Content uploaded by Gergo Baranyi
Author content
All content in this area was uploaded by Gergo Baranyi on Mar 12, 2020
Content may be subject to copyright.
American Journal of Epidemiology
© The Author(s) 2020. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of
Public Health. This is an Open Access article distributed under the terms of the Creative Commons Attribution
Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0), which permits non-commercial re-use,
distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use,
please contact journalpermissions@oup.com.
Vol . 00 , N o. 00
DOI: 10.1093/aje/kwz209
Advance Access publication:
Systematic Reviews and Meta- and Pooled Analyses
The Longitudinal Associations of Perceived Neighborhood Disorder and Lack
of Social Cohesion With Depression Among Adults Aged 50 Years or Older: An
Individual-Participant-Data Meta-Analysis From 16 High-Income Countries
Gerg ˝
o Baranyi∗, Stefan Sieber, Stéphane Cullati, Jamie R. Pearce, Chris J. L. Dibben, and
Delphine S. Courvoisier
∗Correspondence to Dr. Gerg˝
o Baranyi, Center for Research on Environment, Society and Health, School of GeoSciences,
University of Edinburgh, Drummond Street, Edinburgh EH8 9XP, United Kingdom (e-mail: gergo.baranyi@ed.ac.uk).
Initially submitted November 2,2018; accepted for publication June 18,2019.
Although residential environment might be an important predictor of depression among older adults, systematic
reviews point to a lack of longitudinal investigations, and the generalizability of the findings is limited to a few
countries. We used longitudinal data collected between 2012 and 2017 in 3 surveys including 15 European
countries and the United States and comprising 32,531 adults aged 50 years or older. The risk of depression
according to perceived neighborhood disorder and lack of social cohesion was estimated using 2-stage individual-
participant-data meta-analysis; country-specific parameters were analyzed by meta-regression. We conducted
additional analyses on retired individuals. Neighborhood disorder (odds ratio (OR) =1.25) and lack of social
cohesion (OR =1.76) were significantly associated with depression in the fully adjusted models. In retirement,
the risk of depression was even higher (neighborhood disorder: OR =1.35; lack of social cohesion: OR =1.93).
Heterogeneity across countries was low and was significantly reduced by the addition of country-level data
on income inequality and population density. Perceived neighborhood problems increased the overall risk of
depression among adults aged 50 years or older. Policies, especially in countries with stronger links between
neighborhood and depression, should focus on improving the physical environment and supporting social ties in
communities, which can reduce depression and contribute to healthy aging.
cohort studies; depression; mental health; meta-analysis; multicenter studies; residence characteristics
Abbreviations: CES-D, Center for Epidemiologic Studies Depression; CI, confidence interval; ELSA, English Longitudinal Study
of Ageing; HRS, Health and Retirement Study; IPD, individual participant data; OR, odds ratio; SHARE, Survey of Health,
Ageing and Retirement in Europe.
Depression is one of the leading causes of disability
worldwide, affecting 1 out of 5 individuals during their
lifetime (1), and it is associated with a large economic
burden (2). Approximately 13.5% of people over the age
of 50 years suffer from clinically relevant depressive symp-
toms (3), and this percentage rises dramatically among the
oldest old (age ≥80 years) (4). Because of global aging,
the number of people older than 65 years is expected to
grow almost 3-fold by 2050 (5), which will significantly
increase the disease burden related to depression. These
processes present a range of challenges for social, economic,
and health-care systems and require age-specific adaptations
to support healthy aging (6).
In aging individuals, psychosocial and health-related
determinants become more prominent risk factors for the
incidence (7) and recurrence (8) of depression. Because
of increasing morbidity, functional decline, and life-course
transitions (e.g., retirement), older people tend to spend
more time in their local area, which affects the pathways
through which physical and social characteristics influence
their social and psychological well-being (6,9). Exposure to
adverse neighborhood conditions, such as vandalism, crime,
littering, and heavy traffic, have been found to increase the
risk of depression through direct and indirect pathways (10,
11), while social cohesion or social capital buffers individual
distress and weakens the risk of depression (12,13).
1Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
2Baranyi et al.
Although there is a growing body of literature on neigh-
borhoods and mental health, relatively few studies have
assessed the longitudinal associations for this age group
(9), and evidence is based on a low number of (mainly
Anglo-Saxon) economies, limiting the generalizability of
the findings. Examining the evidence of possible neigh-
borhood effects in different settings will provide further
insights into the public health significance of the residential
environment. In addition, the inclusion of several countries
enables the consideration of between-country heterogeneity
in neighborhood effects. Although previous studies have
shown that the prevalence of depression (14) and its
association with social inequality (15) differs by welfare
regime (i.e., typology indicating how states manage their
economies and provide social protection and income
transfers; originally introduced by Esping-Andersen (16)),
there is no evidence of differential neighborhood effects.
Moreover, because micro- and meso-level social and envi-
ronmental factors (e.g., population density, green space, air
pollution) have been previously associated with mental
health and also interact with each other (9), it is feasible
that they will modify neighborhood effects on mental health
between countries. An understanding of how country-level
social, environmental, or welfare-state differences influence
the link between neighborhood and mental health can help
in prioritizing public health policies and interventions at the
national level.
Our primary aims in this individual-participant-data
(IPD) meta-analysis were the following. First, we exam-
ined the longitudinal associations (2 years) of perceived
neighborhood disorder and social cohesion with depressive
symptoms among adults aged 50 years or older, estimating
the risk in a wide range of European and North American
countries. Second, using meta-regression, we explored effect
modification by type of welfare regime and other macro-
level social or environmental indicators in the country-
specific neighborhood effects. In a secondary analysis,
we investigated the robustness of our findings for retired
individuals, a subgroup for whom we assumed that there
would be stronger associations than in the general sample,
since this group tends to spend more time in a residential
environment.
