ARTICLE IN PRESS

0277-9536/$ - se

doi:10.1016/j.so

�Tel.: +44 24

E-mail addr

Social Science & Medicine 62 (2006) 1250–1259

www.elsevier.com/locate/socscimed

Explaining the social gradient in smoking in pregnancy: Earlylife course accumulation and cross-sectional clustering of social

risk exposures in the 1958 British national cohort

Nick Spencer�

School of Health and Social Studies and Warwick Medical School, University of Warwick, Coventry CV4 7AL, UK

Available online 29 August 2005

Abstract

Smoking in pregnancy is a major determinant of low birthweight and a range of adverse infant health outcomes.

There is a well-established social gradient in smoking in pregnancy in the US and northern Europe. Social gradients in

health-related behaviours may result from longitudinal accumulation and cross-sectional clustering of social risk

exposures. There is, however, no published confirmation of this explanation in empirical data with smoking in

pregnancy as the outcome. This study aimed to test the effects of longitudinal accumulation and cross-sectional

clustering of social risk exposures on smoking in pregnancy using data on the first pregnancies of 3163 female members

of the 1958 British national cohort. Social class at birth and aged 11 years was used to create three dichotomous

variables representing cumulative social class (both manual, one manual and one non-manual, both non-manual) early

in the lifecourse. Cross-sectional clustering of social risk was represented by four dichotomous variables created from

combinations of maternal age (o20 vs. 20+), own social class (manual vs. non-manual) and educational attainment

(low vs. other). Cumulative social class in early childhood was associated with smoking in pregnancy in bivariate

analysis but not after adjustment for cross-sectional clustering of social risk exposures. However, women who had been

in the manual social groups at birth and 11 years were at increased risk of cross-sectional clustering of social risk

exposures around pregnancy suggesting a pathway from early lifecourse risk exposure to social risk factors associated

with a high risk of smoking in pregnancy. These findings suggest that the social gradient in smoking in pregnancy

results from longitudinal accumulation and cross-sectional clustering of social risk exposures. Interventions aimed at

reducing social inequalities in smoking in pregnancy need to account for cumulative and cross-sectionally clustered

effects of social risk exposures.

r 2005 Published by Elsevier Ltd.

Keywords: Pregnancy smoking; Social gradient; Cumulative social risk; Cross-sectional clustering of social risk; UK

Introduction

Smoking in pregnancy has been associated with a

range of adverse outcomes for the mother and her infant

including increased risk of obstetric complications

e front matter r 2005 Published by Elsevier Ltd.

cscimed.2005.07.026

76 523 167; fax: +44 2476 524 415.

ess: n.j.spencer@warwick.ac.uk.

(Cnattingius, 2004; Himmelberger, Brown, & Cohen,

1978; Wong & Baumann, 1997), low birthweight

(Kramer, Seguin, Lydon, & Goulet, 2000; Meis et al.,

1997), preterm delivery (Kramer, 1987) and neonatal

morbidity and mortality (Cnattingius, 2004; Meyer,

Jonas, & Tonascia, 1976; Salihu, Aliyu, Pierre-Louis, &

Alexander, 2003). The adverse effects of smoking in

pregnancy stretch into infancy with an increased risk of

ARTICLE IN PRESSN. Spencer / Social Science & Medicine 62 (2006) 1250–1259 1251

sudden unexpected infant death (Blair, Fleming, &

Becon, 1996), hospital admission (Royal College of

Physicians of London, 1992; WHO, 1999) and respira-

tory illness (Chen, Li, & Yu, 1986). Although the

magnitude and independence of some of these effects

has been challenged on the grounds of the potential for

residual confounding by socioeconomic status (Davey

Smith & Phillips, 1992; Logan & Spencer 1996), there

can be little doubt that smoking in pregnancy carries

negative health consequences.

Smoking among women in many countries is a

relatively recent phenomenon occurring initially among

educated, emancipated women (Graham, 1996). Since

the 1950s, when concerns about the health impact of

smoking started to emerge, smoking amongst women

has become increasingly associated with social disad-

vantage particularly in the USA, and Northern Europe

(Graham, 1996). In Southern Europe, smoking among

women is a more recent phenomenon with the result that

the gradient associated with social disadvantage is only

just starting to emerge (Schiaffino et al., 2003). Smoking

initiation in adolescence shows a less marked though still

detectable social gradient (Sweeting & West, 2001) but

cessation in late adolescence and early adulthood is

strongly socially patterned (Jarvis & Wardle, 1999).

