Content uploaded by Kristine Engemann
Author content
All content in this area was uploaded by Kristine Engemann on Mar 14, 2019
Content may be subject to copyright.
Residential green space in childhood is associated with
lower risk of psychiatric disorders from adolescence
into adulthood
Kristine Engemann
a,b,c,1
, Carsten Bøcker Pedersen
c,d,e
, Lars Arge
f
, Constantinos Tsirogiannis
f
, Preben Bo Mortensen
c,d,e
,
and Jens-Christian Svenning
a,b
a
Section for Ecoinformatics & Biodiversity, Department of Bioscience, Aarhus University, 8000 Aarhus C, Denmark;
b
Center for Biodiversity Dynamics in a
Changing World, Department of Bioscience, Aarhus University, 8000 Aarhus C, Denmark;
c
Centre for Integrated Register-based Research, Aarhus University,
8210 Aarhus V, Denmark;
d
National Centre for Register-based Research, School of Business and Social Sciences, Department of Economics and Business
Economics, Aarhus University, 8210 Aarhus V, Denmark;
e
The Lundbeck Foundation Initiative for Integrative Psychiatric Research, Aarhus University, 8210
Aarhus V, Denmark; and
f
Center for Massive Data Algorithmics, Department of Computer Science, Aarhus University, 8200 Aarhus N, Denmark
Edited by Terry Hartig, Uppsala University, Uppsala, Sweden, and accepted by Editorial Board Member Susan T. Fiske January 14, 2019 (received for review
May 2, 2018)
Urban residence is associated with a higher risk of some psychi-
atric disorders, but the underlying drivers remain unknown. There
is increasing evidence that the level of exposure to natural
environments impacts mental health, but few large-scale epide-
miological studies have assessed the general existence and
importance of such associations. Here, we investigate the pro-
spective association between green space and mental health in the
Danish population. Green space presence was assessed at the
individual level using high-resolution satellite data to calculate the
normalized difference vegetation index within a 210 ×210 m
square around each person’s place of residence (∼1 million people)
from birth to the age of 10. We show that high levels of green
space presence during childhood are associated with lower risk of
a wide spectrum of psychiatric disorders later in life. Risk for sub-
sequent mental illness for those who lived with the lowest level of
green space during childhood was up to 55% higher across various
disorders compared with those who lived with the highest level of
green space. The association remained even after adjusting for
urbanization, socioeconomic factors, parental history of mental illness,
and parental age. Stronger association of cumulative green space
presence during childhood compared with single-year green space
presence suggests that presence throughout childhood is important.
Our results show that green space during childhood is associated with
better mental health, supporting efforts to better integrate natural
environments into urban planning and childhood life.
geographic information systems
|
mental health
|
psychological ecosystem
services
|
remote sensing
|
urban planning
The number of people living in cities is increasing, and, glob-
ally, more than 50% of the human population is now city
dwellers, with no sign of this trend slowing down (1). Urban
living often offers good sanitation, access to health care, nutri-
tion, and education (2), but has also been associated with adverse
health effects (3, 4). In some societies, urban residents have al-
most 50% higher risk of developing psychiatric disorders such as
anxiety and mood disorders compared with their rural counter-
parts (5–7), and schizophrenia risk is 200% higher for children
growing up in the most urban environments (8, 9). Different
hypotheses have been proposed to explain these urban−rural
gradients in mental health, including selective migration (10),
social stress processing (4), higher exposure to infections (8), and
reduced exposure to nature (11). However, the mechanistic links
are not well understood (5, 12).
Low exposure to nature, or green space, has been proposed in
recent years as a potential environmental risk factor for various
mental health outcomes. Exposure to green space has been
suggested to lower depression (13, 14) and schizophrenia risk
(15), improve children’s cognitive development (16), and reduce
neural activity linked to psychiatric disorders (11, 17). Dose–
response relationships from other studies show that higher doses
of green space are associated with better mental health (13), and
long-lasting positive effects of moving to greener areas (18)
suggest causation, albeit most of these studies are small and
cross-sectional. Whether the association with green space is
specific or applies to a broader spectrum of psychiatric disorders
and to what degree the association is independent from urban-
ization in general or just mirrors the urban−rural gradient are
unknown. Access to green space partially depends on socioeco-
nomic factors, such as housing prices, and the extent to which
this explains associations between green space and mental ill-
nesses is another unknown aspect.
Exposure to green space may influence mental health through
both psychological and physiological pathways, as green spaces
serve as settings for individual and social behavior and can
mitigate negative influences of other aspects of the physical en-
vironment (19). These mechanistic pathways could vary for different
psychiatric disorders, with green space as a shared risk-decreasing
Significance
Growing up in urban environments is associated with risk of
developing psychiatric disorders, but the underlying mecha-
nisms are unknown. Green space can provide mental health
benefits and possibly lower risk of psychiatric disorders. This
nation-wide study covering >900,000 people shows that chil-
dren who grew up with the lowest levels of green space had
up to 55% higher risk of developing a psychiatric disorder in-
dependent from effects of other known risk factors. Stronger
association between cumulated green space and risk during
childhood constitutes evidence that prolonged presence of
green space is important. Our findings affirm that integrating
natural environments into urban planning is a promising ap-
proach to improve mental health and reduce the rising global
burden of psychiatric disorders.
Author contributions: K.E., C.B.P., P.B.M., and J.-C.S. designed research; K.E. performed
research; L.A. and C.T. contributed new reagents/analytic tools; K.E. analyzed data; and
K.E., C.B.P., P.B.M., and J.-C.S. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission. T.H. is a guest editor invited by the Editorial Board.
