Exposure to Greenness and Mortality in a Nationwide Prospective Cohort Study of Women

Abstract

Background:
Green, natural environments may ameliorate adverse environmental exposures (e.g. air pollution, noise, and extreme heat), increase physical activity and social engagement, and lower stress.

Objectives:
We aimed to examine the prospective association between residential greenness and mortality.

Methods:
Using data from the US-based Nurses' Health Study prospective cohort, we defined cumulative average time-varying seasonal greenness surrounding each participant's address using satellite imagery (Normalized Difference Vegetation Index (NDVI)). We followed 108,630 women and observed 8,604 deaths between 2000-2008.

Results:
In models adjusted for mortality risk factors (age, race/ethnicity, smoking, and individual- and area-level socioeconomic status), women living in the highest quintile of cumulative average greenness (accounting for changes in residence during follow-up) in the 250m area around their home had a 12% lower rate of all-cause non-accidental mortality (95% CI 0.82, 0.94) compared to those in the lowest quintile. Results were consistent for the 1,250m area, although the relationship was slightly attenuated. These associations were strongest for respiratory and cancer mortality. Findings from a mediation analysis suggest that the association between greenness and mortality may be at least partly mediated by physical activity, particulate matter less than 2.5 micrometers, social engagement, and depression.

Conclusions:
Higher levels of green vegetation were associated with decreased mortality. Policies to increase vegetation may provide opportunities for physical activity, reduce harmful exposures, increase social engagement, and improve mental health. While planting vegetation may mitigate effects of climate change, evidence of an association between vegetation and lower mortality rates suggests it also might be used to improve health.

Full text

1
Exposure to Greenness and Mortality in a Nationwide Prospective Cohort Study of Women
Authors:
Peter James,1,2 Jaime E. Hart,1,3 Rachel F. Banay,2 Francine Laden1,2,3
1Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston,
Massachusetts, USA
2Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston,
Massachusetts, USA
3Channing Division of Network Medicine, Department of Medicine, Brigham and Women’s
Hospital and Harvard Medical School, Boston, Massachusetts, USA
Corresponding Author:
Peter James
Department of Epidemiology and Environmental Health, Harvard T.H. Chan School of Public
Health
401 Park Dr
3rd Floor West
Boston, MA 02215 USA
617-525-2567
pjames@hsph.harvard.edu
Running Title: Greenness and mortality in a nationwide cohort
Acknowledgements
The research conducted for this manuscript was supported by the Harvard NHLBI
Cardiovascular Epidemiology Training Grant T32 HL 098048 and NIH Grants UM1 CA186107
and R01 ES017017.
Competing financial interests
The authors declare no competing financial interests.

2
Abstract
Background: Green, natural environments may ameliorate adverse environmental exposures
(e.g. air pollution, noise, and extreme heat), increase physical activity and social engagement,
and lower stress.
Objectives: We aimed to examine the prospective association between residential greenness and
mortality.
Methods: Using data from the US-based Nurses’ Health Study prospective cohort, we defined
cumulative average time-varying seasonal greenness surrounding each participant’s address
using satellite imagery (Normalized Difference Vegetation Index (NDVI)). We followed 108,630
women and observed 8,604 deaths between 2000-2008.
Results: In models adjusted for mortality risk factors (age, race/ethnicity, smoking, and
individual- and area-level socioeconomic status), women living in the highest quintile of
cumulative average greenness (accounting for changes in residence during follow-up) in the
250m area around their home had a 12% lower rate of all-cause non-accidental mortality (95%
CI 0.82, 0.94) compared to those in the lowest quintile. Results were consistent for the 1,250m
area, although the relationship was slightly attenuated. These associations were strongest for
respiratory and cancer mortality. Findings from a mediation analysis suggest that the association
between greenness and mortality may be at least partly mediated by physical activity, particulate
matter less than 2.5 micrometers, social engagement, and depression.
Conclusions: Higher levels of green vegetation were associated with decreased mortality.
Policies to increase vegetation may provide opportunities for physical activity, reduce harmful
exposures, increase social engagement, and improve mental health. While planting vegetation

3
may mitigate effects of climate change, evidence of an association between vegetation and lower
mortality rates suggests it also might be used to improve health.

4
Introduction
The biophilia hypothesis suggests that human beings have evolved to prefer certain natural
environments that are essential to their thriving (Wilson 1984). Researchers are increasingly
exploring how neighborhood greenness, or vegetation, may affect health behaviors and outcomes
(Hartig et al. 2014; James et al. 2015a). Empirical research suggests that greenness may reduce
obesity and promote physical activity (Lachowycz and Jones 2011), as well as improve
cardiovascular health (Pereira et al. 2012), mental health (Alcock et al. 2014; Gascon et al.
2015), and birth outcomes (Hystad et al. 2014). Greenness has been hypothesized to benefit
health by lowering exposure to air pollution, extreme heat, and noise; increasing opportunities
for physical activity; providing a location for social engagement; and decreasing psychological
stress and depression through direct contact with nature (Hartig et al. 2014; Hystad et al. 2014;
Taylor et al. 2015).
Evidence suggests that exposure to greenness may lower mortality rates, although many of these
studies relied on aggregated data (Jonker et al. 2014; Lachowycz and Jones 2014; Richardson et
al. 2010; Richardson and Mitchell 2010), which limit inferences about the effect of greenness on
individual health. Many mortality studies relied on cross-sectional data and could not estimate
exposure over time (Hu et al. 2008; Mitchell and Popham 2008), while others could not account
for important potential confounding by race/ethnicity, individual-level smoking, and area-level
socioeconomic factors such as median home value (Takano et al. 2002; Villeneuve et al. 2012).
Some studies have observed contradictory findings. One ecological study conducted on the city-
level found that all-cause mortality was higher in greener cities (Richardson et al. 2012). An
ecological analysis across the entire United Kingdom found that higher greenness was associated

5
with lower cardiovascular and respiratory mortality among males; however, no significant
associations were found amongst women (Richardson and Mitchell 2010). A recent analysis of
greenness and mortality in male and female stroke survivors living in the Boston area found that
higher exposure to greenness was associated with higher survival rates (Wilker et al. 2014). To
our knowledge, no study has examined time-varying greenness and mortality in a nationwide
prospective cohort of women while accounting for important potential confounding factors and
addressing potential mediators. Our objective was to examine the association between greenness
and all-cause mortality, as well as cause-specific mortality, in a large prospective study of
women across the entire contiguous United States. We hypothesized that higher levels of
surrounding greenness would be associated with lower rates of all-cause, cancer, respiratory, and
cardiovascular mortality; that these associations would differ by mortality cause; and that
findings would support mediation by air pollution, physical activity, social engagement, and
mental health.
Methods
Population
The Nurses’ Health Study (NHS) is a prospective cohort study assessing risk factors for chronic
disease among women. In 1976, 121,701 female registered nurses ages 30-55 years from 11
states (California, Connecticut, Florida, Maryland, Massachusetts, Michigan, New Jersey, New
York, Ohio, Pennsylvania, and Texas) returned an initial questionnaire, which ascertained a
variety of health-related exposures and medical diagnoses. The cohort has been continuously
followed with biennial questionnaires. Response rates at each questionnaire cycle have
consistently been ≥90% (calculated as the number of women who successfully returned a
questionnaire in each cycle among the women who were still alive). Residential addresses from

