Preliminary evidence suggests associations between neighborhood park access and better late-life cognition and reduced Alzheimer’s disease (AD) risk.
Objective:
Examine associations between neighborhood park access and longitudinal change in cognition among U.S. older adults without dementia.
Methods:
We used 2000–2018 observational data from the population-based, multi-site Multi-Ethnic Study of Atherosclerosis (n = 1,733). Measures included proportion of neighborhood park space (park access), distance to nearest park, and 6-year dichotomous and continuous change in scores on the Cognitive Abilities Screening Instrument (CASI; global cognition) and Digit Symbol Coding task (processing speed). Multivariable random intercept models tested main associations and mediation by depressive symptoms, physical activity, and PM2.5 exposure. Effect modification by race (African Americans/Blacks versus Whites) was tested using interaction terms.
Results:
Greater park access (equivalent to 10%more in 1/2-mile around home) was associated with maintained/improved CASI score over six years independent of several covariates including individual- and neighborhood-level socioeconomic status (Odds ratio: 1.04; 95%confidence interval: 1.00–1.08). No other associations were observed with the dichotomous or continuous measures of cognitive change and no mediators were found. While a borderline association was seen between greater park access and maintained/improved CASI for African Americans/Blacks but not for Whites, effect modification was not confirmed by testing interaction terms.
Conclusion:
Neighborhood park access may help maintain/improve late-life global cognition. However, our findings need replication in other population-based studies and regions. Additionally, studies are needed to determine if associations between park access and change in cognition vary by race/ethnicity to inform intervention efforts.
INTRODUCTION
Established risk factors for cognitive decline include but are not limited to older age, lower educational achievement/socioeconomic status (SES), cardiovascular disease and risk factors, and genetic risk factors for Alzheimer’s disease (AD) (e.g., having≥1 apolipoprotein ɛ4 allele) [1, 2]. In contrast, lifestyle factors including healthy diet and physical activity have been associated with slower cognitive decline and reduced AD risk [3, 4].
Neighborhood environments that encourage healthy behaviors may serve as modifiable factors that can be promoted at the population level to prevent or delay the onset of cognitive decline and AD in late life. Although research on neighborhood built environments and cognition in older adults is limited, preliminary evidence suggests that neighborhood walkability (environments conducive to walking) may be associated with better cognitive function [5, 6]. Similarly, few studies have investigated associations between neighborhood park access and cognition, but among those that have, some suggest that greater access and exposure to parks is associated with better cognition and slower cognitive decline in older age [7–10]. For instance, using data from the population-based Multi-Ethnic Study of Atherosclerosis (MESA), authors from this current study found that having greater park space in the 1-mile surrounding the residence was cross-sectionally associated with better processing speed but not global cognition [10]. Parks offer community spaces not only for physical activity, but also stress reduction and social engagement, which have been associated with improved cognition among older adults [11, 12]. Time spent in green space may improve mental health by reducing anxiety and depression [13], conditions associated with cognitive decline in older age [14]. In addition, time spent in parks and living near parks may be associated with a reduction in traffic-related air pollution exposure that has been associated with change in cognition over time [15].
Given the limited evidence for park-cognition associations in older adults, this study aimed to investigate associations between neighborhood park access and change in cognitive function in late life, and whether these associations were mediated by depressive symptoms, physical activity, or air pollution exposure or moderated by race. We hypothesized that: 1) greater access to neighborhood parks and shorter distance to the nearest park would be associated with maintained/improved cognition over time; 2) the associations would be mediated by lower depressive symptoms, greater physical activity, and lower air pollution exposure; and 3) the associations would be stronger in African Americans/Blacks versus Whites. The overarching goal of this research is to determine community-level intervention opportunities to improve cognitive health and lower AD risk.
