ArticlePDF Available

Abstract and Figures

Background Social isolation is known to be associated with poorer health amongst adults, including coronary heart disease. It is hypothesized that this association may be mediated by inflammation. There has been little prospective research on the long-term impact of social isolation in childhood on adult health or the pathways which might be involved. The aim of this study was to investigate whether social isolation in childhood is associated with increased adult inflammation and the mechanisms involved across the life course. Methods This study used multiply-imputed data on 7,462 participants of the National Child Development Study in Great Britain. The association between child social isolation (7-11 yrs) and levels of C-reactive protein (CRP) in middle age (44 yrs) was examined. We additionally investigated the role of adult social isolation, psychological distress, health behaviors and socioeconomic factors as potential mediators using path analysis and concurrent measurements made across the life course. Results Socially isolated children had higher levels of C-reactive protein in mid-life (standardized coefficient= 0.05, p≤0.001). In addition children who were socially isolated tended to have lower subsequent educational attainment, be in a less advantaged social class in adulthood, were more likely to be psychologically distressed across adulthood and were more likely to be obese and to smoke. All of these factors partially explained the association between childhood social isolation and CRP. However this association remained statistically significant after considering all mediators simultaneously. Conclusions Social isolation in childhood is associated with higher levels of C-reactive protein in mid-life. This is explained in part through complex mechanisms acting across the life course. Identification and interventions targeted towards socially isolated children may help reduce long-term adult health risk.
Content may be subject to copyright.
Psychoneuroendocrinology
(2014)
50,
85—94
Available
online
at
www.sciencedirect.com
ScienceDirect
j
ourna
l
h
om
epa
ge
:
www.elsevier.com/locate/psyneuen
Social
isolation
in
childhood
and
adult
inflammation:
Evidence
from
the
National
Child
Development
Study
Rebecca
E.
Laceya,,
Meena
Kumaria,b,
Mel
Bartleya
aDepartment
of
Epidemiology
and
Public
Health,
University
College
London,
London
WC1E
6BT,
United
Kingdom
bInstitute
of
Social
&
Economic
Research,
University
of
Essex,
Colchester
CO4
3SQ,
United
Kingdom
Received
19
May
2014;
received
in
revised
form
12
August
2014;
accepted
13
August
2014
KEYWORDS
Social
isolation;
NCDS;
Cohort
study;
Inflammation;
Path
analysis;
Life
course
Summary
Background:
Social
isolation
is
known
to
be
associated
with
poorer
health
amongst
adults,
including
coronary
heart
disease.
It
is
hypothesized
that
this
association
may
be
mediated
by
inflammation.
There
has
been
little
prospective
research
on
the
long-term
impact
of
social
isolation
in
childhood
on
adult
health
or
the
pathways
which
might
be
involved.
The
aim
of
this
study
was
to
investigate
whether
social
isolation
in
childhood
is
associated
with
increased
adult
inflammation
and
the
mechanisms
involved
across
the
life
course.
Methods:
This
study
used
multiply-imputed
data
on
7462
participants
of
the
National
Child
Development
Study
in
Great
Britain.
The
association
between
child
social
isolation
(7—11
yrs)
and
levels
of
C-reactive
protein
(CRP)
in
middle
age
(44
yrs)
was
examined.
We
additionally
investigated
the
role
of
adult
social
isolation,
psychological
distress,
health
behaviors
and
socioeconomic
factors
as
potential
mediators
using
path
analysis
and
concurrent
measurements
made
across
the
life
course.
Results:
Socially
isolated
children
had
higher
levels
of
C-reactive
protein
in
mid-life
(standard-
ized
coefficient
=
0.05,
p
0.001).
In
addition,
children
who
were
socially
isolated
tended
to
have
lower
subsequent
educational
attainment,
be
in
a
less
advantaged
social
class
in
adult-
hood,
were
more
likely
to
be
psychologically
distressed
across
adulthood
and
were
more
likely
to
be
obese
and
to
smoke.
All
of
these
factors
partially
explained
the
association
between
childhood
social
isolation
and
CRP.
However,
this
association
remained
statistically
significant
after
considering
all
mediators
simultaneously.
Corresponding
author
at:
Department
of
Epidemiology
and
Public
Health,
University
College
London,
1-19
Torrington
Place,
London
WC1E
6BT,
United
Kingdom.
Tel.:
+44
0207
679
1676;
fax:
+44
0203
108
3354.
E-mail
address:
Rebecca.lacey@ucl.ac.uk
(R.E.
Lacey).
http://dx.doi.org/10.1016/j.psyneuen.2014.08.007
0306-4530/©
2014
The
Authors.
Published
by
Elsevier
Ltd.
This
is
an
open
access
article
under
the
CC
BY
license
(http://creativecommons.org/licenses/by/3.0/).
86
R.E.
Lacey
et
al.
Conclusions:
Social
isolation
in
childhood
is
associated
with
higher
levels
of
C-reactive
protein
in
mid-life.
This
is
explained
in
part
through
complex
mechanisms
acting
across
the
life
course.
Identification
and
interventions
targeted
toward
socially
isolated
children
may
help
reduce
long-
term
adult
health
risk.
©
2014
The
Authors.
Published
by
Elsevier
Ltd.
This
is
an
open
access
article
under
the
CC
BY
license
(http://creativecommons.org/licenses/by/3.0/).
1.
Introduction
Chronic
low-grade
inflammation
is
known
to
be
associated
with
adverse
health
outcomes,
such
as
type
II
diabetes
(Bassuk
et
al.,
2004),
depression
(Danner
et
al.,
2003),
and
coronary
heart
disease
(Danesh
et
al.,
2004).