6
2000-2008 questionnaires were matched (geocoded) to obtain latitude and longitude.
Approximately 90% of all addresses were successfully matched to the street segment level
(within a range of house numbers along one side of the street). This analysis was conducted
among all women who were alive in 2000 and had at least one residential address geocoded to
the street segment level. Geocoded residence locations in 2000 are shown in Figure 1a; there are
at least ten nurses in each of the contiguous United States. The study was approved by the
Institutional Review Board of Brigham and Women's Hospital, Boston, MA, and informed
consent was implied through return of the questionnaires.
Exposure
Exposure to vegetation around each participant’s home address was estimated using a satellite-
image based vegetation index. Chlorophyll in plants absorbs visible light (0.4-0.7 µm) for use in
photosynthesis, while leaves reflect near-infrared light (0.7-1.1 µm). The Normalized Difference
Vegetation Index (NDVI) calculates the ratio of the difference between the near-infrared region
and red reflectance to the sum of these two measures, and ranges from -1.0 to 1.0, with larger
values indicating higher levels of vegetative density (Kriegler et al. 1969). For this study, we
used data from the Moderate-resolution Imaging Spectroradiometer (MODIS) from NASA’s
Terra satellite. MODIS provides images every 16 days at a 250m resolution (Carroll et al. 2004).
We used geographic information systems (GIS) software from ArcMap (ESRI, Redlands, CA) to
estimate the mean NDVI value inside radii of 250 and 1,250m buffers around each participant’s
home. We chose the 250m radius as a measure of greenness directly accessible outside each
home, and the 1,250m radius as a measure of greenness within a 10-15 minute walk based on
prior work within the Nurses’ Health Study cohorts on neighborhood environments and health

7
behaviors (James et al. 2014). We created a seasonally time-varying measure based on the NDVI
for a representative month in each season (January, April, July, and October) (Figure 1b-d). Two
exposure metrics were calculated for each radius: contemporaneous NDVI (the greenness value
for the current season) to reflect short term exposure to greenness, as well as cumulative average
NDVI (updated based on changes in seasonal NDVI, as well as changes in address) to reflect
longer term exposure to greenness. For both exposure metrics, exposures were updated both as
NDVI changed over time, as well as when participants moved to new residential addresses
(updated based on the receipt of a biennial questionnaire with a new residential address).
Outcome
We assessed death occurring between the return of the 2000 questionnaire and June 1, 2008.
Deaths were usually reported by families, and deaths among non-respondents were identified by
searching the National Death Index, which has been validated in prior studies in this cohort
(Rich-Edwards et al. 1994). A physician reviewed death certificates and medical records to
classify the primary cause of death according to the International Classification of Diseases,
version 9. Our primary outcome was death from all non-accidental causes. Secondary cause-
specific analyses were conducted for the following categories (infectious and parasitic diseases
ICD9 Codes 0-139; cancer 140-208; diabetes 250; neurodegenerative disease 290, 332, 335, 340,
342, 348; coronary heart disease 390-429, 440-459; stroke 430-438; respiratory diseases 460-
519; kidney disease 580-593; and all other causes). Other deaths (including accidental ICD9
Codes E800-E999) were included as negative control outcomes to detect potential confounding
bias (Lipsitch et al. 2010).

8
Statistical Analysis
Person-months of follow-up were accrued from the return date of the 2000 questionnaire until
either death or the end of follow-up (May 31, 2008), whichever came first. We fit time-varying
Cox proportional hazards models to compute hazard ratios (HRs) and 95% confidence intervals
(CIs) for associations between each NDVI exposure measure and each mortality outcome. We
examined the following covariates as potential confounders, effect modifiers, or mediators (all
covariates time-varying unless otherwise indicated): fixed race/ethnicity (White non-Hispanic
versus other), smoking status (current, past, never), pack-years of smoking, fixed individual-level
socioeconomic status (SES), area-level SES, weight status (normal (body mass index (BMI)
18.5-24.9), overweight (BMI 25-29.9), obese (BMI > 30)), region, urbanicity, whether a
participant had changed addresses over follow-up, physical activity, air pollution, social
engagement, and mental health. Current smoking status and pack-years smoked was updated at
each biennial questionnaire. To account for fixed individual SES, we included information on
self-reported parental occupation for the participant’s mother and father, whether the participant
had a registered nursing degree, marital status, and husband’s highest educational attainment (<
high school, high school graduate, >high school, missing or not married) as reported in 1992. We
examined area-level SES by including information on Census tract median home value and
Census tract median income based on the Census tract containing the residential address at each
questionnaire response and 2000 Census data. Urbanicity was determined by the participant’s
residence in a metropolitan (urban area ≥50,000 people), micropolitan (urban cluster of 10,000-
49,999), or small town / rural (urban cluster of <10,000) Census tract (Morrill et al. 1999).
Physical activity was evaluated based on a validated biennial measure of self-reported total
physical activity in the past year (Wolf et al. 1994). Although the specific activities varied on

9
each questionnaire, questions included the average time per week spent walking, jogging (>10
min per mile), running (≤10 min per mile), bicycling, lap swimming, playing tennis, playing
squash or racquet ball, using a rowing machine, and engaging in calisthenics, aerobics, or aerobic
dance. Each participant also reported the number of flights of stairs that she climbed daily and
her usual walking pace. We multiplied the reported time spent weekly at each activity by its
typical energy expenditure requirements expressed in metabolic equivalents (METs), then
summed all the activity figures to yield a MET hours per week score and categorized total
physical activity as less than 3, 3-8.9, 9-17.9, 18-26.9, and 27 or more MET hours per week.
These cutpoints were chosen to correspond to the equivalent of less than 1, 1 to less than 3, 3 to
less than 6, 6 to less than 9, and 9 or more hours per week of walking at an average pace,
consistent with prior analyses in this cohort (Colditz et al. 2003; Meyerhardt et al. 2006). Air
pollution was quantified as quintiles of residential address-level 12-month average particulate
matter less than 2.5 microns in aerodynamic diameter (PM2.5) predicted from a spatio-temporal
generalized additive mixed model (Yanosky et al. 2014). Social engagement was evaluated twice
(2000 and 2004) over follow-up based on responses to the question “How many hours/week do
you participate in groups?” (e.g. social, community, charity, etc.). We dichotomized social
engagement based on whether participants reported participating in groups one hour per week or
more. Mental health was based on biennial self-reports of doctor-diagnosed depression or regular
antidepressant use.
Analyses were stratified by age of follow-up (months) and time period, and adjusted for
race/ethnicity, smoking status, pack-years of smoking, and individual-level SES measures based
on questionnaire responses, as well as Census tract area-level SES measures. We used the

