Neighbourhood Environment and Stroke: A Follow-Up Study in Sweden
Citation: Hamano T, Kawakami N, Li X, Sundquist K (
Neighbourhood Environment and Stroke: A Follow-Up Study in Sweden
Tsuyoshi Hamano 0
Naomi Kawakami 0
Xinjun Li 0
Kristina Sundquist 0
Qing Song, Morehouse School of Medicine, United States of America
0 1 Center for Community-based Health Research and Education, Organization for the Promotion of Project Research, Shimane University , Izumo , Japan , 2 Department of Environmental and Preventive Medicine, Shimane University School of Medicine, Izumo, Japan, 3 Waseda Institute of Sport Sciences, Waseda University , Tokorozawa , Japan , 4 Center for Primary Health Care Research, Lund University , Malmo , Sweden, 5 SPRC , Stanford School of Medicine, Stanford University , Stanford, California , United States of America
Background: In recent years, research on the association between physical environments and cardiovascular disease outcomes has gained momentum with growing attention being paid to Geographic Information Systems (GIS). This nationwide study is the first to examine the effect of neighbourhood physical environments on individual-level stroke, using GIS-based measures of neighbourhood availability of potentially health-damaging (fast food restaurants and pubs/bars) and health-promoting (physical activity and healthcare) resources. Methods: The study population comprised a nationwide sample of 2,115,974 men and 2,193,700 women aged 35-80 years who were followed between 1 December 2005 and 31 December 2007 in Sweden. Totally 42,270 first-ever strokes (both morbidity and mortality) were identified. Multilevel logistic regression models were used to estimate the association between neighbourhood availability of four different resources (fast food restaurants, pubs/bars, physical activity and healthcare) and individual-level stroke. Principal Findings: There were significant associations between neighbourhood availability of the four types of neighbourhood resources and individual-level stroke. The significant odds ratios varied between 1.06 and 1.12 for men and 1.07 and 1.24 for women. After adjustment for age, income, and neighbourhood-level deprivation, the increased odds remained statistically significant for neighbourhood availability of fast food restaurants in both men and women. Conclusions: Specific neighbourhood availability of resources were associated with individual-level stroke but most of these associations were explained by individual-level sociodemographic factors and neighbourhood-level deprivation.
During the last decade, there have been great efforts to identify
the effect of neighbourhood environments on cardiovascular
disease outcomes . Most of previous studies have mainly
focused on two broad types of neighbourhood environments:
physical environments (e.g., air pollution, traffic noise) and social
environments (e.g., socioeconomic deprivation, social cohesion)
. More recently, research on the association between physical
environments and cardiovascular disease outcomes has gained
momentum with growing attention being paid to Geographic
Information Systems (GIS) [2,7]. GIS allow us to combine a
variety of data, and calculate availability of resources in order to
characterize the physical environments in each neighbourhood
Although previous research has analysed the possible
associations between physical environments, assessed by GIS, and risk of
coronary heart disease using multilevel analysis , very little is
known about the association between physical environments and
stroke. To date, there are limitations to existing work on
GISbased measures of physical environments and stroke. Specifically,
previous research concluded that further work is needed to
examine the independent effect of number of fast food restaurants
on stroke after adjustment for individual-level risk factors in a
multilevel fashion . Only one study from the United States
showed a significant association between the number of fast food
restaurants and neighbourhood rates of stroke .Therefore, this
nationwide study was designed to examine the effect of
neighbourhood physical environments on individual-level stroke
risk, using GIS-based measures of neighbourhood availability of
potentially health-damaging (fast food restaurants and pubs/bars)
and health-promoting (physical activity and healthcare) resources.
The first aim of this nationwide multilevel study was to examine
whether GIS-based measures of neighbourhood availability of
health-damaging resources (fast-food restaurants, bars/pubs) or
health-promoting resources (physical activity and healthcare
facilities) was associated with individual-level stroke. The second
aim was to test if these possible associations remained significant
after adjustment for neighbourhood-level deprivation and
individual-level sociodemographic characteristics.
The study was approved by the Ethics Committee of Lund
University. Data handling and analysis were performed by
assigning serial numbers to data to preserve anonymity.
The study population comprised a nationwide sample of
2,115,974 men and 2,193,700 women aged 3580 years who
were followed between 1 December 2005 and 31 December 2007
in Sweden. All individuals were included in a national Swedish
research dataset, constructed at the Center for Primary Health
Care Research at Lund University. This dataset contained
nationwide individual-level medical diagnoses from the Swedish
Hospital Discharge Register (obtained from the National Board of
Health and Welfare) and the Cause of Death Register. These data
were linked to the Swedish Population Register (census) data
obtained from Statistics Sweden, the Swedish government-owned
statistics bureau. The Population Register includes individual-level
data on sociodemographic factors such as age and income.