Smoking in pregnancy reflects the same social gradient

as that seen for women’s smoking being both more

prevalent and heavier in more disadvantaged groups

(Graham, 1994; Haglund, Cnattingius, & Nordstrom,

1993; Kleinman & Kopstein, 1987). Women in higher

socioeconomic groups are more likely to quit in

pregnancy (Dolan-Mullen, 1999).

Because of its multiple adverse effects on infant and

child health as well as its relatively high prevalence and

steep social gradient, smoking in pregnancy is likely to

make a major contribution to social inequalities in

pregnancy outcome (Kramer et al., 2000) and in infancy

and childhood. The impairment of intrauterine growth

associated with smoking (Kramer, 1987; Kramer et al.,

2000) leads to lower birthweight that has been linked to

a range of adverse later childhood and adult health

outcomes (Barker, 1992, 1998). Kramer et al. (2000)

suggest that smoking in pregnancy is responsible for

about a quarter of intrauterine growth retardation and a

tenth of preterm birth in developed countries with

prevalence of smoking in pregnancy of around 25%.

Understanding of the role of smoking in pregnancy as a

determinant of health inequalities in infancy and child-

hood has been hindered by the conclusion that smoking

in pregnancy ‘explains’ or ‘accounts for’ social differ-

ences in pregnancy and childhood outcomes (Brooke,

Anderson, Bland, Peacock, & Stewart, 1989; Mitchell et

al., 1993). This conclusion depends on the elimination of

an independent effect of social factors on the outcome of

interest after adjustment for smoking in pregnancy.

However, this approach to studying the association

between health, social circumstances and health beha-

viours, as well as being vulnerable to residual confound-

ing (Davey Smith & Phillips, 1992; Logan & Spencer,

1996), fails to account for the mechanisms by which

social factors exert their influence on health (Kuh,

Power, Blane, & Bartley, 1997; Rose, 1992; Spencer,

2003). Poor socioeconomic conditions do not directly

‘cause’ adverse health outcomes but exert their effects

through a range of mediating factors one of which is

smoking.

Given the importance of smoking as a mediator of

social gradients in pregnancy and later outcomes, an

understanding of how social differences in pregnancy

smoking arise and are perpetuated may assist in

addressing social disparities in health. The development

of lifecourse epidemiology (Kuh & Ben-Shlomo, 1997)

has generated interest in the cumulative effects of risk

exposures over time and their contribution to social

gradients in a range of health outcomes (Blane, 1999;

Davey Smith, Hart, Blane, Gillis, & Hawthorne, 1997;

Hertzman, Power, Matthews, & Manor, 2001; Kuh et

al., 1997; Lynch, Kaplan, & Shema, 1997; Power,

Manor, & Fox, 1991). Davey Smith et al. (1997)

demonstrated a stepwise gradient in all-cause mortality

among men in the West of Scotland Collaborative Study

using a cumulative social class measure—compared with

men who were in non-manual social class on all three

occasions, those in two non-manual and one manual

had an age-adjusted relative death risk of 1.29, those in

two manual and one non-manual, a risk of 1.45 and

those in all three manual, a risk of 1.71. Berney, Blane,

Davey Smith, and Holland (2000) examined the effects

of lifetime hazard exposure scores based on socially and

biologically plausible risk exposures on health in early

old age and showed a graded effect on health consistent

with accumulation of risk exposures. In addition to

accumulation longitudinally, risk exposures also cluster

cross-sectionally (Bartley, Blane, & Montgomery, 1997).

Hertzman et al. (2001), based on data from 1958 British

national cohort (the same dataset as I have used for this

study), construct an integrated model of the determi-

nants of adult health combining cumulative lifecourse

factors and contemporary circumstances. I have not

been able to find any other studies that set out to analyse

cumulative and cross-sectional effects on health using

data from the same study.

Together, cumulative and cross-sectional clustering of

social risk exposures provide a plausible explanation for

the development and perpetuation of social gradients.

They also help to explain the persistence of social

gradients in health outcomes in countries with high

levels of relative social equality such as Sweden (Elmen,

Hoglund, Karlberg, Niklasson, & Nilsson, 1996; Vagero

& Lundberg, 1989).

