Community greenness, blood pressure, and hypertension in urban dwellers: The 33 Communities Chinese Health Study

Background: Living in greener areas has many health benefits, but evidence concerning the effects on blood pressure remains mixed. We sought to assess associations between community greenness and both blood pressure and hypertension in Chinese urban dwellers, and whether the associations were mediated by air pollution, body mass index, and physical activity. Methods: We analyzed data from 24,845 adults participating in the 33 Communities Chinese Health Study, which was conducted in Northeastern China during 2009. We measured each participant's blood pressure according to a standardized protocol. We assessed community greenness using two satellite-derived vegetation indexes – the Normalized Difference Vegetation Index (NDVI) and the Soil Adjusted Vegetation Index (SAVI). Particulate matter ≤2.5μm and nitrogen dioxide were used as proxies of ambient air pollution. We applied generalized linear mixed models to investigate the association between greenness and blood pressure. We also performed mediation analyses. Results: Living in greener areas was associated with lower blood pressure and hypertension prevalence; an in- terquartile range increase in both NDVI 500-m and SAVI 500-m were significantly associated with reductions in systolic blood pressure of 0.82mmHg (95% CI: −1.13, −0.51) and 0.89mmHg (95% CI: −1.21, −0.57), respectively. The same increases in greenness were also significantly associated with a 5% (95% CI: 1%, 8%) and 5% (95% CI: 1%, 9%) lower odds of having hypertension, respectively. These associations remained consistent in sensitivity analyses. The associations were stronger among women than men. Air pollutants and body mass index partly mediated the associations, but there was no evidence of mediation effects for physical activity. Conclusions: Our findings indicate beneficial associations between community greenness and blood pressure in Chinese adults, especially for women. Air pollution and body mass index only partly mediated the associations.


Introduction
Hypertension has been consistently identified as the leading contributor to cardiovascular disease burden, and its prevalence continues to increase in low-and middle-income countries (Forouzanfar et al., 2016). The prevalence of hypertension reached 27.9% in Chinese adults between 2012 and 2015 according to the most recent national survey data, which represents over 250 million hypertensive individuals . Effective intervention strategies are therefore needed to stem the rising tide of hypertension in China. The etiology of hypertension is complicated and includes both genetic and environmental factors (Poulter et al., 2015;Staessen et al., 2003). From a public health perspective, identification of environmental factors may be particularly important, as many such factors are modifiable through behavior changes or government level policy changes.
China has experienced rapid urbanization during the past four decades (Guan et al., 2018;United Nations, 2015), which poses a critical challenge in providing sufficient green spaces for urban dwellers (Nieuwenhuijsen et al., 2017). Therefore, understanding the associations between greenness and blood pressure in China could provide new and locally-relevant evidence to be used by public health and urban planning authorities. Additionally, due to urbanization, industrialization, and lifestyle changes, the Chinese population experiences severe air pollution (Guan et al., 2016), as well as epidemics of physical inactivity (Kohl III et al., 2012) and obesity (World Bank, 2011). Thus, China is an ideal setting for exploring the effects of greenness on human health and the underlying biological mechanisms. We hypothesized that exposure to greenness would be associated with blood pressure in the Chinese population, and that the associations would be mediated by air pollution, physical activity, and adiposity. We used data from the 33 Communities Chinese Health Study (33CCHS) to test these hypotheses (Yang et al., 2018b(Yang et al., , 2018c.

Study area
Between April 1 and Dec 31, 2009, we conducted the 33CCHS, a large population-based cross-sectional study, in Liaoning province, which is in Northeastern China (Yang et al., 2018b(Yang et al., , 2018c. Liaoning province is one of the largest industrial centers in China, with > 20 million permanent residents, over 64% of whom reside in urban areas. Air pollution is a substantial environmental and public issue in Liaoning province, due to abundant industrial and motor vehicle emissions, as well as household solid fuel combustion for heating during the cold winter. The prevalence of hypertension (37.7%) in this area is reported to be the highest in China (Li et al., 2018).