Many
studies
have
shown
associations
between
childhood
adver-
sities
and
raised
C-reactive
protein
(CRP)
levels
a
reliable
marker
of
low-grade
inflammation
(Pepys
and
Hirschfield,
2003).
Childhood
adversities
known
to
be
associated
with
increased
adult
inflammation
include
socioeconomic
disad-
vantage
(Phillips
et
al.,
2009;
Carroll
et
al.,
2011;
Miller
and
Cole,
2012),
abuse
and
neglect
(Danese
et
al.,
2007),
parental
divorce
(Lacey
et
al.,
2013),
as
well
as
combined
adverse
childhood
events
scores
(Slopen
et
al.,
2010,
2013).
One
childhood
adversity
which
is
not
frequently
investi-
gated
in
relation
to
adult
health
is
social
isolation.
Humans
have
a
fundamental
need
to
be
socially
connected
to
and
supported
by
others
(Baumeister
and
Leary,
1995).
Social
isolation
in
adulthood
is
known
to
be
detrimental
to
health
being
associated
with,
for
example,
coronary
heart
disease
(Kawachi
et
al.,
1996).
Chronic
inflammation
is
thought
to
be
one
pathway
linking
psychosocial
diversity
to
heart
disease
(Uchino,
2006).
Few
studies
have
addressed
longer-
term
processes
between
childhood
social
isolation
and
adult
health.
Chronic
isolation
has
been
linked
to
school
drop-
out,
problem
drinking,
and
depression
(Asher
and
Paquette,
2003)
risk
factors
for
poorer
adult
health.
However,
little
research
has
explored
whether
childhood
social
isolation
is
associated
with
objective
markers
of
poorer
adult
health,
such
as
inflammation.
Tw o
studies
using
the
Dunedin
birth
cohort
have
begun
to
investigate
this
association.
Caspi
and
colleagues
(2006)
found
that
childhood
social
isolation,
as
indicated
by
items
of
the
Rutter
behavior
scale,
was
associated
with
cardiovas-
cular
risk
factors
at
age
26
(overweight,
hypertension,
raised
glycated
hemoglobin
concentration,
low
maximum
oxygen
consumption
and
elevated
total
cholesterol).
Danese
et
al.
(2009)
found
that
children
with
high
levels
of
social
isolation
had
a
60%
increased
risk
of
a
CRP
value
>3
mg/L
at
age
32,
compared
to
children
who
experienced
a
very
low
level
of
isolation.
A
number
of
pathways
are
hypothesized
to
be
impor-
tant
between
child
isolation
and
adult
health.
Firstly,
poor
social
relations
in
childhood
may
result
in
lower
educa-
tional
attainment
and
this
has
many
consequences
including
adverse
trajectories
of
occupational
and
social
position
(Brown
and
Taylor,
2008).
Secondly,
child
social
isolation
may
increase
psychological
distress
in
adult
life
(Katz
et
al.,
2011;
Takizawa
et
al.,
2014).
Danese
et
al.
(2009)
also
found
that
severe
child
social
isolation
was
associated
with
increased
depressive
symptoms
in
adulthood.
Thirdly,
child-
hood
social
isolation
may
increase
the
risk
of
adult
social
isolation
through
the
development
of
social
and
emotional
mal-adaptation
(Coplan
et
al.,
2012).
Caspi
and
colleagues
(2006)
found
that
child
social
isolation
was
strongly
asso-
ciated
with
adolescent
and
adult
isolation,
and
that
adult
social
isolation
was
in
turn
strongly
associated
with
cardio-
vascular
risk
factors
at
the
age
of
26.
Finally,
social
isolation
in
childhood
may
lower
self-esteem
and
increase
the
risk
of
uptake
and
maintenance
of
adverse
health
behaviors,
such
as
smoking
(Niemela
et
al.,
2011),
problem
alcohol
consump-
tion
(Zimmerman
et
al.,
1997)
and
overeating
(Ackard
et
al.,
2003).
All
of
these
factors
in
turn
have
been
associated
with
poor
health
in
adulthood.
For
instance,
an
association
between
adult
socioeconomic
position
and
inflammation
has
been
shown
numerous
times,
e.g.
Ramsay
et
al.
(2008),
as
have
associations
between
adult
social
isolation
(Shankar
et
al.,
2011),
psychological
distress
(Taylor
et
al.,
2006),
smoking
(Koenig
et
al.,
1999),
alcohol
misuse
(Albert
et
al.,
2003),
and
BMI
(Festa
et
al.,
2001),
with
inflammation.
Unlike
previous
studies
we
propose
that
mechanisms
act-
ing
between
child
social
isolation
and
inflammation
are
linked
in
complex
ways.
We
therefore
consider
pathways
in
combination
with
each
other.
We
extend
previous
work
by
utilizing
a
British
birth
cohort
with
follow-up
into
middle-
age
(more
than
a
decade
greater
than
previously
examined),
by
accounting
for
missing
data
and
by
explicitly
modeling
the
mechanisms
involved.
We
conceptualize
social
isolation
in
this
study
as
social
withdrawal
or
social
rejection,
and
this
is
reflected
in
our
measure
of
child
social
isolation.
Our
hypothesis
is
that
children
who
are
socially
isolated
have
higher
CRP
levels
in
midlife,
and
that
this
is
accounted
for
by
complex
pathways
across
the
life
course
acting
through
adult
social
isolation,
health
behaviors,
material
disadvan-
tage
and
psychological
distress.
Fig.
1
shows
the
conceptual
model
tested
in
this
study.