Published under the PNAS license.
Data deposition: The NDVI maps used to estimate green space presence are available for
download at https://bios.au.dk/en/about-bioscience/organisation/ecoinformatics-and-
biodiversity/data.
1
To whom correspondence should be addressed. Email: engemann@bios.au.dk.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
1073/pnas.1807504116/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1807504116 PNAS Latest Articles
|
1of6
SUSTAINABILITY
SCIENCE
ENVIRONMENTAL
SCIENCES
factor. Green space can promote mental health by supporting
psychological restoration, encouraging exercise, improving social
coherence, decreasing noise and air pollution affecting cognition
and brain development, and improving immune functioning (19–
22). Given that green space can promote mental health and
quality of life in urban populations, urban planning and policy
will benefit from more information on the likely generality of
effects on psychiatric disorders and how those effects are re-
alized over the life course. Here, we investigate whether green
space presence during childhood is associated with the risk of
developing any of a broad range of psychiatric disorders later in
life, by combining nationwide population data with individual-
level green space presence data. We determine the strength and
shape of the association between green space and a spectrum of
mental health outcomes to clarify whether dose–response rela-
tionships exist and, if this is the case, whether the associations
are linear or asymptotic. We also examine whether risk of psy-
chiatric disorders is more strongly associated with green space
presence at a specific age during childhood.
We use data extracted from multiple Danish population-based
registers together with high-resolution satellite images. Unlike most
previous studies on green space, the current study includes all
members of the national population who met our criteria for in-
clusion. Our study population thus includes all persons born in
Denmark from 1985 to 2003 and living in Denmark on their 10th
birthdayforwhomwehavelongitudinal data on mental health
outcomes, socioeconomic status, and place of residence (n=
943,027). We constructed a dataset of yearly individual-level green
space presence within 210 ×210 m, 330 ×330 m, 570 ×570 m, and
930 ×930 m squares around each person’sresidencebasedonthe
normalized difference vegetation index (NDVI). NDVI was cal-
culated from remotely sensed 30-m-resolution Landsat satellite
images covering the entire country for the years 1985–2013. We
combined these two datasets for two purposes: (i)toexaminethe
specific contribution of green space presence during childhood to
the risk of diverse, adolescent into adult psychiatric disorders, over
and above the contributions of correlated risk factors including
urbanization, socioeconomic factors, parental history of mental
illness, and parental age; and (ii) to explore how green space is
affecting mental health by estimating these associations at different
ages and by determining the shape of the relationship green space
has with each of the psychiatric disorders.
Results
Relative risk, estimated as incidence rate ratios (IRR), was
higher for persons living at the lowest NDVI compared with
those living at the highest levels of NDVI for all psychiatric
disorders, except intellectual disability (IRR: 1.04; 95% CI: 0.95
to 1.14) and schizoaffective disorder (IRR: 1.33; 95% CI: 0.98 to
1.82) (Fig. 1). Adjusting for urbanization, parents’socioeco-
nomic status, family history, parental age, municipal socio-
economic factors, and a combination of all five potential
confounding factors only changed the risk estimates slightly, with
no change to the overall association with NDVI, except for
borderline type, anorexia, and bipolar disorder for which
adjusting for all five factors made green space presence in-
significant (Fig. 1 and SI Appendix, Fig. S1). Population attrib-
utable risk estimates showed that the association between NDVI
and psychiatric disorder risk was, in general, comparable in
magnitude to that of family history and parental age, higher than
urbanization, and slightly lower than parents’socioeconomic
status (SI Appendix, Table S1). Substance abuse disorders, spe-
cific personality disorder, borderline type, and intellectual dis-
ability risk were mostly associated with parents’socioeconomic
status, while mood disorder, single and recurrent depressive
disorder, and neurotic, stress-related, and somatic disorder risk
were mostly associated with NDVI, although the last has an as-
sociation of similar strength as parents’socioeconomic status.
The relative risk of developing any psychiatric disorder was re-
latedtoNDVIinadose–response relationship across urbanization
levels, with risk declining incrementally with higher doses of green
space, although nonmonotonically for the capital center region
(Fig. 2). Mean NDVI was lowest for the capital center area, but the
range of NDVI values was represented across each urbanization
category (SI Appendix,TableS2). The strongest association be-
tween relative risk and the lowest decile of green space presence
was for the capital center region (NDVI decile 1; IRR
1
: 1.60; 95%
CI
1
: 1.42 to 1.80) and the weakest association was for rural areas
(IRR
1
: 1.27; 95% CI
1
: 1.22 to 1.33). Although a Cox regression
modelwithaninteractiontermshowed that the association with
NDVI varied significantly across the different degrees of urbani-
zation (P=0.001, chisq =97.8, df =36), the general pattern of
lower NDVI being associated with higher risk was similar within
each degree of urbanization. Adjusting for urbanization and par-
ents’socioeconomic status only slightly lowered estimates.