Study participants
The 33CCHS design, recruitment, and inclusion criteria have been described in detail elsewhere (Yang et al., 2018b(Yang et al., , 2018c. Briefly, we used a four-stage stratified clustering sampling scheme to obtain a representative sample of the study population ( Fig. 1). First, we selected three cities (Shenyang, Anshan, and Jinzhou) out of 14 provincial cities in Liaoning province. Second, there were a total of eleven city districts in the three cities (five districts in Shenyang, and three each in Anshan and Jinzhou); each district had a single air quality monitoring station. From each district, we randomly selected three communities that were located within a 1-km radius from the air-monitoring station. The size of the study communities ranged from 0.25 to 0.64 km 2 , and the distances between the three communities in each district were < 1.5 km. Third, we randomly selected 700-1000 household from each community. Fourth, we randomly selected one adult from each household.
In total, 28,830 potential study participants were invited. Of those, 24,845 (response rate = 86.2%) were included in the current analysis, based on the following six criteria: (1) blood pressure measurements were available; (2) aged 18-74 years; (3) resided at the study address for at least five years; (4) no pre-existing severe illness (e.g. terminal cancer); (5) not pregnant; and (6) completed a valid study questionnaire. The study protocols were approved by the Human Studies Committee of Sun Yat-sen University. All participants gave informed consent prior to data collection.

Outcome assessment
Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured by nurses using the standard mercuric-column sphygmomanometer method. Nurses were trained according to the American Heart Association recommendations to standardize the measurements (Pickering et al., 2005). This protocol was described in detail previously (Yang et al., 2017;Dong et al., 2013). In brief, SBP and DBP were measured with participants in a sitting position, after a minimum of five minute rest, and > 30 min after exercising, tobacco smoking, and consuming tea, coffee, or alcohol. The measurements were repeated three times after an interval of at least 2 min. The average of the three measurements was recorded. Information on anti-hypertensive treatments was collected by questionnaire. Hypertension was defined as mean SBP ≥ 140 mm Hg or DBP ≥ 90 mm Hg, or reported use of anti-hypertensive medicine (Chobanian et al., 2003).

Exposure assessment
We used the Normalized Difference Vegetation index (NDVI; Tucker, 1979) and the Soil Adjusted Vegetation Index (SAVI; Huete, 1988), which were derived from the Landsat 5 Thematic Mapper satellite images at 30 m × 30 m resolution (http://earthexplorer. usgs.gov), to determine community greenness. The derivation of both NDVI and SAVI indexes are based on land surface reflectance of the visible red and near-infrared parts of the light spectrum. SAVI additionally incorporates a correction factor to minimize soil background. Both NDVI and SAVI values range from −1 to +1, with higher values indicating more greenness. We obtained two cloud-free Landsat 5 Thematic Mapper satellite images during August 2010, the greenest month in Northeastern China and the year closest to our collection of health data from study participants (i.e., between April 1 and Dec 31, 2009). For each study community's centroid, greenness was defined as the average of NDVI or SAVI in 500-m and 1000-m buffers. Considering the small size of the study communities (0.25-0.64 km 2 ), the short distance between communities in each study district (1.5 km), as well as the radius of buffers used in recent studies (Markevych et al., 2014b;Dadvand et al., 2014), we used NDVI values in the 500-m buffer for the main analysis. However, we also reported other metrics and buffers in sensitivity analyses. These calculations were performed using ArcGIS 10.4 (ESRI, Redlands, CA, USA).