2.
Methods
2.1.
Sample
This
study
used
data
from
the
National
Child
Develop-
ment
Study
(NCDS)
which
aimed
to
recruit
all
babies
born
in
Great
Britain
during
one
week
of
1958,
achieving
a
sample
of
17,414
(98.2%)
(Power
and
Elliott,
2005).
Par-
ticipants
were
surveyed
at
the
following
ages:
7,
11,
16,
23,
33,
42,
44,
46,
and
50
yrs.
Information
was
collected
from
multiple
sources
on
educational,
social,
medical,
eco-
nomic
and
aspects
of
participants’
lives.
Informed
consent
was
sought
from
respondents
for
each
survey
and
ethical
approval
was
obtained
from
the
South
East
and
London
mul-
ticenter
research
ethics
committees
(Shepherd,
2012).
The
age
44
survey
took
the
form
of
a
biomedical
assessment,
during
which
blood
samples
were
taken
on
a
sub-sample
Child
social
isolation
and
adult
inflammation
87
Social isol
ao
n
(7-11 yrs)
Inamma
on
(44 yr
s)
Educaonal
aainment
(23 yr
s)
Adult social class
(42 yr
s)
Psychological dis
tress
(23-42 yr
s)
Health behaviours
(smoking &
alcohol
misuse)& BMI
(42 yr
s)
Social isolaon
(33 yr
s)
Figure
1
Proposed
conceptual
model
of
pathways
between
child
social
isolation
and
adult
inflammation.
of
participants.
We
used
information
on
participants
from
childhood
through
to
age
44
from
whom
blood
samples
were
taken
(n
=
8233,
87.8%
of
target).
2.2.
Measures
2.2.1.
Adult
inflammation
CRP
was
measured
in
citrated
plasma
by
high-sensitivity
nephelometric
analysis
of
latex
particles
coated
with
CRP-
monoclonal
antibodies
(Elliott
et
al.,
2008).
Inter
and
intra-assay
coefficients
of
variation
were
<10%.
This
is
the
only
inflammatory
marker
measured
in
this
cohort.
Partici-
pants
with
values
of
10
mg/L
(n
=
230),
indicative
of
recent
infection
and
pathology
(Pepys
and
Hirschfield,
2003),
were
removed
from
the
analyses.
CRP
values
were
positively
skewed
and
were
therefore
log-transformed.
2.2.2.
Childhood
social
isolation
As
previously
mentioned
we
conceptualized
social
isola-
tion
as
social
rejection
or
withdrawal.
Taking
Caspi
et
al.’s
(2006)
approach
two
items
from
the
Rutter
behavior
scale
A
(parent-reported)
were
taken
from
the
survey
at
age
7
and
the
same
two
items
also
taken
from
age
11.
The
two
items
were
‘prefers
to
do
things
on
his/her
own
rather
than
with
others’
(social
withdrawal)
and
‘is
bullied
by
other
children’
(social
rejection).
Responses
to
both
items
at
both
surveys
were
‘does
not
apply’
(coded
0),
‘applies
somewhat’
(coded
1)
and
‘certainly
applies’
(coded
2).
An
isolation
score
was
created
which
summed
the
responses
across
all
four
vari-
ables
(two
items
across
both
surveys),
creating
a
variable
with
a
range
of
0
(equivalent
to
no
problems
at
age
7
or
11)
to
8
(equivalent
to
severe
problems
at
age
7
and
11).
Scores
of
7
and
8
were
combined
into
a
single
category
as
there
were
few
participants
with
a
score
of
8
(n
=
10).
Unfor-
tunately
items
directly
comparable
to
those
used
in
Caspi
et
al.
(2006)
were
not
available
until
age
16
in
this
cohort,
as
they
used
a
later
form
of
the
Rutter
behavior
scale
which
was
not
available
in
our
study
in
1965
and
1969.
2.2.3.
Pathway
variables
2.2.3.1.
Socioeconomic
factors.
Educational
attainment
was
taken
as
the
highest
qualification
achieved
by
age
23
(no
qualifications,
Certificate
of
Secondary
Education
(CSE)
or
Ordinary-level
(O-level),
Advanced-level
(A-level),
or
degree/higher
qualification).
For
reference,
CSE/O-level
qualifications
are
broadly
comparable
to
US
high
school
(or
general
equivalency)
degree,
A-level
is
equivalent
to
‘some
college’
and
degree
plus
is
equivalent
to
a
com-
pleted
bachelor’s
degree
or
higher
(Crosby
and
Hawkes,
2007).
Adult
social
class
(highest
in
the
household)
was
measured
using
the
Registrar
General’s
Social
Class
(RGSC)
schema,
based
upon
occupation
(I
professional,
II
man-
agerial/technical,
IIINM
skilled
non-manual,
IIIM
skilled
manual,
IV
semi-skilled
manual
and
V
unskilled).
2.2.3.2.
Psychological
distress.
Psychological
distress
was
measured
using
Rutter’s
Malaise
Inventory
at
ages
23,
33
and
42,
comprising
24
yes/no
items
regarding
emotional
and
88
R.E.
Lacey
et
al.
somatic
symptoms
(Rutter,
1970).
A
hierarchical
factor
anal-
ysis
was
conducted,
firstly
deriving
two
factors
for
each
age
(one
emotional
and
one
somatic)
which
have
previously
been
identified
(Rodgers
et
al.,
1999).
Secondly,
an
overall
factor
was
derived
using
just
the
three
emotional
factors
derived
in
the
previous
step.