We found no consistent sign of green space presence being
associated with any particularly sensitive age across all disorders
(SI Appendix, Fig. S2). Alcohol abuse, specific personality dis-
orders, and borderline type diverged from the general pattern,
with a tendency toward stronger protective associations
0.81.01.21.41.61.8
Relative risk
Any psychiatric disorder
Substance abuse
Alcohol
Cannabis
Schizophrenia and related disorders
Schizophrenia
Schizoaffective disorder
Mood disorders
Bipolar disorder
Recurrent depressive disorder
Single and recurrent depressive disorder
Neurotic, stress−related, and somatic disorders
Obsessive compulsive disorder
Eating disorders
Anorexia nervosa
Specific personality disorders
Borderline type
Intellectual disability
Adjusted urbanization
Adjusted SES
Unadjusted
Adjusted ALL
Fig. 1. The association between childhood green space presence and the
relative risk of developing a psychiatric disorder later in life. Green space
presence was measured as the mean NDVI within a 210 ×210 m square
around place of residence (n=943,027). Low values of NDVI indicate sparse
vegetation, and high values indicate dense vegetation. Relative risk esti-
mates are relative to the reference level (set to the highest decile) for NDVI
fitted as numeric deciles in classes of 10. Estimates above the dashed line
indicate higher risk of developing a given psychiatric disorder for children
living at the lowest compared with the highest values of NDVI. Three ad-
ditional models were fitted to adjust for the effect of urbanization, parental
socioeconomic status (SES), and the combined effect of urbanization, pa-
rental and municipal socioeconomic factors, parental history of mental ill-
ness, and parental age at birth on risk estimates. All estimates were adjusted
for age, year of birth, and gender and plotted with 95% CIs.
2of6
|
www.pnas.org/cgi/doi/10.1073/pnas.1807504116 Engemann et al.
occurring at age 3 y to 4 y based on the nonoverlapping confi-
dence intervals between individual estimates. We compared the
association with cumulated green space presence by fitting
models with deciles of mean NDVI at the 10th birthday and as
cumulated NDVI from birth to the 10th birthday. Mean and
cumulated NDVI were both associated with risk in a dose–
response relationship, with children living at the lowest green
space presence having the highest risk of developing a psychiatric
disorder (SI Appendix, Fig. S3). Risk estimates were generally
higher for cumulated green space presence than presence mea-
sured at age 10 y, again suggesting an accumulating dose–response
relationship. About half of all cases across all disorders were di-
agnosed in adulthood (age >19 y) (SI Appendix,TableS3). Splitting
data between persons diagnosed in adolescence (age 13 y to 19 y)
and adulthood showed a stronger association between risk of any
psychiatric disorder and green space for the former (IRR
1
: 1.64;
95% CI
1
: 1.58 to 1.70) than the latter age category (IRR
1
: 1.43;
95% CI
1
:1.35to1.52),withnochangeinthedirectionoftheas-
sociation (SI Appendix,Fig.S4). Associations mostly did not differ
between the two age groups for individual disorders, with the ex-
ception of substance and cannabis abuse. Both substance and
cannabis abuse showed stronger associations between risk and
green space for persons diagnosed in adolescence; however, can-
nabis abuse diagnosed in adulthood was statistically insignificant
due to low sample size in the first half of the sampling period. We
found no strong difference in the association between risk and
green space measured at 210 ×210 m, 330 ×330 m, 570 ×570 m,
or 930 ×930 m presence zones (SI Appendix,TableS4). For both
the age sensitivity and cumulated green space analysis, adjusting
for urbanization and parents’socioeconomic status only slightly
lowered risk estimates.
Discussion
Our results show that high levels of childhood green space are as-
sociated with lower risk of developing any of a spectrum of ado-
lescent into adult psychiatric disorders. Living at the lowest levels of
green space compared with living at the highest levels of green space
was associated with 15 to 55% higher risk, except for intellectual
disability and schizoaffective disorder. The protective association
remained after adjusting for other known risk factors including ur-
banization, socioeconomic factors, family history of mental illness,
and parental age, indicating an independent association with green
space. Our results are in line with previous reports of positive im-
pacts on mental health from green space (19, 20, 23, 24). Further-
more, the association with NDVI was comparable in magnitude to
or even higher than those of other known risk factors, including
parents’socioeconomic status, history of mental illness, and age.
A number of psychological and physiological mechanisms might
link elements of green space to decreased risk of psychiatric dis-
orders. Each of the hypothetical mechanisms considered below may
be of greater importance for some disorders than for others. Indi-
viduals with urban upbringing have high neural activity linked to
stress processing, which could lead to higher risk of psychiatric
disorders in adults (4). Green space can enhance psychological
restoration, can affect brain structure through positive associations
with amygdala integrity, and could mitigate negative effects from
the socially dense and noisy city environment that heighten stress
(17, 19, 23, 24). The present study offers some evidence bearing
specifically on green space as a support for psychological restora-
tion. Previous studies have shown mental health benefits for
children with better access to green space (25), with, for example,
nearby nature buffering the negative impact of life stress for rural
children (26). Neurotic, stress-related, and somatic disorders, as
Capital center
Capital suburb
Provincial city
Provincial town
Rural area
Green space age 10
1.0 1.2 1.4 1. 6 1.8
NDVI deciles
Relative risk
Capital center
1.0 1.2 1.4 1. 6
NDVI deciles
Relative risk
Capital Suburb
2 4 6 810
1.0 1.1 1.2 1.3 1.4
NDVI deciles
Relative risk
Provincial city
2 4 6 810
1.00 1.10 1.20 1. 30
NDVI deciles
Relative risk
Provincial town
1.01.11.21.31.41.5
NDVI deciles
Relative risk
2 4 6 810 2 4 6 810
2 4 6 810
Rural
Green space age 10 adjusted
Fig. 2. The association between relative risk of developing any psychiatric disorder and childhood green space presence across urbanization levels. Data were
split between each of the five urbanization classes (Capital center n=56 650, Capital suburb n=124,193, Provincial city n=90,648, Provincial town n=
265,570, and Rural n=376,525). NDVI was recalculated as deciles, and separate models, shown in black, were fitted within each urbanization class to de-
termine the shape of the association between green space and mental health. Integer values on the xaxis refer to decile ranges, i.e., 1 corresponds to decile
0 to 10%. An additional model, shown in grey, was fitted for each urbanization class to adjust for urbanization and parents’socioeconomic status. Estimates
of relative risk from all five models were adjusted for year of birth and gender and plotted with 95% CIs within each degree of urbanization.