MATERIALS AND METHODS
Sample
We obtained de-identified, longitudinal data from MESA, a study of subclinical cardiovascular disease that enrolled 6,814 participants in 2000–2002. MESA enrollees resided in six U.S. cities (Forsyth County, North Carolina; New York, New York; Baltimore, Maryland; Saint Paul, Minnesota; Chicago, Illinois; Los Angeles, California), and to date have completed six exams over 18 years. At each exam, data have been collected on demographics, health history and health status, anthropometry, medications, self-reported physical activity, as well as a number of medical assessments and procedures (e.g., imaging, blood pressure). Further details about the MESA cohort are found elsewhere [16, 17]. The primary analyses were restricted to participants with longitudinal cognitive exam data from Exam 5 (2010–2012) and Exam 6 (2016–2018), usable Cognitive Abilities Screening Instrument (CASI) scores (excluded scores < 20 lacking face validity and CASI missing≥3 items), and non-missing data from Exam 5 on proportion of neighborhood park space and distance to the nearest park. Supplementary Figure 1 provides a diagram of the primary measures used by exam/year.
Sample selection flow chart.
Park access measures
The ancillary MESA Neighborhood Study [18] calculated numerous neighborhood characteristics for each participant based on residential addresses. New addresses were accounted for when calculating neighborhood characteristics based on participants’ addresses at a given exam. Neighborhood variables were developed for Exam 1 to Exam 5. In this study, we operationalized park access as the proportion of the land surrounding participants’ residences composed of parks and straight-line (Euclidian) distance in kilometers to the nearest park. Park amenities in 36 counties at the six MESA sites were determined by contacting municipal/county departments, searching online resources, and visiting parks in order to exclude ornamental or dog parks. Additional details on the park data are provided elsewhere [18, 19]. The 1/2-mile area (i.e., radial buffer) surrounding the residence was the primary neighborhood boundary of interest. We hypothesized that the area closer to the home better represented the area visited by older adults, who may experience shrinking life space and greater disability/frailty with age. In addition, research suggests individuals walk approximately half a mile to neighborhood destinations on average [20]. Given the lack of a standardized neighborhood boundary definition across studies, we also evaluated park access within 1-mile radial buffers. The neighborhood park measures at Exam 5 were the primary measures of interest because these data were available for participants originating from all six sites.
Other neighborhood measures
Neighborhood SES for each US Census tract was previously calculated using a principal components analysis of seven US Census American Community Survey variables (2007–2011) [21]. The variables included neighborhood median home value, median household income, and percentage rental income, as well as percentage of neighborhood residents with annual household income > $50,000, a managerial occupation, a high school degree, and a bachelor’s degree. Neighborhood population density (persons/km2) for the 1/2-mile area surrounding the residence was determined using 2010 Census block data. Assuming an equal distribution of the population per block, the population density for the radial buffer was calculated by adding up the population density for the included blocks determined based on the percentage of a 1/2-mile radial buffer in each block.
Change in cognition
MESA participants received cognitive testing at Exams 5 and 6 (no cognitive testing done prior to Exam 5). Two cognitive measures were used, the CASI (version 2) [22] and the Digit Symbol Coding (DSC) task [23]. CASI is a global measure of cognitive function assessing attention, concentration, judgment, orientation, abstraction, short-term and long-term memory, verbal fluency, language, and visual construction (range: 0–100), and the DSC measures processing speed (range: 0–133). Higher scores equate with better cognitive function. The Digit Span test, a measure of working memory, was less consistently associated with built environment measures compared to the CASI and DSC in prior cross-sectional studies [6, 24] and therefore was excluded to reduce multiple testing.
To create dichotomous measures of longitudinal change in CASI and DSC scores, we calculated continuous measures of annual change on the tests from Exam 5 to 6 by running an unadjusted linear regression for each participant with the cognitive test as the outcome and time as the predictor (Exam 5 = 0, Exam 6 = years since Exam 5). We then derived the dichotomous outcomes for the regression analyses by categorizing the resulting continuous measures as maintained/improved score (e.g., change in CASI = 0 or increase from Exam 5 to 6) versus decline in score from Exam 5 to 6. For the regression analyses focused on continuous measures of cognition, raw CASI and DSC scores were first transformed into z-scores by subtracting from each participant’s score the mean score of the sample and dividing the result by the standard deviation of the sample.