10
missing indicator method to account for missing covariate data. We used cubic regression splines
to determine the linearity of exposure-response relationships (Durrleman and Simon 1989). Tests
for non-linearity used the likelihood ratio test, comparing the model with only the linear term to
the model with the linear and the cubic spline terms. We report results for both continuous NDVI
and NDVI quintiles. We examined the linear test for trend using the ordinal rank for each
quintile. To test for violations of the proportional hazards assumption, we included interaction
terms of each exposure and calendar time and performed likelihood ratio tests to determine
statistically significant violations. Additionally, we tested for effect modification of the
relationship between greenness and mortality by race/ethnicity, smoking, Census tract median
income, Census tract median home value, weight status (defined by normal weight (BMI 18.5-
24.9), overweight (BMI 25-29.9), obese (BMI > 30), physical activity, air pollution, region,
urbanicity, and whether a participant had changed addresses over follow-up. We modeled
interaction terms between continuous cumulative average NDVI in a 250m buffer and each
potential effect modifier and used Likelihood Ratio tests to assess statistical significance. We
also obtained strata-specific effect estimates from stratified analyses. P-values under 0.05 were
used to define statistical significance. Data were analyzed in SAS 9.3 (Cary, NC).
We explored potential mechanisms through which greenness might affect mortality by
evaluating the mediating effect of factors such as physical activity, air pollution exposure, social
engagement, or mental health. We calculated the mediation proportion and its 95% CI using the
publicly available %Mediate macro (http://www.hsph.harvard.edu/donna-
spiegelman/software/mediate/) (Lin et al. 1997). Briefly, the macro compares the exposure effect
estimate from the full model that includes the exposure, one or more potential intermediate

11
variables, and any covariates to the exposure effect estimate obtained from a partial model that
leaves out the intermediate variable or variables. The mediation proportion is the proportion of
reduced mortality explained by higher exposure to greenness that can be attributed to elevated
levels of physical activity, air pollution exposure, social engagement, or mental health, as well as
the joint effect of all of these mediators combined. Confidence intervals for the mediation
proportion were calculated by using the data duplication method (Lin et al. 1997). Mediation
analyses assumed that there was no unmeasured exposure–outcome confounding; no unmeasured
mediator–outcome confounding; no unmeasured exposure–mediator confounding; and no
mediator–outcome confounder affected by exposure (VanderWeele 2015). While these
assumptions are unverifiable, we included major confounders in our mediation analyses and
therefore we believe our assumptions are reasonable.
Results
Participants were primarily White non-Hispanic, normal weight, and had low levels of physical
activity (Table 1). The majority of participants lived in metropolitan areas, and half the sample
lived in the Northeastern US. Those living in areas with higher levels of greenness were slightly
younger, more likely to be White non-Hispanic, had husbands with higher levels of education,
and lived in neighborhoods with higher SES. Areas with higher greenness had lower levels of air
pollution.
We observed 8,604 deaths over 627,008 person-years of follow-up among the 108,630 eligible
cohort members. Table 2 shows HRs for the relationship between cumulative average greenness
exposure and non-accidental mortality. Analyses showed a consistent relationship between

12
higher greenness and decreased mortality that was robust to adjustment for individual- and area-
level covariates. In fully adjusted models, those living in the highest quintile of cumulative
average greenness in the 250m area around their home had a 12% lower rate of mortality (95%
CI 0.82, 0.94) compared to those in the lowest quintile. Results were consistent for the 1,250m
radius, although the relationship was slightly attenuated. Likelihood ratio tests from cubic
regression spline analyses indicated that relationships between greenness and mortality rate were
linear (data not shown). Continuous analyses indicated an inverse association between greenness
and mortality as well, with a 12% lower rate of mortality (95% CI 0.82, 0.94) in fully-adjusted
models based on a 0.1 increase in cumulative average NDVI in the 250m area around
participants’ homes. Again, the association was attenuated in the 1,250m buffer. Results from
models of contemporaneous NDVI were weaker, but showed a generally consistent inverse
association between greenness and mortality.
Cause-specific mortality analyses demonstrated that the associations were strongest for
respiratory, cancer, and kidney-disease mortality, and we observed negative HRs for stroke
mortality that were not statistically significant (Table 3). We estimated that those living in the
highest quintile of cumulative average greenness in the 250m area around their home had a 34%
lower rate of respiratory disease-related mortality (95% CI 0.52, 0.84), a 13% lower rate of
cancer mortality (95% CI 0.78, 0.97), and a 41% lower rate of kidney-disease mortality (95% CI
0.33, 1.05) compared to those in the lowest quintile. We did not observe any statistically
significant associations between greenness and mortality from coronary heart disease, diabetes,
or infections. In addition, associations for our negative control outcome of other deaths
(including accidental, 185 out of the 1,219 other deaths) were null.

13
Stratified Analyses
We observed no statistically significant differences in the association between greenness and
mortality by race/ethnicity, smoking status, Census tract median income, Census tract median
home value, PM2.5, weight status, region, urbanicity, or between movers and nonmovers (see
Supplemental Material, Figure S1). Stratified analyses for physical activity, urbanicity, Census
tract median income, and PM2.5 are shown in Figure 2. We did observe a stronger association
between greenness and mortality among participants with higher physical activity levels,
although there was no statistical evidence of a difference in associations (p=0.14). More precise
relationships were observed for greenness in metropolitan areas compared to micropolitan and
rural areas; however, we did not observe statistical evidence of a difference in associations
across levels of urbanicity. This likely reflects that 84% of our sample lived in urban areas. There
was no statistical evidence of a difference in associations across different levels of Census tract
median income (interaction p-value = 0.99).
Mediation
Estimates of the proportion of the association between greenness and mortality that might be
mediated by other factors (assuming underlying assumptions of the mediation analyses hold)
were statistically significant for physical activity, PM2.5, social engagement, and mental health
(Table 4). The largest proportion of mediation was doctor-diagnosed depression or
antidepressant use, which was estimated to explain 30.6% (95% CI 15.5%, 51.4%) of the
association with cumulative average greenness in a 250m buffer and mortality, and social
engagement, which was estimated to explain 19.1% (95% CI 10.0%, 33.3%). Although the
estimated proportions explained were smaller, they were also statistically significant for physical
activity (2.1% explained (95% CI 0.2%, 19.3%)) and air pollution (4.4% explained (95% CI

14
2.4%, 7.7%)). The estimate from the joint mediation analysis suggested that all four mediators
combined might explain 27.1% (95% CI 14.7%, 44.6%) of the association between greenness
and all-cause mortality. Findings were generally similar for greenness in a 1,250m radius.
Supplemental Material Table S1 shows mediation results for cancer, respiratory, and kidney-
disease mortality, where we observed the strongest estimate of mediation by the mental health
pathway and the weakest estimate of mediation by physical activity.
Discussion
In this nationwide study of adult women, higher levels of greenness around each participant’s
address were associated with lower rates of all-cause, non-accidental mortality, regardless of
adjustment for age, race/ethnicity, smoking, individual-level SES, and area-level SES. These
findings were strongest for cancer, respiratory, and kidney-disease mortality. Results were
consistent when focusing on the area immediately around each residence (250m buffer) versus a
larger radius (1,250m buffer) around each participant’s home. Results were strongest when
examining cumulative average exposure to greenness versus contemporaneous greenness,
suggesting a larger benefit of chronic exposure to greenness for health. The association between
greenness and mortality was not statistically significantly different by race/ethnicity, physical
activity, smoking status, area-level SES, air pollution exposure, weight status, region of the US,
whether a participant lived in a rural or urban area, or whether a participant moved over follow-
up. Assuming that assumptions of the mediation analysis hold, our estimates suggest that a large
proportion of the association between greenness and mortality may be explained through mental
health pathways of depression risk and social engagement, which subsequently impacted
mortality.