METHODS
Data sources
Data were drawn from 3 representative longitudinal panel
surveys of aging adults: the English Longitudinal Study of
Ageing (ELSA) (17), the Health and Retirement Study (HRS)
(18), and the Survey of Health, Ageing and Retirement in
Europe (SHARE) (19). All of the studies have compara-
ble designs and contain information on noninstitutionalized
community-dwelling adults aged 50 years or over (51 years
for HRS (18)), as well as details on their partners, irrespec-
tive of their age. Participants are followed up approximately
every 2 years, with regular refreshment samples being added
to compensate for attrition bias and to balance the age
structure. The initial HRS cohort was recruited in the United
States in 1992 (20) and served as an exemplar for subse-
quent aging studies. ELSA, with a representative sample for
England, was set up in 2002 (17). The first wave of SHARE
was conducted in 2004/2005, and the most recent wave was
conducted in 2015; it includes 17 European countries and
Israel (19). ELSA, HRS, and SHARE data are harmonized,
allowing cross-national comparisons.
Our analytical sample comprised individuals who pro-
vided valid measurements of depression at 2 consecutive
study waves and answered at least 1 question on perceived
neighborhood characteristics at the baseline wave. We
excluded participants if they had depression at baseline,
were living in a nursing home, were younger than 50 years,
moved to a new residential address between baseline and
follow-up, or had missing values for baseline covariates.
Because data on the neighborhood were not usually col-
lected in all waves, we used the most recently available
sweeps in compliance with our criteria: for ELSA, wave 7
(2014/2015) and wave 8 (2016/2017); for SHARE, wave 5
(2013) and wave 6 (2015). In the HRS, since 2006 approxi-
mately 50% of the sample has been selected for an enhanced
face-to-face interview, while the other half is interviewed via
telephone; the survey mode alternates in each wave. Neigh-
borhood perception is part of the psychosocial questionnaire,
which is administered after the face-to-face interviews, once
every 4 years for the same person (18). Therefore, in order
to have information for the entire HRS sample, we extracted
exposure data from 2 consecutive waves (wave 11 in 2012
and wave 12 in 2014) and linked them with matching follow-
ups (wave 12 in 2014 and wave 13 in 2016). The rates
of attrition between baseline and follow-up were 16% for
ELSA, 12% and 16% for the 2 HRS subsamples, and 15%
(Switzerland) to 32% (Luxemburg) in SHARE.
Neighborhood
For the measures of perceived neighborhood disorder and
lack of social cohesion, we used 4 similarly operational-
ized items asking participants about the “local area, that
is, everywhere within a 20-minute walk or about a mile
[kilometer in SHARE] of your home.” Neighborhoods were
assessed in ELSA and HRS on a 7-point bipolar scale in
the self-completion part of the questionnaire, while SHARE
applied a 4-point Likert scale in the interview denoting
agreement or disagreement with the opposing statement.
A priori,we assigned 2 items to the neighborhood dis-
order domain, capturing information on 1) vandalism and
crime/graffiti and 2) the cleanliness of the area. Lack of
social cohesion included items on 1) feeling part of the
area and 2) receiving help if in trouble. Principal compo-
nents analysis did not confirm the 2-component structure
but indicated 1 underlying score, which provided satisfying
internal consistency (Cronbach’s α= 0.57–0.82). In order
to make neighborhood variables comparable across studies,
we first dichotomized scores for all items (SHARE: 0–1
vs. 2–3; ELSA and HRS: 0–3 vs. 4–6) to obtain similar
response patterns between cohorts. Scales were computed
by calculating the average value of the respective items,
which ranged between 0 and 1, with higher numbers indi-
cating more problems and less cohesion in the residential
area.
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
IPD Meta-Analysis of Neighborhood and Depression 3
Depression
Depressive symptoms were assessed with 2 self-report
symptom scales: the Center for Epidemiologic Studies
Depression (CES-D) Scale (21) and the EURO-D Scale
(22). The CES-D was implemented in ELSA and HRS, and
the EURO-D was implemented in SHARE. The original
CES-D Scale, containing 20 items, was developed to detect
depressive symptomatology in the general population during
the week preceding the interview (21). In ELSA and HRS,
a short version of the CES-D was used, with 8 items
asking respondents whether (in the past week) they had
felt depressed, felt that everything was an effort, had restless
sleep, were happy, were lonely, enjoyed life, felt sad, or
could not get going. The EURO-D Scale consists of 12 items
measuring the presence of depression, pessimism, wishing
for death, guilt, sleep, interest, irritability, appetite, fatigue,
concentration, enjoyment, and tearfulness in the last month
(22). Both scales have high internal consistency and test-
retest reliability, provide a valid measurement of depression,
and show high correlation within the same population (22,
23). Binary answers, indicating the presence or absence
of depressive symptoms, were summed, with increasing
scores indicating higher levels of depressive symptoms. For
approximation of a clinically significant level of depressive
symptoms, a cutoff score of ≥3 was applied for CES-D (23)
and a cutoff score of ≥4 was applied for the EURO-D (22,
23); these thresholds were also used in a recent comparative
study (24).
Baseline covariates
We adjusted for several sociodemographic and health-
related confounders measured at baseline that were relevant
to the neighborhood-depression association (10,12,25,26).
In addition to sex (male, female), age (because of a nonlinear
relationship with depression, this variable was categorized
as 50–59, 60–69, 70–79, and ≥80 years), and immigration
(born in the country of interview or not), we included 3 indi-
cators of socioeconomic status: educational attainment, total
equalized household net wealth, and economic activity. For
education, we used the International Standard Classification
of Education classification (27) from the harmonized data
sets and grouped the highest educational attainment into 3
categories: primary (levels 0 and 1), secondary (levels 2–
4), and tertiary (levels 5 and 6). Household nonpension net
wealth included financial, physical, and housing wealth after
all debt had been subtracted. We calculated an equalized
Figure 1. Selection of participants from 3 studies (the English Longitudinal Study of Ageing (ELSA), the Health and Retirement Study (HRS),
and the Survey of Health, Ageing and Retirement in Europe (SHARE)) for a pooled data set on neighborhood perception and depression. The
pooled data set contained information assessed in 16 different countries between 2012 and 2017. Note that the HRS collects information on
neighborhood perception from half of the sample in each study wave. Because the survey mode alternates between waves, we extracted and
merged data from both subsamples.