While we define cognitive change dichotomously and continuously in our analyses, the dichotomous measures were a priori of primary interest because the study aimed to determine if having greater park access was associated with either maintained or improved cognition over time. A maintenance of cognition over six years may be clinically significant among older adults who frequently experience subtle declines in cognition due to normal aging [25].
Covariates
Participant demographics included age in years, sex, race/ethnicity (White, Chinese American, African American/Black, Hispanic), education (< high school degree, high school degree, some college, bachelor’s degree or higher) and family income (≥$30,000/year versus < $30,000/year). Health-related variables included having≥1 apolipoprotein ɛ4 allele (APOE ɛ4, genetic risk factor for AD [26]), and self-reported measures of diabetes, arthritis (any type), cardiovascular disease, and cerebrovascular disease (stroke or transient ischemic attack). Participant appraisals of neighborhood safety walking day or night and safety from crime were reported on a 5-point scale (1 = strongly agree to 5 = strongly disagree). Lastly, the number of residential moves during MESA follow-up were categorized (none, one, two or more). Table 1 presents the MESA exam for which these covariates were based.
Participant and neighborhood characteristics
SD, standard deviation; APOE, apolipoprotein E; CES-D, Center for Epidemiologic Studies Depression scale; COPD, chronic obstructive pulmonary disease; TIA, transient ischemic attack; SD, standard deviation; BMI, body mass index; PM, particular matter; O3, ozone; CASI, Cognitive Abilities Screening Instrument. aMissing: education, n = 5; income, n = 41; APOE, n = 93; CES-D, n = 17; diabetes, n = 11; arthritis, n = 9; cardiovascular/cerebrovascular disease, n = 1; physical activity, n = 5; neighborhood socioeconomic status, n = 29; PM2.5, n = 73. bSelf-reported; tertiles: 1 = < 2505 MET-min; 2 = 2505 to < 6015 MET-min; 3 =≥6015 MET-min. cpositive = worse neighborhood SES; negative = better neighborhood SES.
Potential mediators
Individual-level air pollution exposures, depressive symptoms, and physical activity were evaluated as potential mediators as they have been previously associated with cognition in older adults [27–29] and are conceivably related to neighborhood park space exposure. The MESA Air ancillary study [30] derived annual measures of participant exposure to particular matter < 2.5μm in size (PM2.5) from 1999–2016. Our air pollution exposure measure was based on the mean of the participant’s 1999–2016 PM2.5 values. Severity of depressive symptoms at Exam 5 was determined using the Center for Epidemiological Studies Depression Scale [31] ([CES-D] score≥16). Total Metabolic Equivalent (MET)-minutes of moderate and/or vigorous physical activity was calculated from a self-reported questionnaire [32] collected at Exam 5, which included all reported moderate/vigorous activity achieved via any means (e.g., purposeful exercise, household chores, lawn/garden, childcare, transportation, work).
Analyses
We described the sample overall and stratified by the dichotomous measure of change in CASI score from Exam 5 to 6 (maintained/improved score versus decline) using descriptive statistics (e.g., means, standard deviations [SD]). Scatterplots were examined to confirm that there was not a specific threshold for the associations between the park and cognition variables.
We tested associations between the continuous park measures and dichotomous cognitive change measures using unadjusted and multivariable random intercept logistic regression (accounting for data clustering at the US Census tract level). In addition, unadjusted and multivariable random intercept linear regression models (clustered on Census tract and participant) were run using the continuous cognitive z-scores and including variables for time (years since Exam 5) and interaction terms between the park variables and time (e.g., proportion park space×years). We tested each combination of park measure (distance and proportion) and cognitive measure (CASI and DSC) in separate regression models. We ran models for proportion of park space using both the 1/2-mile and 1-mile buffers. Minimal models controlled for age, sex, education, race/ethnicity, income, ≥1 APOE ɛ4 allele, number of residential moves, neighborhood SES, neighborhood population density, and site. Full models additionally controlled for appraisals of neighborhood safety walking and safety from crime, arthritis, cardiovascular and cerebrovascular disease, and diabetes. The covariates included in the multivariable analyses were chosen a priori based on previous associations with cognition or because they were hypothesized confounders of the park-cognition associations. Race was assessed as a potential effect modifier by including multiplicative interaction terms between the park variables and race in the multivariable models (excluding Saint Paul, Minnesota, sample because it included no African American/Black participants).