The men and women were followed between 1 December 2005
(the start of follow-up) and 31 December 2007 for the outcome
variable, i.e., first hospitalisation during the study period for stroke
(both morbidity and mortality). The disease codes were based on
the 10th version of the International Classification of Diseases
(ICD-10) , and I60 to I69 were used to classify the outcome. In
this study, men and women with pre-existing stroke were
excluded. Totally 42,270 stroke cases (23,782 men and 18,488
women) occurred during the follow-up period.
Health-damaging and health-promoting resources
Four categories of neighbourhood resources that could be
regarded as potentially health-damaging or health-promoting were
selected . The four categories were fast food restaurants (e.g.,
pizzerias and hamburger joints), bars/pubs, physical activity
facilities (e.g., swimming pools, gyms, ski facilities), and healthcare
facilities (e.g., healthcare centres, public hospitals, dentists,
pharmacies). Neighbourhood availability of the four categories
was measured as counts per predefined administrative areas (Small
Area Market Statistics, SAMS) by the use of GIS. We employed
ArcGIS/ArcInfo 9.2 software from ESRI, which offers various
ready-to-use spatial-analysis tools .
The ready to-use nationwide GIS dataset of business contacts
(i.e., health-damaging and health-promoting resources) for
November 2005 was provided to us by the Swedish company
Teleadress . Teleadress was created when the former
government-owned company Telecom was divided into several
subcompanies. It is a leading aggregator, processor and provider of
Swedish contact information, delivering all available business
telephone numbers, addresses and geographical coordinates in
Sweden. The data included all business information in the Swedish
Telephone Book (i.e., the Yellow Pages), in accordance with
previous studies [13,14].
SAMS units are predefined areas and were used as proxies for
neighbourhoods, as has been done previously [5,1518]. Each
SAMS unit has an average of about 1,000 residents. This study
examined only those SAMS units that overlap with localities or
urban areas. In Sweden, localities (which are defined by Statistics
Sweden every fifth year) represent any village, town or city with a
minimum of 200 residents and adjacent areas where the houses
are no more than 200 m apart . We chose to include only
SAMS units overlapping with localities because more rural SAMS
units have very few goods, services and resources. In 2005, 1940
Swedish localities were identified by Statistics Sweden. GIS were
used to overlay the SAMS boundaries with the locality boundaries.
Of a total of 9,617 SAMS units in Sweden, 7,945 overlapped with
localities and were therefore selected. Together they accounted for
84% of the Swedish population. SAMS units with fewer than 50
people were excluded on the basis that they might yield unreliable
statistical estimates in the calculation of the neighbourhood
deprivation index. A final total of 7,033 SAMS units was included
in the present study.
For gender, analyses for men and women were conducted
separately. Age was categorised as 3544, 4554, 5564, 6574
and 7580 years. Family income was categorised as empirical
quartiles based on the distribution. The family income variable
took the number of people in the family into account as well as the
ages of the family members (children were given lower
consumption weights than adults).
Neighbourhood-level deprivation was also included as a
covariate because previous research has shown that
neighbourhood deprivation is an important environmental disease
determinant . The neighbourhood deprivation index was constructed
using 2005 census data provided by Statistics Sweden. A summary
index was used to determine neighbourhood-level deprivation
using the following four deprivation indicators for individuals aged
2564 years (the working population): low educational status (,10
years of formal education); low income (income from all sources,
including that from interest and dividends, ,50% of the median
individual income); unemployment (not employed, excluding
fulltime students, those completing compulsory military service and
early retirees); and social welfare . Each of the four variables
loaded on the first principal component with similar loadings
(+0.47 to +0.53) and explained 52% of the variation between these
variables. A z score was calculated for each SAMS
neighbourhood. The z scores, weighted by the coefficient for the
eigenvectors, were then summed to create the index . The
index was categorized into three groups: below one standard
deviation (SD) from the mean (low deprivation), above one SD
from the mean (high deprivation), and within one SD of the mean
(moderate deprivation). Higher scores reflect more deprived
Age-standardised incidence proportions (proportions of subjects
who became cases among those who entered the study time
interval) were calculated separately for men and women by direct
age standardisation using 10-year age groups, with the entire
Swedish population of men or women aged 3580 as the standard
Multilevel logistic regression models were created with
incidence proportions as the outcome variables. These models are a
good approximation of multilevel Cox proportional hazards
models under conditions such as ours: a large sample size, low
incidence rates, risk ratios of moderate size and relatively short
follow-up . The first model included neighbourhood
availability of each of the four categories of neighbourhood resources in
order to determine the crude odds ratio (OR) of stroke with 95%
confidence interval (CI). The second model also included
neighbourhood-level deprivation. The third model included
neighbourhood availability for each resource, neighbourhood-level
deprivation, and individual-level age and income. The analyses
were performed using MLwiN . A p-value of ,0.05 was
considered statistically significant.