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
4Baranyi et al.
measure by dividing the household sum by the square
root of benefiting members (28) and categorized it into
country-specific tertiles (low, medium, or high wealth).
Economic activity described whether the respondent was
working (employed, self-employed), retired, or out of the
labor force (homemaker, unemployed, permanently sick or
disabled). We included information on partnership (married
or cohabiting vs. neither) and on current smoking (yes,
no). A binary variable described whether the respondent
reported at least 2 out of 7 physician-diagnosed chronic
diseases or conditions (arthritis, cancer, cardiovascular
disease, diabetes, high blood pressure, lung disease, and
Tab le 1. Baselineaand Follow-upbCharacteristics (%c) of Adults Aged 50 Years or Older (n= 32,531) From 3 Surveys Included in a Study of
Neighborhood Perception and Depression, 2012–2017
Characteristic
Survey
Pooled Data
ELSA
(n= 4,634)
HRS
(n= 8,646)
SHARE
(n= 19,251)
Sex
Male 46.3 42.0 45.5 44.7
Female 53.7 58.0 54.5 55.3
Age group, years
50–59 18.2 26.2 28.4 26.4
60–69 43.3 30.2 37.1 36.1
70–79 29.1 31.1 24.8 27.1
≥80 9.3 12.6 9.7 10.4
Country of birth
Born in country of interview 94.1 88.6 89.5 89.9
Born outside country of interview 5.9 11.4 10.5 10.1
Educational attainment
Primary (ISCED levels 0 and 1) 19.2 12.8 17.3 16.4
Secondary (ISCED levels 2–4) 46.2 60.1 55.9 55.6
Ter tiary (ISCED levels 5 and 6) 34.6 27.1 26.8 28.0
Tertile of equalized household wealth
Low 26.1 22.4 30.8 27.9
Medium 35.7 36.2 33.4 34.5
High 38.2 41.4 35.8 37.6
Economic activity
Employed 30.4 35.4 29.6 31.3
Retired 62.6 49.7 58.7 56.8
Out of labor force 7.0 14.9 11.7 11.9
Partnership status
In a couple 77.1 68.5 63.9 67.0
Alone 22.9 31.5 36.1 33.0
Current smoking
No 91.6 89.5 82.8 85.9
Yes 8.4 10.5 17.2 14.1
No. of chronic diseases or conditions
<2 76.941.576.3 67.2
≥2 23.1 58.5 23.7 32.8
No. of ADL/IADL functional limitations
0 83.4 90.6 90.7 89.6
≥1 16.6 9.4 9.3 10.4
Table continues
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
IPD Meta-Analysis of Neighborhood and Depression 5
Tab le 1. Continued
Characteristic
Survey
Pooled Data
ELSA
(n= 4,634)
HRS
(n= 8,646)
SHARE
(n= 19,251)
Neighborhood disorderd,e0.13 (0.004) 0.12 (0.003) 0.15 (0.002) 0.14 (0.002)
Lack of social cohesiond,f0.09 (0.003) 0.13 (0.003) 0.08 (0.001) 0.09 (0.001)
Composite neighborhood scored,g 0.11 (0.003) 0.13 (0.003) 0.11 (0.001) 0.12 (0.001)
Depression at follow-uph
Yes 10.4 10.4 15.2 13.2
No 89.6 89.6 84.8 86.8
Abbreviations: ADL, Activities of Daily Living; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; IADL,
Instrumental Activities of Daily Living; ISCED, International Standard Classification of Education; SHARE, Survey of Health, Ageing and
Retirement in Europe.
aBaseline measures: ELSA, 2014/2015; HRS, 2012 and 2014; SHARE, 2013.
bFollow-up measures: ELSA, 2016/2017; HRS, 2014 and 2016; SHARE, 2015.
cPercentages may not sum to 100 because of rounding.
dValues are expressed as mean (standard error).
eThe neighborhood disorder measure captured perceived 1) vandalism and crime/graffiti and 2) cleanliness of the residential area; values
ranged between 0 and 1.
fLack of social cohesion included 1) not feeling part of the neighborhood and 2) not receiving help if in trouble; values ranged between 0
and 1.
gThe composite neighborhood score comprised all 4 perceived neighborhood characteristics; values ranged between 0 and 1.
hThe Center for Epidemiologic Studies Depression Scale (21) was implemented in ELSA and HRS, and the EURO-D Scale (22)was
implemented in SHARE.
stroke). Finally, a measure of functional limitations indicated
whether the respondent had at least 1 disability affecting
Activities of Daily Living or Instrumental Activities of Daily
Living (29).
Country-level indicators
Countries were grouped into 5 types of welfare regimes
based on an expanded classification (15) of Ferrera’s typol-
ogy (30), which is considered a state-of-the-art typology and
is often used in cross-national surveys (15). The 1) Scandi-
navian welfare regimes (Denmark, Sweden) are described
as having universal coverage and generous social transfers;
the 2) Bismarckian regimes (Austria, Belgium, France, Ger-
many, Luxembourg, Switzerland) have earnings-related ben-
efits administered by the employer and familialism; the 3)
Anglo-Saxon regimes (England, Israel, United States) have
minimum welfare provisions and a strong emphasis on the
market (15); the 4) Southern European regimes (Italy, Spain)
are characterized as “rudimentary,” with services ranging
from generous to limited and with high reliance on the fam-
ily (15,30); and the 5) Eastern European regimes consist of
postcommunist countries (Czech Republic, Estonia, Slove-
nia) which have experienced shifts towards marketization
from more universalist communist welfare states (15).