Engemann et al. PNAS Latest Articles
|
3of6
SUSTAINABILITY
SCIENCE
ENVIRONMENTAL
SCIENCES
well as single and recurrent depressive disorder, had some of the
highest relative risk and population attributable risk estimates
associated with NDVI, which could reflect the role of green spaces
as restorative environments. Strong associations for substance
abuse disorders could indicate the development of better stress-
processing ability with more green space, resulting in less need for
self-medication later in life. Genetically vulnerable individuals
could be particularly at risk from stress-triggered expression of
psychiatric disorders, and gene−environment interactions should be
investigated next, e.g., using polygenic risk scores or genome-wide
association studies.
In our study, the relative risk of developing any psychiatric
disorder was associated with green space in a dose–response
relationship within all urbanization levels, showing that the associ-
ation with green space presence is present after adjusting for
urbanization. We found the highest relative risk for the capital
center region and the lowest for rural areas, consistent with
previous findings (27, 28). Higher pace of life and social stress in
the most urbanized areas could create a stronger need for re-
storative environments such as urban green space. This finding
also suggests that the highly urbanized capital center area could
benefit most from additional green space as an early intervention
tool in healthy city planning and development. In contrast, the
nonmonotonic decrease of risk in the capital center area could
indicate that capital residents residing in high-income neighbor-
hoods receive risk-decreasing benefits from the urban environment,
e.g., better schools or lower crime rates, not captured by our mu-
nicipal socioeconomic adjustment and that these benefits, at some
level, become more important than green space.
We found no consistent sign across all psychiatric disorders of
green space presence being associated with any particularly
sensitive age during childhood. NDVI at age 10 y and cumulated
NDVI were both associated with risk across all psychiatric dis-
orders in a dose–response relationship that could reflect causa-
tion. We might expect neurodevelopmental disorders such as
schizophrenia to be more strongly associated with green space
during the earliest years of life when brain development is most
vulnerable compared with, e.g., substance abuse disorders.
Schizophrenia risk peaked slightly at age 3 y, but the pattern was
not stronger than for other disorders. Interestingly, substance
and alcohol abuse showed slight increases in the protective as-
sociation during both the earliest and latest years, suggesting that
multiple pathways may influence these particular associations.
Comparing associations for persons diagnosed in adolescence
versus in adulthood may indicate that childhood green space is
somewhat more strongly associated with developing a psychiatric
disorder in adolescence. We hypothesize that, during the earliest
years of childhood, pathways related to passive exposure, such as
noise reduction or air pollution removal, may be important,
whereas pathways related to use, such as exercise and social
interactions, may become increasingly important as a child
becomes more independent. Green space around children’s
schools likely becomes increasingly important with age. Further-
more, parents’actions will influence children’s visits to parks and
other green spaces. The use of different high-resolution measures
of green space such as vegetation height and structure should be
investigated further to understand which aspects of g ree n sp ace
affect mental health and through which mechanisms.
Risk estimates were generally higher for a measure of cumu-
lated green space presence, compared with presence measured
at age 10 y, supporting a dose–response relationship over time.
Furthermore, this suggests that the protective association with
green space builds up over time and that green space presence is
important all through childhood. Our results are consistent with
previous work showing improved mental well-being and cogni-
tive development among children exposed to more green space
(16, 25). The relation to cognitive development found by Dad-
vand et al. (16) was partly mediated by air pollution, and this
result is supported by other studies on air pollution and psychi-
atric disorders (29, 30). The role of green spaces as natural filters
of environmental pollution should be investigated further in re-
lation to psychiatric disorders. Another promising hypothesis is
that exposure to green space, especially biodiverse green space,
and animal contact lead to better immune functioning, which has
been linked to mental health (22, 31). Training the human immune
systems seemingly requires prolonged exposure to microbiota dur-
ing childhood (22), consistent with our finding of accumulative
associations with green space presence. For neurodevelopmental
disorders, such as intellectual disability and schizophrenia, poor
immune functioning and consequently greater risk of infections
could explain some of the association with green space.
Although we found a strong association between green space
and psychiatric disorder risk, our study also has some limitations.
At this point, we cannot completely dismiss that choice of resi-
dential location is somewhat affected by genetic confounding. A
recent study using polygenic risk scores showed that people with
higher genetic loading for schizophrenia lived in denser urban
areas (32). However, another study for Denmark showed that the
association between urbanicity and the risk of schizophrenia was
not explained by genetic liability (33). Hence, the association
between green space and mental health is unlikely to be entirely
driven by genetically determined choice of residential location,
but gene−environment interactions could still play a role. Se-
lection bias from parents of higher socioeconomic status choosing
to move to greener areas could also influence the results, al-
though, for schizophrenia, selective migration has been shown to
only partly explain the association in Denmark (34). Although we
adjusted for municipal and parents’socioeconomic status, our
results may be influenced by unmeasured socioeconomic factors
such as lower-quality green space, higher crime rates, and fewer
social advantages in deprived neighborhoods.