Lynch, Kaplan, & Salonen (1997) report an associa-

tion between adult health-related behaviour and early

ARTICLE IN PRESSN. Spencer / Social Science & Medicine 62 (2006) 1250–12591252

social risk exposures based on a longitudinal in study of

Finnish men. Similar findings were reported in a

Norwegian study (Arnesen & Forsdahl, 1985). Ermisch,

Francesconi, and Pevalin (2001) analysed the results of

the British Household Panel Survey 1991–1999 and

reported that smoking by young men was related to

having lived in persistent poverty (being in a household

with an income o60% of the median household income

in three out of four years) in their preschool years. The

effect in young women suggested a relationship but was

not statistically significant at the 5% level. Hobcraft

(2003), in his analysis of the 1958 British national

cohort, reported an association of childhood disadvan-

tage with risk of smoking at age 33 but not age 23 as did

Power and Hertzman (1999). When a series of inter-

mediate factors such as educational attainment at 16 and

frequent school absences are added into the multiple

logistic regression model, the effects of childhood

disadvantage are lost (Hobcraft, 2003). Other studies

(Blane et al., 1996; Notkola, Punsar, Karvonen, &

Haapakoski, 1985) failed to find any influence of

childhood socioeconomic position on smoking beha-

viour in adult life. These studies suggest that smoking in

adulthood is more dependent on social position in

adulthood than parental social position. Few studies

report on cumulative or intergenerational influences on

smoking in pregnancy. An intergenerational study in

Sweden (Hofvendahl, Hofvendahl, & Wallender, 1997)

reported a two-fold increase in risk of smoking in

pregnancy among women whose own mothers smoked

in pregnancy after adjustment for a social environment

index.

This paper aims to explore both the cumulative and

cross-sectionally clustered effects of social risk exposure

on smoking in the first pregnancy of female members of

the 1958 British national cohort study (an intergenera-

tional cohort following families across more than one

generation) (Ferri, 1993) who have had at least one

pregnancy by the age of 33 years. The results should

help to inform health and social policy initiatives aimed

at reducing smoking in pregnancy and reducing inequal-

ities in pregnancy and infancy outcomes (Department of

Health (UK), 1998, 2002). Improved understanding of

the antecedents of the social gradient in pregnancy

smoking may enable policy makers to make more

effective policy choices and set realistic targets.

Methods

Study population

The 1958 British national cohort study (National

Child Development Study) (Ferri, 1993) collected data

on 17,414 births (8960 girls and 9593 boys) during 3–9

March 1958. Subsequent sweeps were undertaken when

cohort members were 7, 11, 16, 23 and 33 years old.

Data were collected through birth records and face-to-

face interviews with parents at birth, face-to-face inter-

views with parents only when the cohort members were

7, 11 and 16 years and face-to-face interviews with

cohort members themselves at 23 and 33 years. Data

were available on 15,468 cohort members (7564 girls and

7904 boys) at 7, 15,503 at 11 (7565 girls and 7938),

14,761 at 16 (7159 girls and 7602 boys), 12,537 at 23

(6519 women and 6018 men) and 10,414 at 33 (5467

women and 4947 men). Social class data were collected

at birth for cohort members and at subsequent sweeps.

Data from birth and the sweeps at 11 and 23 years were

used to derive the cumulative and clustered social risk

exposures. Social class at birth was based on the

occupation of the cohort member’s father at the time

of her birth. Educational attainment was derived from

questions in the sweep at 23 years related to highest

school and post-school qualifications. All pregnancies to

female cohort members up to the age of 33 years were

recorded. Data on age of first pregnancy and smoking in

first pregnancy were collected in the 23 and 33 year

sweeps. At least one pregnancy (including abortions and

miscarriages) was reported by 4673 cohort members.

This study reports the results on 3163 first pregnancies

for which complete data were available. Data on social

class at age 23 were missing in 958 women (20.5% of all

first pregnancies) and on social class at 11 years in 552

women (11.8% of all first pregnancies). Social class

distribution in the sample was similar at birth (23% non-

manual) and at 11 years (31.8% non-manual) to that of

all female members of the original cohort (25.2% non-

manual at birth and 33.9% at 11 years) and the women

who were recorded as having at least one pregnancy

(23.1% non-manual at birth and 32.5% at 11 years).