Macro-levelsocial and environmental indicators were
extracted from the World Bank Database (31) for the closest
year of data collection (see Web Table 1, available at https://
academic.oup.com/aje):grossdomestic productatpurchasing
power parity per capita (in current international dollars), Gini
index of income inequality (32), population density (number
of people per km2), urbanization rate (percentage of the
population that is urban), forest coverage (percentage of
land area), and annual mean air pollution level (particulate
matter less than or equal to 2.5 μm in diameter, measured
in μg/m3). Before including these variables in the models,
we standardized all external raw data. Correlations between
indicators are shown in Web Table 2.
Statistical analysis
We conducted a 2-stage IPD meta-analysis to estimate
the overall associations between perceived neighborhood
characteristics and depression (33). First, we fitted sepa-
rate logistic regression models for each country, includ-
ing perceived neighborhood characteristics as a continuous
independent variable, to obtain odds ratios for depression
with 95% confidence intervals. Second, we derived effect
estimates and their variance and pooled them using meta-
analysis. Heterogeneity between countries was quantified
with the I2statistic, indicating the percentage of variance
explained by individual countries (34). Because the hetero-
geneity was low (I2<25%), we fitted fixed-effects models
with inverse variance pooling, assuming a single underlying
true association across countries (33). We present results
from 2 sets of models: The first set of models controlled for
age and sex, and the second set adjusted for all confounders
(age, sex, country of birth, education, wealth, economic
activity, partnership status, current smoking, chronic dis-
eases or conditions, and functional limitations). Prior to the
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
6Baranyi et al.
Figure 2 Continues
main analyses, we tested the linearity assumption by imput-
ing neighborhood variables in categorical form into the mod-
els, which was confirmed by the stepwise increasing gradi-
ents. Interaction models did not reveal significantly different
neighborhood associations among male and female partici-
pants; therefore, no sex-stratified results were prepared.
Although heterogeneity was relatively low, we still exam-
ined whether between-country variation in the risk estimates
might be explained by sample (sample size, percentage of
female participants) or country (e.g., welfare regime, Gini
index, air pollution) characteristics. We first retained log
odds and their standard errors from the fully adjusted logis-
tic models and then performed univariable random-effects
meta-regression. Models were fitted by the restricted max-
imum likelihood method and corrected with the Hartung-
Knapp variance estimator.
Because multicenter studies can be analyzed in various
ways (35), in the sensitivity analyses we derived risk esti-
mates pooled by 1) 2-stage IPD with random-effects models
and estimated with 2) 1-stage IPD with random intercepts
(multilevel logistic models) and 3) 1-stage IPD with fixed
country effects (logistic models). Although we expected
only small differences (33), we report results from the 2-
stage IPD meta-analysis as the main results, because in
multilevel models at least 30 countries would be required to
accurately estimate the country-level parameters (36). Find-
ings on neighborhood disorder and lack of social cohesion
are presented in the Results section of the text, while findings
from analyses of the composite neighborhood problems
score are shown in the Web material (Web Tables 3 and 4,
Web Figure 1). We provide stage 1 results of the IPD meta-
analysis (i.e., covariate-adjusted logistic models by country)
for the composite neighborhood problems score in Web
Tabl e 3 .
All analyses were performed using STATA 13 (StataCorp
LLC, College Station, Texas).
RESULTS
After application of all inclusion and exclusion criteria
(Figure 1), the pooled analytical sample contained 32,531
participants from 16 countries: Austria (n= 1,448),
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
IPD Meta-Analysis of Neighborhood and Depression 7
Figure 2. Country-specific and pooled associations of A) perceived neighborhood disorder and B) perceived lack of social cohesion with
depression among adults aged 50 years or older in 16 high-income countries, 2012–2017. Results were adjusted for age, sex, country of birth,
education, wealth, economic activity, partnership status, current smoking, chronic diseases or conditions, and functional limitations. Countries
are grouped by type of welfare regime. The size of each gray square is proportional to the relative weight of the sample in the meta-analysis;
diamonds represent the pooled estimates.Odds ratios (ORs) greater than 1 indicate increased risk of depression, while ORs less than 1 indicate
decreased risk. The overall I2values were 0.0% (P= 0.53) for perceived neighborhood disorder and 23.7% (P= 0.19) for perceived lack of social
cohesion. Bars, 95% confidence intervals (CIs). ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; SHARE, Survey
of Health, Ageing and Retirement in Europe.
Belgium (n= 1,875), the Czech Republic (n= 1,645),
Denmark (n= 1,491), England (n= 4,634), Estonia
(n= 1,713), France (n= 1,250), Germany (n= 1,819),
Israel (n= 561), Italy (n= 1,157), Luxemburg (n= 456),
Slovenia (n= 1,144), Spain (n= 1,742), Sweden (n= 1,640),
Switzerland (n= 1,310), and the United States (n= 8,646).
Tabl e 1 shows the characteristics of study participants by
survey. For the total sample, 55.3% were female, and
the mean age was 66.7 years. Over half of the sample
(56.8%) was retired at the time of data collection. Although
household wealth was defined in terms of 3 equally large
categories within countries, in the analytical data set there
was underrepresentation of individuals from the low wealth
group, partly because of censoring of depression cases at
baseline. After 2 years, the incidence of depression was
13.2%, with large variation by country (P<0.001), ranging
between 8.1% (Denmark) and 22.7% (Estonia).