These limitations point to several follow-up questions. First, the
deviations from the general dose–response relationship warrant
further studies into the epidemiology of certain psychiatric disorders
such as alcohol abuse, anorexia, and intellectual disability. Second,
despite the strong longitudinal design of our study, our risk estimates
fundamentally only show correlations. Causation is generally hard to
infer from observational studies and is difficult to prove when the
etiology of psychiatric disorders is unknown (35). New knowledge
from genetic and neurobiological research (e.g., refs. 17 and 36)
could guide future combinations of longitudinal and experimental
studies. Third, the mechanistic links between green space and psy-
chiatric disorders remain to be identified. Using NDVI from satellite
images allowed us to estimate individual-level green space presence
at a fine resolution but captured no information about other aspects
of natural environments such as blue space, biodiversity, the pres-
ence of animals, or quality of vegetation. Also, NDVI provides no
information about the use of green space. Effects of different
qualitative aspects and use of green space is important to identify
a possible mechanistic link (37). Restorative environment studies
assessing the positive, restoration-promoting (i.e., salutogenic) char-
acteristics of green space in combination with the absence of risk
factors such as noise and pollution are promising ways to determine
whether and how natural environments provide benefits that pro-
mote mental health (38). Tracking people’s use of green space
through GPS (39) or possibly through social media could provide
more information about pathways related to use of green space.
Our results complement other studies showing positive asso-
ciations between nature and mental health (19–21, 23, 24), while
showing a consistent, protective association with individual-level
childhood green space presence for many of a spectrum of psy-
chiatric disorders after adjusting for urbanization and covering a
large proportion of the population. As shown by the dose–response
relationship between green space presence and psychiatric disorder
risk, we found no sign of the positive association with green space
reaching an upper limit. Hence, finding ways to provide high green
4of6
|
www.pnas.org/cgi/doi/10.1073/pnas.1807504116 Engemann et al.
space exposure during childhood should be encouraged in sus-
tainable urban planning. Population attributable risk estimates
suggest that green space might contribute large health benefits
across the population. Future studies should address the extent to
which benefits arise from passive exposure versus active use, and
whether the source of benefits differs for different psychiatric dis-
orders. Cumulated green space presence was more strongly asso-
ciated with risk, and this should be considered in future studies of
green space effects on mental health.
In conclusion, our data show a consistent association between
higher levels of green space during childhood and a lower risk of
developing any of a multitude of psychiatric disorders later in
life. These findings contribute to our understanding of the urban
environment as an important environmental risk factor for
mental health and can guide the design of healthy city environ-
ments, as well as institutions and programs affecting childhood
life, for example, school systems. Ensuring access to green space
and enhancing opportunities for a diverse range of uses, espe-
cially in densifying urban environments, could be an important
tool for managing and minimizing the global burden of disease
increasingly dominated by psychiatric disorders. Lower access to
green space could be an added risk factor for mental health
among vulnerable groups of society (40, 41). Loss of human−
nature interactions presents a health risk, and it can also reduce
peoples’appreciation of natural environments, creating negative
feedback loops (42). In contrast, positive experiences, such as psy-
chological restoration or social cohesion, can motivate positive
ecological behaviors (43). Increasing urban nature could potentially
provide mental health benefits while simultaneously protecting
biodiversity and ecosystem services of natural environments.
Methods
Study Population and Assessment of Psychiatric Disorders. Denmark is a small,
relatively homogeneous country with a population of 5.8 million people and
a total area of about 43,000 km
2
. Distances within the country are small, with
most people living within 25 km of a city with >30,000 inhabitants and a
psychiatric hospital or department. The Danish Civil Registration System was
established in 1968 and contains a personal identification number (PIN) and
information on gender, place of birth, vital status, parents’PINs, and con-
tinuously updated information on vital status and place of residence for all
Danish citizens. All national registers use the PIN, linking each individual to
all other national registers, e.g., containing information on health, contex-
tual, and socioeconomic information. The study population included all
persons born in Denmark from 1985 to 2003 and who were alive and re-
siding in Denmark at their 10th birthday (943,027 persons).
We linked all individuals from the study population and their parents and
siblings with the Danish Psychiatric Central Research Register (44) to obtain
information about psychiatric disorders. The Register contains information on
all admissions to Danish psychiatric in-patient facilities since April 1, 1969, and,
since 1995, all out-patient visits to psychiatric departments or emergency care.
There are no private psychiatric in-patient facilities in Denmark, and treatment
is free. From 1969 to 1993, the diagnostics system used was the Danish modi-
fication of the International Classification of Diseases (ICD-8) (45) and, from
1994 and on, the ICD-10 (46). Individuals were classified with a psychiatric dis-
order if they had been admitted to a psychiatric facility, received outpatient
care, or visited a psychiatric emergency care unit with a diagnosis of one of 18
psychiatric disorders (SI Appendix,TableS5). The date of onset was defined as
the first contact at which any of the above-defined diagnoses were applied.
Quantification of Green Space. We calculated mean green space covering the
period of 1985 to 2013 from the NDVI, obtained from 30-m-resolution remote
sensing images from the Landsat archive (earthexplorer.usgs.gov/, accessed
February 2, 2016). NDVI is the difference between absorbed (red) and
reflected (near-infrared) light by vegetation following
NDVI =NIR −RED
NIR +RED,
where NIR is the near-infrared and RED is the red band. NDVI is a commonly
used and effective measure of green space (47, 48). Low values indicate
sparse vegetation, and high values indicate dense vegetation.
