Association of Neighborhood Physical Activity Opportunities with Incident Cardiovascular Disease in the Cardiovascular Health Study

Abstract

We determined associations of cumulative exposures to neighborhood physical activity opportunities with risk of incident cardiovascular disease (CVD). We included 3595 participants from the Cardiovascular Health Study recruited between 1989-1993 (mean age=73; 60% women; 11% black). Neighborhood environment measures were calculated using Geographic Information Systems (GIS) and annual information from the National Establishment Time Series database, including the density of (1) walking destinations and (2) physical activity/recreational facilities in a 1- and 5-km radius around the respondent’s home. Incident CVD was defined as the development of myocardial infarction, stroke, or cardiovascular death and associations with time to incident CVD were estimated using Cox proportional hazards models. A total of 1986 incident CVD cases occurred over a median follow-up of 11.2 years. After adjusting for baseline and time-varying individual and neighborhood-level confounding, a one standard deviation increase in walking destinations and physical activity/recreational facilities within 5 km of home was associated with a respective 7% (95% confidence interval (CI)=0.87-0.99) and 12% (95% CI=0.73-1.0) decreased risk of incident CVD. No significant associations were noted within a 1-km radius. Efforts to improve the availability of physical activity resources in neighborhoods may be an important strategy for lowering CVD.

INTRODUCTION

While the overall cardiovascular disease (CVD) prevalence in US adults is approximately 9%, this number rises dramatically in older individuals. It is estimated that 43% of men and 31% of women over the age of 80 have at least one form of CVD.¹ Nearly five percent of all major CVD events are directly attributed to not meeting American Heart Association physical activity (PA) recommendations.² This is particularly important for older adults as over 70% of individuals 65 years of age or older do not meet these PA recommendations.³ Studies in older adults have demonstrated that even modest improvements, however, in physical activity can decrease overall mortality by over 20% and CVD incidence by as much as 50%^4–6

While the emphasis on CVD prevention has traditionally focused on an individualized approach to improve risk factors, including PA, reducing the overall burden of CVD remains an important public health priority. Public policies that make it easier to incorporate physical activity into daily life may be effective in this regard and has been advocated for by the US surgeon general.⁷ Objectively measured attributes allow control over the spatial unit representing the home neighborhood to better understand the environmental factors most relevant for different health-related processes. In this respect, objectively measured attributes may have direct relevance to policy or community-driven action. Geographic Information Systems (GIS) based measures affords the ability to assess relationships between objectively measured neighborhood physical environments and CVD.^{8, 9}

Older individuals may be particularly affected by aspects of their physical and social neighborhood environment.^{10, 11} The space in which individuals live their lives constricts with age and associated life events (e.g., retirement), consequently, adults typically spend more time within their neighborhoods as they age.¹² As a result, they may rely more heavily on their immediate environment for physical resources, such as food shopping and recreational facilities, and social interaction.^{13, 14}

Better neighborhood walkability, street connectivity, and access to or availability of recreational facilities are associated with higher PA levels as well as slower longitudinal declines in PA levels over time in older adults.^15–18 A meta-analysis that included over 30 longitudinal studies found that attributes of the built environment that promote physical activity were independently linked to a reduced presence of important cardiac risk factors—obesity, diabetes, and hypertension.¹⁹ Therefore, we hypothesized that neighborhood characteristics conducive to physical activity, therefore, may decrease the incidence of CVD.

While research to understand how neighborhood environments impact older residents’ physical health has grown, the literature with respect to CVD is limited. Self-reported measures of the walking/physical activity environment and GIS-based measures of proximity to physical activity resources have been associated with better cardiovascular health and a lower estimated long-term cardiovascular disease risk, but not a slower progression of subclinical coronary atherosclerosis.^20–22 More importantly, a GIS-based assessment of proximity to physical activity facilities in a national Swedish cohort was not associated with a reduced incident CVD risk.^23–26 These studies were performed in a population that spanned a broad age range and was not well-characterized at baseline. In addition, longitudinal measures of neighborhood characteristics were not incorporated. Neighborhoods are dynamic environments, and not accounting for changes over time limits the ability to develop valid estimates of neighborhood effect on health.