The causal steps approach was carried out using the R “lme4” package to test whether the association between proportion of park space (X) and longitudinal change in CASI (Y) was mediated by depressive symptoms, total MET-minutes of physical activity/week, or PM2.5 exposure (M) [33]. The three potential mediators were tested independently by running three multivariable random intercept regression models to test: 1) the association between X and Y; 2) the association between X and Y while controlling for M, and 3) the association between X and M. We tested for mediation by the variables using both linear and logistic regression and controlling for the covariates from the previously specified minimal models.
We conducted several sensitivity analyses. First, we calculated propensity scores based on age, sex, education, race/ethnicity, income, marital status, site, and comorbidities at Exam 1 to address potential selection/attrition bias. The propensity scores were incorporated into inverse probability weights that were applied to the full logistic and linear regression models. Second, we examined associations between the neighborhood park measures at Exam 1 (available for participants from Saint Paul, Chicago, Los Angeles) and dichotomous cognitive measures, to determine if the time period measured affected the associations. Third, we restricted the analyses of dichotomous cognitive measures to individuals who did not move during MESA follow-up, to assess whether moving to new neighborhoods may help explain observed associations. In the two final sensitivity analyses, the cut point for the dichotomous cognitive change measures were changed to > 1 SD increase in score from Exam 5 to 6 (versus≤1 SD increase) and to > 1 SD decrease in score (versus≤1 SD decrease) to determine if the results changed depending on the chosen cut points. The > 1 SD increase cut point helped address potential learning effects that result in improvement in cognitive test scores over time due to learning the test and not necessarily due to the exposure of interest. The > 1 SD decrease cut point provides the outcome in terms of clinically significant decline instead of maintenance/improvement in cognition over time. A score > 1 SD below the mean for healthy individuals is often used to diagnose cognitive impairment [34].
All regression analyses were conducted in R version 4.4.0 using the “lme4” package (“lm” function).
RESULTS
After exclusions the sample included 1,733 participants (Fig. 1). Participants were an average of 67 years old (SD = 8), 53%were female, and 75%had at least some college education (Table 1). Twenty-nine percent were African American/Black, 17%Hispanic, 11%Chinese American, and 43%White. Twenty-eight percent were APOE ɛ4 carriers. The majority (73%) did not move residences during follow-up (2000–2012), while 19%moved once and 8%moved twice or more. Compared to non-movers, movers more often experienced a > 1SD change (increase or decrease) in proportion of neighborhood park space and distance to the nearest park from Exam 1 to 5 (Supplementary Table 1). Eighty-eight percent reported feeling safe walking in their neighborhood day or night (agree, strongly agree) and 73%reported feeling safe from crime. On average, participants lived in neighborhoods with 4.9%park space (SD = 8.3%, range = 0.0–52.2%) and lived 0.80 km (1/2-mile) to the nearest park (SD = 1.19 km, range = 0.0–20.8 km) (Supplementary Figure 2). The average time between Exam 5 and 6 was 6.3 years (period in which cognitive change was measured) (SD = 0.5). Table 1 presents additional participant characteristics.
Participants had mean scores of 89.8 on the CASI (SD = 6.9) and 55.2 (SD = 16.9) on the DSC at Exam 5 (Supplementary Table 2). Their six-year mean change in scores from Exam 5 to 6 was 0.03 (SD = 1.02) for the CASI and –0.73 (SD = 1.83) for the DSC (Supplementary Table 2, Supplementary Figure 2).