This
ensures
that
an
association
seen
with
CRP
is
not
driven
by
physical
symptoms
captured
by
the
Malaise
Inventory
and
also
measures
psychological
distress
across
adulthood.
2.2.3.3.
Adult
social
isolation.
Social
isolation
was
mea-
sured
at
age
33.
Participants
were
asked
to
list
up
to
four
people
who
could
provide
help
in
six
situations
indicative
of
different
forms
of
support
personal
advice,
confiding
sup-
port,
distress
support,
domestic
help,
financial
support
and
household
DIY.
Following
the
approach
by
Matthews
et
al.
(1999)
responses
were
summed
for
the
emotional
(personal
advice,
confiding
support
and
distress
support)
and
practical
support
items
(household
DIY,
financial
support
and
domestic
help).
Participants
with
<3
sources
of
emotional
or
practical
support
were
classified
as
socially
isolated.
2.2.3.4.
Health
behaviors.
Smoking
status
was
collected
at
age
42
as
‘never
smoked’,
‘ex-smoker’
or
‘current
smoker’.
The
CAGE
questionnaire
was
used
to
identify
prob-
lem
drinking
at
age
42.
Participants
with
a
CAGE
score
of
1
or
more
were
identified
as
alcohol
misusers.
Body
mass
index
(BMI)
was
self-reported
at
age
42.
A
sensitivity
analysis
sug-
gested
that
this
correlated
highly
with
measured
BMI
at
ages
33
and
44
(r
=
0.739
BMI33—BMI42,
r
=
0.821
BMI42—BMI44,
r
=
0.783
BMI33—BMI42).
We
used
BMI
from
age
42
as
this
fitted
well
in
our
conceptual
model,
being
prior
to
CRP
mea-
surement
at
age
44
but
after,
or
concurrent
to,
many
other
pathway
variables.
2.2.4.
Covariates
Gender,
childhood
BMI
and
parental
divorce
were
included
as
covariates.
Information
on
parental
divorce
by
age
7
(prior
to
the
childhood
social
isolation
measures)
was
collected
from
the
parental
interview.
Parental
divorce
in
childhood
is
known
to
be
associated
with
raised
CRP
in
adulthood
(Lacey
et
al.,
2013)
and
also
greater
peer
isolation
(Teja
and
Stolberg,
1994).
We
additionally
controlled
for
child
BMI
measured
at
age
7.
Child
BMI
is
also
known
to
be
associated
with
both
increased
adult
health
risk
(Biro
and
Wien,
2010)
and
increased
risk
of
social
isolation
(Strauss
and
Pollack,
2003).
2.3.
Statistical
analysis
Missing
data
were
accounted
for
by
multiple
imputation
by
chained
equations,
imputing
missing
information
on
all
vari-
ables
for
those
participants
who
had
CRP
values.
Imputation
models
included
all
analysis
variables,
variables
predictive
of
missingness
(e.g.
indicators
of
social
disadvantage)
and
the
same
measures
from
preceding
and
subsequent
waves,
where
available.
20
imputed
datasets
were
created
and
the
estimates
from
regression
models
for
each
dataset
were
combined
using
Rubin’s
rules
(1987).
Table
1
shows
a
com-
parison
of
observed
and
imputed
data.
The
imputed
and
observed
data
look
similar,
therefore
suggesting
that
the
imputation
has
been
conducted
appropriately.
The
final
sam-
ple
for
analysis
was
7462.
Multivariate
linear
regression
models
were
run,
firstly
testing
the
association
between
childhood
isolation
and
CRP,
controlling
for
gender,
child
BMI
and
parental
divorce.
As
a
preliminary
step
to
the
analysis
of
pathways
successive
linear
models
were
run
including
each
group
of
pathway
variables
one
at
a
time,
followed
by
a
final
model
control-
ling
for
all
pathway
variables
simultaneously.
To
formally
test
the
proposed
mechanisms
path
analysis
was
conducted.
The
model
was
refined
according
to
modification
indices
and
statistically
non-significant
associations
were
removed
start-
ing
with
that
closest
to
1.
The
‘total
effect’
of
childhood
social
isolation
was
then
decomposed
into
its
‘direct’
and
‘indirect’
components.
To
clarify,
the
path
modeling
term
of
‘indirect
effect’
refers
to
the
proportion
of
the
associa-
tion
between
social
isolation
and
CRP
which
is
attributable
to
the
mechanisms
considered
in
the
model
(e.g.
adult
social
isolation,
health
behaviors,
socioeconomic
factors
and
psy-
chological
distress).
The
‘direct
effect’
does
not
necessarily
mean
a
causal
effect,
but
could
also
represent
residual
con-
founding
or
other
mechanisms
which
were
not
considered
in
the
present
analyses.
The
95%
confidence
intervals
for
the
indirect
and
direct
effects
were
calculated
using
a
bias-
corrected
bootstrapping
method
with
1000
replications.
This
method
is
recommended
to
reduce
bias
associated
with
the
assumption
of
multivariate
normality
when
testing
media-
tion
and
the
standard
errors
of
indirect
effects
(Preacher
and
Hayes,
2008).
3.
Results
3.1.
Childhood
isolation
and
inflammation
Table
1
shows
the
distribution
of
social
isolation,
CRP,
pathway
and
control
variables
in
the
sample.
Results
of
preliminary
regression
analyses
(standardized
coefficients)
testing
the
association
between
childhood
isolation
and
adult
CRP
are
shown
in
Supplementary
Table.
Model
1
shows
that
a
one
standard
deviation
increase
in
child
isolation
increased
adult
CRP
by
0.06
standard
deviations
(equiv-
alent
to
1.07
mg/L
increase
in
CRP)
after
accounting
for
gender,
parental
divorce
and
child
BMI.