The IPD meta-analyses models showed significantly ele-
vated odds ratios for clinically relevant depressive symptoms
by neighborhood disorder (odds ratio (OR) = 1.44, 95%
confidence interval (CI): 1.28, 1.61) and lack of social
cohesion (OR = 1.99, 95% CI: 1.75, 2.26) after adjustment
for sex and age (Web Figure 2). In the fully adjusted models
(Figure 2), the pooled odds ratio for neighborhood disorder
was 1.25 (95% CI: 1.11, 1.41), with individual odds ratios
ranging between 0.52 and 2.11 and significantly higher than
1 in the Czech Republic, Denmark, and the United States.
Lack of social cohesion had a pooled odds ratio of 1.76
(95% CI: 1.54, 2.01), with individual odds ratios ranging
from 0.91 to 5.36 and significantly elevated in Belgium,
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
8Baranyi et al.
Tab le 2. Associations of Perceived Neighborhood Disorder and Lack of Social Cohesion With Depression in 16 High-Income Countries (Meta-
Regression Analysis), 2012–2017
Country-Level Indicatora
Neighborhood DisorderbLack of Social Cohesionc
βSE PValue βSE PValue
Age 50 Years or Older
Sample size (no. of individuals) 0.021 0.038 0.60 −0.059 0.047 0.23
% of female participants 0.098 0.060 0.13 −0.044 0.094 0.65
GDP PPP per capita, CID −0.065 0.088 0.47 −0.052 0.128 0.69
Gini index of income inequalityd−0.026 0.054 0.64 −0.174 0.061 0.01
Population density (no. of people/km2) 0.002 0.059 0.97 0.089 0.073 0.24
% of population that is urban 0.016 0.074 0.84 −0.064 0.104 0.55
% of forest coverage −0.062 0.080 0.45 0.055 0.110 0.63
Air pollution (PM2.5 level), μg/m30.044 0.062 0.49 0.152 0.083 0.09
In Retirement
Sample size (no. of individuals) 0.054 0.056 0.35 −0.097 0.063 0.15
% of female participants 0.084 0.083 0.33 −0.186 0.121 0.15
GDP PPP per capita, CID 0.061 0.114 0.60 −0.003 0.158 0.99
Gini index of income inequality 0.019 0.070 0.79 −0.188 0.082 0.04
Population density (no. of people/km2) 0.044 0.078 0.58 0.194 0.087 0.04
% of population that is urban 0.133 0.094 0.18 0.096 0.123 0.45
% of forest coverage −0.175 0.099 0.10 −0.102 0.125 0.43
Air pollution (PM2.5 level), μg/m30.038 0.078 0.64 0.205 0.102 0.07
Abbreviations: CID, current international dollars; GDP PPP, gross domestic product at purchasing power parity; PM2.5, par ticulate matter less
than or equal to 2.5 μm in diameter; SE, standard error.
aRaw data were standardized before meta-regression.
bAssociations between neighborhood disorder and depression did not differ by welfare regime (individuals aged ≥50 years: F(4, 11) = 1.29,
P= 0.33; retired individuals: F(4, 11) =1.18,P= 0.37).
cAssociations between lack of social cohesion and depression did not differ by welfare regime (individuals aged ≥50 years: F(4, 11) = 1.73,
P= 0.21; retired individuals: F(4, 11) = 0.71, P= 0.60).
dScores on the Gini index can range from 0 to 100, with higher numbers indicating more inequality (32).
the Czech Republic, Estonia, France, Germany, Slovenia,
England, and the United States. Meta-regression indicated
stronger associations between lack of social cohesion and
depression in more equal countries (β=−0.174, P= 0.01),
as measured by Gini index. Furthermore, there was a ten-
dency for associations between lack of social cohesion and
depression to be stronger in countries with higher levels of
air pollution (β= 0.152, P= 0.09) (Tabl e 2 ).
We repeated the analyses for retired individuals. In the
sex- and age-adjusted models, neighborhood disorder had an
odds ratio of 1.48 (95% CI: 1.28, 1.71), while the odds ratio
for lack of social cohesion was 2.06 (95% CI: 1.73, 2.45)
(Web Figure 3). Although the pooled odds ratios decreased
after adjustment for all covariates, they remained higher in
this subsample than in the full sample. The pooled odds ratio
for neighborhood disorder was 1.35 (95% CI: 1.16, 1.57)—
10% higher when including only participants at retirement
compared with all participants aged 50 years or older. The
pooled odds ratio for lack of social cohesion was 1.93 (95%
CI: 1.61, 2.30), indicating 17% higher odds of depression
during retirement (Web Figure 4). Meta-regression analy-
ses found significantly elevated risk of depression by lack
of social cohesion in more equal countries (β=−0.188,
P= 0.04) and in countries with higher population density
(β= 0.194, P= 0.04) (Tabl e 2 ). There was a tendency for
associations between neighborhood disorder and depres-
sion to be weaker in countries with more forest coverage
(β=−0.175, P= 0.099) and for associations between
lack of social cohesion and depression to be stronger in
countries with higher levels of air pollution (β= 0.205,
P= 0.07).
The pooled neighborhood associations were robust and
did not significantly differ when estimated in 1-stage IPD
meta-analysis (random or fixed country effects) or in
random-effects 2-stage IPD meta-analysis (Web Table 5).
Analyses that used the composite neighborhood problems
score produced risk estimates comparable to those calcu-
lated for lack of social cohesion (full sample: OR = 1.74,
95% CI: 1.49, 2.03; in retirement: OR = 1.96, 95% CI 1.60,
2.40) (Web Figure 1). Similarly to the main analysis, we
found stronger associations between neighborhood prob-
lems and depression in more equal countries (β=−0.160,
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
IPD Meta-Analysis of Neighborhood and Depression 9
P= 0.04). In the subsample of retirees, there was a
tendency toward weaker associations between neighborhood
problems and depression in countries with more forest
coverage (β=−0.248, P= 0.095) and toward stronger
associations in countries with higher population density
(β= 0.202, P= 0.07) (Web Table 4).