Overall, small differences were observed in participant characteristics between those with maintained/improved versus declining CASI scores (Table 1). Compared to individuals with declining scores, those with maintained/improved CASI scores were younger (mean of 66 years versus 69 years), more often had a college degree (46%versus 43%), were more often Chinese American (13%versus 9%), had higher income (77%versus 72%), less often had≥1 APOE ɛ4 allele (26%versus 29%), and less often were in the lowest tertile of MET-minutes of moderate/vigorous physical activity (32%versus 35%). On average, individuals with maintained/improved CASI scores had greater proportions of neighborhood park space and shorter distances to the nearest park (5.3%and 0.74 km, respectively) compared to those with declining scores (4.4%and 0.86, respectively).
In the unadjusted regression analyses, proportion of park space in the 1/2-mile surrounding the residence was associated with maintained/improved CASI score (OR: 1.03; 95%CI: 1.00–1.06) (Supplementary Table 3). Proportion of park space was not associated with change in DSC score and distance to the nearest park was not associated with CASI or DSC. The minimal and full models resulted in similar findings, with every 0.10 increase in proportion of park space (i.e., 10%) associated with maintained/improved CASI score (fully model OR: 1.04; 95%CI: 1.00–1.08) (Table 2). In the linear regression analyses, proportion of park space and distance to the nearest park were not associated with longitudinal change in CASI and DSC z-scores (Table 3). Park measures based on 1-mile buffers were not associated with the cognitive measures. Depressive symptoms, total MET-minutes of physical activity/week, and PM2.5 exposure were not statistical mediators of the associations between proportion of park space and the continuous measure of change in CASI z-score (Supplementary Figure 3), or for the association between proportion of park space and the dichotomous measure of change in CASI z-score (data not shown).
Multivariable associations between continuous park measures and dichotomous measures of cognitive change
OR, odds ratio; CI, confidence interval; CASI, Cognitive Abilities Screening Instrument; DSC, Digit Symbol Coding. aProportion park space (in 1/2-mile radial area surrounding residence) is per 0.10 (i.e., 10%); range observed in sample: 0 to 52%; distance to nearest park in kilometers; range observed in sample: 0.00–20.8 km. bControlling for age, sex, education, race/ethnicity, income, neighborhood socioeconomic status, site, APOE ɛ4 carrier, number of residential moves, neighborhood population density. cControlling for covariates from minimal model plus perception of neighborhood safety walking, perception of neighborhood crime, arthritis, cardiovascular disease, cerebrovascular disease, diabetes.
Multivariable associations between continuous park measures and continuous measures of cognitive change
CI, confidence interval; CASI, Cognitive Abilities Screening Instrument; DSC, Digit Symbol Coding; y, years. aProportion park space (in 1/2-mile radial area surrounding residence) is per 0.10 (i.e., 10%); range observed in sample: 0 to 52%; distance to nearest park in kilometers; range observed in sample: 0.00–20.8 km. bControlling for age, sex, education, race/ethnicity, income, neighborhood socioeconomic status, site, APOE ɛ4 carrier, number of residential moves, neighborhood population density. cControlling for covariates from minimal model plus perception of neighborhood safety walking, perception of neighborhood crime, arthritis, cardiovascular disease, cerebrovascular disease, diabetes.
The racial/ethnic composition of the sample by study site is provided in Supplementary Table 4. As most sites were missing representation of all the racial/ethnic groups, the examination of effect modification by race was restricted to African Americans/Blacks and Whites and excluded individuals from Saint Paul, Minnesota, where no African Americans/Blacks contributed to the sample. Race was not an effect modifier based on the interaction terms included in the logistic regression models (Table 4) or in post hoc random intercept linear regression models that included the interaction terms (data not shown). Nonetheless, when examining stratified associations between every 0.10 point increase (i.e., 10%) in proportion of park space (1-mile buffer) and maintained/improved CASI score, we found borderline associations for African Americans/Blacks (OR: 1.07; 95%CI: 1.00, 1.14) but not Whites (OR: 1.04, 95%CI: 0.96, 1.12) (Table 4).