Approximately
1.3%
of
the
variability
in
CRP
was
accounted
for
by
these
fac-
tors
(R-squared
=
1.25%).
The
second
model
is
additionally
adjusted
for
socioeconomic
variables
(educational
attain-
ment
and
social
class).
The
association
between
social
isolation
and
CRP
changed
very
little
and
remained
statisti-
cally
significant
(ˇ
=
0.05,
p
=
0.001,
R-squared
=
4.2%).
This
was
also
the
case
in
model
3
which
controlled
for
adult
psy-
chological
distress
(ˇ
=
0.06,
p
=
0.001,
R-squared
=
1.7%).
Adult
social
isolation
appeared
to
be
less
important;
adding
this
variable
in
model
4
did
not
change
the
estimate
for
the
main
association,
nor
was
adult
social
isolation
associated
with
adult
CRP
(p
=
0.922).
Model
5
controlled
for
health
behaviors.
The
association
between
childhood
social
isolation
and
CRP
remained
statistically
significant
(ˇ
=
0.05,
p
=
0.001,
R-squared
=
1.3%).
The
proportion
of
CRP
variance
explained
rose
to
15.1%
in
this
model,
sug-
gesting
that
health
behaviors
and
adult
BMI
explain
more
than
other
factors
in
previous
models.
After
controlling
for
all
pathway
variables
simultaneously
(model
6)
child-
hood
social
isolation
was
still
associated
with
an
increase
Child
social
isolation
and
adult
inflammation
89
Table
1
Sample
characteristics
and
a
comparison
of
observed
and
imputed
data.
%
missing
Observed,
N
(%)
Imputed
(%)a
Main
exposure
Childhood
social
isolation,
median
[IQR]
26.4
2
[1,
4]
2
[1,
4]
Main
outcome
CRP,
median
[IQR],
mg/L
(44
yrs)
0b0.94
[0.46,
2.08]
0.94
[0.46,
2.08]
Socioeconomic
pathway
Educational
attainment
(23
yrs)
No
qualifications
16.7
674
(10.9)
11.7
CSE
2-5/O-level 3080
(49.6) 49.7
A-level
1157
(18.6) 18.3
Higher
qualification/degree 1302
(21.0) 20.4
Household
social
class
(42
yrs)
I
8.7
618
(9.1)
8.7
II
3240
(47.6)
46.4
IIINM
2147
(31.5)
31.8
IIIM
635
(9.3)
10.2
IV
152
(2.2)
2.5
V
18
(0.3)
0.3
Psychological
distress
Adult
malaise
score
(23—42
yrs)
Mean
of
hierarchical
factor
score
[SD]
26.0
0.04
[0.8]
0.01
[0.8]
Adult
social
isolation
Socially
isolated
(33
yrs)
No
11.1
5352
(80.7)
80.1
Yes
1280
(19.3)
19.9
Health
behaviors
Smoking
status
(42
yrs)
Never
smoker
3.2
3302
(45.7)
45.7
Ex-smoker
1887
(26.1)
26.1
Current
smoker
2034
(28.2)
28.3
Problem
drinking
(42
yrs)
CAGE
score
=
0
4.3
4943
(69.2)
69.2
CAGE
score
1
2196
(30.8)
30.8
BMI,
mean
[SD],
kg/m2(42
yrs)
9.5
25.6
[4.5]
25.7
[4.5]
Control
variables
Gender
(0
yrs)
Male
0
3768
(50.5)
50.5
Female
3694
(49.5)
49.5
Parental
divorce
(7
yrs)
No
16.3
6026
(96.5)
96.4
Yes
221
(3.5)
3.6
Child
BMI,
mean
[SD],
kg/m2(7
yrs)
23.8
15.8
[1.7]
15.8
[1.7]
Abbreviations:
IQR,
interquartile
range;
CRP,
C-reactive
protein;
CSE,
Certificate
of
Secondary
Education;
O-level,
Ordinary-level;
A-
level,
Advanced-level;
NM,
non-manual;
M,
manual;
SD,
standard
deviation;
CAGE,
cut
down
annoyed
guilty
eye-opener
items
used
to
screen
for
problem
alcohol
consumption.
aOnly
%s
are
given
for
imputed
data
as
Ns
vary
across
the
20
imputed
datasets.
bNo
missing
values
as
analysis
is
on
those
with
CRP
values.
in
adult
CRP
(ˇ
=
0.04,
p
=
0.002,
R-squared
=
16.1%).
In
all
models
pathway
variables,
with
the
exception
of
adult
social
isolation,
remained
independent
predictors
of
adult
CRP.
Supplementary
table
related
to
this
article
can
be
found,
in
the
online
version,
at
http://dx.doi.org/10.1016/
j.psyneuen.2014.08.007.
3.2.
Pathways
between
childhood
isolation
and
inflammation
In
order
to
test
the
mechanisms
between
childhood
social
isolation
and
adult
CRP
path
analysis
was
conducted.
The
final
model
and
estimates
are
shown
in
Fig.
2
and
Table
2.
Fig.
2
differs
from
Fig.
1
as
associations
were
90
R.E.
Lacey
et
al.
Figure
2
Final
model
of
pathways
linking
child
social
isolation
and
adult
inflammation
(effect
estimates
and
path
descriptions
given
in
Table
2).
Table
2
Standardized
path
coefficients
for
pathways
between
child
social
isolation
and
adult
CRP.