DISCUSSION
This cross-national longitudinal study provides evidence
for a link between perceived neighborhood disorder and lack
of social cohesion and depression among adults aged 50
years or older. These findings are based on analyses of data
from 3 representative panel surveys including 32,531 partic-
ipants across 16 high-income countries. Risk estimates were
10%–17% higher, on average, in a subsample containing
only retired individuals than in the total sample. We identi-
fied low country-level variation between risks of depression
by neighborhood problems, which could be partly explained
by macro-level indicators such as income inequality, popu-
lation density, forest coverage, and air pollution.
Our findings are in line with those of previous cross-
sectional studies (9) and longitudinal studies exploring the
possible effect of perceived neighborhood disorder (10,11,
26) and social cohesion/social capital (12,13,25,37)on
the risk of depression in older age. As people age and then
retire, the geographical extent of their mobility space tends
to decrease, and they often become more reliant on their
community and local services (9). At the same time, depres-
sion trajectories widen by neighborhood quality in aging
individuals (12), leading to stronger associations between
neighborhood and depression among retired individuals.
These findings suggest that the broader social, economic,
and environmental context of a country might modify
the association between neighborhood characteristics and
depression. In Southern European countries, neighborhood
disorder and lack of social cohesion did not increase the
risk of depression, while in Eastern European and Anglo-
Saxon countries we often found strong and significant
associations. Type of welfare regime did not statistically
explain differences, which may have been due to the
low number of countries in each welfare-regime group.
However, other unexplored social norms and cultural values
predicting source of social support (community vs. family
and close relatives) and ways of coping with residential
stressors might be better predictors of modification of the
relationship.
Meta-regression analysis estimated stronger risks of
depression by lack of social cohesion, when people were
living in economically more equal countries. Egalitarian
countries tend to have better health outcomes, which might
be linked via social capital or other aspects of social
organization (38). Perceived lack of social cohesion in more
equal economies, therefore, violates the normative rules
of the society and the general expectation of people with
regard to their neighborhoods and neighbors. This perceived
discrepancy between reality and expectations might cause
insecurity and lead to higher levels of psychological distress.
There was also weak evidence for a modifying role of air
pollution in the link between social cohesion and depression,
which seems to be important in more polluted countries,
where social cohesion can buffer the distress induced by
air pollution (39). In addition to income inequality and air
pollution, findings among retired individuals revealed that in
countries where people live in closer proximity to each other,
lack of social cohesion predicted depression more strongly.
The value of the immediate community increases with
higher population density, especially for individuals who are
more reliant on their surroundings. Finally, neighborhood
disorder tended to be associated with higher risk of mental
health problems in countries with less forest coverage.
Exposure to nature may be protective for mental health
by reducing the hazardous effect of environmental distress
(40) caused by (for example) neighborhood disorder, traffic
noise, or air pollution.
This study had several strengths. We report here (for the
first time, to our knowledge) pooled risks of depression for
neighborhood disorder and lack of social cohesion among
adults aged 50 years or older based on data from several
high-income countries, many of them (e.g., Southern and
Eastern European countries) often neglected in research.
The presented analyses were based on longitudinal data
with baseline and follow-up measures of outcome, placing
this among the few prospective studies in the neighbor-
hood literature. Effect estimates from 16 different coun-
tries were pooled together by IPD meta-analyses, taking
into account demographic, socioeconomic, and health con-
founders. Moreover, we have provided possible explanations
for country-level differences in the risk of depression by
neighborhood problems.
The study also had limitations. First, the exposure,
outcome, and covariates were all self-reported measures.
Although we excluded possible depression cases at baseline
to avoid the potential for underlying depression to distort
the perceptions of neighborhood or covariates, we could
not completely rule out reverse causation or an unmeasured
psychological mechanism (e.g., reporting behavior) leading
to biased estimates (41). Second, despite the high correlation
between outcome measures, they have relevant differences
(23): The CES-D Scale tends to have stronger associations
with social and demographic factors than the EURO-D,
indicating a more extreme pool of cases, and it captures a
shorter time interval (1 week vs. 1 month). Third, there was
a significant number of missing values for neighborhood
perception. Although the sample size was not related
to the variation between effect parameters, nonresponse
bias might have influenced the results. Missing values
for neighborhood originated from the survey method in
ELSA and HRS (e.g., leave-behind questionnaire), while in
SHARE only part of the sample (household respondents)
was asked about their residential area, providing very
different reasons for missingness in the pooled data set.
Fourth, because neighborhood perception was not assessed
in each wave, we could not include the same year of
baseline and follow-up for all surveys, which meant that it is
possible that unknown macroeconomic or societal changes
may have affected the results. Fifth, several European
and North American countries were not included in this
study, due to either a lack of data or insufficient data
harmonization. We cannot exclude the possibility that the
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
10 Baranyi et al.
absence of these countries may have influenced the study’s
findings.
Future research should make use of comparable multicen-
ter surveys (e.g., Gateway to Global Aging Data) and extend
its focus to low- and middle-income countries. Although
there are cross-sectional multicenter studies on residen-
tial environment and health available in different coun-
try settings (42), longitudinal evidence is needed to better
understand how macro-level social and environmental indi-
cators shape neighborhood effects. In addition, using objec-
tively measured neighborhood exposure would overcome
possible bias related to the use of self-report measures.
Neighborhood environment is a significant determinant of
mental health and has the potential to reduce the negative
effects of socioeconomic inequalities on health (42). More-
over, it is modifiable and therefore offers policy-makers
opportunities for intervention to enhance health among older
adults (6). Policies, especially in countries with a stronger
link between neighborhood and depression, should focus
on improving the physical qualities of residential areas and
supporting social ties in communities, which can reduce
mental health problems and related disability and make
positive contributions to healthy aging.