Multivariable associations between continuous park measures and dichotomous measures of cognitive change by race
OR, odds ratio; CI, confidence interval; CASI, Cognitive Abilities Screening Instrument; DSC, Digit Symbol Coding. Excluding Saint Paul site, which included no African Americans/Blacks (see Supplementary Table 3). aControlling for age, sex, education, income, neighborhood socioeconomic status, site, APOE ɛ4 carrier, perception of neighborhood safety walking, perception of neighborhood crime, arthritis, cardiovascular disease, cerebrovascular disease, diabetes, number of residential moves, neighborhood population density. bProportion park space (in 1/2-mile radial area surrounding residence) is per 0.10 (i.e., 10%); range observed in sample: 0–52%.
Sensitivity analyses
The use of inverse probability weights to account for attrition/selection bias resulted in little change in the main findings of the logistic regression (Supplementary Table 5) and linear regression analyses (Supplementary Table 6). Focusing on park measures from Exam 1 (Supplementary Table 7) and restricting to individuals who did not move during MESA follow-up (Supplementary Table 8) resulted in comparable odds ratios as the main analyses, but the reduced sample size resulted in wider confidence intervals. After dichotomizing the cognitive outcome as a > 1SD increase in score (versus≤1 SD increase), proportion of park space was not associated with CASI or DSC, but greater distance to the nearest park was associated with lower odds of improved CASI scores (OR = 0.98; 95%CI: 0.96–0.99) (Supplementary Table 9). In the final sensitivity analysis of individuals with a > 1SD decrease in score (versus≤1 SD decrease), proportion of park space and distance to the nearest park were not associated with change in CASI or DSC (Supplementary Table 10).
DISCUSSION
Proportion of neighborhood park space was associated with maintained/improved global cognition (versus decline) over six years among older adults. Depressive symptoms, total physical activity/week, and PM2.5 exposure did not mediate the association between proportion of park space and global cognition. No associations were found between residential distance to the nearest park and cognitive change. Analyses yielded comparable odds ratios when using different neighborhood boundaries (1/2- versus 1-mile), excluding movers, and focusing on a different period of park space assessment (Exam 1 versus Exam 5), but not when altering the definition of longitudinal cognitive change (i.e., when using a continuous measure or different cut points). Finally, race was not found to be an effect modifier based on interaction terms, although when analyses were stratified by race, borderline associations were found between greater proportion of park space and maintained/improved global cognition for African Americans/Blacks but not Whites.
It must be noted that the logistic regression results appeared sensitive to grouping. In sensitivity analyses, the positive association between park access and change in global cognition became null when altering the cut point to a > 1SD increase in score (versus≤1SD increase) or a > 1SD decrease in score from Exam 5 to 6 (versus≤1SD decrease). These 1 SD cut points, while similar to those used to detect the presence of cognitive impairment [34], may not be appropriate for studying associations with subtler changes in cognition among individuals without impairment at baseline, as was the goal of this study. Thus, our findings must be corroborated in other cohorts without dementia and further work is needed to identify the potential influence of retesting effects (improvement with practice) when attempting to assess clinically significant and positive influences of park access on cognition over time.
Neighborhood park access may be associated with increased cognitive reserve, resilience, or resistance [35] to neuropathology that typically results in cognitive decline. Compared to individuals with no neighborhood parks, those with 52%of the neighborhood composed of park space (highest percentage observed in sample) were 1.2 times more likely to experience maintained/improved CASI scores over six years. In addition, the borderline significant finding for the continuous CASI measure also suggested that compared to those with no neighborhood parks, those with a greater percentage of neighborhood park space experienced a slower decline on the CASI over 6 years. However, given the tentative nature of our findings, an evaluation of the clinical significance is limited until our results are replicated in other studies, including those employing additional cognitive assessments, other neuropsychological tests sensitive to longitudinal changes in cognition, and in other racially/ethnically and geographically diverse cohorts.