Measures of social risk

Registrar General’s social class at cohort member’s

birth, in mid-childhood (11 years) and at 23 years were

collapsed to manual (social classes IIIm, IV,V and

unclassified) and non-manual (social classes I, II and

IIInm) in order to simplify the construction of categories

of cumulative social class and to ensure sufficient

numbers for analysis. The Registrar General’s social

class is based on occupation of the head of the

household. Manual social class in the Registrar Gen-

eral’s classification is a reliable proxy for low income

and relative social disadvantage—manual workers have

lower average incomes and are less likely to own their

houses (Reid, 1998). Three dichotomous variables were

created to represent cumulative social class in the early

lifecourse from birth to 11 years (manual twice, manual

once and non-manual twice) using the method described

by Katz (1999) and used by Davey Smith et al. (1997) in

their study of all-cause mortality among Scottish men.

ARTICLE IN PRESSN. Spencer / Social Science & Medicine 62 (2006) 1250–1259 1253

Educational attainment by the age of 23 was dichot-

omised to no school or post-school qualifications vs. any

qualifications. This dichotomy was chosen partly to

identify the most educationally disadvantage group of

women and also because of difficulties establishing a

hierarchy among relatively low-level school qualifica-

tions and post-school trade qualifications. Teenage

pregnancy was defined as pregnancy before the age of

20 years. Four dichotomous variables were created to

represent cross-sectional clustering of social risk with

various combinations of social class at 23 years, teenage

pregnancy and educational qualifications—all three risk

exposures, two risk exposures, one risk exposure and no

risk exposures.

Smoking measures

Data were available on the cohort member’s self-

reported daily consumption of cigarettes in her first

pregnancy. Women were asked if they had smoked

throughout pregnancy and if so, how many cigarettes

they smoked per day. For the purposes of this study,

smoking was defined as consumption of at least one

cigarette per day throughout pregnancy. This variable

was dichotomised to smoking in pregnancy vs non-

smoking in pregnancy. Data were not available to

distinguish never from former smokers.

The smoking status of the cohort member’s mother

during her pregnancy carrying the cohort member was

extracted from data collected at the time of the cohort

member’s birth. This variable was dichotomised in the

same way as cohort member’s smoking and introduced

into the logistic regression models as a potential

confounder.

Statistical analysis

Cumulative social risk from birth to early childhood

To estimate the effects of cumulative social class early

in the lifecourse on smoking in pregnancy, odds ratios

with 95% confidence intervals of the cohort member

smoking in pregnancy by cumulative social class at birth

and 11 years were calculated in a logistic regression

model. As odds ratios overestimate the effect size of risk

exposures, they were converted to relative risks using the

formula proposed by Zhang and Yu (1998):

RR ¼ OR=ð1� PoÞ þ ðOR� PoÞ,

where RR is the relative risk, OR the odds ratio and Po

the prevalence of pregnancy smoking.

Cross-sectional clustering of social risk around the time of

pregnancy

The effect of cross-sectional clustering of social risk

on smoking in pregnancy was estimated in a logistic

regression model including the four dichotomised

variables created from combinations of social class at

23 years, educational attainment and teenage pregnancy.

Constructing a model of cumulative and clustered social

risk exposures on smoking in pregnancy

In order to study the direct and indirect effects of

cumulative low social class (manual at birth and 11

years) and cross-sectional clustering of social risk

exposures at the time of pregnancy on smoking in

pregnancy, a model, using the concept of the conceptual

hierarchical framework described by Victora, Huttly,

Fuchs, & Olinto (1997), was constructed in the following

steps:

1.

The independent (direct) effects of cumulative low

social class early in the lifecourse and clustering of

social risk exposures around pregnancy on smoking

in pregnancy were tested in a logistic regression

model.

2.

To test for an indirect effect of cumulative low social

class early in the lifecourse on smoking in pregnancy

acting through the clustering of social risk exposures

around pregnancy, logistic regression models were

fitted on each level of clustered social risk exposures

(all three, two, one and none) with cumulative social

class early in the lifecourse as the independent

variable.

All analyses were undertaken in SPSS version 10

(SPSS, 1999).

Results

The final study sample included 3165 female cohort

members who had given birth by the age of 33 years with

complete data. The characteristics of the sample are

shown in Table 1. There was a steady fall from birth

through age 11 to 23 years in the proportions of women

in the sample in the manual social class groups. This

pattern is consistent with the reduction in manual social

class groups noted in the UK between the early 1960s

and the 1980s (Heath & Payne, 1999).