This paper estimated the longitudinal associations between neighborhood exposures to (1) walking destinations and (2) physical activity resources with risk of incident cardiovascular disease in an older population that was well-characterized at baseline.

METHODS

Study population

The Cardiovascular Health Study (CHS) is a community-based, longitudinal observational study of adults aged 65 and older at baseline designed to evaluate risk factors for the development and progression of cardiovascular disease. The study’s primary objectives and design have been reported previously.^{27, 28} An initial cohort of 5,201 individuals was recruited in 1989-1990, and a supplemental cohort of 687 predominantly African American participants was recruited in 1992-1993. All participants were residents of two predominantly urban communities (Sacramento (Sacramento County), CA, Pittsburgh (Alleghany County), PA) and two predominantly rural communities (Hagerstown (Washington County), MD; Winston-Salem (Forsyth County), NC), using a random sample of Medicare eligibility lists. The CHS received approval from institutional review boards of all participating centers and all participants provided written informed consent. Self-reported health behaviors, history of diseases, anthropometric measures, current medication use, seated blood pressure readings, electrocardiogram recordings, and fasting blood chemistry measures were obtained during the baseline interview and clinical examination.

Individual neighborhoods were defined by geocoding residential addresses of CHS participants within the contiguous US and generating 1- and 5-km radial buffers around the geocoded locations, removing areas of hydrography. Characteristics of the buffer areas were measured through a spatial overlay with census tracts (using 2010 boundaries) for population and housing variables and through a point-in-polygon process for counts of business establishments.

The analytic sample was limited to CHS decedents as of Aug 12, 2016 (n=5384), in order to mitigate issues of personal data security during linkage of CHS data with potentially identifiable personal and geographic data (described below). Participants were also excluded if they had a non-valid address for geocoding (n=556) or a history of CVD at baseline (n=1233). The analytic sample comprised 3595 participants.

Neighborhood Measures

Residential addresses assembled during the course of cohort follow-up were complemented through linkage to previously validated LexisNexis personal profile data.²⁹ The addresses from January of each year were then linked with other time-varying or interpolated data sources to describe neighborhood characteristics. Measured neighborhood characteristic changes over time were thus recorded whenever neighborhood resources changed, or a participant moved.

Neighborhood socio-economic and built environment characteristics considered as covariates included % unemployed, % vacant housing, median home value, median household income, and median block perimeter. Measures were defined based the Longitudinal Tract Database (LTDB) assembled by Brown University to characterize change in 2010 census-tract areas across decades since 1990,^{30, 31} which we have cleaned, linearly interpolated across intervening years, and carried forward through 2014.

Neighborhood-level retail characteristics were incorporated using annual information from the National Establishment Time Series (NETS) database. NETS is considered one of the most comprehensive establishment data sources available, serving as a virtual census of American businesses.³² For each business, company name and standard industrial classification (SIC) codes are maintained (among other characteristics), allowing for the categorization of the type of business.³³ While the North American Industry Classification System (NAICS) was established in 1997 to expand the SIC system, SIC codes are still used in datasets like NETS to study longitudinal changes over time (e.g., pre- and post-1997).³⁴

The two previously validated neighborhood-level exposures derived from NETS data were: (1) physical activity resources, representing physical activity/recreational establishments, including multi-use, moderate physical activity, vigorous physical activity facilities, and (2) walking destinations such as banks, groceries, restaurants, other food shops, museums, and libraries.^{35, 36} A detailed description of this dataset, including a comprehensive list of all categories has been published.³⁴ Variables were operationalized as continuous standardized densities, calculated as the count per square kilometer. Different spatial scales can produce different results depending on how well relevant geographic spaces are captured, therefore, we measured densities using both 1-km and 5-km buffers.^{37, 38} To avoid the influence of extreme values and improve interpretability, each variable was truncated at the 99^th percentile and standardized to mean=0; standard deviation=1.