Table 1 shows the number of subjects in the study population
with availability/no availability of the four types of resources,
number of stroke events and age-standardised incidence
proportions (%) by neighbourhood deprivation and neighbourhood
availability of the potentially health-damaging and
healthpromoting resources. Nearly half of the study population had at
least one fast food restaurant in their neighbourhoods. Most
people lived in neighbourhoods with no bars/pubs. Around 40%
of subjects lived in neighbourhoods with availability to physical
activity facilities or healthcare facilities. The age-standardised
incidence of stroke increased with increasing neighbourhood
deprivation. For the total study population, the incidence of stroke
for men was 0.9% in low-deprivation neighbourhoods and 1.2%
and 1.3%, respectively, in moderate- and high-deprivation
neighbourhoods. The corresponding incidence proportions for
women were 0.6%, 0.9% and 1.0%, respectively.
Results of the associations between the four categories of
neighbourhood availability and stroke are presented in Table 2.
For men, there were statistically significant higher odds of stroke
for those living in neighbourhoods with availability of fast food
restaurants, bars/pubs, physical activity facilities and healthcare
facilities in the unadjusted model (Model 1). The significant ORs
varied between 1.06 and 1.12. After adjusting for
neighbourhoodlevel deprivation (Model 2), neighbourhood availability of bars/
pubs was no longer significantly associated with stroke. After
further adjustment for individual-level age and income (Model 3),
only neighbourhood availability of fast food restaurants continued
to be significantly associated with higher odds of stroke.
For women, a similar pattern was observed, with significantly
higher ORs of stroke for those living in neighbourhoods with
availability of fast food restaurants, bars/pubs, physical activity
facilities and healthcare facilities (Model 1). The significant ORs
varied between 1.07 and 1.24. These ORs remained significant
after adjustment for neighbourhood deprivation (Model 2). After
further adjustment for individual-level age and income (Model 3),
only neighbourhood availability of fast food restaurants and health
care facilities continued to be significantly associated with higher
odds of stroke. For both men and women, the ORs that remained
significant in the full model were only slightly increased and varied
between 1.02 and 1.03.
There were no interactions between neighbourhood-level
availability of the four resources and neighbourhood-level
deprivation (data not shown).
To the best of our knowledge, no study has examined the effect
of neighbourhood availability of potentially health-damaging and
health-promoting resources on individual-level stroke risk in a
multilevel fashion. The main finding of this study is that specific
potentially health-damaging as well as health-promoting
neighbourhood resources imply higher odds of stroke. Although the
ORs of the associations are not large, our study included more
than four million people aged between 35 and 80 years.
Our results are consistent with recent work from the United
States, where an association between the number of fast food
restaurants in the neighbourhood and stroke was found . This
association might be explained through high salt and caloric intake
from fast food consumption leading to hypertension and obesity
[9,23,24], which in turn may increase the risk of stroke. Our data
did not, however, allow us to establish whether individuals with
stroke consumed more fast food than those without stroke. This is
a main limitation of the present study. Future work is needed to
investigate dietary habits as potentially mediating pathways
between availability of neighbourhood fast food restaurants and
The reason why high neighbourhood availability to healthcare
facilities was associated with stroke is, however, uncertain. It might
be explained by lack of timeliness . This lack of timeliness can
result in emotional distress, so that residents who live far from
healthcare facilities may avoid seeking a medical treatment. As a
result, the number of stroke patients in such neighbourhoods
might be underestimated. To date, few studies have found barrier
effects of spatial access to health care, with greater distance
resulting in less health care utilization . Although more
research is needed to examine reasons for why such an association
was shown in women, these results indicate that health policies
might be focused on providing more equitable health care
These results extend previous findings on the association
between GIS-based measures of physical environments and stroke.