The Landsat archive contains satellite data of Earth acquired by six satellites
over more than 40 y. Over the years, the purpose and spatial focus of the Landsat
program has changed, and, as a result, the availability of the data varies. The
Landsat satellites provide images of 4 to 11 bands a t 30- to 120-m resolution on a
16- to 18-d revisit cycle. We aimed to obtain images from the growing season in
June, July, or August with none to low cloud cover for the entire country each
year. The best data coverage comes from the later years, whereas data avail-
ability of the earliest years fluctuates. For example, the satellite images from
1978 to 1983 only cover parts of Zealand and the island of Bornholm. Also, due
to technical difficulties, some years in the time period are only partly covered or
not covered at all (see SI Appendix,TableS6for details of each year).
All Landsat images were atmospherically corrected and converted to Top
of Atmosphere (TOA) reflectance using ENVI version 5.1 to remove atmo-
spheric effects from water vapor and the position of the Sun. Despite our best
efforts to find cloud-free images, some images were partly covered by clouds.
For images with severe cloud cover (∼5 to 30%), we downloaded several
images covering the same area at different dates and merged them to ob-
tain a single complete image. Clouds were identified and masked loosely
following a previously published approach (49). Clouds on images from
Landsat 8 (only for year 2013) were identified and masked using the Quality
Assessment band. Manual assessment revealed an acceptable match be-
tween the cloud masks and clouds on the images, although with slight im-
precisions with unidentified thin clouds (hazes) and small patches of bare soil
wrongly identified as clouds. Lastly, all images were processed with histogram
matching using the best (low cloud cover and large land area) image for each
year, and large water bodies were masked out before calculating NDVI. The
NDVI maps for Denmark can be downloaded from https://bios.au.dk/en/about-
bioscience/organisation/ecoinformatics-and-biodiversity/data/.
NDVI images were mosaicked into a single image for each year with bilinear
interpolation. Missing values were interpolated using simple linear interpolation
for cells with a minimum of three measurements across all years. The mean of the
NDVI values was then calculated for quadratic areas (presence zones) of 210 ×
210 m, 330 ×330 m, 570 ×570 m, and 930 ×930 m (7, 11, 19, and 31 cells, re-
spectively) around each address for the years 1985–2013. Cumulated NDVI is the
mean of mean NDVI from birth to the 10th birthday for each cohort member
with at least 10 y of observations. As the Danish residence database is continu-
ously updated, NDVI values for each residence were used for children that moved
within the 10-y timespan. Unless stated otherwise, we refer to estimates from
green space presence measured on the year of the 10th birthday, when children
are relatively independent and likely to be exposed to surrounding green space
through outdoor activities. For each presence zone, place of residence was
located in the center of the quadrat. Calculating the mean of the NDVI for this
large number of addresses was a challenging computational task. Therefore,
we performed these computations using efficient algorithms to process large
amounts of geographic data within a reasonable amount of time (50).
Statistical Analyses. Cohort members were followed for the development of
psychiatric disorders from their 10th birthday until first treatment contact for
any of the psychiatric disorders, death, emigration from Denmark, or De-
cember 31, 2013, whichever came first. IRR of psychiatric disorders were
estimated in time-to-event analyses using Cox regressions, using age as the
underlying timescale with separate baselines for each gender (51). NDVI
values were linked to each cohort member with the addresses. Mean NDVI
at age 10 y was fitted as deciles for all four presence zone sizes, and cu-
mulated NDVI was likewise fitted as deciles (cutoff values are presented in SI
Appendix, Table S7). We adjusted models for potential confounding by
other known risk factors identified by previous studies. Urbanization was
fitted as a categorical variable with five levels: capital center, capital suburb,
provincial city, provincial town, or rural area, as previously described (34).
We controlled for a range of individual and socioeconomic confounding
factors, including year of birth, gender, parents’education (three levels:
primary school, high school\vocational training, or higher education), parents’
income (gross income divided in quintiles and adjusted for inflation and gender
differences), parents’employment status (three levels: employed, unemployed,
or outside workforce), parents’age (seven levels: 12 y to 19 y, 20 y to 24 y, 25 y
to 29 y, 30 y to 34 y, 35 y to 39 y, 40 y to 44 y, and 45 y or older), and parents’
previous diagnosis with any psychiatric disorder up to and including the child’s
10th birthday (two levels: yes or no) (52). We also controlled for differences in
residential areas’socioeconomic status at the smallest available administrative
unit using the Danish municipalities. Many important political decisions are
made at the municipal level, e.g., budgets for schools, parks, and early pre-
vention action programs. Danish municipalities are also responsible for imple-
mentation of action programs and the practical management of schools and
parks. We calculated a measure of socioeconomic deprivation for each mu-
nicipality as the average income as defined above, the proportion of individuals
Engemann et al. PNAS Latest Articles
|
5of6
SUSTAINABILITY
SCIENCE
ENVIRONMENTAL
SCIENCES
with a low education level (primary school compared with high school or
higher education), and the proportion of individuals outside the workforce
(unemployed or outside workforce compared with employed). Estimates ad-
justed for each possible confounding factor separately are presented for easy
comparison of each factor’s association and to avoid potential bias from
overadjustment (SI Appendix,Fig.S1). The calendar year of the end of study for
each cohort member was treated as a time-dependent variable to account for
different hazard rates over time and categorized as 1995–2000, 2001–2005,
2006–2010, and 2011–2013. All other variables were treated as independent
of time.