Cumulative social class from birth to 11 years and

clustered cross-sectional social risk exposures both

demonstrate stepwise gradients for smoking in preg-

nancy in bivariate analysis (Table 2). When cumulative

social class and clustered cross-sectional risk exposure

were adjusted for each other in a multivariate logistic

regression model fitted on pregnancy smoking and

controlled for own mother’s smoking in pregnancy,

only clustered cross-sectional risk exposures retained a

significant relationship with pregnancy smoking (Table

2). Cumulative social risk early in the lifecourse does not

have an effect on pregnancy smoking independent of

ARTICLE IN PRESSN. Spencer / Social Science & Medicine 62 (2006) 1250–12591254

cross-sectional clustered social risk exposures around

the time of pregnancy.

Table 3 shows the relationship of cumulative social

risk exposure in early childhood and cross-sectional

social risk exposures around pregnancy. There is a

Table 2

Corrected relative risk of smoking in pregnancy by cumulative socia

sectional clustering of social risk exposures around pregnancy

U

sm

Cumulative social class (birth to 11 years)

Manual social class at birth and 11 years (n ¼ 2007) 1

Manual social class once only (n ¼ 576) 1

Non-manual social class at birth and 11 years (n ¼ 580) 1

w2 for linear trend (p value) 9

Clustered social risk exposures around pregnancy

All three social risk exposures (n ¼ 113) 2

Two social risk exposures (n ¼ 485) 1

One social risk exposure (n ¼ 1259) 1

No social risk exposure (n ¼ 1306) 1

w2 for linear trend (p value) 3

�Cumulative social class (birth to 11) and clustered social risk arou

mother’s smoking in pregnancy.

Table 1

Characteristics of women in the sample

Characteristics of women in the

sample (n ¼ 3163)

Number (%)

Social class at birth

Manual 2434 (76.9)

Non-manual 729 (23.1)

Social class at 11

Manual 2156 (68.2)

Non-manual 1007 (31.8)

Social class at 23

Manual 1680 (53.2)

Non-manual 1483 (46.8)

Age at first pregnancy

o20 408 (12.9)

20+ 2755 (87.1)

Educational attainment

No qualifications 480 (15.2)

Qualifications 2683 (84.8)

Smoked in pregnancy

Yes 1152 (36.4)

No 2011 (63.6)

Mother smoked in pregnancy

Yes 1160 (36.7)

No 2003 (63.3)

stepwise increase in risk of exposure to a cluster of all

three social risk exposures around pregnancy from

women who were in non-manual social classes at birth

and 11 years through women who moved between

manual and non-manual groups in early childhood and

those who were in manual social class families at both

times in early childhood. Similar stepwise risk patterns

are seen for exposure to clusters of two social risk

factors and exposure to a single social risk factor around

pregnancy.

The model shown in the Fig. 1 illustrates the effects of

cumulative early social risk exposure on pregnancy

smoking. Although cumulative early social risk exposure

has no direct effect on pregnancy smoking independent

of cross-sectional clustered social risk exposure, it exerts

an indirect effect on pregnancy smoking through its

effect on social risk exposures around the time of

pregnancy. Women in manual social class households at

birth and 11 years of age were twice as likely, as women

in non-manual social class households at both times in

early childhood, to experience all three clustered social

risk exposures around pregnancy, twice as likely to

experience two social risk exposures and 33% more

likely to experience one social risk exposure.

Discussion

Cumulative social risk exposure early in the lifecourse

appears to have little direct effect on risk of smoking in

pregnancy once cross-sectional clustered social risk

exposures have been accounted for. At first glance, this

would appear to support the conclusion that smoking in

adulthood is related to current not past social risk

exposure (Blane et al., 1996). Undoubtedly, the findings

l class early in the life course (birth to 11 years) and by cross-

nadjusted RR (95% CI) of

oking in pregnancy

Adjusted RR (95% CI) of

smoking in pregnancy�

.34(1.20,1.45) 1.11(0.97,1.26)

.27(1.09,1.44) 1.16(0.99,1.33)

.00 (reference) 1.00 (reference)

.14 (p ¼ 0:003)

.09 (1.85,2.27) 2.05(1.82,2.25)

.87 (1.74,2.01) 1.85(1.70,1.98)

.45 (1.34,1.57) 1.44(1.31,1.56)

.00 (reference) 1.00 (reference)

3.0 (po0:0001)

nd pregnancy adjusted for each other and for cohort member’s

ARTICLE IN PRESS

Table 3

Relationship of cumulative social class at age 11 with cross-sectional social risk exposures around pregnancy