Path
label
Description
ˇaP
a
Child
social
isolation
CRP
0.03
0.002
b
Child
social
isolation
educational
attainment
0.11
0.001
c
Child
social
isolation
psychological
distress
0.07
0.001
d
Child
social
isolation
never
smoker
Ref
Child
social
isolation
ex-smoker
0.08
0.001
Child
social
isolation
current
smoker
0.03
0.280
Child
social
isolation
BMI
0.02
0.036
e
Educational
attainment
psychological
distress
0.19
0.001
f
Educational
attainment
social
class
0.53
0.001
g
Social
class
psychological
distress
0.09
0.001
h
Social
class
CRP
0.12
0.001
i
Psychological
distress
BMI
0.04
0.001
j
Psychological
distress
CRP
0.06
0.001
k
Never
smoker
CRP
Ref
Ex-smoker
CRP
0.02
0.182
Current
smoker
CRP
0.13
0.001
BMI
CRP 0.34
0.001
l
Educational
attainment
BMI 0.14
0.001
Abbreviations:
CRP,
C-reactive
protein;
BMI,
body
mass
index;
Ref,
reference
category.
Model
fit:
TLI
=
0.965;
CFI
=
0.976;
RMSEA
=
0.018.
aConfidence
intervals
are
not
available
with
STDYX
estimates
with
imputed
data
in
MPlus.
Path
model
is
adjusted
for
gender,
parental
divorce
and
child
BMI.
Child
social
isolation
and
adult
inflammation
91
added
in
response
to
modification
indices
and
some
of
the
hypothesized
associations
were
removed
as
they
were
not
statistically
significant.
In
particular,
paths
operating
through
adult
social
isolation
and
problem
alcohol
consump-
tion
were
removed
as
neither
variable
was
associated
with
adult
CRP
(p
=
0.969
and
p
=
0.636,
respectively).
Tw o
addi-
tional
paths
were
added
between
educational
attainment
and
adult
BMI,
and
between
psychological
distress
and
adult
BMI
in
response
to
modification
indices.
Looking
firstly
at
the
socioeconomic
pathway
we
found
that
children
who
were
socially
isolated
tended
to
have
lower
educational
attainment
by
age
23.
Educational
attain-
ment
was
strongly
associated
with
occupational
social
class
at
age
42.
Social
class
was
in
turn
associated
with
CRP
levels
at
age
44
(CRP
levels
decrease
with
increasing
disadvantage
as
indicated
by
social
class).
Adult
psychological
distress
was
also
found
to
be
an
important
mediator;
as
child
iso-
lation
increases
psychological
distress
scores
also
increase.
Psychological
distress
was
in
turn
associated
with
increased
CRP
levels
at
age
44.
Health
behaviors
were
also
found
to
be
important
mediators
of
the
association
between
child
isolation
and
adult
CRP.
In
particular,
socially
isolated
chil-
dren
were
more
likely
to
be
ex-smokers
compared
to
never
smokers,
though
no
differences
were
observed
for
current
smoking
status.
Smoking
status
was
associated
with
CRP,
with
current
smokers
having
higher
CRP
levels
on
average.
Child
social
isolation
was
not
found
to
be
associated
with
alcohol
misuse,
as
mentioned
above.
In
addition,
child
social
isolation
was
associated
with
having
a
higher
BMI
at
age
42,
a
risk
factor
for
increased
CRP.
The
path
model
suggests
that
the
mechanisms
operating
across
the
life
course
are
particularly
complex
as
each
group
of
pathway
variables
are
also
interlinked.
For
instance,
edu-
cational
attainment
is
not
only
associated
with
social
class,
but
also
independently
with
adult
psychological
distress
and
adult
BMI.
Similarly
adult
psychological
distress
is
negatively
related
to
adult
BMI
(as
psychological
distress
increases,
BMI
decreases).
Even
after
considering
all
pathways
simultaneously
child-
hood
social
isolation
was
still
associated
with
increased
CRP
many
years
later
(ˇ
=
0.03,
p
=
0.002,
equivalent
to
1.03
mg/L
increase
per
standard
deviation
of
child
social
iso-
lation),
consistent
with
the
preliminary
regression
results.
The
‘direct’
and
‘indirect
effects’
were
decomposed
and
the
results
suggest
that
the
indirect
mechanisms
investi-
gated
accounted
for
more
than
half
of
the
‘total
effect’
of
child
social
isolation
on
adult
CRP
(total
effect
=
0.058
95%
CI:
0.036,
0.080,
indirect
effect
=
0.031
95%
CI:
0.022,
0.040
therefore
the
%
of
total
effect
explained
by
indirect
mecha-
nisms
=
0.031/0.058
=
0.535
or
53.5%).
The
overall
‘indirect
effect’
was
statistically
significant
(p
0.001).
More
specif-
ically
the
indirect
mechanisms
operating
via
adult
BMI
(child
isolation
adult
BMI
CRP)
and
that
operating
via
socioeconomic
factors
(child
isolation
educational
attain-
ment
social
class
CRP)
appear
to
be
the
most
important
(data
not
shown).
4.
Discussion
These
findings
provide
further
evidence
of
the
life-long
impact
of
childhood
circumstances,
and
adversities,
on
adult
health.
In
particular,
observations
suggest
that
childhood
isolation
is
related
to
an
inflammatory
marker
in
mid-life,
only
part
of
this
relationship
being
accounted
for
by
mediat-
ing
psycho-social
and
behavioral
factors.
C-reactive
protein
has
repeatedly
been
described
as
a
predictor
of
adverse
health
outcomes
(Danesh
et
al.,
2004).
Our
results
therefore
provide
evidence
of
potential
mechanisms
by
which
child-
hood
social
isolation
might
be
related
to
poorer
health
in
adulthood.