A study that examined the number of fast food outlets and stroke
from the United States used census tracts as a proxy for
neighbourhoods . In the current study, however, SAMS units
that has an average of about 1,000 residents was used to define a
neighbourhood. SAMS units are much smaller than, for example,
census tracts. This enabled us to assess availability in each
individuals immediate neighbourhood, which increases the
probability that the individuals were actually exposed to the
potentially health-damaging and promoting resources in their
daily lives. To date, there is little evidence on which
neighbourhood definition is most appropriate in order to consider the effect
of GIS-based measures of the physical environment on stroke. In
addition, our study is the first of its kind, and our findings must be
confirmed in other settings to provide more robust evidence in the
formulation of efficient neighbourhood health policies.
The present study has several strengths. To our knowledge, this
is the first large-scale study of an entire national population to
examine the effect of potentially health-damaging and
healthpromoting neighbourhood resources on individual-level stroke in a
multi-level fashion. The large-scale design allowed us to include
42,270 cases of stroke. Moreover, our study included adjustment
for individual-level and neighbourhood-level covariates in the
multilevel framework. This study also has certain limitations. First,
our data did not allow for the assessment of other important risk
factors for stroke, such as smoking, poor dietary habits, and
physical inactivity. Second, we had no data on whether those
individuals with stroke actually utilized the examined resources in
their neighbourhoods. Third, the follow-up period was only two
years. However, this means that the potential changes during the
study period in neighbourhood availability of resources were most
likely minor and that most people probably remained in their
Our results suggest that specific neighbourhood-level resources
are associated with individual-level stroke but that some
associations seem to be explained by individual-level sociodemographic
factors and neighbourhood-level deprivation. Caution is, however,
warranted in the interpretation of these findings. Future studies
should consider the actual use of neighbourhood resources in
8 3 8 4
t n 8 3 5 5 4 0 3 4 2
ms e ,4 1 7 84 ,6 95 ,5 36 ,1
u f v 8 ,4 ,0 , 5 , 0 , 0
) N o e 1 9 9 2 1 7 1 8 1
2 6 6 0 5 7 2
t n 8 3 4 2 8 5 2 0 0
ms e ,7 ,2 ,5 20 ,5 ,0 ,7 78 ,0
u f v 3 1 2 , 0 0 3 , 4
N o e 2 1 1 3 2 1 1 9 1
) ) ) )
) .5 ) .3 ) .0 ) .4
.5 4 .7 7 .0 9 .6 1
5 (5 2 (8 1 (5 8 (6
(4 1 (1 2 (4 3 (3 1
3 9 2 2 1 2 3 3
8 ,4 5 ,4 5 ,5 8 ,1
,426 ,351 ,586 ,748 ,476 ,842 ,461 ,992 100
9 1 2 1 8 1 8 1 .t
rsce ts iliite s .e0n
u n c e o
rseo rtaau ility ility iftya liity iliitca liity .lranp
roohuod frseood iliiltyaab liaaavb subp ililityaab ilaaavb iltccaav iliiltyaab liaaavb frcaeh ililityaab ilaaavb j/173ou
hb ts vA oN r/s vA oN isy vA oN ltae vA oN .10
ieg aF aB Ph H :i1
Table 2. Associations between stroke and neighbourhood availability of potentially health-promoting and health-damaging
Men (n = 2,115,974)
Women (n = 2,193,700)
21.05 1.18 1.14
21.17 1.03 1.00
21.05 1.16 1.10
21.17 1.00 0.96
21.05 1.07 1.03
21.09 1.02 0.99
21.05 1.24 1.19
21.22 1.03 1.00
OR odds ratio, CI confidence interval.
aModel 1 is unadjusted.
bModel 2 is adjusted for neighbourhood-level deprivation.
cModel 3 is adjusted for neighbourhood-level deprivation and individual-level age and income.
individuals with stroke and compare their use with healthy
Conceived and designed the experiments: TH NK KS. Performed the
experiments: NK KS. Analyzed the data: XL. Contributed reagents/
materials/analysis tools: KS. Wrote the paper: TH NK KS.
1. Kawachi I , Berkman LF , editors ( 2003 ) Neighborhoods and Health . New York, NY : Oxford University Press.
2. Diez Roux AV , Mair C ( 2010 ) Neighborhoods and health . Ann N Y Acad Sci 1186 : 125 - 145 .
3. Miller KA , Siscovick DS , Sheppard L , Shepherd K , Sullivan JH , et al. ( 2007 ) Long-term exposure to air pollution and incidence of cardiovascular events in women . N Engl J Med 356 : 447 - 458 .
4. Babisch W , Ising H , Gallacher JE , Sweetnam PM , Elwood PC ( 1999 ) Traffic noise and cardiovascular risk: the Caerphilly and Speedwell studies, third phase: 10-year follow up . Arch Environ Health 54 : 210 - 216 .
