Effects of Air Pollutants on Mortality of Chronic Kidney Disease Patients with Green Spaces Exposure: A Large Observational Study

With increasing air pollution, the association between green space exposure and health outcomes is a global health concern. The relationship between air pollution and the survival of patients with chronic kidney disease (CKD) who are exposed to residential greenness is yet to be elucidated. This study aimed to determine this relationship in Seoul between 2002 and 2015. A time-varying survival analysis was conducted to investigate the association between long-term exposure to air pollutants and mortality in 29,602 patients with CKD living in residential environments with small and large green infrastructure. The low and high index groups were dened using continuous and percentile thresholds of the satellite data—the derived average Normalized Difference Vegetation Index within 250 m and 1,250 m of residence, respectively. During the observation, 3,863 (14%) deaths occurred. The effect of air pollution exposure on mortality was worse in the low index group than in the high index group. Particularly, exposure to SO 2 was associated with increased mortality risk, regardless of the greenness threshold. Consistent results were observed in co-pollutant models. High greenery exposure signicantly reduced the risk of air pollution related mortality. Our results emphasize the need for creating environmental infrastructures that include green spaces.


Introduction
Chronic kidney disease (CKD), which is indicated by a glomerular ltration rate less than 60 mL/min/1.73 m 2 for more than 3 months, is a condition in which the kidney function or structure becomes weakened [1] and is recognized as a major public health problem worldwide. The prevalence rates of CKD in Norway and the United States were estimated as 10.2% and 11.0% in the 1990s, respectively, in a large population cohort study [2] . Unlike traditional risk factors such as glomerulonephritis, hypertension, and diabetes mellitus [3] , environmental exposure to pollutants has increasingly become a cause of kidney disease.
Moreover, the kidney is susceptible to air pollution because of its ltration function [4] . A study conducted on a cohort of US veterans found that long-term exposure to particulate matter less than 10 µm [PM 10 ], nitrogen dioxide [NO 2 ], and carbon monoxide [CO], increased the risk of CKD incidence by 7%, 9%, and 10%, respectively [5] . In addition, research conducted in Korea between 2007 and 2011 showed that increases in annual mean concentrations of PM 10 and NO 2 were associated with a decrease in the estimated glomerular ltration rate (eGFR) by 46% and 15%, respectively [6] . As shown in the abovementioned research, exposure to ambient air pollution has been one of the major risk factors for the loss of kidney function.
Considering the threat of air pollutant exposure to humanity, adaptation of green infrastructure is emerging as a solution; generally, this is directly related to the ability to reduce extreme heat and precipitation [7] . Furthermore, it is widely understood that exposure to greenery has positive effects on health by moderating ground temperature [8] , ltering pollutants from the air [9] , improving cardiovascular and mental health issues [10] , and providing space for exercise and social interaction [11] . In particular, previous studies have investigated the preventive role of exposure to green spaces on the adverse health effects of ambient air pollution. A study conducted in Barcelona from 2008 to 2009 found that 54 pregnant women with residential exposure to green spaces were exposed to lower levels of outdoor PM 2.5 [12] . Another study with a cohort of 12,873 participants showed that the hazard ratio (HR) of allcause mortality was 1.07 (1.02-1.11) in the highest tertile of the Normalized Difference Vegetation Index (NDVI; 0.53-0.96), with a 10 µg/m 3 increase in the 3-year average PM 2.5 concentration, compared with 1.10 (1.06-1.14) in the lowest NDVI (-0.14 to 0.30) [13] . However, to the best of our knowledge, only few studies have reported the association between exposure to air pollution and mortality from CKD based on patients' exposure to green spaces.
In this study, we aimed to examine the association between exposure to ambient air pollutants and the risk of mortality in CKD patients with high and low greenness exposures, using a large observational cohort in Seoul, Korea.

Ethical aspects
The study complied with the Declaration of Helsinki and the Declaration of Istanbul and received full

Study population and covariates
The retrospective cohort consisted of 32,949 participants diagnosed with CKD in three medical centers (Seoul National University Hospital, Seoul National University Bundang Hospital, and Seoul National University Boramae Medical Center) between January 2000 and December 2015. We excluded patients with less than three months of cohort participation (n =7) for the diagnosis of CKD.
We collected data on 32,949 participants since 2000; however, the number of participants was limited because, although air pollutant data except that for particulate matter less than 2. We obtained demographic and medical data on the participants retrospectively such as coordinates of their residence, sex, age, and eGFR, which indicated kidney function, hemoglobin, and categories of diagnoses of diabetes and hypertension (0: not diagnosed, 1: presence), at enrolment. The mortality date was con rmed from Statistics Korea. Additionally, we assigned area-level characteristics obtained from Statistics Korea, such as the population density, the number of hospital beds, and the rate of nancial independence, to each participant, based on the district of the residential address of each participant in the enrollment year.