Our
findings
extend
those
from
the
Dunedin
study
in
which
CRP
was
measured
at
age
32
(Danese
et
al.,
2009)
and
indicate
that
effects
persist
later
into
life.
Our
findings
also
complement
Caspi
et
al.
(2006)
who
find
that
socially
isolated
children
have
heightened
risk
of
cardiovascular
diseases
as
indicated
by
hypertension,
obesity,
hypercholes-
terolemia,
raised
glycated
hemoglobin
concentration
and
low
maximum
oxygen
consumption.
In
addition,
we
extend
these
previous
observations
to
enhance
the
understanding
of
mediating
pathways
between
child
isolation
and
adult
inflammation.
Our
mediation
model
was
developed
using
previous
literature
linking
childhood
social
isolation
with
adult
health
outcomes,
particularly
drawing
on
Caspi
et
al.
(2006).
Their
hypothesis
was
that
isolated
children
had
a
higher
risk
of
cardiovascular
dis-
ease
because
they
were
more
likely
to
be
socially
isolated
across
the
life
course,
were
more
likely
to
engage
in
risky
health
behaviors
such
as
smoking,
heavy
drinking
and
phys-
ical
inactivity,
and
were
more
likely
to
be
depressed
as
adults.
These
factors
in
turn
were
thought
to
be
associ-
ated
with
indicators
of
increased
cardiovascular
risk.
We
therefore
included
these
same
mediators
in
our
model,
but
additionally
included
a
path
involving
adult
socioeconomic
factors.
There
is
evidence
from
the
same
cohort
study
that
socially
isolated
children
tended
to
have
lower
educational
attainment,
which
has
consequences
for
adult
socioeco-
nomic
attainment
(Brown
and
Taylor,
2008).
Our
findings
suggest
that
this
association
is
indeed
explained
in
part
by
a
complex
web
of
life
course
mech-
anisms.
Tw o
pathways
involved
socioeconomic
factors
and
adult
psychological
distress.
In
particular
social
isolation
in
childhood
predicted
lower
educational
attainment
and
increased
psychological
distress,
as
has
been
previously
described
for
this
cohort
(Katz
et
al.,
2011;
Ammermueller,
2012;
Takizawa
et
al.,
2014).
In
addition
child
social
isolation
was
associated
with
the
uptake
of
risky
health
behaviors,
with
increased
adult
BMI
and
smoking,
as
shown
by
other
studies
of
pathways
from
childhood
adversities
to
adult
inflammation.
For
instance,
Raposa
et
al.
(2014)
found
that
the
association
between
early
adversity
and
adult
inflamma-
tion
was
largely
explained
by
smoking
and
BMI.
Similar
to
our
study
they
found
that
alcohol
consumption
was
not
involved
as
a
mediator,
although
they
looked
at
frequency
of
drinking
rather
than
alcohol
misuse.
The
association
of
ever
smoking
but
not
current
smoking
at
age
42
suggests
that
child
social
isolation
is
associated
with
smoking
initiation
rather
than
cessation.
The
finding
that
child
isolation
is
not
associated
with
adult
isolation
disagrees
with
the
notion
that
childhood
social
isolation
represents
a
developmental
mal-adaption
(Hymel
et
al.,
1990).
However,
this
finding
may
relate
to
the
measure
of
adult
isolation
used
in
this
study.
Recent
data
suggest
that
bullying
in
childhood
is
associated
with
isolation
in
adult
life,
when
participants
in
this
cohort
were
92
R.E.
Lacey
et
al.
aged
50
(Takizawa
et
al.,
2014).
Unfortunately
we
were
not
able
to
use
the
same
measure
of
adult
isolation
as
this
was
only
available
after
measurement
of
CRP.
After
considering
all
mediators
child
social
isolation
was
still
associated
with
increased
CRP
in
mid-life.
After
consid-
ering
all
mediators,
child
social
isolation
was
still
associated
with
increased
CRP
in
mid-life
in
accordance
with
Danese
et
al.
(2009).
The
effect
size
seen
is
clinically
significant
as
a
one
standard
deviation
increase
in
child
social
isola-
tion
(equivalent
to
a
score
of
1.5
on
this
variable)
produced
an
increase
in
CRP
of
1.07
mg/L.
For
comparison,
Danesh
et
al.
(1998)
reported
a
70%
increased
risk
of
coronary
heart
disease
amongst
those
in
the
highest
tertile
of
the
CRP
distri-
bution
(mean
CRP
of
this
group
=
2.4
mg/L).
This
mean
CRP
level
is
approximately
equivalent
to
that
produced
by
a
two
standard
deviation
increase
in
child
social
isolation
in
our
study,
assuming
the
same
associations
of
CRP
and
event
rate.
It
is
possible
that
child
isolation
exerts
a
direct
effect
upon
physiological
functioning
and
that
our
findings
provide
further
evidence
of
the
‘biological
embedding’
of
early
life
stress
to
enact
permanent
changes
in
health
(Hertzman,
1999).
The
mechanisms
by
which
this
occurs
may
include
changes
in
direct
beta
cell
mediated
inflammation
(Nance
and
Sanders,
2007;
Irwin
and
Cole,
2011).
Recent
data
from
animals
and
humans
suggest
that
the
adverse
social
envi-
ronment
(social
disadvantage
in
man
and
‘social
defeat’
in
mice)
results
in
changes
in
blood
cell
composition
and
pro-inflammatory
gene
expression
(Powell
et
al.,
2013)
potentially
leading
to
mechanisms
that
may
predispose
to
changes
such
as
cytokine
induced
‘sickness
behavior’
(Dantzer
and
Kelley,
2007)
or
adverse
health
(Hingorani
and
Casas,
2012).