Cumulative social class at age

11

Relative risk (95% CI) of

exposure to all three cross-

sectional risk factors

Relative risk (95% CI) of

exposure to two cross-

sectional risk factors

Relative risk (95% CI) of

exposure to one cross-sectional

risk factor

Both manual (n ¼ 2007) 2.00 (1.47,2.38) 2.06 (1.82,2.25) 1.33 (1.19,1.47)

One manual and one non-

manual (n ¼ 576)

1.64(0.99,2.19) 1.57(1.23,1.87) 1.27(1.10,1.44)

Both non-manual (n ¼ 580) 1.00 (reference) 1.00 (reference) 1.00 (reference)

w2 for linear trend (p value) 18.3 (po0:0001) 94.2 (po0:0001) 21.0 (po0:0001)

2.00*** 2.05***

2.06***

1.85***

1.33**

1.44***

NS

*** p<0.0001 ** p<0.005 NS = non-significant pathway

Manual social class at birth & 11 years Smoking in

pregnancy

All 3 social risk exposures

2 social riskexposures

1 social risk exposure

Fig. 1. Model using a conceptual hierarchical framework: from

cumulative low social class early in the lifecourse through

clustered social risk exposures around pregnancy to smoking in

pregnancy—associations between variables in pathway repre-

sented by relative risks and p values compared with the most

advantaged category. Relative risks of pregnancy smoking by

cross-sectional clustered social risk exposures are adjusted for

cumulative social class from birth to 11 years.

N. Spencer / Social Science & Medicine 62 (2006) 1250–1259 1255

of this study suggest that social risk exposure around the

time of pregnancy is a powerful determinant of smoking

in pregnancy. However, the results also show that

cumulative social risk, represented here by cumulative

social class from birth to mid-childhood (11 years), has

an indirect effect on smoking in pregnancy through its

effect on social risk exposure at the time of pregnancy.

The greater the early lifecourse exposure to social risk

the greater the chances of women experiencing cross-

sectional clustered social risk exposures that are directly

associated with an increased risk of smoking in

pregnancy. Exposure to three social risk factors around

the time of pregnancy increases the risk of pregnancy

smoking twofold, exposure to two risk factors by 85%

and exposure to one by 44% compared with exposure to

none. Manual social class at birth and 11 years increases

two fold the risk of experiencing all three social risk

exposures around the time of pregnancy, experiencing

two social risk exposures twofold and experiencing one

risk exposure by 33%. Thus, early childhood cumulative

social risk sets up a pathway to clustered cross-sectional

risk that, in turn, is associated with an increased risk of

pregnancy smoking.

Existing studies of the social patterning of smoking in

pregnancy tend to focus on current circumstances.

Although a number of studies have reported increased

risk of smoking associated with cross-sectional cluster-

ing of risk exposures (Graham & Blackburn, 1998;

Jarvis & Wardle, 1999; Marsh & McKay, 1994), none

have explored the links between cross-sectional cluster-

ing and cumulative social risk. The study reported here

seems to be the first to explore the social gradient in

pregnancy smoking by examining both longitudinal

accumulation and cross-sectional clustering of social

risk exposures.

Limitations of the study

The relatively high attrition rate may jeopardise the

validity of the findings. However, as discussed above the

social composition of the sample studied appears to be

similar to that of all female cohort members and all

women with recorded pregnancies at birth and age 11

years. Differential attrition in longitudinal studies leads

to loss of lower social groups. The effect of such attrition

in this study would be to bias the results towards the

null.

Misclassification bias may have been introduced by

the use of social class at 23 years as the measure of adult

social class. Some women will have had their first

pregnancy some time before or after the age of 23 years

and their social class may have changed. As the general

trend is towards upward social mobility among these

women, it is probable that, if misclassification has

occurred, non-manual women would be classified as

manual. This would also tend to bias towards the null.

ARTICLE IN PRESSN. Spencer / Social Science & Medicine 62 (2006) 1250–12591256

Smoking data were self-reported and no biochemical

measure was available with the result that some smokers

are likely to have been misclassified as non-smokers. If,

however, misclassification is not systematically biased by

social group as Graham and Owen (2003) have

demonstrated, the use of self-reported smoking in

pregnancy in this study is unlikely to bias the findings

of this study. The effects of social risk exposures on

smoking in pregnancy may also have been confounded

by inclusion in the study sample of women who

subsequently had an abortion. It is possible that women

contemplating an abortion would be less likely to quit

smoking once they became aware of their pregnancy.