CVD Outcomes

The primary CVD endpoint was defined by development of any of the following: myocardial infarction (MI), stroke, or cardiovascular death. Secondary CVD endpoints were considered, defined as any primary endpoint or angina pectoris, congestive heart failure (HF), and transient ischemic attack (TIA). The criteria, identification, and adjudication process for qualifying incident CVD events have been previously described.^{27, 39} Briefly, provisional diagnoses of CVD were assigned by the CHS Field Center principal investigator and reviewed and adjudicated at periodic meetings of the CHS Events Subcommittee. This committee classified incident morbid events for the six end points and assigned cause of death for fatal events. The cardiac (MI, angina, and HF) endpoints were adjudicated by the Cardiac Subcommittee and cerebrovascular end points (stroke and TIA) were adjudicated by the Stroke Subcommittee.

Confounding Variables

All confounders were selected a priori based on directed acyclic graphs (DAG’s) and only those which met the formal confounding definition were included. Self-reported baseline individual-level confounders included: age (65-69, 70-74, 75+), gender, race (non-Hispanic Black, non-Hispanic White, other race/ethnicity), education (<HS degree, HS degree, any vocational school, any college, college graduate or more), income (< $5,000; $5-7,999; $8-11,999; $12-15,999; $16-24,999; $25-34,999; $35-49,999; ≥ $50,000), clinic site (Sacramento, CA; Hagerstown, MD; Winston-Salem, NC; Pittsburgh, PA), and enrollment wave (1989 vs. 1993). To account for potential confounding throughout follow-up, we also included time-dependent confounders, including marital status (married, widowed, divorced, separated, never married), and several neighborhood-level confounders (% unemployed, % vacant housing, median home value, median household income, density of unhealthy food establishments, and urban vs. non-urban area).

Several additional individual and clinical variables may have been related to study attrition, potentially introducing bias into hazard estimates, either through collider stratification bias or through introducing additional confounding pathways.⁴⁰ To account for informative censoring, study participants were weighted by their inverse probability of attrition based on baseline health covariates, including baseline smoking status (pack years), alcohol use (any vs. no use), body mass index (BMI), hypertension (SBP ≥140/90 or taking HTN meds at baseline), and diabetes status (yes/no).⁴¹

We also considered the role of baseline mobility as a potential effect modifier of the main study associations. Baseline mobility was defined based on the energy in kilocalories expended in weekly household and leisure-time physical activity using the Minnesota Leisure Time Activities Questionnaire.

Statistical analysis

We estimated the risk of incident CVD associated with neighborhood exposures over time with a series of time-varying weighted Cox proportional hazard models. These models accounted for unequal individual follow-up periods, sources of time-dependent confounding,⁴² and violations of the proportional hazards assumption.⁴³ To specify these models, we coded a risk set for each individual, determined by the length of time between follow-up interviews, from enrollment until the participant experienced the outcome or was censored. The data structure comprised one row per individual per time interval (i.e., days of follow-up), with corresponding covariate values for that interval. The cumulative hazard function was then estimated as a function of the hazard at each interval, weighted by the event times. These methods have been used in similar research.⁴⁴ All models were additionally weighted by the inverse probability of attrition, based on baseline individual health indicators. We completed all analyses for both 5-km and 1-km buffers separately, representing relatively large and small neighborhood boundaries, respectively.

To examine the role of baseline mobility of the study sample in a supplementary analysis, we restricted the main analytic models to only those in the top 80% of physical activity, in accordance with previous definitions.⁴⁵ Analyses were implemented in R (version 3.6.2) and time-dependent Cox regression models were implemented with the ‘survival’ package.⁴⁶

Finally, to determine how the number of facilities divided by the population size in the area is related to the outcome, we repeated the primary analyses defining each exposure as a population density (i.e., the count per capita).

Among the analytic sample, covariate data were missing for 749 participants (21%). To minimize this loss of information, missing data were multiply imputed. Twenty models were imputed using predictive mean matching, based on all observed exposure, covariate, and outcome data, and combined with corrected standard errors.⁴⁷ Imputed model estimates were compared to unimputed estimates in order to examine the robustness of analytic models to the degree of missing data. All analyses were conducted in 2021.