Few studies have investigated associations between neighborhood park access and cognition in older adults [8, 37]. Greater neighborhood park access was associated with better processing speed (DSC) but not global cognition in a cross-sectional study using the MESA cohort [10]. In addition, greater neighborhood park area (mi2) was not associated with scores on the Telephone Interview Cognitive Status (TICS) in a cross-sectional study of older Chicago adults [36]. Two other studies examined associations between neighborhood park access in early-, mid-, and late-life and longitudinal change of Moray House Test scores (measure of intelligence) among the Lothian Birth Cohort study in Scotland. The first found that greater percentage of neighborhood park space in childhood and adulthood was associated with slower cognitive decline from ages 70 to 76 [8]. The second found that percentage of park space near the home, school, and route to school at age 11–18 was associated with slower cognitive decline from ages 70 to 76 [37]. Our study makes a significant contribution to the extant literature on the topic by finding associations between proportion of neighborhood park space during older age and maintained/improved cognition over time. Results from this study taken together with the prior studies suggest that park exposure throughout the life course may be beneficial to late-life cognitive health. The mechanisms are yet to be determined but may involve associated life-long health behaviors that promote cognitive reserve, resilience, and resistance and protect against late-life cognitive decline [35].
Two other studies are relevant to our findings. The first found that greater neighborhood greenness (i.e., healthy vegetation) measured using the normalized difference vegetation index was associated with slower decline in global cognition, reasoning, and fluency among middle-age and older adults in the UK [9]. Although the results parallel our findings in some ways, the use of a neighborhood greenness measure targets a potentially different construct because it captures green spaces and amount of vegetation around homes, businesses, and public areas, while park measures focus on publicly available green space. Yet, the two measures naturally overlap. Further work is needed to determine whether park access (and associated amenities) or greenness is the stronger or only predictor of cognitive change in older age.
The second study found that, among middle-aged and older adults in Europe (Spain, the Netherlands, and UK), greater distance to the nearest outdoor natural environment was associated with longer time to complete the Color Trails Test [38]. The Color Trails Test, a measure of visual attention and effortful executive processing, was measured cross-sectionally and distance to the natural environment was not restricted to park spaces. In contrast, our study was based on longitudinal differences in other cognitive domains (global cognition, processing speed). These differences may explain our lack of findings for an association between distance to nearest park and cognitive change. Additional studies are needed to determine if distance to the nearest park is associated with longitudinal cognitive decline, including various cognitive domains such as attention that have been found to be associated with both park space exposure and the associated health behaviors/exposures.
This study has a number of methodological limitations. While the overall retention rate has been estimated at 80%, previous comparisons of Exam 5 participants to Exam 1 participants not included in Exam 5 showed that those lost to follow-up were more often racial/ethnic minorities and had lower SES [39]. In addition, approximately one-third of the sample did not receive repeat cognitive testing at Exam 6. The exclusion of individuals missing park data may have biased findings and excluded individuals living in rural areas. While attrition and selection bias may have affected our findings, our use of inverse probability weighting helps reduce those concerns. Most of the sample had at least some college education, and thus findings may not be generalizable to those with lower education/SES. It is possible that our findings are biased by residual confounding from factors such as SES or unmeasured neighborhood characteristics (e.g., community resources) that may not have been fully accounted for in our analyses. For instance, neighborhood SES measured at a smaller scale (e.g., Census block group) was not available for this study but may better control for potential confounding by this factor. Associations were observed using dichotomous but not continuous measures of cognitive change and multiple comparisons may have resulted in the significant association between proportion of neighborhood park space and maintained/improved CASI score. Our sensitivity analyses suggested that the choice of cutpoints for the dichotomous outcome measures made a difference to the findings. Future studies could employ methods such as converting the cognitive test outcomes to interval scaled measures to understand potential threshold or floor and ceiling effects. Our mediation analyses were exploratory. New studies would benefit from more advanced mediation methods (e.g., testing mediation and interaction simultaneously [40]) and by evaluating other potential mediators such as objectively-measured physical activity (i.e., accelerometers) and social engagement. In addition, we could not eliminate the possibility of reverse causality, in which individuals who have better cognition (or are more likely to exhibit learning effects when taking repeated cognitive tests) are more likely to live and/or remain in neighborhoods with more park space. Thus, further research will need to replicate our findings and account for potential reverse causality and self-selection.