Potential confounders and mediators
As suggested by Jager et al. (2008), we first considered potential confounders based on the following three criteria: (1) it should be a risk factor for hypertension or elevated blood pressure; (2) it must be antecedent to the greenness exposure (i.e., a "cause" of the greenness exposure) and unequally distributed between exposed and unexposed groups (i.e., groups with different greenness levels); and (3) it must not be an "effect" of greenness exposure, nor be an intermediate factor in the causal pathway of hypertension. Then, we constructed a directed acyclic graph (DAG, Fig. S1) representing the existing literature to select a minimally sufficient set of covariates to adjust for confounding (Greenland et al., 1999), by employing DA-Gitty v2.3 software (www.dagitty.net). Based on the DAG, the following variables were retained as confounders in our statistical models: age (years), sex (man vs. woman), ethnicity (Han vs. others), household income levels (< 10,000 Yuan vs. > 10,000 Yuan), and district-level gross domestic product (GDP, Chinese Yuan) (Fig. S2).
Information on these selected confounders was mainly collected using a questionnaire. GDP in each district was obtained from each city's Statistical Yearbooks.
Also, based on the DAG, air pollution (i.e., PM 2.5 and NO 2 ), physical activity, and body mass index (BMI) were selected as candidate mediators. We previously described the PM 2.5 and NO 2 assessments in detail (Yang et al., 2018b(Yang et al., , 2018c. Briefly, we downloaded two types of daily aerosol optical depth (AOD) data (i.e., Deep Blue (DB) and Dark Target (DT)) from the Aqua Atmosphere Level 2 Product Collection 1 at a 0.1°× 0.1°spatial resolution. We then combined DB and DT AOD data using an inverse variance weighting method. Further, we developed a generalized additive model to link AOD data with ground-level PM 2.5 measurements, meteorological data, land use information, vegetation data, and other spatial predictor (i.e., fire spot, elevation, calendar month). To test the validity of PM 2.5 predictions, we employed a 10-fold cross-validation procedure, indicating that the adjusted R 2 and root mean squared error were 75% and 15.1 μg/m 3 . To incorporate the impact of air pollutants in areas adjacent to each study community, the value of PM 2.5 at each point was linearly interpolated from the values of the four nearest grid cells. Consequently, different communities that fell into the same grid cell showed different predicted PM 2.5 values (i.e., contingent on levels from the nearby grid cells). NO 2 concentrations were measured using chemiluminescence and reported hourly by airmonitoring stations (Yang et al., 2017). The continuous NO 2 measures were then averaged into daily concentrations. Using the daily measures, we calculated three-year (2006-08) average PM 2.5 and NO 2 concentrations for each of the study communities. We collected information on regular physical activity using self-reported questionnaire (yes (exercised ≥180 min per week) vs. no (exercised < 180 min per week)). Body mass index (BMI, kg/m 2 ) was calculated using measured body weight and height.

Statistical analysis
Pair-wise correlations between NDVI, SAVI, and air pollutants were tested by the Spearman rank correlation test. Based on recent published studies of greenness and blood pressure Lane et al., 2017;Brown et al., 2016), we hypothesized a linear relationship between greenness and blood pressure metrics in the main analysis. Due to the hierarchical structure of our data, we used generalized linear mixed models with a random intercept for community to assess the associations of greenness exposure (per IQR increase in NDVI 500m buffer) with SBP and DBP levels (PROC MIXED in SAS) and hypertension prevalence (PROC GLIMMIX in SAS) (Yang et al., 2018b(Yang et al., , 2018c. The results generated from PROC MIXED procedure and the PROC GLIMMIX procedure were presented as regression coefficients (β) and odds ratios (ORs), and their corresponding 95% confidence intervals (CIs), respectively. We used two levels of covariate adjustments. Crude models were not adjusted. In the main models, we adjusted for the covariates that were selected using the DAG (i.e., age, sex, ethnicity, income, and district-level GDP).
To test the robustness of our estimates, we performed several sensitivity analyses. First, we repeated the analyses using a 1000-m buffer for NDVI, as well as SAVI in buffers of 500-m and 1000-m. Second, we estimated the greenness-blood pressure association after excluding participants who reported taking anti-hypertensive medicine, participants who had cardiovascular diseases (defined as selfreported heart diseases and stroke), or hypotensive participants (defined as SBP ≤90 mm Hg and DBP ≤60 mm Hg (Lim et al., 2003)). Third, we tested non-linear associations between greenness and blood pressure metrics, by categorizing NDVI 500-m levels into quartiles (Q 1 , < 25th percentile; Q 2 , between 25th and 50th percentile; Q 3 , between 50th and 75th percentile; and Q 4 , ≥75th percentile) and examining the effect estimates for each Q 2 , Q 3 , and Q 4 compared with Q 1 , respectively. Next, we tested whether age, sex, and household income levels were potential modifiers of associations between NDVI 500-m and blood pressure metrics. In these tests, age was categorized as older group (≥65 years) and younger group (< 65 years), and household income levels as high income group (≥10,000 Yuan) and low income group (< 10,000 Yuan). Effect modification was tested in subgroup analyses and considered present if a regression cross-product term (i.e., greenness * age or greenness * sex or greenness * income) was statistically significant.
As prior evidence (Markevych et al., 2017) and our DAG suggested that air pollution, physical activity, and adiposity were potential mediators of the effects of greenness on human health, we used these factors in mediation analyses (Schisterman et al., 2009). We used the PROCESS v. 2.16.3 macro for SAS (pre-specified Model 4 in the SAS procedure) to assess the mediating effects of air pollutants and BMI, by calculating bias-corrected 95% CIs of indirect paths (Hayes, 2013). In this analysis, we applied 5000 bootstraps. The proportion of the mediated effect was calculated as: (β indirect effects / β total effects ) × 100%. The two mediators were tested one-at-a-time. We did not account for the multi-level nature of the data. However, we adjusted for community as fixed effects to partially offset the issue of clustering. As our physical activity variable was dichotomous, it could not be assessed in mediating modelling and therefore, we assessed its mediation effects by comparing the effect estimates before and after additional adjustment for physical activity.
All statistical analyses were performed in SAS 9.4 (SAS Institute, Inc. Cary, NC, USA). A two-tailed p value < 0.05 was considered as statistically significant.