15
Our findings were consistent with, yet slightly stronger than, a general population mortality
study that was cross-sectional (Mitchell and Popham 2008), as well as a study that was unable to
adjust for individual-level smoking and sociodemographic characteristics (Villeneuve et al.
2012). Mitchell and Popham (2008) classified the percentage of green space for geographic units
across England and observed lower levels of all-cause and circulatory mortality in the greenest
areas, with an estimated six percent reduction (95% CI 4%, 7%) in all-cause mortality in
geographic units with the highest quintile of green space. A study of residents of Ontario,
Canada found that an IQR difference in NDVI (0.24) was associated with an estimated five
percent (95% CI 3%, 6%) reduced rate of non-accidental mortality (Villeneuve et al. 2012).
Wilker et al. (2014) followed patients in the Boston area following an ischemic stroke and found
that those living in the highest quartile of NDVI had an estimated 22% (95% CI 3%, 37%) lower
rate of all-cause mortality compared to the lowest quartile. Our findings differed from an
ecological study of 49 major cities across the US using the National Land Cover Database to
define greenness (Richardson et al. 2012). This analysis showed that all-cause mortality rates
were highest in the greenest cities; however, the city-level, cross-sectional analysis was unable to
account for any individual-level factors, including smoking. In addition, an ecological cross-
sectional study of urban wards across the United Kingdom found no relationship between
greenness, measured through a land use database and satellite imagery, and cardiovascular,
respiratory, or lung cancer mortality among women (Richardson and Mitchell 2010).
In cause-specific mortality analyses, we observed associations between greenness and
respiratory, cancer, and kidney-disease mortality. The findings for respiratory and cancer
mortality are consistent with the pathway that greenness reduces air pollution exposure and

16
increases physical activity, which are known to lower the risk of these outcomes (Hamra et al.
2014; Hoek et al. 2013; Lee et al. 2012). In addition, our findings on respiratory mortality were
similar to those observed by Villeneuve et al. (2012). While the absolute number of cases was
small, there was a strong relationship between greenness and kidney disease that has not been
observed in previous studies. This relationship could exist because greenness is linked to
physical activity; inadequate physical activity is a risk factor for kidney disease (Stump 2011).
We also observed suggestive evidence for a link between greenness and stroke mortality, which
has been seen in other studies (Hu et al. 2008; Villeneuve et al. 2012). Although prior research
has shown relationships between greenness and coronary heart disease, diabetes, and infections
(Lachowycz and Jones 2014; Mitchell and Popham 2008; Rook 2013), we did not observe
associations for these outcomes. A possible explanation for these discrepant findings is that our
cohort of female nurses may utilize greenness in different ways than cohorts that include men
(Richardson and Mitchell 2010) or individuals from other SES groups. Our finding of no
association between greenness exposure and other deaths (including accidental), which can be
considered negative controls because there is no clear mechanism for an association, lends
confidence that the observed associations are not the result of uncontrolled confounding or other
sources of bias that create a spurious causal inference (Lipsitch et al. 2010).
Our mediation analyses suggest that greenness affects all-cause mortality, as well as cancer,
respiratory, and kidney-disease mortality, through mental health, social engagement, physical
activity and air pollution. There is a foundation for each of these mechanisms in the literature.
An analysis by de Vries et al. (2013) estimated that stress and social cohesion mediated the

17
relationship between streetscape greenery and health. In their study, total physical activity was
not a mediator; however, physical activity that took place in a public space did appear to mediate
the greenery-health relationship. Higher exposure to greenness has been consistently linked to
lower levels of depression, anxiety, and stress (Alcock et al. 2014; Beyer et al. 2014; Gascon et
al. 2015). Studies have demonstrated that views of nature may have a direct psychological
benefit (Fuller et al. 2007). Higher levels of social engagement are also correlated with greenness
exposure (Maas et al. 2009). Greenness appears to buffer exposure to air pollution (Dadvand et
al. 2012; Su et al. 2009), and vegetation has been shown to remove particulate matter, SO2, and
carbon monoxide (Nowak et al. 2006), although evidence is inconsistent for other pollutants
(Kim et al. 2013). Greenness may protect individuals from exposure to harmful noise (Gidlöf-
Gunnarsson and Öhrström 2007), as well as alleviate thermal discomfort during heat stress
(Lafortezza et al. 2009). Greater exposure to greenness has been associated with higher levels of
physical activity (Almanza et al. 2012), but, consistent with our findings, other studies have
shown that physical activity does not fully explain the greenness and health relationship
(Lachowycz and Jones 2014).
This study had a few limitations. The most appropriate scale to measure greenness is unclear
(Mitchell et al. 2011). This uncertainty in the relevant geographic context to study exposure is a
fundamental problem in spatial analyses (Kwan 2012). We explored two geographic scales
(250m and 1,250m buffers) that showed similar results, but it is unclear whether we would have
observed stronger or weaker relationships if we had examined different scales. While satellite-
based measures of vegetation have been used extensively to measure exposure to greenness,
NDVI does not measure quality of greenness. Nevertheless, a validation study has demonstrated

18
that NDVI performs adequately when compared to environmental psychologists’ evaluations of
green spaces (Rhew et al. 2011). Three environmental psychologists examined photographs,
evaluated greenness on a likert scale (none or very little to very high greenness), and compared
this to NDVI measured in a 100m radius around 124 homes in Baltimore/Washington and
Seattle. Interrater reliability was high between environmental psychologist evaluations
(ICC=0.82) and correlations between the photograph ratings and NDVI were high (r=0.69,
p<0.001), indicating that NDVI may be a valid measure of greenness. Because of a lack of time-
varying, nationwide data, we did not examine the association between exposure to major green
spaces (e.g., parks) and mortality. The NHS dataset enabled evaluation of many potential
confounders and mediators; however, we were still limited in our ability to measure exposure to
heat and noise, which could play important roles in how greenness affects health. In addition, the
underlying assumptions required for estimates from the mediation analysis to be valid are
unverifiable, which is a limitation of our mediation analysis. Selection into neighborhoods
according to health status is a consistent concern in studies of geographic context and health. If
participants in better health selected to move to neighborhoods with higher levels of greenness,
this confounding by neighborhood preference could explain the relationship between greenness
and mortality. However, in prior analyses we have demonstrated that BMI and physical activity
levels did not predict neighborhood selection by built environment features in this cohort (James
et al. 2015b), and therefore it is unlikely that neighborhood self-selection is a major concern. The
consistent relationship between greenness and SES measures indicated the potential for strong
confounding; however, we aimed to reduce the likelihood of confounding by adjusting for
multiple individual- and area-level measures of SES. The association between greenness and
mortality was consistent across different levels of area-level SES, which lends additional

19
confidence that the observed relationships were not a consequence of unmeasured confounding
by area-level SES. Additionally, the negative control outcome of other deaths was not associated
with greenness, further decreasing the likelihood that our findings were an artifact of
confounding. Because of the geographic distribution of our sample, it is likely that we had
limited power to assess variation by region and urban/rural differences. Finally, participants in
this study were nurses at the time of recruitment, were over 90% White non-Hispanic, and were
all female. While this does restrict the generalizability of these results to the general population,
it also eliminates confounding by sex, and lowers the potential for confounding by SES and
race/ethnicity.
This study also has a number of notable strengths. To our knowledge, it is the first prospective
examination of the relationship between exposure to greenness and mortality across the entire
US. We were able to construct time-varying measures of exposure to greenness in the area
surrounding each participant’s home address over eight years of follow-up. Additionally, we
were able to control for important confounders, such as smoking and individual- and area-level
SES. Because participants were located in a diverse range of geographic settings across the
country, we were able to test whether the relationship between greenness and mortality was
consistent in different regions, as well as in urban and rural locations. Finally, measurements of
important intermediate exposures and behaviors enabled us to examine the potential mechanisms
through which greenness impacts mortality.