Smoking prevalence was similar in those pregnancies

ending in abortion to those going to term suggesting

that this potential source of confounding is unlikely to

have greatly affected the study findings.

Among women having pregnancies before their 20th

birthday, early pregnancy may have been associated

with reduced chances for upward social mobility.

Teenage pregnancy might then be a cause rather than

a consequence of manual social class at 23 years of age.

There is evidence from the data set that teenage

pregnancy made no difference to the proportion of

women moving from non-manual social class at 11 to

manual at 23 years; however, pregnancy before the 20th

birthday was associated with less likelihood of a move

from manual to non-manual social groups between ages

11 and 23. The possibility of reverse causality may have

biased the study results although the effect is unlikely to

be great.

The extent of social risk exposure at each stage of the

lifecourse is likely to be under-estimated by collapsing

social class groups down to manual and non-manual. As

discussed in the methods section, this decision was

prompted by the need to simplify the categories of

cumulative social risk exposure and to ensure sufficient

numbers in each group for meaningful analysis. The

likely impact on the results is an under-estimation of the

effects of cumulative social risk in early childhood.

The pregnancies studied occurred between the mid-

1970s and 1991. It is possible that the social patterning

of smoking in pregnancy has changed since these

pregnancies were completed with the result that the

findings may only have limited significance for current

approaches to reduce smoking in pregnancy. However,

the published evidence indicates that the social gradient

in pregnancy smoking in the UK has become more not

less marked in recent years (Graham, 1996) suggesting

that the models of risk exposure reported here are likely

to be relevant to current debates on smoking in

pregnancy. Variations in social patterning of pregnancy

smoking between countries and differing patterns of

social mobility are likely to limit the generalisability of

the findings to countries other than the UK. However,

for countries that do have similar social patterning of

pregnancy smoking and similar levels of social risk

exposure over the lifecourse and around pregnancy, the

findings may have more relevance.

Strengths of the study

The main strength of this study is the prospective

collection of social class and other data from birth

through to the first pregnancy. This eliminates problems

of recall bias and retrospective data collection. In

addition, the collection of social class data during

childhood allows a robust measure of cumulative social

class, incorporating key periods in the early lifecourse,

to be derived.

Implications of these findings for policy development and

health promotion interventions

Smoking in pregnancy is a major determinant of

impaired fetal growth particularly among women in

developed countries where levels of cigarette consump-

tion in pregnancy are relatively high (Kramer et al.,

2000). Because of its known social gradient, smoking in

pregnancy is also likely to make a significant contribu-

tion to social disparities in pregnancy outcomes (Kramer

et al., 2000). For this reason, health policy and

promotion initiatives designed to reduce social dispa-

rities in pregnancy outcomes place heavy reliance on

smoking in pregnancy cessation especially amongst low

income women (Department of Health, 2002).

Interventions in pregnancy to promote smoking

cessation are effective (Lumley, Oliver, Chamberlain,

& Oakley, 2001); however, low socioeconomic status

women are least likely to participate and most likely to

drop out of these programmes (Dolan-Mullen, 1999).

The early lifecourse risk exposures shown to be

important in determining risk of pregnancy smoking in

this study may partly explain the differential social

uptake and differential social success rate of smoking

cessation programmes. The social patterning of smok-

ing, that results in earlier initiation, higher cigarette

consumption and greater levels of nicotine dependency

among lower social groups (Jarvis & Wardle, 1999) is

rooted in early social experience and is likely to be

sensitive to cumulative lifecourse social risk exposure.

Low socioeconomic status pregnant women, who have

been exposed to higher cumulative social risks, are likely

to be more dependent on nicotine, less likely to

contemplate cessation and less likely to succeed even if

they attempt to quit. Stress has been implicated as a

factor increasing the risk of smoking in pregnancy

(Kramer et al., 2000; Sheehan, 1998) and low income

women are thought to experience more acute and

chronic stress in pregnancy (Kramer et al., 2000). Social

support to reduce the adverse effects of stress may

influence smoking in pregnancy but has not been shown

ARTICLE IN PRESSN. Spencer / Social Science & Medicine 62 (2006) 1250–1259 1257

to be effective in reducing adverse pregnancy outcomes,

particularly low birthweight and preterm birth, asso-

ciated with smoking (Hodnett & Fredericks, 2003).