Population characteristics
The average age of the study participants was 45.6 years and with similar proportions of women (49%) and men (51%) ( Table 1). Most participants had a household income level of ≥10,000 Yuan per year (76.8%). Thirty percent of the participants were smokers, 23% consumed alcohol, and 31% exercised regularly. Mean SBP and DBP values were 127.28 mm Hg and 81.42 mm Hg, respectively. The prevalence of hypertension was 34.8%, which is similar to the 37.7% reported in a recent national survey (Li et al., 2018). Compared to non-hypertensive participants, hypertensive patients were more likely to be men, older, Han nationality, and to have lower levels of household income (Table S2). Table S3 describes the distribution of greenness indicators in the 33 study communities. Greenness levels varied markedly among the different communities. For example, the median NDVI 500-m value was 0.29, but with a wide range of 0.18 to 0.80. NDVI and SAVI were strongly and positively inter-correlated (r s ranged from 0.67 to 0.98); however, their correlations with air pollutants (PM 2.5 and NO 2 ) were relatively weak and negative, with r s ranging from −0.29 to −0.14 ( Table S4). Table 2 shows the associations for NDVI 500-m with blood pressure and hypertension prevalence. In the unadjusted model, higher levels of NDVI 500-m were consistently and significantly associated with lower SBP and DBP levels, as well as with lower hypertension prevalence. After adjustment for age, sex, ethnicity, income, and arealevel GDP (Table 2, adjusted model), the associations were attenuated but remained statistically significant for SBP and hypertension prevalence. Specifically, an IQR (0.17 unit) increase in NDVI 500-m was significantly associated with lower SBP of 0.82 mm Hg (95% CI: −1.13 to −0.51) and 5% (95% CI: 1-8%) lower odds for having hypertension.

Sensitivity analyses
The associations were generally consistent in sensitivity analyses using 1000-m NDVI and 500-m and 1000-m SAVI buffers (Tables 2 and  S5). In addition, estimates did not differ substantially after excluding participants taking anti-hypertensive medicines, participants who reported cardiovascular diseases, or those who had hypotension (Table  S6). Further, when treating NDVI 500-m as quartiles of exposure, we also documented significant trends, in which the effect estimates decreased with increasing NDVI 500-m levels (Fig. 2). Table 3 shows associations for greenness exposures, blood pressure measures, and hypertension prevalence according to potential effect modifiers. We detected evidence of modification of effect by sex and age. More specifically, associations of NDVI 500-m with SBP, DBP, and hypertension prevalence were statistically significant in women, but not in men. The effect modification of age was not consistent; while the association between NDVI 500-m and SBP was stronger in older participants than in younger participants, an opposite pattern was observed for NDVI 500-m and DBP (i.e., stronger association in the younger group). There was no evidence of effect modification by household income level.

Mediation analyses
Since DBP was not associated with greenness in the main model, we only explored the mediating roles of air pollution, physical activity, and BMI on associations for greenness and SBP (Table 4). We observed that PM 2.5 , NO 2 , and BMI significantly mediated 13.9%, 16.0%, and 39.9% of the estimated associations between greenness and SBP, respectively. Additional adjustment by physical activity did not change effect estimates (Table S7).