Air pollution data
Hourly air pollution data (PM 2.5 , PM 10

Normalized Difference Vegetation Index
The NDVI has been used as a representative indicator of greenness exposure in previous epidemiological studies [14,15] . NASA's Earth Observing System provided the NDVI obtained from the Terra Moderate Resolution Imaging Spectroradiometer (MODIS) Vegetation Indices (MOD13Q1) Version 6, which is generated every 16 days at a 250 m spatial resolution [16] . The NDVI is calculated using visible (red) and near infrared (NIR) light, based on the characteristic that plants re ect most of the NIR light and not much visible red light [(NIR-red)/(NIR+red)]. Therefore, the NDVI ranges from -1 to 1, with -1 indicating water, and 1 indicating a dense green environment. We assigned the average exposure to residential greenness for each participant's geocoded residence within 250 m and 1,250 m radii during the summer months (June-August), 1 year before cohort enrolment, because the NDVI experiences seasonality. The average greenness value within 250 m and 1,250 m radii of the participant's residence indicated a measurement of the greenness directly accessible outside each home and the greenness within a 10-15 minute walk of the home, respectively.

Statistical analysis
To investigate the long-term effects of air pollutants (PM 2.5 , PM 10 , NO 2 , SO 2 , CO, and O 3 ) on the mortality of CKD patients, while considering the strati ed effects of greenness, a time-dependent Cox model was used. The annual NDVI exposures for each participant during the summer months were classi ed into higher and lower green areas using continuous and categorical NDVI criteria. Using these greenery distributions, the NDVI standards were divided based on continuous values of 0.3, 0.35, and 0.4, and percentiles of 50%, 75%, and 90%. The Hazard Ratios (HRs) and 95% con dence intervals (CIs) were estimated in fully adjusted models strati ed by sex and 5-year age groups and adjusted by eGFR at baseline, hemoglobin, diabetes, hypertension, population density, number of hospital beds, and nancial independence rate. Unlike the NDVI exposure, we re ected the time-varying characteristics of air pollution from 1 year before enrollment to the censoring date. In addition, we calculated survival time from the date of cohort entry to the censoring date, censoring patients at the date of death or end date of follow-up (December 31, 2015). We constructed co-pollutant models based on the correlation among pollutants to con rm their robustness. The results are presented as HRs with interquartile range (IQR) increase, which is the difference between the 75th and 25th percentiles of air pollutant concentration. All statistical analyses were conducted using R version 3.6.1 (Foundation for Statistical Computing, Vienna, Austria). Table 1 shows the baseline characteristics of the 29,602 CKD patients in Seoul, Korea. The NDVI exposure around the 1,250 m buffer, which was within walking distance, was higher overall, compared to the 250 m buffer. The average was calculated to be 0.31 at 250 m (dotted) and 0.33 at 1,250 m (twodashed) radii ( Figure. 1). When patients were divided according to a 50% concentration of the 250 m buffer, people living in areas with more greenness had less deaths, older age, and lower population density than participants in the area with less greenness. We found no differences in sex, eGFR, or diabetes diagnosis between the two groups. When classifying the NDVI group by percentile and continuous threshold, the group differences of air pollutant concentration between low and high NDVI spaces were all signi cant (Table S1). The average exposure concentrations of air pollutants in the high NDVI areas were lower than those in the low NDVI areas, except for O 3 . decreased in patients with high greenery exposure, compared to those with low greenery exposure. Consistent results were found in the high-NDVI environment classi ed by continuous criteria ( Figure. 3 and Table S3).

Results
Co-pollutant models were used to con rm the robustness of the main analyses. Before implementing the two-pollutant models, we calculated the correlation between the air pollutants ( Figure S1). When the correlation coe cient was 0.7 or more, we excluded the variable from the co-pollutant model because of collinearity. For example, when analyzing the effect of PM 2.5 on the mortality of CKD patients, PM 10 was excluded from the two-pollutant model because the correlation with PM 10 was large at 0.88. Therefore, the effects of PM 2.5 , associated with mortality risk were analyzed in a single model and in a two-pollutant model with NO 2 and SO 2 , respectively. Robust results were shown in the co-pollutant model (Figures S2-S7). HRs tended to increase in high NDVI areas by some greenery standards and buffers. However, the IQR increase in the concentration of each air pollutant had an increased effect on mortality in low NDVI areas than in high NDVI areas, and its signi cance disappeared, generally. For example, when using the NDVI threshold within the 250 m buffer by 0.35, HRs in the low NDVI group were 1. .08), were found for the high NDVI group, respectively. The protective effects of greenery exposure against air pollution were greater when exposed at a 250 m radius rather than a 1,250 m radius.