20
Conclusions
In this nationwide cohort of adult women, we observed that those living in the highest quintile of
satellite-measured green vegetation around their home had a lower mortality rate when compared
to those living in the lowest quintile of greenness. Findings were consistent across all regions of
the US, as well as in urban and rural areas, and we observed no threshold at which greater
greenness exposure was not associated with lower mortality rates. Mediation analyses suggested
that the association between greenness and mortality was explained primarily by improving
mental health and increasing social engagement, as well as by lowering air pollution exposure
and increasing physical activity. While more research is required on the relationship between
other natural environments and health (e.g., blue spaces (Gascon et al. 2015)), these findings
suggest that green vegetation has a protective effect, and that policies to increase vegetation in
both urban and rural areas may provide opportunities for physical activity, reduce harmful
exposures, increase social engagement, and improve mental health. While the recognized
benefits of planting vegetation include reducing wastewater loads, sequestering carbon, and
mitigating the effects of climate change (Jesdale et al. 2013), evidence of an association between
vegetation and lower mortality rates suggests a potential co-benefit to improve health, presenting
planners, landscape architects, and policy-makers with an actionable tool to grow healthier
places.

21
References
Alcock I, White MP, Wheeler BW, Fleming LE, Depledge MH. 2014. Longitudinal effects on
mental health of moving to greener and less green urban areas. Environ Sci Technol 48:1247-
1255.
Almanza E, Jerrett M, Dunton G, Seto E, Pentz MA. 2012. A study of community design,
greenness, and physical activity in children using satellite, gps and accelerometer data. Health
Place 18:46-54.
Beyer KM, Kaltenbach A, Szabo A, Bogar S, Nieto FJ, Malecki KM. 2014. Exposure to
neighborhood green space and mental health: Evidence from the survey of the health of
Wisconsin. Int J Environ Res Public Health 11:3453-3472.
Carroll ML, DiMiceli CM, Sohlber RA, Townshend JRG. 2004. 250m MODIS Normalized
Difference Vegetation Index. College Park, MD.
Colditz GA, Feskanich D, Chen WY, Hunter DJ, Willett WC. 2003. Physical activity and risk of
breast cancer in premenopausal women. Br J Cancer 89:847-851.
Dadvand P, de Nazelle A, Triguero-Mas M, Schembari A, Cirach M, Amoly E, et al. 2012.
Surrounding greenness and exposure to air pollution during pregnancy: An analysis of personal
monitoring data. Environ Health Perspect 120:1286-1290.
de Vries S, van Dillen SM, Groenewegen PP, Spreeuwenberg P. 2013. Streetscape greenery and
health: Stress, social cohesion and physical activity as mediators. Soc Sci Med 94:26-33.
Durrleman S, Simon R. 1989. Flexible regression models with cubic splines. Stat Med 8:551-
561.
Fuller RA, Irvine KN, Devine-Wright P, Warren PH, Gaston KJ. 2007. Psychological benefits of
greenspace increase with biodiversity. Biology letters 3:390-394.
Gascon M, Triguero-Mas M, Martinez D, Dadvand P, Forns J, Plasencia A, et al. 2015. Mental
health benefits of long-term exposure to residential green and blue spaces: A systematic review.
Int J Environ Res Public Health 12:4354-4379.
Gidlöf-Gunnarsson A, Öhrström E. 2007. Noise and well-being in urban residential
environments: The potential role of perceived availability to nearby green areas. Landscape and
Urban Planning 83:115-126.
Hamra GB, Guha N, Cohen A, Laden F, Raaschou-Nielsen O, Samet JM, et al. 2014. Outdoor
particulate matter exposure and lung cancer: A systematic review and meta-analysis. Environ
Health Perspect 122:906-911.
Hartig T, Mitchell R, de Vries S, Frumkin H. 2014. Nature and health. Annu Rev Public Health
35:207-228.
Hoek G, Krishnan RM, Beelen R, Peters A, Ostro B, Brunekreef B, et al. 2013. Long-term air
pollution exposure and cardio- respiratory mortality: A review. Environ Health 12:43.
Hu Z, Liebens J, Rao KR. 2008. Linking stroke mortality with air pollution, income, and
greenness in northwest Florida: An ecological geographical study. Int J Health Geogr 7:20.
Hystad P, Davies HW, Frank L, Van Loon J, Gehring U, Tamburic L, et al. 2014. Residential
greenness and birth outcomes: Evaluating the influence of spatially correlated built-environment
factors. Environ Health Perspect 122:1095-1102.
James P, Berrigan D, JE H, Hoehner CM, Kerr J, Major J, et al. 2014. Effects of buffer size and
shape on associations between the built environment and energy balance. Health Place 27:162-
170.

