Exploring associations of greenery, air pollution and walkability with cardiometabolic health in people at midlife and beyond

To examine associations of neighborhood greenery, air pollution and walkability with cardiometabolic disease in adults aged ≥45 years in the Frankston–Mornington Peninsula region, Victoria, Australia.


Introduction
Cardiometabolic disease, including heart disease, stroke and type 2 diabetes, makes the greatest contribution to the global burden of disease. 1 In 2019, 41 million people died from non-communicable diseases, worldwide. 2Cardiovascular disease was the cause of 17.9 million of these deaths, while a further 2 million deaths were caused by diabetes. 2Although interventions that aim to lower modifiable risk factors for cardiometabolic disease are traditionally applied at the individual level, there is growing interest in the associations of local neighborhoods with these risk factors. 3Socioecological systems theory posits that there are multiple layers of influence centered on the individual related to their health. 4,5There is increased awareness that built environment characteristics, such as clean air, the presence of trees and parks, and pedestrian/cycling infrastructure that promotes incidental physical activity, are important for the health of those who live, work and play there. 6As adults grow older and work less, they might spend more time in neighborhood settings that potentially influence their health.
8][9] For example, exposure to nitrogen dioxide (NO 2 ), a key component of traffic emissions, 10 has been associated with type 2 diabetes. 11In addition, exposure to fine particulate matter of diameter ≤2.5 μm (PM 2.5 ), resulting from fossil fuel combustion, is considered a major contributor to global disease burden and reduced life expectancy due to cardiometabolic disease. 12Almost one-third of all anthropogenic PM 2.5 exposures come from vehicle emissions 10 in urban areas of Australia, where car travel is the dominant transport mode. 138][9] Even though there are generally lower concentrations of air pollutants in Australia, compared with low-to middle-income countries, there is no safe threshold for pollutants. 14Therefore, examination of varying concentrations, even at lower levels, and associations with health-related outcomes is warranted.Potentially, air pollution might be offset, at least in part, by greenery. 6Views of, and access to, greenery in urban areas are associated with lower levels of stress. 15,16and higher levels of physical activity, 16 which might protect against cardiometabolic disease.Public parks, in particular, provide settings for physical activity and interaction with nature. 17nother neighborhood characteristic that might promote residents' cardiometabolic health is walkability, which considers street connectivity, population density, and pedestrian access to shops and services. 18Walkable neighborhoods facilitate walking for transport to local destinations. 18A systematic review of longitudinal studies of the built environment and associations with cardiometabolic disease found that adults residing in more walkable neighborhoods were less likely to develop type 2 diabetes or hypertension. 19Another systematic review reported inverse associations between neighborhood walkability and risk/prevalence of type 2 diabetes, and between greenery and diabetes. 20Although parks and walkable neighborhoods show potential to promote cardiometabolic health, few studies have examined these environmental exposures in combination with air pollution. 21Further investigation is required, as more walkable neighborhoods might include more complex street networks with higher traffic volumes. 21Hence, it is vital to construct multi-exposure models with health-related outcomes to examine potential associations between co-existing exposures.
To address these knowledge gaps, we aimed to explore multiple neighborhood environmental exposures in relation to the prevalence of self-reported cardiometabolic disease among mid-to-older aged adults (aged ≥45 years) in the Frankston-Mornington Peninsula region, Australia (described under "Setting and sample").Our specific questions were: 1. Are air pollutants positively associated with the prevalence of type 2 diabetes, heart disease and stroke?2. Is greenery negatively associated with the prevalence of type 2 diabetes, heart disease and stroke?3. Is greater walkability negatively associated with the prevalence of type 2 diabetes, heart disease and stroke?
We also explored the interaction between walkability and air pollution in relation to cardiometabolic disease prevalence, and air pollution as a potential mediator of associations between greenery and cardiometabolic disease prevalence.