We performed the following sensitivity analyses to determine the best
models for evaluating the association with NDVI by fitting additional Cox
regressions: assessing (i) the association with different green space presence
zone sizes (SI Appendix, Table S4), (ii) the potential modifying association of
gender (SI Appendix, Table S8), (iii) the association with NDVI measured at
different ages from birth to the 10th birthday (SI Appendix,Fig.S2), (iv)the
potential modifying effect of urbanization (Figs. 1 and 2), and (v) the associa-
tion with green space measured as mean cumulated NDVI from birth to the
10th birthday for each cohort member with observations for least 10 y out of
11 y (SI Appendix,Fig.S3). We calculated the population attributable risk from
IRR to compare the magnitude of the associations with NDVI to other known
risk factors. In the main text, unless stated otherwise, NDVI was measured
within quadratic presence zones of 210 ×210 m at place of residence at the
10th birthday, as we found no strong difference in measuring green space of
different presence zone sizes or when measured at other specific ages.
All data processing and statistics were performed in R (53) using packages
data.table, Hmisc, landsat, lubridate, plyr, raster, rgdal, RStoolbox, sp, and
survival. The only exception was the TOA conversion, performed using the
ENVI software, version 5.1 (Exelis Visual Information Solutions). Data on the
study population are not publicly available due to privacy protection. To
request access to the data, contact P.B.M. at pbm@econ.au.dk.
ACKNOWLEDGMENTS. We thank S. Nielsen and A. Pearcy for proofreading
and comments on a previous draft of the manuscript. We thank M. Thygesen
and A. Timmermann for data administrative help. This study was funded by the
Stanley Medical Research Institute and the Centre for Integrated Register-based
Research at Aarhus University. J.-C.S. considers this work a contribution to his
VILLUM Investigator project (VILLUM FONDEN Grant 16549).
1. United Nations, Department of Economic and Social Affairs, Population Division
(2014) World urbanization prospects: The 2014 revision, highlights (ST/ESA/SER.A/
352). Available at https://esa.un.org/unpd/wup/Publications/Files/WUP2014-High-
lights.pdf. Accessed February 15, 2019.
2. Dye C (2008) Health and urban living. Science 319:766–769.
3. Lambert KG, Nelson RJ, Jovanovic T, Cerdá M (2015) Brains in the city: Neurobiological
effects of urbanization. Neurosci Biobehav Rev 58:107–122.
4. Lederbogen F, et al. (2011) City living and urban upbringing affect neural social stress
processing in humans. Nature 474:498–501.
5. Peen J, Schoevers RA, Beekman AT, Dekker J (2010) The current status of urban-rural
differences in psychiatric disorders. Acta Psychiatr Scand 121:84–93.
6. Vassos E, Agerbo E, Mors O, Pedersen CB (2016) Urban-rural differences in incidence
rates of psychiatric disorders in Denmark. Br J Psychiatry 208:435–440.
7. March D, et al. (2008) Psychosis and place. Epidemiol Rev 30:84–100.
8. Mortensen PB, et al. (1999) Effects of family history and place and season of birth on
the risk of schizophrenia. N Engl J Med 340:603–608.
9. Vassos E, Pedersen CB, Murray RM, Collier DA, Lewis CM (2012) Meta-analysis of the
association of urbanicity with schizophrenia. Schizophr Bull 38:1118–1123.
10. Mortensen PB (2000) Urban–rural differences in the risk for schizophrenia. Int J Ment
Health 29:101–110.
11. Bratman GN, Hamilton JP, Hahn KS, Daily GC, Gross JJ (2015) Nature experience re-
duces rumination and subgenual prefrontal cortex activation. Proc Natl Acad Sci USA
112:8567–8572.
12. Adli M, et al. ( 2017) Neurourbanism: Towards a new discipline. Lancet Psychiatry 4:183–185.
13. Shanahan DF, et al. (2016) Health benefits from nature experiences depend on dose.
Sci Rep 6:28551.
14. Maas J, et al. (2009) Morbidity is related to a green living environment. J Epidemiol
Community Health 63:967–973.
15. Engemann K, et al. (2018) Childhood exposure to green space—A novel risk-
decreasing mechanism for schizophrenia? Schizophr Res 199:142–148.
16. Dadvand P, et al. (2015) Green spaces and cognitive development in primary
schoolchildren. Proc Natl Acad Sci USA 112:7937–7942.
17. Kühn S, et al. (2017) In search of features that constitute an “enriched environment”in
humans: Associations between geographical properties and brain structure. Sci Rep 7:11920.
18. Alcock I, White MP, Wheeler BW, Fleming LE, Depledge MH (2014) Longitudinal ef-
fects on mental health of moving to greener and less green urban areas. Environ Sci
Technol 48:1247–1255.
19. Hartig T, Mitchell R, de Vries S, Frumkin H (2014) Nature and health. Annu Rev Public
Health 35:207–228.
20. James P, Banay RF, Hart JE, Laden F (2015) A review of the health benefits of
greenness. Curr Epidemiol Rep 2:131–142.
21. Twohig-Bennett C, Jones A (2018) The health benefits of the great outdoors: A sys-
tematic review and meta-analysis of greenspace exposure and health outcomes.
Environ Res 166:628–637.
22. Rook GA (2013) Regulation of the immune system by biodiversity from the natural envi-
ronment: An ecosystem service essential to health. Proc Natl Acad Sci USA 110:18360–18367.