Key findings
The results of our large population-based study suggest that higher community greenness levels were significantly associated with lower SBP levels and decreased hypertension prevalence. These associations were robust to a series of sensitivity analyses. In addition, we observed that sex modified the association between greenness and blood pressure. While BMI mediated 39.9% of the association, air pollutants only mediated up to 16% of the associations and no mediation effect was observed for physical activity. To our knowledge, this is one of the few epidemiological studies to report the association of exposure to community greenness with blood pressure, especially in a developing country.
environments or of viewing natural environments on blood pressure levels. The majority (n = 11) of these studies detected a beneficial effect of greenness exposures on SBP and/or DBP (Calogiuri et al., 2016;Grazuleviciene et al., 2015;Hartig et al., 2011;Li et al., 2011;Ochiai et al., 2015;Park et al., 2010;Parsons et al., 1998;Toda et al., 2013;Tsunetsugu et al., 2013;Tsunetsugu et al., 2007), and only two of them reported no association Sung et al., 2012). Although it is difficult to directly compare our results to the experimental studies due to the different study designs, these prior findings are roughly in line with our current findings and provide important support for our hypothesis. Collectively, although results from our current study and those previously published are not completely consistent, the overall evidence generally supports an association between higher greenness levels and lower blood pressure levels (specifically SBP) and hypertension prevalence. However, it is notable that most of these studies were either crosssectional by design or had small sample sizes, and thus a well-designed and large longitudinal investigation is needed to validate our findings.

Underlying mechanism
Although the mechanisms by which greenness benefits health remain unclear, several biopsychosocial pathways have been proposed (Markevych et al., 2017). First, green spaces can reduce ambient air pollution levels (Hirabayashi and Nowak, 2016), which have been associated with an increased risk for hypertension (Yang et al., 2018a). We tested this mechanism in mediation analyses and found that beneficial effects of greenness on blood pressure were largely independent of air pollution, which is in line with some earlier studies Vienneau et al., 2017). Second, living close to green spaces, such as parks, was associated with a higher likelihood of engaging in conducting physical activity (Lachowycz and Jones, 2011), which is a strong protective factor for hypertension (Oparil et al., 2018). In the mediation analyses, we did not detect evidence of mediating effects of physical activity. Our findings were in agreement with Markevych et al.'s (2014a) results that the effect of greenness on children's blood pressure was independent of physical activity. However, Jia et al. (2018) reported that approximately 50% of the association between NDVI and hypertension was mediated by physical exercise. Third, evidence suggests that green environments can reduce adiposity (Sarkar, 2017), which is a well-documented risk factor for hypertension. Our results support this hypothesis in that BMI mediated a large proportion of the association linking greenness to blood pressure. Fourth, evidence has suggested that greenness was also associated with psychological and physiological stress alleviation, increased social cohesion, greater and more diverse microbial exposure, and reduced exposure to noise and heat (Markevych et al., 2017;Rook, 2013). However, the absence of these data in our study prevented us from investigating these mechanisms as potential mediators of the association between greenness and blood pressure. More mechanistic studies are therefore needed to validate the underlying mechanisms.

Susceptible populations
We found that the greenness-blood pressure associations were stronger in women than in men. The evidence concerning sex modification of greenness-blood pressure association is limited and the results have been mixed. Most prior studies were conducted in women or men only or reported sex-adjusted effects. Of four previous studies that performed sex-stratified analysis Jia et al., 2018;Morita et al., 2011;Vienneau et al., 2017), one study reported a stronger association in men than in women (Jia et al., 2018), and the remaining three did not detect modification by sex Morita et al., 2011;Vienneau et al., 2017). Nevertheless, our results were not unexpected. One possible explanation may be that Chinese women may have a higher likelihood of using green spaces than Chinese men. Square dancing, for instance, is one of the most popular exercises among Chinese adults, but is practiced mostly by women in surrounding green spaces (Gao et al., 2016). In addition, evidence has shown that women may also spend more time around the Table 3 Associations between per IQR increase in NDVI 500-m (0.17 unit) and blood pressure metrics by age, sex, and household income (n = 24,845). Abbreviations: β, unstandardized regression coefficient; CI, confidence interval; DBP, diastolic blood pressure; NO 2 , nitrogen dioxide; NDVI, normalized difference vegetation index; OR, odds ratio; PM 2.5 , particle with aerodynamic diameter ≤ 2.5 μm; SAVI, soil adjusted vegetation index; SBP, systolic blood pressure. a Adjusted for age, sex, ethnicity, household income level, and district gross domestic product level. b Statistically significant interaction (p < 0.05). Abbreviations: β, unstandardized regression coefficient; BMI, body mass index; NO 2 , nitrogen dioxide; NDVI, normalized difference vegetation index; PM 2.5 , particle with aerodynamic diameter ≤ 2.5 μm; SBP, systolic blood pressure. a Adjusted for age, sex, ethnicity, household income level, and district gross domestic product level, and community.
home and in nearby parks in associations with child care obligations and part-time employment (Tamosiunas et al., 2014). Another plausible explanation for the sex-specific association may be a preponderance of hazardous occupational exposures among men, which dominated to such an extent that any beneficial effects of greenness were masked. We also detected modification of greenness-blood pressure associations by age, but the patterns were different for different blood pressure metrics. While the association between greenness and SBP was stronger in the older group, its association with DBP and hypertension prevalence was only statistically significant in the younger group. In line with our findings, Jia et al. (2018) observed a stronger association between greenness and hypertension in middle aged adults than in older adults. However, the picture is not clear-  did not detect modification by age. Further investigations therefore remain needed to validate our age-specific results.