RESULTS

The mean 5-km buffer exposure densities by baseline demographic are presented in Table 1, and the correlation coefficients for neighborhood exposures and covariates are presented in Table 2. The mean 5-km unstandardized numbers of walking destinations and physical activity/recreational resources were 12.7 (SD=4.7) and 0.30 (SD=0.92), respectively. In general, the availability of walking destinations varied by marital status (range: 11.9-14.6) and across clinic site (range: 11.9-13.5/km) (Table 1). For physical activity resources, there were no clear patterns for baseline characteristics, though resources were higher among NH White (0.31; SD=0.99) compared with NH Black (0.25; SD=0.93) participants (Table 1). Walking destinations and physical activity resources were negatively correlated with median household income (alpha=−0.211) and positively correlated with unhealthy food (alpha=0.266) and population density (0.591) (Table 2). Physical activity resources were negatively correlated with % unemployed (alpha=−0.337) and positively correlated with unhealthy food (alpha=0.312) and population density (0.419) (Table 2). Mean 1-km exposure patterns were generally similar to 5-km densities (Supplementary Tables 1 and 2).

Table 1.

Mean 5-km buffer exposure densities by baseline demographic and neighborhood covariates

	Total		Walking destinations (5km)		Physical activity resources (5km)
	n; mean	%; SD	Mean	SD	Mean	SD
Total	3595		12.7	4.7	0.30	0.92
Individual characteristics
Age at baseline
65-69	1185	33.0%	12.5	4.8	0.31	0.99
70-74	1201	33.4%	12.6	4.6	0.30	0.98
75+	1209	33.6%	12.8	4.8	0.30	0.97
Race/Ethnicity
NH White	3175	88.3%	12.6	4.8	0.31	0.99
NH Black	396	11.0%	12.7	4.2	0.25	0.93
Other	24	0.7%	13.1	6.6	0.29	0.98
Gender
Women	2161	60.1%	12.7	4.8	0.31	0.99
Men	1434	39.9%	12.5	4.6	0.30	0.98
Education
<HS degree	1013	28.2%	12.5	5.1	0.29	0.98
HS degree	1031	28.7%	12.5	4.7	0.31	0.97
Any vocational school	308	8.6%	13.1	5.0	0.33	0.25
Any college	896	24.9%	12.9	4.5	0.31	0.97
College or more	347	9.7%	12.6	3.8	0.29	0.95
Income
< $5,000	172	4.8%	13.1	4.9	0.29	0.92
$5,000-$7,999	300	8.3%	12.6	4.6	0.30	0.97
$8,000-$11,999	406	11.3%	12.8	5.0	0.31	0.91
$12,000-$15,999	594	16.5%	12.3	4.6	0.29	0.97
$16,000-$24,999	707	19.7%	12.6	4.9	0.30	0.98
$25,000-$34,999	571	15.9%	12.8	4.9	0.31	0.98
$35,000-$49,999	381	10.6%	12.7	4.6	0.31	0.98
>$50,000	464	12.9%	12.9	4.2	0.31	0.97
Marital status
Married	2374	66.0%	12.6	4.8	0.31	0.98
Widowed	880	24.5%	12.7	4.5	0.30	0.96
Divorced	151	4.2%	11.9	4.5	0.27	0.91
Separated	27	0.8%	14.6	6.6	0.32	0.94
Never Married	163	4.5%	13.4	4.6	0.32	0.92
Clinic site
Bowman Gray	963	26.8%	13.5	5.5	0.33	0.92
Davis	983	27.3%	12.0	4.5	0.27	0.96
Hopkins	797	22.2%	13.3	5.3	0.34	0.99
Pittsburgh	852	23.7%	11.9	2.9	0.27	0.92
Cohort
Original	3351	93.2%	12.6	4.7	0.31	0.98
Supplemental	244	6.8%	12.7	4.3	0.27	0.94
Non-Metro area	3549	98.7	11.9	5.4	0.32	0.34
Metro area	46	1.3	19.3	6.8	0.87	1.19

Table 2.