ACKNOWLEDGMENTS
Author affiliations: Center for Research on Environment,
Society and Health, School of GeoSciences, University of
Edinburgh, Edinburgh, United Kingdom (Gerg˝
o Baranyi,
Jamie R. Pearce, Chris J. L. Dibben); Swiss National
Centre of Competence in Research “LIVES—Overcoming
Vulnerability: Life Course Perspectives,” Center for the
Interdisciplinary Study of Gerontology and Vulnerability,
University of Geneva, Geneva, Switzerland (Stefan Sieber,
Stéphane Cullati); and Department of Readaptation and
Geriatrics, Faculty of Medicine, University of Geneva,
Geneva, Switzerland (Stéphane Cullati, Delphine S.
Courvoisier).
This work was supported by the European Union’s
Horizon 2020 research and innovation program under the
Marie Skłodowska-Curie Actions (agreement 676060
(LONGPOP) with G.B., J.R.P., and C.J.L.D.) and the Swiss
National Centre of Competence in Research “LIVES—
Overcoming Vulnerability: Life Course Perspectives,”
which is financed by the Swiss National Science
Foundation (grant 51NF40-160590 to S.S. and S.C.).
The English Longitudinal Study of Ageing (ELSA) was
developed by a team of researchers based at University
College London, the United Kingdom National Centre for
Social Research, and the Institute for Fiscal Studies.
Funding was provided by the US National Institute of
Aging and a consortium of United Kingdom government
departments coordinated by the Office for National
Statistics. ELSA data were collected by the National Centre
for Social Research and made available through the UK
Data Archive.
The Health and Retirement Study (HRS) was sponsored
by the US National Institute on Aging (grant
U01AG009740) and conducted by the University of
Michigan. This analysis used Early Release data (for wave
2016), which have not been cleaned and may contain errors
that will be corrected in the Final Public Release version of
the data set.
Data collection in the Survey of Health, Ageing and
Retirement in Europe (SHARE) was funded by the
European Commission through the Fifth Framework
Programme (grant QLK6-CT-2001-00360), the Sixth
Framework Programme (projects SHARE-I3 (grant
RII-CT-2006-062193), COMPARE (grant
CIT5-CT-2005-028857), and SHARELIFE (grant
CIT4-CT-2006-028812)), the Seventh Framework
Programme (projects SHARE-PREP (grant GA 211909),
SHARE-LEAP (grant GA 227822), and SHARE M4 (grant
GA 261982)), and Horizon 2020 (projects SHARE-DEV3
(grant GA 676536) and SERISS (grant GA 654221)) and
by the European Commission’s Directorate-General for
Employment, Social Affairs and Inclusion. Additional
funding was received from the German Ministry of
Education and Research, the Max Planck Society for the
Advancement of Science, the US National Institute on
Aging (grants U01_AG09740-13S2, P01_AG005842,
P01_AG08291, P30_AG12815, R21_AG025169,
Y1-AG-4553-01, IAG_BSR06-11, OGHA_04-064, and
HHSN271201300071C), and various national funding
sources (see http://www.share-project.org). This analysis
used data from SHARE waves 5 and 6 (10.6103/SHARE.
w5.610,10.6103/SHARE.w6.610);
see Börsch-Supan et al. (19) for methodological details.
This publication reflects only the authors’ views, and the
Research Executive Agency of the European Commission
is not responsible for any use that may be made of the
information it contains.
Conflict of interest: none declared.
REFERENCES
1. Hasin DS, Sarvet AL, Meyers JL, et al. Epidemiology of
adult DSM-5 major depressive disorder and its specifiers
in the United States. JAMA Psychiatry. 2018;75(4):
336–346.
2. Ferrari AJ, Charlson FJ, Norman RE, et al. Burden of
depressive disorders by country, sex, age, and year: findings
from the Global Burden of Disease Study 2010. PLoS Med.
2013;10(11):e1001547.
3. Beekman AT, Copeland JR, Prince MJ. Review of
community prevalence of depression in later life. Br J
Psychiatry. 1999;174(4):307–311.
4. Luppa M, Sikorski C, Luck T, et al. Age- and gender-specific
prevalence of depression in latest-life—systematic review and
meta-analysis. J Affect Disord. 2012;136(3):212–221.
5. World Health Organization. Global Health and Aging.
Bethesda, MD: National Institutes of Health; 2011. (NIH
publication no. 11-7737).
6. Rechel B, Grundy E, Robine JM, et al. Ageing in the
European Union. Lancet. 2013;381(9874):1312–1322.
7. Buchtemann D, Luppa M, Bramesfeld A, et al. Incidence of
late-life depression: a systematic review. J Affect Disord.
2012;142(1–3):172–179.
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
IPD Meta-Analysis of Neighborhood and Depression 11
8. Deng Y, McQuoid DR, Potter GG, et al. Predictors of
recurrence in remitted late-life depression. Depress Anxiety.
2018;35(7):658–667.
9. Barnett A, Zhang CJP, Johnston JM, et al. Relationships
between the neighborhood environment and depression in
older adults: a systematic review and meta-analysis. Int
Psychogeriatr. 2017;30(8):1153–1176.
10. Bierman A. Marital status as contingency for the effects of
neighborhood disorder on older adults’ mental health. J
Gerontol B Psychol Sci Soc Sci. 2009;64B(3):425–434.
11. Roberts RE, Kaplan GA, Shema SJ, et al. Does growing old
increase the risk for depression? Am J Psychiatry. 1997;
154(10):1384–1390.
12. Ruiz M, Scholes S, Bobak M. Perceived neighbourhood
social cohesion and depressive symptom trajectories in older
adults: a 12-year prospective cohort study. Soc Psychiatry
Psychiatr Epidemiol. 2018;53(10):1081–1090.
13. Ehsan AM, De Silva MJ. Social capital and common mental
disorder: a systematic review. J Epidemiol Community
Health. 2015;69(10):1021–1028.