22
James P, Banay RF, Hart JE, Laden F. 2015a. A review of the health benefits of greenness. Curr
Epidemiol Rep 2:131-142.
James P, Hart JE, Arcaya MC, Feskanich D, Laden F, Subramanian SV. 2015b. Neighborhood
self-selection: The role of pre-move health factors on the built and socioeconomic environment.
Int J Environ Res Public Health 12:12489-12504.
Jesdale BM, Morello-Frosch R, Cushing L. 2013. The racial/ethnic distribution of heat risk-
related land cover in relation to residential segregation. Environ Health Perspect 121:811-817.
Jonker MF, van Lenthe FJ, Donkers B, Mackenbach JP, Burdorf A. 2014. The effect of urban
green on small-area (healthy) life expectancy. J Epidemiol Community Health 68:999-1002.
Kim S, Guenther A, Apel E. 2013. Quantitative and qualitative sensing techniques for biogenic
volatile organic compounds and their oxidation products. Environmental science Processes &
impacts 15:1301-1314.
Kriegler F, Malila W, Nalepka R, Richardson W. Preprocessing transformations and their effects
on multispectral recognition. In: Proceedings of the Proceedings of the Sixth International
Symposium on Remote Sensing of Environment, 1969. Ann Arbor, MI, 97-131.
Kwan M. 2012. The uncertain geographic context problem. Annals of the Association of
American Geographers 102:958-968.
Lachowycz K, Jones AP. 2011. Greenspace and obesity: A systematic review of the evidence.
Obes Rev 12:e183-189.
Lachowycz K, Jones AP. 2014. Does walking explain associations between access to greenspace
and lower mortality? Soc Sci Med 107:9-17.
Lafortezza R, Carrus G, Sanesi G, Davies C. 2009. Benefits and well-being perceived by people
visiting green spaces in periods of heat stress. Urban Forestry and Urban Greening 8:97-108.
Lee IM, Shiroma EJ, Lobelo F, Puska P, Blair SN, Katzmarzyk PT. 2012. Effect of physical
inactivity on major non-communicable diseases worldwide: An analysis of burden of disease and
life expectancy. Lancet 380:219-229.
Lin DY, Fleming TR, De Gruttola V. 1997. Estimating the proportion of treatment effect
explained by a surrogate marker. Stat Med 16:1515-1527.
Lipsitch M, Tchetgen Tchetgen E, Cohen T. 2010. Negative controls: A tool for detecting
confounding and bias in observational studies. Epidemiology 21:383-388.
Maas J, van Dillen SM, Verheij RA, Groenewegen PP. 2009. Social contacts as a possible
mechanism behind the relation between green space and health. Health Place 15:586-595.
Meyerhardt JA, Heseltine D, Niedzwiecki D, Hollis D, Saltz LB, Mayer RJ, et al. 2006. Impact
of physical activity on cancer recurrence and survival in patients with stage III colon cancer:
Findings from calgb 89803. J Clin Oncol 24:3535-3541.
Mitchell R, Popham F. 2008. Effect of exposure to natural environment on health inequalities:
An observational population study. Lancet 372:1655-1660.
Mitchell R, Astell-Burt T, Richardson EA. 2011. A comparison of green space indicators for
epidemiological research. J Epidemiol Community Health 65:853-858.
Morrill R, Cromartie J, Hart G. 1999. Metropolitan, urban, and rural commuting areas: Toward a
better depiction of the United States settlement system. Urban Geography 20:727-748.
Nowak DJ, Crane DE, Stevens JC. 2006. Air pollution removal by urban trees and shrubs in the
United States. Urban Forestry and Urban Greening 4:115-123.
Pereira G, Foster S, Martin K, Christian H, Boruff BJ, Knuiman M, et al. 2012. The association
between neighborhood greenness and cardiovascular disease: An observational study. BMC
Public Health 12:466.

23
Rhew IC, Vander Stoep A, Kearney A, Smith NL, Dunbar MD. 2011. Validation of the
normalized difference vegetation index as a measure of neighborhood greenness. Ann Epidemiol
21:946-952.
Rich-Edwards JW, Corsano KA, Stampfer MJ. 1994. Test of the national death index and
equifax nationwide death search. Am J Epidemiol 140:1016-1019.
Richardson E, Pearce J, Mitchell R, Day P, Kingham S. 2010. The association between green
space and cause-specific mortality in urban New Zealand: An ecological analysis of green space
utility. BMC Public Health 10:240.
Richardson EA, Mitchell R. 2010. Gender differences in relationships between urban green
space and health in the United Kingdom. Soc Sci Med 71:568-575.
Richardson EA, Mitchell R, Hartig T, de Vries S, Astell-Burt T, Frumkin H. 2012. Green cities
and health: A question of scale? J Epidemiol Community Health 66:160-165.
Rook GA. 2013. Regulation of the immune system by biodiversity from the natural environment:
An ecosystem service essential to health. Proc Natl Acad Sci U S A 110:18360-18367.
Stump CS. 2011. Physical activity in the prevention of chronic kidney disease. Cardiorenal Med
1:164-173.
Su JG, Jerrett M, Beckerman B. 2009. A distance-decay variable selection strategy for land use
regression modeling of ambient air pollution exposures. Sci Total Environ 407:3890-3898.
Takano T, Nakamura K, Watanabe M. 2002. Urban residential environments and senior citizens'
longevity in megacity areas: The importance of walkable green spaces. J Epidemiol Community
Health 56:913-918.
Taylor MS, Wheeler BW, White MP, Economou T, Osborne NJ. 2015. Research note: Urban
street tree density and antidepressant prescription rates—a cross-sectional study in London, UK.
Landscape and Urban Planning 136:174-179.
VanderWeele T. 2015. Explanation in causal inference : Methods for mediation and interaction.
New York:Oxford University Press.
Villeneuve PJ, Jerrett M, Su JG, Burnett RT, Chen H, Wheeler AJ, et al. 2012. A cohort study
relating urban green space with mortality in Ontario, Canada. Environ Res 115:51-58.
Wilker EH, Wu CD, McNeely E, Mostofsky E, Spengler J, Wellenius GA, et al. 2014. Green
space and mortality following ischemic stroke. Environ Res 133:42-48.
Wilson EO. 1984. Biophilia. Cambridge:Harvard University Press.
Wolf AM, Hunter DJ, Colditz GA, Manson JE, Stampfer MJ, Corsano KA, et al. 1994.
Reproducibility and validity of a self-administered physical activity questionnaire. Int J
Epidemiol 23:991-999.
Yanosky JD, Paciorek CJ, Laden F, Hart JE, Puett RC, Liao D, et al. 2014. Spatio-temporal
modeling of particulate air pollution in the conterminous United States using geographic and
meteorological predictors. Environ Health 13:63.

24
Table 1. Age-Adjusted Nurses’ Health Study Participant Characteristics by Quintiles of Cumulative Average NDVI within 250m Buffers
from 2000-2008 (N=108,630)
Characteristic Total
Greenness
Quintile 1
Greenness
Quintile 2
Greenness
Quintile 3
Greenness
Quintile 4
Greenness
Quintile 5
Cumulative Average NDVI
(250m Buffer) (Mean ± SD)
0.47 ± 0.12
0.29 ± 0.08
0.42 ± 0.03
0.48 ± 0.03
0.54 ± 0.03
0.62 ± 0.05
Age (years)a (Mean ± SD)
68.98 ± 7.29
69.89 ± 7.26
69.42 ± 7.32
68.94 ± 7.30
68.49 ± 7.24
68.16 ± 7.18
White non-Hispanic (%)
94
90
94
94
95
95
BMI (Mean ± SD)
25.83 ± 7.35
25.33 ± 8.26
25.91 ± 7.43
26.08 ± 7.18
25.98 ± 7.08
25.82 ± 6.79
Weight Status (%)
Normal Weight (BMI 18.5-
24.9)
39
38
38
38
39
41
Overweight (BMI 25-29.9)
33
32
33
33
33
33
Obese (BMI >30)
23
23
23
24
23
21
Missing BMI
6
8
6
5
5
5
Total Physical Activity, MET
hrs/week (%)
<3
23
23
23
23
22
21
3 to <9
21
20
21
22
21
20
9 to <18
18
18
19
18
19
19
18 to <27
11
11
11
11
11
12
≥27
19
18
19
19
19
21
Missing
8
11
8
7
7
7
Smoking Status (%)
Never smoker
44
44
45
45
44
44
Past smoker
45
45
45
45
46
47
Current smoker
10
11
11
10
10
10
Have RN degree (%)
73
69
73
74
74
74
Married (%)
64
58
63
65
66
68
Husband’s Highest Education
(%)
< high school
4
4
4
4
4
3
High school grad
26
25
27
27
26
24
> high school education
35
32
33
34
36
41