Correlation coefficients for neighborhood exposures (count of facilities within 5-km buffers) and covariates

	Walkable destinations	Physical activity	% Unemployed	Median household income	% vacant housing	Median home value	Unhealthy food	Population density
Walkable destinations	1	0.492	0.053	−0.211	0.159	−0.033	0.266	0.591
Physical activity		1	−0.337	0.193	−0.029	0.075	0.312	0.419
% Unemployed			1	−0.686	0.363	−0.355	0.013	0.325
Median household income				1	−0.391	0.734	−0.0894	0.061
% vacant housing					1	−0.308	0.681	0.508
Median home value						1	−0.170	0.012
Unhealthy food							1	0.210
Population density								1

Cox proportional hazards models for 1-km densities

Attrition weights used in the Cox proportional hazards models ranged from 0.2-25.54 (mean=1.02; SD=0.45). The hazards of primary and secondary cardiac events related to continuous walking destination neighborhood (1-km) densities are presented in Table 3. In the fully adjusted models, walking destination hazard ratios suggested reduced incident primary or secondary CVD rates, however confidence intervals included the null value among the whole population (primary CVD HR=0.97; 95% CI=0.93, 1.01; secondary CVD HR=0.96; 95% CI=0.90, 1.02). Estimates were nearly identical when limited to those without impairments in daily physical functioning. Hazard ratios for cardiac events related to continuous physical activity resource (1-km) densities are presented in Table 3. In the fully adjusted models, physical activity resources were not associated with incident primary or secondary outcomes, among the whole population (primary CVD HR=0.95; 95% CI=0.61, 1.29; secondary CVD HR=0.95; 95% CI=0.65, 1.25). Estimates were nearly identical when limited to those without impairments in daily physical functioning. The complete model results including HRs for all covariates are presented in Supplementary Tables 3 and 4.

Table 3.

Cox proportional hazards of primary and secondary cardiac events related to 1-km walking destinations and physical activity facility densities

	Any primary	Any Secondary
Walking destinations (per SD increase)	HR (95% CI)	HR (95% CI)
Unadjusted
	0.98 (0.96, 1.02)	0.97 (0.93, 1.01)
Adjusted1
	0.97 (0.93, 1.01)	0.96 (0.90, 1.02)
Only among those not in bottom 20% physical functioning*
	0.99 (0.96, 1.03)	1.00 (0.97, 1.02)
	Any primary	Any Secondary
Physical activity resources (per SD increase)	HR (95% CI)	HR (95% CI)
Unadjusted
	0.90 (0.63, 1.18)	0.93 (0.73, 1.13)
Adjusted1
	0.95 (0.61, 1.29)	0.95 (0.65, 1.25)
Only among those not in bottom 20% physical functioning*
	0.97 (0.70, 1.14)	0.96 (0.65, 1.25)

¹Adjusted for: age at baseline, race/Ethnicity, sex, education, income, marital status, clinic site, CHS study cohort, % unemployed, median home income, % vacant housing, median home value, % unhealthy food density, population density, urban vs. non-urban area

Cox proportional hazards models for 5-km densities

Hazard ratios estimating the effects of 5-km walking destinations on CVD risk are presented in Table 4. In the fully adjusted models, there was a 7% decrease in risk of primary (95% CI=0.87, 0.99) and 11% decreased risk of secondary (95% CI=0.82, 0.96) CVD outcomes. Hazard ratios were the same among only those without impairments in daily physical functioning. The hazards on CVD events among 5-km physical activity resources are also presented in Table 4. In the fully adjusted models, there was a 12% decrease in risk of primary CVD outcomes, though confidence intervals did include the null value (95% CI=0.73, 1.0) and 15% decreased risk of secondary (95% CI=0.74, 0.98) CVD outcomes. Hazard ratios were generally similar among those meeting the baseline threshold for physical mobility, though model standard errors were slightly larger. The complete model results including HRs for all covariates are presented in Supplementary Tables 5 and 6.

Table 4.

Cox proportional hazards of primary and secondary cardiac events related to 5-km walking destinations and physical activity facility densities

	Any primary	Any Secondary
Walking destinations (per SD increase)	HR (95% CI)	HR (95% CI)
Unadjusted
	0.96 (0.87, 1.06)	0.97 (0.89, 1.05)
Adjusted1
	0.93 (0.87, 0.99)	0.89 (0.82, 0.96)
Only among those not in bottom 20% physical functioning*
	0.90 (0.81, 0.99)	0.87 (0.8, 0.94)
	Any primary	Any Secondary
Physical activity resources (per SD increase)	HR (95% CI)	HR (95% CI)
Unadjusted
	0.89 (0.80, 0.98)	0.86 (0.77, 0.96)
Adjusted1
	0.88 (0.73, 1.0)	0.85 (0.74, 0.98)
Only among those not in bottom 20% physical functioning*
	0.89 (0.76, 1.03)	0.85 (0.74, 0.99)