Another limitation was the lack of complete racial/ethnic representation from each MESA study site. This limited our ability to stratify findings by race/ethnicity and examine effect modification because any observed racial/ethnic difference in associations may actually be the result of site-based differences in park access and quality. Nevertheless, the borderline findings when stratifying by race in sites with both African Americans/Blacks and Whites suggest that greater park access may be associated with better cognition in African Americans/Blacks but not in Whites. Cost of exercise facilities and lack of nearby activity spaces such as parks are known barriers to physical activity among African Americans/Blacks [41], and physical activity has been shown to improve cognition [4]. The potential cognitive benefit among African Americans/Blacks is important to clarify in future studies, as improvements in park access could provide a population-based approach to reduce known racial/ethnic disparities in late-life cognitive function [42].
Future studies could build upon our preliminary work using more refined exposure and outcome measures. Park exposures from childhood and middle adulthood were unavailable and may demonstrate stronger associations with cognition compared to late-life measures. Future studies will benefit from objective measures of time spent in park spaces, such as those obtained from global positioning system devices, and from measures of the quality of park spaces (e.g., amenities) that may be most beneficial to cognition. The cognitive domains investigated were limited and the available cognitive tests do not represent the full battery of tests that dementia specialists typically use to diagnose dementia. A more comprehensive set of tests should be incorporated in new studies. Lastly, future work is needed to understand differences in park access and quality and how this may be related to racial/ethnic disparities in cognitive decline.
The strengths of the study include the use of a population-based, longitudinal cohort with racial/ethnic and geographic diversity. Rich data were available from MESA, allowing for the control for important confounders and examination of mediators. We accounted for number of residential moves in our analyses, which was not accounted for in the prior studies of park space and late-life cognition. Multiple radial buffers were employed to determine if associations varied depending on the definition of the neighborhood boundary. Lastly, this is the first known study to investigate associations between neighborhood park access and cognitive decline in US older adults. This is significant in that the associations may vary by geographical context.
Overall, the evidence for associations between neighborhood park access and cognition is suggestive but limited and currently insufficient to base policy or intervention recommendations to improve cognitive health in older adults. However, the findings from this study can serve as the basis for future studies of park access, cognitive decline, and AD risk based on population-based cohorts and using life course methods.
Footnotes
ACKNOWLEDGMENTS
The objective neighborhood data used in this study were developed as part of Ana Diez Roux’s MESA Neighborhood Study (R01 HL071759). This research was supported by contracts 75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, N01-HC-95163, 75N92020D00004, N01-HC-95164, 75N92020D00007, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168 and N01-HC-95169 from the National Heart, Lung, and Blood Institute, and by grants UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420 from the National Center for Advancing Translational Sciences (NCATS).
The authors thank the other investigators, the staff, and the participants of the MESA study for their valuable contributions. A full list of participating MESA investigators and institutions can be found at 
.
This publication was developed under a STAR research assistance agreements, No. RD831697 (MESA Air) and RD-83830001 (MESA Air Next Stage), awarded by the U.S Environmental Protection Agency. It has not been formally reviewed by the EPA. The views expressed in this document are solely those of the authors and the EPA does not endorse any products or commercial services mentioned in this publication.
This work was supported by National Institute on Aging grants K01AG063895 (L.M. Besser), 3R01AG049970-04S1 (Y. Michael); P30AG049638, R01AG054069, R01AG053938, R01AG054491, R01AG058969, RF1NS110043, R03AG064569, and U01HL096812 (T. Hughes), R01AG040211 and R01NS101483 (J. Galvin); HHSN271-2011-00004C, HHSN268201600004C, AG055606, AG050657, P30AG049638, HL133684, R01AG058969, R01AG058571, 1U19AG065188, U10CA, and R01 HL127659 (S. Rapp); National Institutes of Health grants RF1AG057033 and P30ES007033 (J. Kaufman) and R01HL131610 (J. Hirsch); and US Environmental Protection Agency grants RD831697 and RD-83830001 (J. Kaufman).