25
Missing or not married
35
39
35
34
34
32
Census 2000 tract median
income (Mean ± SD)
$63K ± $24K
$57K ± $22K
$60K ± $22K
$62K ± $22K
$66K ± $24K
$72K ± $29K
Census 2000 tract median home
value (Mean ± SD)
$170K ± $125K
$175K ± 138K
$159K ± $121K
$156K ± $115K
$167K ± $113K
$194K ± $133K
12-month average PM
2.5
(µg/m3)
(Mean ± SD)
12.03 ± 2.80
12.51 ± 3.74
12.23 ± 2.79
12.05 ± 2.45
11.88 ± 2.28
11.47 ± 2.38
Census Tract Urbanicity (%)
Metropolitan (urban area
≥50,000 people)
84
86
83
83
84
83
Micropolitan (urban cluster
of 10,000-49,999)
10
9
10
10
9
10
Small town or rural (urban
cluster of <10,000)
7
5
7
7
6
7
Region (%)
Northeast
50
35
42
52
63
56
Midwest
17
15
24
22
15
8
West
14
34
16
8
5
7
South
19
16
18
17
17
28
Moved over follow-up (%)
32
33
30
30
30
35
Depression (Doctor Diagnosed
or Antidepressant Use) (%)
12
11
12
12
11
11
Social Engagement
(Participation in Groups >1 hour
per week) (%)
68
67
69
69
68
69
a Value is not age-adjusted

26
Table 2. Hazard ratios (HR) and 95% confidence intervals (CI) for greenness and non-accidentala all-cause mortality in the Nurses’
Health Study (N=108,630, with 8,604 deaths from 2000-2008)
250m Buffer
1250m Buffer
Exposure Metric
Cumulative Average
Greenness
Contemporaneous
Greenness
Cumulative Average
Greenness
Contemporaneous
Greenness
Age-
Adjusted
HR
(95% CI)
Fully
Adjusted
HR
(95% CI)
b
Age-
Adjusted
HR
(95% CI)
Fully
Adjusted
HR
(95% CI)
b
Age-
Adjusted
HR
(95% CI)
Fully
Adjusted
HR
(95% CI)
b
Age-
Adjusted
HR
(95% CI)
Fully
Adjusted
HR
(95% CI)
b
Quintile 1
Ref
Ref
Ref
Ref
Ref
Ref
Ref
Ref
Quintile 2
0.91
(0.85, 0.97)
0.92
(0.86, 0.98)
0.96
(0.90, 1.03)
0.96
(0.90, 1.02)
0.96
(0.90, 1.02)
0.95
(0.89, 1.01)
0.96
(0.90, 1.02)
0.95
(0.89, 1.01)
Quintile 3
0.88
(0.83, 0.94)
0.90
(0.84, 0.96)
0.93
(0.87, 0.99)
0.93
(0.87, 0.99)
0.95
(0.89, 1.01)
0.94
(0.88, 1.01)
0.95
(0.89, 1.01)
0.94
(0.88, 1.01)
Quintile 4
0.91
(0.85, 0.97)
0.94
(0.88, 1.00)
0.87
(0.82, 0.93)
0.89
(0.83, 0.95)
0.93
(0.87, 1.00)
0.94
(0.88, 1.01)
0.93
(0.87, 1.00)
0.94
(0.88, 1.01)
Quintile 5
0.83
(0.77, 0.88)
0.88
(0.82, 0.94)
0.89
(0.84, 0.96)
0.93
(0.87, 0.99)
0.86
(0.81, 0.92)
0.89
(0.83, 0.96)
0.86
(0.81, 0.92)
0.89
(0.83, 0.96)
P for Trendc
<0.0001
0.002
<0.0001
0.003
<0.0001
0.004
<0.0001
0.004
Continuous (per 0.1 unit)
0.83
(0.78, 0.89)
0.88
(0.82, 0.94)
0.97
(0.94, 1.00)
0.99
(0.95, 1.02)
0.84
(0.79, 0.90)
0.89
(0.83, 0.95)
0.96
(0.93, 1.00)
0.98
(0.94, 1.01)
a Excludes accidental ICD9 Codes E800-E999
b Hazard ratios are adjusted for age and calendar year, race/ethnicity, smoking status, pack-years smoked, parental occupation, RN degree, marital
status, husband’s highest education, Census tract median home value, and Census tract median income
c Based on linear test for trend using the ordinal rank for each quintile

27
Table 3. Hazard ratios (HR) and 95% confidence intervals (CI) for cumulative average NDVI (250m buffer) and cause-specific mortality
in the Nurses’ Health Study (N=108,630)
Outcome Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for Trend b
Continuous
(per 0.1 unit)
Infectious and Parasitic Diseases (304 Cases)
Cases
72
66
67
48
51
Adjusted HR (95% CI)a
Ref
0.99 (0.71, 1.40)
1.07 (0.76, 1.51)
0.82 (0.56, 1.19)
0.92 (0.63, 1.33)
0.411
0.78 (0.55, 1.10)
Cancer (3,363 Cases)
Cases
778
695
654
650
586
Adjusted HR (95% CI)a
Ref
0.93 (0.84, 1.03)
0.90 (0.81, 1.00)
0.93 (0.83, 1.03)
0.87 (0.78, 0.97)
0.024
0.85 (0.76, 0.94)
Diabetes (145 Cases)
Cases
37
32
20
33
23
Adjusted HR (95% CI)a
Ref
0.90 (0.56, 1.46)
0.59 (0.34, 1.02)
1.05 (0.65, 1.71)
0.81 (0.47, 1.38)
0.625
0.85 (0.52, 1.39)
Neurodegenerative Diseases (827 Cases)
Cases
188
173
173
146
147
Adjusted HR (95% CI)a
Ref
0.97 (0.78, 1.19)
1.03 (0.83, 1.27)
0.93 (0.74, 1.16)
0.98 (0.78, 1.22)
0.767
0.93 (0.75, 1.15)
Coronary Heart Disease (1,420 Cases)
Cases
320
299
264
308
229
Adjusted HR (95% CI)a
Ref
0.99 (0.85, 1.16)
0.94 (0.79, 1.11)
1.19 (1.02, 1.40)
0.97 (0.81, 1.15)
0.474
1.02 (0.87, 1.20)
Stroke (606 Cases)
Cases
155
114
140
109
88
Adjusted HR (95% CI)a
Ref
0.76 (0.59, 0.97)
0.99 (0.79, 1.26)
0.86 (0.67, 1.11)
0.77 (0.59, 1.01)
0.195
0.79 (0.62, 1.01)
Respiratory (766 Cases)
Cases
213
163
158
129
103
Adjusted HR (95% CI)a
Ref
0.84 (0.69, 1.04)
0.86 (0.69, 1.06)
0.75 (0.60, 0.94)
0.66 (0.52, 0.84)
<0.001
0.73 (0.59, 0.90)
Kidney (139 Cases)
Cases
39
34
27
21
18
Adjusted HR (95% CI)a
Ref
0.93 (0.58, 1.48)
0.77 (0.47, 1.27)
0.64 (0.37, 1.11)
0.59 (0.33, 1.05)
0.029
0.63 (0.38, 1.04)
Other (1,219 Cases)c
Cases
290
260
215
223
231
Adjusted HR (95% CI)a
Ref
0.95 (0.80, 1.12)
0.82 (0.69, 0.99)
0.91 (0.76, 1.09)
1.01 (0.84, 1.20)
0.810
1.01 (0.85, 1.20)