Discussion
This study explored the association between long-term exposure to air pollution and the mortality risk of approximately 30,000 CKD patients exposed to high and low levels of vegetation in Seoul, Korea. The risk effects of air pollution associated with mortality in low NDVI areas were greater than those in high NDVI areas, indicating that green environment could reduce the mortality risk associated with air pollution. Our ndings are consistent with the results observed in co-pollutant models.
An approximately 10-year study found an association between mortality from respiratory diseases in 66,820 elderly individuals aged 65 years or older, and air pollutants (PM 2.5 , PM 10 , NO 2 , and O 3 ) in areas with low and high greenness, using a time-strati ed case-crossover design [17] . The percent excess risk and 95% CI per 10 µg/m 3 increase in a 4-day moving average of PM 2.5 , NO 2 , and O 3 for pneumonia were and low-level green areas, respectively [18] . Although a higher risk of non-accidental disease was shown in greener spaces, cardiovascular mortality rate was estimated to increase by 7.46% (3.97-11.07%) in the high-green area and 11.23% (7.28-15.32%) in the low-green area. Likewise, studies examining the health effects associated with air pollution and residential greenness were in the implementation stage, and ndings regarding the protective effects of greenery exposure were inconsistent Two representative theories could explain our results showing that the effects of air pollution could be reduced using green infrastructure [19] . First, urban vegetation typically removes a few percent of air pollutants by depositing them onto the leaves. However, this effect is not often apparent in urban areas and is usually seen in forested areas. Dispersion, however, is more important than deposition for reducing air pollution exposure in downtown areas. Air pollutants can be dispersed and diluted because green environment can control the speed, direction, and dispersion distance of air pollutants [20] . In other words, exposure to green spaces can play a role in maintaining the distribution of air pollutants, and reduced air pollution may be associated with decreased mortality risk in areas with more residential greenness. The underlying mechanisms of the effects of air pollution associated with greenness in patients with CKD are unknown. However, this may be because the health bene ts of green space exposure are related to various pathways; these bene ts include reducing air pollution and noise and enabling stress alleviation and relaxation [21] .
Our study compared the average exposure rate to air pollutants between patients living in areas with low and high levels of green infrastructure. Reduced concentrations of air pollutants in high vegetation areas were observed, except for O 3 . A previous study conducted in Spain supported the contribution of urban forests regulating services to the mitigation of pollution (305.6 tons [t] of removed air pollutants and 19036 t of CO 2 equivalent per year) [22] . In addition, a study conducted in the United States showed that one of the ways that urban trees affect air quality is pollution removal, showing an estimated total removal of 711,000 metric tons ($3.8 billion value) of pollutants (O 3 , PM 10 , NO 2 , SO 2 , and CO) [23] . However, the results of research on the reduction effect of O 3 on greenery exposure are controversial. Previous studies have reported that certain tree types can emit isoprene in areas with enough pollutants during periods of extensive heat and weak winds, which increases the concentration of O 3 in the atmosphere [24] . Furthermore, yearly O 3 removal differed according to tree functional groups (evergreen broadleaves, deciduous broadleaves, conifers, and actual functional groups), classi ed by tree cover, leaf season length, and physiology [25] . Therefore, O 3 concentration depends on vegetation characteristics, such as tree species, physiological status, and surrounding environment, such as climate and distribution of atmospheric chemicals [26] . Although O 3 concentration was high in green areas, the effects of health risk associated with air pollutants in high vegetation areas were lower than in the low vegetation areas. There were no clear explanations for these associations. However, the health effects may differ depending on the O 3 source (emissions from leaves in plants or created by chemical reactions between oxides of nitrogen [NOx] and volatile organic compounds [VOCs] from cars and industrial sources). Another reason why the risk size of ambient O 3 concentration on health was smaller is due to the complex effect of greenery exposure on human health.
Our study showed distinct risk effects of SO 2 associated with the mortality of CKD patients living in low vegetation areas, regardless of the buffer and criteria of residential NDVI. SO 2 emitted from fossil fuel combustion by both industrial processes and power plants induces stomatal closure at high concentrations, resulting in photosynthesis reduction and general water stress in leaves [27] . These mechanisms protect plants from rapid detoxi cation to avoid injury. Therefore, visible symptoms such as abscission of older leaves and tip necrosis in owers were commonly observed due to SO 2 pollution [28] .
Weakened plant functions may interfere with the pathways of residential greenness, affecting health at high SO 2 concentrations in low vegetation areas.
Our study has some limitations. First, the results from epidemiologic studies may contain exposure measurement errors for actual exposure and ambient air pollution. Second, we could not adjust for potential confounders, such as physical activity and lifestyle, because of lack of data availability. Finally, our data could not explain the underlying mechanism of the negative health effects associated with O 3 in low greenery areas, despite low concentrations. However, we observed protective effects of vegetations related to air pollutants in a large observational cohort in a metropolitan city.

Conclusions
In conclusion, green infrastructure could reduce the concentrations of air pollutants, except O 3 , and provide positive health bene ts to CKD patients through various pathways. Therefore, it is necessary to create a planned and phased green environment.

Abbreviations
Con dence interval, CI; Chronic kidney disease, CKD; carbon monoxide, CO; hazard ratio, HR; interquartile range, IQR; Normalized Difference Vegetation Index, NDVI; particulate matter less than 2.5 μm, PM2.5; particulate matter less than 10.0 μm, PM10. Association between IQR increase in air pollution and risk of mortality by percentile Association between IQR increase in air pollution and risk of mortality by continuous value

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.