Setting and sample
We carried out this analysis utilizing the health and environmental data infrastructure of the National Centre for Healthy Aging, an initiative funded by the Commonwealth Government of Australia. 22The study setting is the Frankston-Mornington Peninsula region (identified as Mornington Peninsula Statistical Area 4 [SA4] by the Australian Statistical Geography Standard), 23 which contains a mixture of urban and regional locations. 24This area is shown in Appendix I, Figure A1.The total area is 853.8 km 2 , with most of this considered by the Australian Bureau of Statistics to be part of Melbourne's Significant Urban Area, comprising groups of urban centers (with population ≥10 000).However (as shown in Appendix I, Figure A2), much of the lower part of the Mornington Peninsula is considered to be regional.
The total population at the most recent 5-yearly Census of Population and Housing (carried out in 2021) was 308 108, with 150 866 residents aged ≥45 years. 25We limited our age group of interest to ≥45 years, as cardiometabolic disease is more prevalent at midlife or older age. 1 Our unit of analyses was the neighborhood; that is, we examined disease prevalence by neighborhood.Neighborhoods were operationalized at Statistical Area 1 (SA1) level, the smallest reporting unit of the Australian Census, representing approximately 200 households. 23

Outcome measures: Cardiometabolic disease prevalence in neighborhood
The 2021 Census 25 was the first to ask respondents to report whether (yes/no) they had specific long-term health conditions, diagnosed by a health professional.The following cardiometabolic disorders were included: "diabetes (non-gestational)," "heart disease" and "stroke."The number of people aged ≥45 years in each SA1 who reported having each disorder/disease was divided by the population in that age group, to give the disease prevalence, by SA1.Although the Census did not distinguish between type 1 and type 2 diabetes, almost all reported cases were assumed to be type 2, as it accounts for >96% of all diabetes cases globally. 26

Greenery
Overall greenery (including tree canopy, grass) was assessed using the Normalized Difference Vegetation Index and satellite imagery (NASA Landsat 8, red and near-infrared band).We included in our analysis mean annual Normalized Difference Vegetation Index values for each SA1 in 2019.These publicly available Normalized Difference Vegetation Index values were computed by Ramsay & Mavoa 27 using Landsat sensor data, adjusted for variation in solar and atmospheric characteristics.

Air pollution
Exposures to annual average PM 2.5 and NO 2 were estimated for each SA1 using two validated satellite-based land-use regression models, developed for the Australian continent and described in detail elsewhere. 28,29These model predictions were used to | 209 estimate exposures in 2019 at mesh block centroids (mesh block is the smallest area defined by ASGS), which aggregate to form whole SA1s, and we used the mesh blocks to average over each SA1 in our study area. 28,29

Walkability
The walkability of each SA1 was measured using Walk Score, a free online tool that computes a score (0-100) indicative of pedestrian access to local destinations (e.g.shops) from a chosen address. 30Walk Score assesses walking routes to destinations, with raw scores being assigned according to proximity (within 400 m, then up to 2.4 km), and accounts for population density and intersection density (a measure of street connectivity and route choice). 30Walk Score assigns higher scores to areas with better pedestrian access to local amenities, and has been validated internationally. 31,32Walk Scores were generated for the address listed in the Geocoded National Address File 33 closest to the centroid of each SA1, using the Walk Score Application Programming Interface with the R programming language. 34

Covariates
To control for the socioeconomic status of the SA1, deciles of the Index of Relative Socioeconomic Disadvantage and Advantage were included. 35The Index of Relative Socioeconomic Disadvantage and Advantage is derived from Census data based on area-level attributes of disadvantage (e.g. % unemployed) and advantage (e.g. % of employed classified as "Professionals"). 35Also, given that some disease risk factors increase with age and might vary by sex, 1 our analyses were adjusted for the percentage of people aged ≥65 years and women.

Statistical analysis
All statistical modelling (with significance set at P < 0.05) was carried out using IBM SPSS v26.0 (IBM Corporation, Armonk, NY, USA).Stroke prevalence was not normally distributed and underwent transformation (square root) before inclusion in analyses.Multicollinearity among exposure variables was assessed by examining variance inflation factors, and was considered to be present if variance inflation factors were >5. 36Initially, associations between each exposure and disease prevalence were examined in single exposure models using univariable linear regression.Subsequent multivariable regression analyses assessed cross-sectional associations between environmental exposures and disease prevalence, controlling for the covariates described above.Further analyses were run for PM 2.5 controlling for covariates and NO 2 , and vice versa.Potential interactions between walkability and air pollutants, proposed by a recent study, 21 were examined in relation to disease prevalence, with effects scaled to one unit of each exposure.
MacKinnon's product of coefficients test 37 was used to examine PM 2.5 and NO 2 as potential mediators of associations between greenery and disease prevalence.The standard methods 37 are described in Appendix II, with pathways shown in Appendix II, Figure A3.
Ethical approval to carry out this study was not required, because the data had no individual identifiers, and were publicly available in the database of the 2021 Census of Population and Housing, in Australia. 25