5. Sundquist J , Johansson SE , Yang M , Sundquist K ( 2006 ) Low linking social capital as a predictor of coronary heart disease in Sweden: a cohort study of 2.8 million people . Soc Sci Med 62 : 954 - 963 .
6. Clark CJ , Guo H , Lunos S , Aggarwal NT , Beck T , et al. ( 2011 ) Neighborhood cohesion is associated with reduced risk of stroke mortality . Stroke 42 : 1212 - 1217 .
7. Kawakami N , Li X , Sundquist K ( 2011 ) Health-promoting and health-damaging neighbourhood resources and coronary heart disease: a follow-up study of 2 165 000 people . J Epidemiol Community Health 65 : 866 - 872 .
8. Rushton G ( 2003 ) Public health, GIS, and spatial analytic tools . Annu Rev Public Health 24 : 43 - 56 .
9. Morgenstern LB , Escobar JD , Sanchez BN , Hughes R , Zuniga BG , et al. ( 2009 ) Fast food and neighborhood stroke risk . Ann Neurol 66 : 165 - 170 .
10. World Health Organization ( 1992 ) International statistical classification of disease and related health problems, 10th Revision . Geneva, Switzerland: World Health Organization.
11. ESRI ( 2006 ) ArcGIS/ArcInfo 9.2 Software . Redlands, CA: Environmental Systems Research Institute , Inc.
12. Teleadress . Available: http://www.teleadress.se/htdocs/omoss/english.htm. Accessed 2009 Aug 8.
13. Pollack CE , Cubbin C , Ahn D , Winkleby M ( 2005 ) Neighbourhood deprivation and alcohol consumption: does the availability of alcohol play a role? Int J Epidemiol 34 : 772 - 780 .
14. Chuang YC , Cubbin C , Ahn D , Winkleby MA ( 2005 ) Effects of neighbourhood socioeconomic status and convenience store concentration on individual level smoking . J Epidemiol Community Health 59 : 568 - 573 .
15. Sundquist K , Malmstrom M , Johansson SE ( 2004 ) Neighbourhood deprivation and incidence of coronary heart disease: a multilevel study of 2.6 million women and men in Sweden . J Epidemiol Community Health 58 : 71 - 77 .
16. Sundquist K , Winkleby M , Ahlen H , Johansson SE ( 2004 ) Neighborhood socioeconomic environment and incidence of coronary heart disease: a follow-up study of 25,319 women and men in Sweden . Am J Epidemiol 159 : 655 - 62 .
17. Sundquist K , Theobald H , Yang M , Li X , Johansson SE , et al. ( 2006 ) Neighborhood violent crime and unemployment increase the risk of coronary heart disease: a multilevel study in an urban setting . Soc Sci Med 62 : 2061 - 2071 .
18. Winkleby M , Sundquist K , Cubbin C ( 2007 ) Inequities in CHD incidence and case fatality by neighborhood deprivation . Am J Prev Med 32 : 97 - 106 .
19. Statistics Sweden. Localities 2005 : populations and buildings . Available: http:// www.scb. se/Pages/PressRelease____225977.aspx. Accessed 2009 Aug 8.
20. Gilthorpe MS ( 1995 ) The importance of normalisation in the construction of deprivation indices . J Epidemiol Community Health 49 (suppl2): S45 - S50 .
21. Callas PW , Pastides H , Hosmer DW ( 1998 ) Empirical comparisons of proportional hazards, poisson, and logistic regression modeling of occupational cohort data . Am J Ind Med 33 : 33 - 47 .
22. Rasbash J , Browne W , Goldstein H , Yang M , Plewis I , et al. ( 2000 ) A User's Guide to MLwiN . Version 2. 1 edn. London: Multilevel Models Project, Institute of Education, University of London.
23. Schmidt M , Affenito SG , Striegel-Moore R , Khoury PR , Barton B , et al. ( 2005 ) Fast-food intake and diet quality in black and white girls: The National Heart, Lung, and Blood Institute Growth and Health study . Arch Pediatr Adolesc Med 159 : 626 - 631 .
24. Rosenheck R ( 2008 ) Fast food consumption and increased caloric intake: A systematic review of a trajectory towards weight gain and obesity risk . Obes Rev 9 : 535 - 547 .
25. National Healthcare Disparities Reports . Rockville: Agency for Healthcare Research and Quality web site . Available: http://www.ahrq.gov/qual/qrdr09. htm. Accessed 2012 Feb 2.
26. Hiscock R , Pearce J , Blakely T , Witten K ( 2008 ) Is neighborhood access to health care provision associated with individual-level utilization and satisfaction? Health Serv Res . 43 : 2183 - 2200 .