¹Adjusted for: age at baseline, race/Ethnicity, sex, education, income, marital status, clinic site, CHS study cohort, % unemployed, median home income, % vacant housing, median home value, % unhealthy food density, population density, urban vs. non-urban area

When compared to the dataset with no imputation, results from a complete case analysis were not meaningfully different in magnitude or direction of the model estimates, although standard errors were slightly smaller overall.

Results were similarly significant when analyses were repeated defining neighborhood exposures as a population density (Supplementary Tables 7 and 8), although magnitude of associations were slightly smaller.

DISCUSSION

In a large community-based cohort of older adults we found that higher density of neighborhood physical activity facilities in a 1-km radius were not associated with a reduced risk of incident CVD events. The reduction in risk, however, was significant for facilities within a 5-kilometer radius.

Residing in disadvantaged neighborhoods, as determined by area-based socioeconomic characteristics derived from US census block group data, has been associated with higher rates of incident coronary heart disease (CHD) and cardiovascular death.^{48, 49} Environments with greater land use mix and more fitness facilities have also been associated with a reduced risk of incident CHD among women.⁵⁰ Prior studies, however, evaluating the association between GIS-based built environment measures and incident CVD are limited and none have specifically evaluated time-varying measures in an older population. In a nationwide Swedish sample of over 4 million individuals, increased proximity to physical activity facilities was associated with a paradoxically increased risk of stroke and coronary heart disease in men and women.²⁶ Associations were strongly weakened for incident stroke and disappeared for incident coronary heart disease after additional adjustment for neighborhood level deprivation. Distance buffers used in the Swedish study were created to include the nearest town/city (if not part of one already) or according to a 1-km radius. Significant limitations in this study preclude generalizability of findings. The study population ranged in baseline age from 35-80 and individuals were not well-characterized at baseline. Information regarding lifestyle factors and laboratory variables were not available. Finally, exposure variables were assessed only at baseline and follow-up was limited to 6 years. Our approach used longitudinal exposure information based on both changing home addresses and changes within US neighborhoods across decades.

While results reported here do not directly explain what underlies this relationship and further investigation is needed to better understand what mediates this association, prior studies do provide insight into potential mechanisms. Greater access to neighborhood physical activity resources and residing in high-walkability neighborhoods have been associated with lower insulin resistance, higher levels of physical activity, lower weight and waist circumference.^{51, 52} Overall cardiovascular health and estimated atherosclerotic CVD risk were also improved with greater access to neighborhood physical activity resources and residing in high-walkability neighborhoods.^{20, 21} These beneficial effects may, therefore, contribute to a lower incidence of cardiovascular risk factors and, ultimately, decreased CVD. While previous literature has demonstrated a relationship between improved neighborhood environment and a lower incidence of diabetes, hypertension, and obesity, only survey-based measures of the physical activity environment were significant.^53–55 Longitudinal measures were not used in these studies and only 1-mile radius buffers were evaluated. Future research should focus on testing these putative mechanisms.