Results
Our sample comprised 699 neighborhoods whose characteristics are described in Table 1.
Associations of environmental exposures with neighborhood prevalence of diabetes, heart disease and stroke among adults aged ≥45 years are presented below (Table 2).Significant associations were found between each of PM 2.5 , NO 2 and greenery, and prevalence of diabetes.Higher levels of PM 2.5 and NO 2 were associated with higher prevalence of diabetes, whereas higher levels of greenery were associated with a lower prevalence of diabetes.Significant associations in the same corresponding directions were observed between these exposure variables and the transformed variable for stroke prevalence.Greenery was inversely associated with prevalence of heart disease in the univariable model.However, this association was no longer statistically significant after controlling for covariates (Table 2).

Interaction between walkability and PM 2.5
Associations between walkability and all three cardiometabolic disease prevalences were highly significant in the univariable model, but not in the hypothesized direction (Table 2).However, when an interaction term for walkability and PM 2.5 was included, the associations of PM 2.5 and walkability with diabetes prevalence Analyses were conducted at SA1 level; however, for background information, the mean population for each SA1 was 440 (SD 152) people, with 215 (SD 90) people aged ≥45 years and 98 (SD 70) people ≥65 years.
A Carver et al.
were statistically significant and in the hypothesized direction (Table 3), and the interaction term itself was significantly associated with diabetes prevalence (Table 3).There was no significant interaction of walkability and PM 2.5 in relation to prevalence of heart disease or stroke; nor was there any significant interaction between walkability and NO 2 in relation to any of the cardiometabolic disease prevalences (results not shown).
Walkability was re-categorized as "low," "medium" or "high" using a tertile split, and the interaction of walkability and PM 2.5 in relation to prevalence of diabetes was plotted (Figure 1).This visualization of data shows that relatively higher walkability was associated with lower diabetes prevalence, only where PM 2.5 levels were lower (Figure 1).

Mediation analysis
Mediation analysis was carried out where greenery was significantly associated with the cardiometabolic disease prevalence, after controlling for covariates (therefore, not carried out for heart disease).Results of mediation analyses that explored PM 2.5 and NO 2 as potential mediators of associations between greenery and diabetes prevalence are presented in Table 4.A significant mediated effect was found in each case; the proportion mediated by PM 2.5 was 92%, and the proportion mediated by NO 2 was 69%.No mediated effects were found for PM 2.5 and NO 2 in the association between greenery and stroke prevalence (square root transformed), as pathway "b" (between the air pollutant and stroke prevalence, adjusting for greenery) was non-significant (P ≥ 0.05; results not shown).