28
a Hazard ratios are adjusted for age and calendar year, race/ethnicity, smoking status, pack-years smoked, parental occupation, RN degree, marital
status, husband’s highest education, Census tract median home value, and Census tract median income
b Based on linear test for trend using the ordinal rank for each quintile
c Includes any cause of death not included in above categories, including accidental causes of death

29
Table 4. Estimated proportion of association between greenness and mortality in the Nurses’
Health Study explained by physical activity, air pollution exposure, social engagement, and mental
healtha, b
Mediator
Proportion of Association
of Cumulative Average
Greenness in 250m Buffer
Explained by Mediator
(95% CI)
Proportion of Association
of Cumulative Average
Greenness in 1,250m
Buffer Explained by
Mediator (95% CI)
Total Physical Activity (<3 metabolic
equivalent of task hrs/week v. ≥3 MET-
hrs/week)
2.1% (0.2%, 19.3%) 1.1% (0.1%, 15.8%)
Air Pollution (Modeled PM
2.5
<9.7 µg/m3
(Quintile 1) v. ≥9.7 µg/m
3
(Quintiles 2-5))
4.4% (2.4%, 7.7%) 5.1% (2.4%, 10.5%)
Social Engagement (Participate in
Groups >1 per week v. ≤1 per week)
19.1% (10.0%, 33.3%) 12.8% (6.4%, 24.0%)
Mental Health (Doctor-Diagnosed or
Antidepressant Use v. None)
30.6% (15.5%, 51.4%) 25.5% (12.8%, 44.4%)
All Mediators Combined
27.1% (14.7%, 44.6%)
19.8% (10.2%, 35.0%)
a Analyses adjusted for age and calendar year, race/ethnicity, smoking status, pack-years smoked, parental
occupation, RN degree, marital status, husband’s highest education, Census tract median home value, and
Census tract median income
b Mediation analyses assume that there is no unmeasured exposure–outcome confounding; no unmeasured
mediator–outcome confounding; no unmeasured exposure–mediator confounding; and no mediator–
outcome confounder affected by exposure.

30
Figure Legends
Figure 1. a) Nurses’ Health Study geocoded residence locations at baseline (2000); NDVI 2000
Values in b) January, c) July, and d) September
Figure 2. Hazard ratios a and 95% confidence intervals for a 0.1 unit increase in cumulative average
NDVI at the 250m buffer and all-cause non-accidental mortality in the Nurses’ Health Study
(N=108,630) stratified by a) levels of physical activity, b) Census tract metropolitan / rural or small
town status, c) Census tract median household income, and d) predicted PM2.5 exposure. Hazard
ratios are adjusted for age and calendar year, race/ethnicity, smoking status, pack-years smoked,
parental occupation, RN degree, marital status, husband’s highest education, Census tract median
home value, and Census tract median income, except when stratifying variable.

a
b
c
d

a) P for interaction: 0.14
b) P for interaction: 0.058
c) P for interaction: 0.99
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
<3 MET hrs/wk 3-8.9 MET hrs/wk 9-17.9 MET hrs/wk 18-26.9 MET hrs/wk 27 MET hrs/wk
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by Physical Activity Categories
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
Metropolitan (urban area ≥50,000
people)
Micropolitan (urban cluster of
10,000-49,999)
Small town or rural (urban cluster
of <10,000)
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by Metropolitan / Micropolitan / Rural
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
Census Tract Median
Income Quintile 1
Census Tract Median
Income Quintile 2
Census Tract Median
Income Quintile 3
Census Tract Median
Income Quintile 4
Census Tract Median
Income Quintile 5
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by Census Tract Median Income

Supplemental Material
Exposure to Greenness and Mortality in a Nationwide Prospective Cohort Study of Women
Authors:
Peter James, Jaime E. Hart, Rachel F. Banay, Francine Laden

Supplemental Material, Table S1. Estimated proportion of association between greenness and mortality in the Nurses’ Health Study
explained by physical activity, air pollution exposure, social engagement, and mental healtha, b
Mediator
Proportion of Association between Cumulative Average Greenness in 250m Buffer and
Mortality Explained by Mediator (95% CI)
Cancer Mortality
Respiratory Mortality
Kidney-Disease Mortality
Total Physical Activity (metabolic
equivalent of task-hrs/week)
2.5% (0.7%, 9.1%)
Too small to calculate reliably
(<1%)
Too small to calculate reliably
(<1%)
Air Pollution (Modeled PM2.5)
3.1% (1.3%, 6.9%)
2.1% (1.2%, 3.7%)
7.2% (2.5%, 18.9%)
Social Engagement (Participate in Groups
>1 per week)
9.7% (4.2%, 21.1%) 4.9% (2.0%, 11.8%) 6.7% (2.4%, 17.3%)
Mental Health (Doctor-Diagnosed or
Antidepressant Use)
19.9% (8.2%, 40.7%) 6.3% (2.1%, 17.1%) 9.6% (3.4%, 24.6%)
All Mediators Combined
15.3% (6.6%, 31.5%)
7.0% (3.1%, 14.9%)
11.2% (4.1%, 27.3%)
a Analyses adjusted for age and calendar year, race/ethnicity, smoking status, pack-years smoked, parental occupation, RN degree, marital status,
husband’s highest education, Census tract median home value, and Census tract median income
b Mediation analyses assume that there is no unmeasured exposure–outcome confounding; no unmeasured mediator–outcome confounding; no
unmeasured exposure–mediator confounding; and no mediator–outcome confounder affected by exposure.

a) P for interaction: 0.23
b) P for interaction: 0.061
c) P for interaction: 0.61
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
Non-White White
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by Race/Ethnicity
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
Northeast Midwest West South
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by Region
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
Census Tract Median
Home Value Quintile 1
Census Tract Median
Home Value Quintile 2
Census Tract Median
Home Value Quintile 3
Census Tract Median
Home Value Quintile 4
Census Tract Median
Home Value Quintile 5
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by Census Tract Median Home Value

d) P for interaction: 0.69
e) P for interaction: 0.73
f) P for interaction: 0.59
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
Particulate Matter
Quintile 1
Particulate Matter
Quintile 2
Particulate Matter
Quintile 3
Particulate Matter
Quintile 4
Particulate Matter
Quintile 5
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by PM2.5 Quintiles
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
Normal Weight (BMI 18.5-24.9) Overweight (BMI 25-29.9) Obese (BMI >30)
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by Weight Status Categories
00.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
Never Moved over Follow-up Moved over Follow-up
Hazard Ratio (95% CI)
Association between a 0.1 Unit Increase in 250m Cumulative Average NDVI and All-
Cause Mortality Stratified by Whether a Participant Moved Over Follow-up

Supplemental Material, Figure S1. Hazard ratios and 95% confidence intervals for a 0.1 unit
increase in cumulative average NDVI at the 250m buffer and all-cause non-accidental mortality in
the Nurses’ Health Study (N=108,630) stratified by a) Race/Ethnicity, b) Census region, c) Census
tract median home value, d) PM2.5 quintiles, e) Weight status, and f) Whether a participant moved
over follow-up. Hazard ratios are adjusted for age and calendar year, race/ethnicity, smoking
status, pack-years smoked, parental occupation, RN degree, marital status, husband’s highest
education, Census tract median home value, and Census tract median income, except when
stratifying variable