14. Chung H, Ng E, Ibrahim S, et al. Welfare state regimes,
gender, and depression: a multilevel analysis of middle and
high income countries. Int J Environ Res Public Health.
2013;10(4):1324–1341.
15. Bambra C, Eikemo TA. Welfare state regimes, unemployment
and health: a comparative study of the relationship between
unemployment and self-reported health in 23 European
countries. J Epidemiol Community Health. 2009;63(2):92–98.
16. Esping-Andersen G. The Three Worlds of Welfare Capitalism.
Cambridge, United Kingdom: Polity Press; 1990.
17. Steptoe A, Breeze E, Banks J, et al. Cohort profile: the
English Longitudinal Study of Ageing. Int J Epidemiol. 2013;
42(6):1640–1648.
18. Fisher GG, Ryan LH. Overview of the Health and Retirement
Study and introduction to the special issue. Work Aging
Retire. 2018;4(1):1–9.
19. Börsch-Supan A, Brandt M, Hunkler C, et al. Data resource
profile: the Survey of Health, Ageing and Retirement in
Europe (SHARE). Int J Epidemiol. 2013;42(4):992–1001.
20. Sonnega A, Faul JD, Ofstedal MB, et al. Cohort profile: the
Health and Retirement Study (HRS). Int J Epidemiol.
2014;43(2):576–585.
21. Radloff LS. The CES-D Scale: a self-report depression scale
for research in the general population. Appl Psychol Meas.
1977;1(3):385–401.
22. Prince MJ, Reischies F, Beekman AT, et al. Development of
the EURO-D scale—a European Union initiative to compare
symptoms of depression in 14 European centres. Br J
Psychiatry. 1999;174(4):330–338.
23. Courtin E, Knapp M, Grundy E, et al. Are different measures
of depressive symptoms in old age comparable? An analysis
of the CES-D and Euro-D scales in 13 countries. Int J
Methods Psychiatr Res. 2015;24(4):287–304.
24. Suemoto CK, Ueda P, Beltran-Sanchez H, et al. Development
and validation of a 10-year mortality prediction model:
meta-analysis of individual participant data from five cohorts
of older adults in developed and developing countries. J
Gerontol A Biol Sci Med Sci. 2017;72(3):410–416.
25. Fone D, White J, Farewell D, et al. Effect of neighborhood
deprivation and social cohesion on mental health inequality: a
multilevel population-based longitudinal study. Psychol Med.
2014;44(11):2449–2460.
26. Stafford M, McMunn A, de Vogli R. Neighbourhood social
environment and depressive symptoms in mid-life and
beyond. Ageing Soc. 2011;31(6):893–910.
27. UNESCO Institute for Statistics. International Standard
Classification of Education: ISCED 2011. Montreal, QC,
Canada: UNESCO Institute for Statistics; 2012.
28. Jürges H. Health inequalities by education, income and
wealth: a comparison of 11 European countries and the US.
Appl Econ Lett. 2010;17(1):87–91.
29. Chan KS, Kasper JD, Brandt J, et al. Measurement
equivalence in ADL and IADL difficulty across international
surveys of aging: findings from the HRS, SHARE, and
ELSA. J Gerontol B Psychol Sci Soc Sci. 2012;67(1):
121–132.
30. Ferrera M. The ‘Southern model’ of welfare in social Europe.
J Eur Soc Policy. 1996;6(1):17–37.
31. The World Bank. World Bank Open Data. World Development
Indicators. Washington, DC: World Bank; 2017. https://data.
worldbank.org/. Accessed January 31, 2019.
32. The World Bank. World Development Indicators. Gini Index
(World Bank Estimate). Washington, DC: World Bank; 2017.
https://data.worldbank.org/indicator/SI.POV.GINI/. Accessed
January 31, 2019.
33. Burke DL, Ensor J, Riley RD. Meta-analysis using individual
participant data: one-stage and two-stage approaches, and
why they may differ. Stat Med. 2017;36(5):855–875.
34. Higgins JP, Thompson SG, Deeks JJ, et al. Measuring
inconsistency in meta-analyses. BMJ. 2003;327(7414):
557–560.
35. Basagaña X, Pedersen M, Barrera-Gomez J, et al. Analysis of
multicentre epidemiological studies: contrasting fixed or
random effects modelling and meta-analysis. Int J Epidemiol.
2018;47(4):1343–1354.
36. Bryan ML, Jenkins SP. Multilevel modelling of country
effects: a cautionary tale. Eur Sociol Rev. 2016;32(1):3–22.
37. Choi N, Kim J, Dinitto DM, et al. Perceived social cohesion,
frequency of going out, and depressive symptoms in older
adults: examination of longitudinal relationships. Gerontol
Geriatr Med. 2015;1:2333721415615478.
38. Kawachi I, Kennedy BP, Lochner K, et al. Social capital,
income inequality and mortality. Am J Public Health.
1997;87(9):1491–1498.
39. Wang RY, Xue DS, Liu Y, et al. The relationship between air
pollution and depression in China: is neighbourhood social
capital protective? Int J Environ Res Public Health.
2018;15(6):Article 1160.
40. Hartig T, Mitchell R, de Vries S, et al. Nature and health.
Annu Rev Public Health. 2014;35:207–228.
41. Diez Roux AV. Neighborhoods and health: where are we and
were do we go from here? Rev Epidemiol Sante Publique.
2007;55(1):13–21.
42. Mitchell RJ, Richardson EA, Shortt NK, et al. Neighborhood
environments and socioeconomic inequalities in mental
well-being. AmJPrevMed. 2015;49(1):80–84.
Am J Epidemiol. 2020;00(00):1–11
Downloaded from https://academic.oup.com/aje/advance-article-abstract/doi/10.1093/aje/kwz209/5579821 by World Health Organization user on 06 March 2020