While hazard ratios for 1-km and 5-km buffer areas shared some overlap, only the 5-km neighborhood exposures were statistically significantly associated with a reduced CVD risk. Prior studies have suggested that older individuals may rely more heavily on their immediate environment to conduct daily activities,^{10, 11} but area-based measurement of the neighborhood environment varies widely within (e.g., based on the variation in size of census tracts) and between studies. The results of the current study suggest that a 5-km radius buffer may be useful in future studies of neighborhood physical activity resources, perhaps capturing relevant commercial amenities that are not contained within 1-km buffers. The urbanicity of CHS participants’ residential neighborhoods varied, and particularly within rural environments there may be reliance on cars or other options to access physical activity resources in a larger area than older adults are expected to access by walking. Facilities inside a 1-km buffer can be accessed within a typical 10-minute walk, but facilities within 5 km could require an hour’s walking time; however, those could easily be accessed within a 10-minute drive. A large majority of older adults, including those aged 85 and older, remain active drivers and only report limit their driving under conditions such as poor weather, night-time driving, long driving distances, or heavy traffic.^{56, 57} Prior research has also demonstrated that driving status, based on driver’s license, car ownership, and feeling comfortable to drive, moderated the association between GIS-measured neighborhood environments and older adults’ PA levels.⁵⁸ It is also possible that a larger neighborhood radius more accurately captured the participant’s true activity-space and included destinations such as large supermarkets or fitness gyms. More research to understand additional factors which may facilitate or hinder access to these resources—including behavioral studies of how adults of varying age access different resources—will help to develop neighborhood measures on the most valid spatial scales. An explanation for why the count of facilities in the 1-km buffers was not associated with reduced risk is that the sample population comprised older individuals with greater prevalence of functional impairment,⁵⁹ which may have further constrained their ability to access proximal neighborhood resources by walking. However, observed patterns were largely consistent in a sensitivity analysis restricting the sample to those with less physical functioning impairment.

This study was strengthened by using a large well-characterized longitudinal sample, which allowed us to establish temporality and assess time-to-incident cardiovascular outcomes, adjusting for a wide range of individual- and neighborhood-level confounders. The use of time-varying exposures allowed us to account for time-varying confounding over the duration of the study.

This study should be interpreted in the context of the following limitations. First, we did not have information about facility use by respondents (e.g., how often individuals went to the gym, grocery store, etc.). We neither accounted for each individual’s true activity-space nor weighted destinations differently according to factor such as destination size, network distance, or travel cost.⁶⁰ Second, the CHS sample represents an aging population, and comorbid health conditions were highly prevalent, which may modify the use of neighborhood amenities even when they are readily available. While we accounted for issues of baseline health status by excluding respondents with prior CVD events, including attrition weights, and stratifying models based on general mobility status, many other health conditions may still be relevant. Third, while we controlled for confounding throughout the study period using measured variables, residual or unmeasured confounding may remain. For example, measures of wealth or neighborhood safety may be important beyond the socioeconomic characteristics included, as both could influence availability and utilization of neighborhood resources. Fourth, the exposures were defined in a way that would yield parsimonious, conceptually meaningful, and statistically stable model estimates. Nevertheless, there may be important heterogeneity within the categories included that could benefit from future efforts to empirically optimize classification or to complement secondary longitudinal data with on-site or qualitative data collection. However, we considered alternative categories such as binary (any vs. none), quartiles, and median-split binary variables, and model estimates were generally consistent in direction and magnitude across all categorization methods. The analytic sample was limited to participants who were deceased as of 2016, which influenced the probability of selection on disease status, as decedents had a higher probability of CVD than non-decedents. However, it is unlikely that selection probability differed by exposure status.

In conclusion, a higher density of walking destinations and physical activity resources within a 5-km radius of home, but not a smaller 1-km radius, were associated with a reduced risk of incident CVD in an older population. Considering older adults are thought to be more heavily reliant on their immediate environment, these findings are important and suggest that public health efforts to improve PA opportunities over larger radii, a potentially easier strategy to implement, may have a benefit in CVD prevention. Multiple areas of further study are still needed. These include using more nuanced measures of the built environment that incorporate gravity-based models or space-time measures, accounting for important characteristics such as neighborhood safety, better understanding the factors that modify this association such as driving status or a rural vs. urban environment, and exploring whether physical activity levels mediate, in part, associations between neighborhood PA opportunities and CVD.

• We determined associations of cumulative exposures to neighborhood physical activity opportunities with risk of incident cardiovascular disease (CVD) in 3595 Cardiovascular Health Study participants (mean age=73; 60% women; 11% black).

• Neighborhood environment measures were calculated using Geographic Information Systems and annual information from the National Establishment Time Series database,

• After adjusting for baseline and time-varying individual and neighborhood-level confounding, a one standard deviation increase in walking destinations and physical activity/recreational facilities within 5 km of home was associated with a respective 7% and 12% decreased risk of incident CVD.

• Efforts to improve the availability of physical activity resources in neighborhoods may be an important strategy for lowering CVD in older individuals.