23. Fong KC, Hart JE, James P (2018) A review of epidemiologic studies on greenness and
health: Updated literature through 2017. Curr Environ Health Rep 5:77–87.
24. van den Berg M, et al. (2015) Health benefits of green spaces in the living environment: A
systematic review of epidemiological studies. Urban For Urban Green 14:806–816.
25. McCormick R (2017) Does access to green space impact the mental well-being of
children: A systematic review. J Pediatr Nurs 37:3–7.
26. Wells NM, Evans GW (2003) Nearby nature: A buffer of life stress among rural chil-
dren. Environ Behav 35:311–330.
27. Sarkar C, Webster C, Gallacher J (2018) Residential greenness and prevalence of major
depressive disorders: A cross-sectional, observational, associational study of 94 879
adult UK biobank participants. Lancet Planet Health 2:e162–e173.
28. Mitchell R, Popham F (2007) Greenspace, urbanity and health: Relationships in Eng-
land. J Epidemiol Community Health 61:681–683.
29. Attademo L, Bernardini F, Garinella R, Compton MT (2017) Environmental pollution and
risk of psychotic disorders: A review of the science to date. Schizophr Res 181:55–59.
30. Oudin A, Bråbäck L, Åström DO, Strömgren M, Forsberg B (2016) Association between
neighbourhood air pollution concentrations and dispensed medication for psychiatric
disorders in a large longitudinal cohort of Swedish children and adolescents. BMJ
Open 6:e010004.
31. Böbel TS, et al. (2018) Less immune activation following social stress in rural vs. urban
participants raised with regular or no animal contact, respectively. Proc Natl Acad Sci
USA 115:5259–5264.
32. Colodro-Conde L, et al. (2018) Association between population density and genetic
risk for schizophrenia. JAMA Psychiatry 75:901–910.
33. Paksarian D, et al. (2018) The role of genetic liability in the association of urbanicity at
birth and during upbringing with schizophrenia in Denmark. Psychol Med 48:305–314.
34. Pedersen CB (2015) Persons with schizophrenia migrate towards urban areas due to
the development of their disorder or its prodromata. Schizophr Res 168:204–208.
35. Miller G (2012) Mysteries of the brain. Why is mental illness so hard to treat? Science
338:32–33.
36. Sekar A, et al.; Schizophrenia Working Group of the Psychiatric Genomics Consortium
(2016) Schizophrenia risk from complex variation of complement component 4.
Nature 530:177–183.
37. Markevych I, et al. (2017) Exploring pathways linking greenspace to health: Theo-
retical and methodological guidance. Environ Res 158:301–317.
38. von Lindern E, Lymeus F, Hartig T (2017) The restorative environment: a comple-
mentary concept for salutogenesis studies. The Handbook of Salutogenesis, eds
Mittelmark MB, et al. (Springer, New York), pp 181–195.
39. Dewulf B, et al. (2016) Associations between time spent in green areas and physical
activity among late middle-aged adults. Geospat Health 11:411.
40. Cabrera LY, Tesluk J, Chakraborti M, Matthews R, Illes J (2016) Brain matters: From
environmental ethics to environmental neuroethics. Environ Health 15:20.
41. Wolch JR, Byrne J, Newell JP (2014) Urban green space, public health, and environmental
justice: The challenge of making cities ‘just green enough.’Landsc Urban Plan 125:234–244.
42. Soga M, Gaston KJ (2016) Extinction of experience: The loss of human-nature inter-
actions. Front Ecol Environ 14:94–101.
43. Hartig T, Kaiser FG, Strumse E (2007) Psychological restoration in nature as a source of
motivation for ecological behaviour. Environ Conserv 34:291–299.
44. Mors O, Perto GP, Mortensen PB (2011) The Danish psychiatric central research reg-
ister. Scand J Public Health 39(7 Suppl):54–57.
45. World Health Organization (1967) International Statistical Classification of Diseases
and Related Health Problems (ICD-8) (World Health Org, Geneva).
46. World Health Organization (1992) The I CD-10 Classification of Mental and Beh avioural
Disorders: Clinical Descriptions and Diagnostics Guidelines (World Health Org, Geneva).
47. 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.
48. Lo C (1997) Integration of landsat thematic mapper and census data for quality of life
assessment. Remote Sens Environ 62:143–157.
49. Martinuzzi S, Gould W, González O (2007) Creating cloud-free Landsat ETM+data
sets in tropical landscapes: Cloud and cloud-shadow removal (US Dep Agric, Portland,
OR), General Technical Report IITF-GTR-32.
50. Arge L, Haverkort H, Tsirogiannis C (2012) Fast generation of multiple resolution
instances of raster data sets. Proceedings of the 20th ACM SIGSPATIAL International
Conference on Advances in Geographic Information Systems (ACM GIS) (ACM Press,
New York), pp 52–60.
51. Andersen PK, Gill RD (1982) Cox ’s regression model for counting processes: A large
sample study. Ann Stat 10:1100–1120.
52. Danmarks Statistik (1991) IDA–En Integreret Database for Arbejdsmarkedsforskning:
Hovedrapport (Denmark Stat, Copenhagen).
53. R Development Core Team (2013) R: A language and environment for statistical com-
puting (R Foundation for Statistical Computing, Vienna), Version 3.5.1.
6of6
|
www.pnas.org/cgi/doi/10.1073/pnas.1807504116 Engemann et al.