However,
long-term
changes
in
gene
expres-
sion
in
this
context
have
yet
to
be
described.
Raised
levels
of
inflammatory
markers
may
also
be
due
to
unmeasured
sub-
clinical
health
such
as
atheroma
(Drakopoulou
et
al.,
2009).
However,
we
were
unable
to
account
for
this
in
our
analy-
ses.
It
is
also
possible
that
this
remaining
‘direct
effect’
was
explained
by
residual
confounding
or
other
mechanisms
not
considered
in
this
study.
This
study
is
not
without
its
limitations.
CRP
was
only
measured
at
age
44
therefore
we
do
not
know
from
which
point
this
may
be
raised;
for
example,
it
is
possible
that
CRP
might
be
chronically
raised
from
early
adulthood,
or
some
other
age,
but
we
are
unable
to
distinguish
the
point
of
change
in
this
study.
There
was
no
formal
measure
of
childhood
social
isolation
available,
however
our
measure
comprises
a
question
relating
to
peer
withdrawal
(isola-
tion)
and
a
question
relating
to
peer
rejection
(bullying)
which
likely
represent
different
aspects
of
social
isolation.
When
we
looked
separately
at
each
of
these
questions,
the
associations
we
saw
were
largely
driven
by
the
bully-
ing
item
although
the
other
item
about
preferring
to
be
alone
was
still
associated
with
raised
CRP
without
consider-
ing
the
bullying
item
(results
not
shown).
These
observations
may
support
the
findings
of
Rubin
et
al.
(1991)
who
sug-
gest
that
the
type
of
isolation
in
childhood
may
determine
consequent
outcomes.
Thus
peer
rejection
may
have
more
adverse
sequelae
than
peer
withdrawal
because
this
may
be
more
outside
of
the
child’s
control.
Therefore
our
find-
ings
implicate
a
role
of
low
control
on
inflammatory
marker
levels.
We
do
not
have
a
multi-dimensional
measure
of
child
isolation,
for
example
covering
the
family,
school,
and
religious
ties.
It
might
be
that
some
children
are
well
inte-
grated
within
one
domain,
such
as
the
family,
but
not
in
others,
such
as
with
peers.
We
additionally
do
not
have
measures
of
the
subjective
experience
of
social
isolation,
such
as
loneliness,
in
this
dataset
in
childhood
or
adult-
hood.
Previous
research
by
Cacioppo
and
other
authors
has
shown
that
the
subjective
experience,
particularly
loneli-
ness,
is
more
likely
to
result
in
adverse
health
outcomes
than
objective
measures
of
isolation
(Uchino
et
al.,
1996;
McDade
et
al.,
2006;
Cornwell
and
Waite,
2009).
However,
we
hypothesized
that
the
socially
withdrawn
or
socially
rejected
child
would
have
a
higher
probability
of
remaining
in
an
environment
that
fails
to
provide
sufficient
social
sup-
port.
This
is
likely
to
be
perceived
as
stressful,
which
in
turn
activates
stress
mechanisms
in
the
body
resulting
in
increased
CRP
levels
by
mid-life.
We
also
do
not
take
account
of
comorbidities
contemporaneous
to
our
measure
of
CRP.
However,
data
from
the
Health
Survey
for
England
(Craig
and
Mindell,
2012)
have
shown
that
the
prevalence
of
disease
is
low
for
this
age
group.
For
example,
the
prevalence
of
diabetes
is
<3%,
ischemic
heart
disease
or
stroke
is
<2%
and
around
6%
have
any
cardiovascular
disease.
It
is
possible
that
the
association
is
confounded
by
sub-clinical
illness
leading
to
fatigue
and
reduced
engagement
with
others.
However,
we
were
not
able
to
capture
this
in
our
study.
This
study
also
has
a
number
of
strengths,
such
as
the
use
of
a
large
sample
representative
of
the
British
population
of
a
similar
age.
Missing
data
were
accounted
for,
thereby
reducing
bias
associated
with
differential
non-response.
The
data
were
collected
prospectively
thereby
minimizing
recall
bias
and
offered
a
long
follow-up
period
of
44
yrs.
In
addition,
path
analysis
is
more
appropriate
than
standard
regression
techniques
for
examining
explanatory
pathways,
therefore
adding
strength
to
the
literature
base
on
the
mechanisms
linking
childhood
adversities
to
later
health.
In
conclusion,
child
social
isolation
is
associated
with
increased
adult
inflammation
as
measured
by
CRP.
This
is
explained
in
part
through
complex
mechanisms
acting
across
the
life
course.
Identification
and
interventions
targeted
toward
isolated
children
may
help
reduce
long-term
adult
health
risk.
In
particular,
interventions
toward
preventing
smoking
uptake,
maintaining
healthy
BMIs
and
encouraging
educational
attainment
may
be
useful.
Further
research
is
also
required
to
investigate
pathways
into
child
isolation,
as
interventions
targeted
toward
these
may
also
help
alleviate
health
risk
in
later
life.
Role
of
funding
source
This
research
was
supported
by
the
European
Research
Coun-
cil
and
the
UK’s
Economic
and
Social
Research
Council.
Conflicts
of
interest
None
of
the
authors
report
any
conflict
of
interest.
Acknowledgements
Thanks
to
participants
of
the
National
Child
Development
Study
and
to
the
study
team.
The
data
are
supplied
by
the
Economic
and
Social
Data
Service
(ESDS).
Those
who
carried