Strengths and limitations
Strengths of our study include a large and population-based sample of Chinese urban dwellers and a high response rate (86.2%), which allowed for sufficient statistical power to detect modest effects. Moreover, blood pressures were measured three times using standard methods suggested by the American Heart Association (Pickering et al., 2005) to reduce variability in the study outcomes. In addition, we adjusted for a parsimonious, yet comprehensive panel of covariates to preclude confounding in our results without introducing further bias. Finally, we adopted a conditional procedure to quantify the mediating effects of air pollution and adiposity on greenness-blood pressure associations. This method tends to be more valid and have higher statistical power, and thus provides more precise estimates for the mediation effects compared with traditional Baron and Kenny approach (Baron and Kenny, 1986).
Our study also has several limitations. First, this study adopted a cross-sectional design, which precludes us from inferring any causal relationship. Reverse causality, i.e., the possibility that participants with hypertension may be less likely to live close to areas with higher greenness levels, cannot be excluded but is very unlikely. Second, greenness exposure levels were assigned to community centroids but not to personal addresses, which means that we only had 33 unique data points for the 24,845 study participants. This may have introduced exposure measurement misclassification. However, the exposure misclassification is likely to have been non-differential with respect to blood pressure and thus bias the results towards the null (Hutcheon et al., 2010). Third, due to the clustering sampling scheme used in our study, the data were aggregated into communities (a type of administrative unit). Additionally, as mentioned above, greenness exposure was measured at community-level. Thus, the modifiable areal unit problem, in which the choice of community boundaries impacts the results, is very likely and may have biased our estimates (Openshaw and Taylor, 1979). Fourth, information on physical activity was collected by dichotomous question (exercise regularly: yes or no), whereas detailed information on exercising time, duration, and location (indoor or outdoor) was unavailable, which might have introduced misclassification and prevented us from including this variable into the standard mediation analysis. Fifth, we used the vegetation indexes NDVI and SAVI to estimate greenness levels, which are sensitive to season, and cannot distinguish between structured greenspaces and vegetation outside them. This prevents us from finding out what aspects of greenspace are most relevant for the associations with blood pressure and hypertension. This may also be a reason for physical activity not mediating the greenness-blood pressure associations in our study. Satellite-derived greenness may include green that is weakly correlated with physical activity. For example, high volume roadways lined with trees are captured by satellite-derived greenness, but people do not like to do exercise along them because of traffic noise and air pollution, especially in China. In addition, we did not exclude blue pixels from the NDVI layer, thus, our greenness estimates can be partially affected in presence of water (Markevych et al., 2017). Sixth, although we considered and adjusted for a number of cofounders, the potential for unmeasured confounding is possible. Finally, although we performed repeated blood pressure measurements to ensure reliable study outcomes, they were carried out by nurses at clinics, so the "white coat effect" cannot be ruled out. In addition, we measured blood pressures at a single point in time, which may not represent long-term patterns of blood pressure.

Conclusion
In summary, higher community greenness levels were associated with lower SBP levels and decreased hypertension prevalence, especially in women. BMI mediated a large proportion of the association between greenness and blood pressure, whereas air pollution only mediated a small proportion. Our findings might help policy makers initiate, maintain, and increase green public areas, which in turn may help in reducing the hypertension burden. However, the beneficial effects of greenness on blood pressure should be further investigated by well-designed longitudinal studies, especially taking greenness types into account, which will help to refine preventive health and urban design strategies.

Declaration of interests
None.