Discussion
The present study makes an important contribution to the growing body of research that examines neighborhood measures of greenery, PM 2.5 , NO 2 and walkability in multi-exposure models of cardiometabolic disease prevalence among mid-to-older aged adults.Our findings extend and further explore those from earlier studies that examined either greenery or air pollution or walkability in relation to cardiometabolic disease, in the absence of the other exposures.
The present findings that PM 2.5 and NO 2 were each positively associated with diabetes prevalence in single exposure models concurred with those of an Italian study. 38In our study, the positive association with diabetes prevalence remained for PM 2.5 after controlling for NO 2 ; however, the association between NO 2 and diabetes prevalence was no longer significant after controlling for PM 2.5 .This supports the argument that NO 2 is, in effect, an indicator of other co-pollutants from vehicle emissions. 39he strengths of the present study are the exploration of interaction between walkability and PM 2.5 , and investigation of air pollution exposures as mediators of association between greenery and diabetes prevalence.Some previous studies that showed Table 2 Associations of environmental exposures with prevalence of cardiometabolic disease among adults aged ≥45 years, by Statistical Area 1 associations of greenery with lower diabetes prevalence, 40 lower hospitalization rates for cardiovascular disease, 41 as well as lower risk of incidence of cardiovascular disease, 42 did not include air pollution or walkability exposures.Chandrabose et al. 21proposed that more walkable neighborhoods have destinations to which some people still drive, creating vehicle emissions that adversely affect health.We found that higher walkability was associated with lower diabetes prevalence only in areas with lower levels of PM 2.5 .Similarly, the CANHEART study, 43 one of few studies examining multiple environmental exposures and cardiometabolic disease, reported that beneficial associations of higher walkability with diabetes were attenuated in more highly-trafficked areas with higher NO 2 levels. 43r identification of both PM 2.5 and NO 2 as mediators of the association between greenery and diabetes prevalence aligns with findings of a study of greenery, air pollution and cardiometabolic disease in the Netherlands. 11Possibly, greenery might offset air pollution by plant-based filtering and/or dispersal of air pollutants, 44 or by occupying land that might otherwise be used for roads. 45bjective measurement would be required for confirmation.The present data are cross-sectional, thus precluding any temporal or causal inference.Furthermore, there is potential in cross-sectional studies for mediating effects to be overestimated. 46It is suggested that for greater understanding of mediators (and of possible reverse causality), future studies should be longitudinal, with measurement Figure 1 Interaction between walkability and fine particulate matter of diameter ≤2.5 μm (PM 2.5 ) in relation to the prevalence of diabetes.This plot is a visual presentation of the interaction of walkability (measured using Walk Score) and PM 2.5 (measured in μg/m 3 ) in relation to the prevalence (%) of diabetes in neighborhoods.Walkability values have been categorized as "low", "medium" or "high" based on a tertile split.
Table 4 Mediation analysis: adjusted † associations of the a, b, c, and c / pathways ‡ and significant mediated effect by particulate matter of diameter ≤2.  of exposure variables, potential mediators and outcome variables at three or more time-points. 46Only area-level data rather than individual-level data are included in our analyses and, therefore, individual-level associations cannot be reported due to the ecological design. 47Consequently, we were unable to adjust for many individual-level factors and behaviors (e.g.smoking) that might affect cardiometabolic health outcomes.We found no significant associations between either PM 2.5 or NO 2 and prevalence of heart disease.This was in contrast to the findings of a systematic review linking both long-and short-term exposure to raised levels of PM 2.5 with higher risks of cardiovascular disease and heart failure, 48 and evidence of a causal association between short-term exposure to NO 2 and ischemic heart disease. 39Possibly, the term "heart disease" included in the Census was too vague, and led to underreporting of some conditions (e.g.arrythmia).Diabetes and stroke might have been reported more accurately if respondents required insulin or had experienced a defined medical event, such as a stroke.
The present study found that PM 2.5 and NO 2 were each adversely associated with stroke prevalence, although these associations were no longer significant when adjusting for each other.Overall, associations of air pollution and stroke are less studied than their associations with cardiovascular disease or diabetes.However, a systematic review 39 reported that short-term exposures to PM 2.5 or NO 2 were associated with stroke-related hospital admissions and deaths from stroke.A more recent review 49 provided strong evidence to support that short-and long-term exposures to PM 2.5 were causally related to incidence of ischemic stroke, but there was inadequate evidence to infer corresponding relationships for NO 2.

49
Further strengths of the present study include the analysis of relatively granular exposure data (at SA1 level) for PM 2.5 , NO 2 , greenery and walkability in single and multi-exposure models for almost 700 geographically diverse neighborhoods, with >150 000 residents aged ≥45 years.However, the conduct of our study in a single region (i.e. the Frankston-Mornington Peninsula region) of Australia might limit the generalizability of findings to other populations.Further study limitations include its cross-sectional design, self-report of professionally-diagnosed disease outcome measures and non-use of age standardized prevalence.Future studies should explore the use of more comprehensive clinical data that might be available across a geographic region.One efficient method would be the use of linked electronic health records 22 to objectively measure the prevalence and incidence of hospitalization due to cardiometabolic disease.
In conclusion, the present study contributes to evolving evidence that neighborhood greenery might be protective against diabetes and stroke, but air pollution appears detrimental.Walkable neighborhoods might promote, health but only if air pollution is lower.Longitudinal studies, including clinicallyconfirmed diagnoses, rather than self-report, are required to confirm our findings, and inform public health policy to promote healthy aging.
Neighborhoods and cardiometabolic health © 2023 The Authors.Geriatrics & Gerontology International published by John Wiley & Sons Australia, Ltd on behalf of Japan Geriatrics Society.
5 μm, and by nitrogen dioxide Pathway c