Lumley et al. (2001) acknowledge in their systematic

review of smoking cessation interventions that the role

of these interventions in reducing social inequality in

pregnancy smoking and in adverse pregnancy outcomes

remains unclear.

Although cumulative social risk in early childhood

has no direct effect on pregnancy smoking, the indirect

effect it exerts on social factors that directly influence

pregnancy smoking suggests the need to take cumulative

social risk into account when trying to reduce social

inequalities in pregnancy smoking. These results are

consistent with the recommendations of the Working

Party on Health Inequalities chaired by Sir Donald

Acheson (Department of Health, 1998) that improving

social conditions in early childhood is the key to

reducing inequalities in future pregnancy outcomes.

This is not to suggest that efforts to reduce current social

disparities should be abandoned but to acknowledge

that the impact of adverse social risk exposure over the

lifecourse may take time to eradicate and may be less

sensitive to interventions directed solely at the index

pregnancy (Spencer, 2003).

This study contributes to knowledge of the ante-

cedents of smoking in pregnancy and adds to the

literature exploring the impact of lifecourse social risk

exposures on behaviour that has a major impact on

health of women and their children. It is also amongst

the few studies (Hertzman et al., 2001) that explore the

combined effects of longitudinal accumulation and

cross-sectional clustering on health and health-related

behaviour in early adulthood. Smoking in pregnancy

has an effect, through its influence on fetal growth and

physiology, on the health of the newborn infant and into

adult life. Thus, early childhood social risk exposures

can be seen to establish an intergenerational chain of

risk stretching from the mother’s own childhood

through to the adult life of her child.

Acknowledgements

I acknowledge the UK Data Archive, University of

Essex, Colchester for kind permission to use the 1958

National Childhood Development Study data and all

those involved in the provision, collection and prepara-

tion of this invaluable dataset.

Funding: none.

Conflict of interest: none.

References

Arnesen, E., & Forsdahl, A. (1985). The Tromso heart study:

Coronary risk factors and their association with living

conditions during childhood. Journal of Epidemiology and

Community Health, 39, 210–214.

Barker, D. J. (Ed.). (1992). Fetal and infant origins of adult

disease. London: BMJ Publications.

Barker, D. J. (1998). Mothers, babies and disease in later life

(2nd ed.). Edinburgh: Churchill Livingstone.

Bartley, M., Blane, D., & Montgomery, S. (1997). Health and

the life course: Why safety nets matter. British Medical

Journal, 314, 1194–1196.

Berney, L., Blane, D., Davey Smith, G., & Holland, P. (2000).

Lifecourse influences on health in early old age. In H.

Graham (Ed.), Understanding health inequalities (pp. 79–95).

Buckingham: Open University Press.

Blair, P. S., Fleming, P. J., Bacon, C., et al. (1996). Smoking

and the sudden infant death syndrome: Results from a case-

control study for the confidential enquiry into stillbirths and

deaths in infancy. British Medical Journal, 313, 195–199.

Blane, D. (1999). The life course, the social gradient, and

health. In M. Marmot, & R. G. Wilkinson (Eds.), Social

determinants of health (pp. 64–80). Oxford, New York:

Oxford University Press.

Blane, D., Hart, C. L., Davey Smith, G., Gillis, C. R., Hole, D.

J., & Hawthorne, V. M. (1996). The association of

cardiovascular disease risk factors with socioeconomic

position during childhood and during adulthood. British

Medical Journal, 313, 1434–1438.

Brooke, O. G., Anderson, H. R., Bland, J. M., Peacock, J. L., &

Stewart, C. M. (1989). The effects on birthweight of

smoking, alcohol, caffiene, socio-economic factors and

psychosocial stress. British Medical Journal, 298, 795–801.

Chen, Y., Li, W., & Yu, S. (1986). Influence of passive smoking

on admissions for respiratory illness in early childhood.

British Medical Journal, 293, 303–306.

Cnattingius, S. (2004). The epidemiology of smoking during

pregnancy: Smoking prevalence, maternal characteristics,

and pregnancy outcomes. Nicotine Tobacco Research,

6(Suppl. 2), S125–S140.

Davey Smith, G., Hart, C., Blane, D., Gillis, C., & Hawthorne,

V. (1997). Socioeconomic position over the life course and

mortality. British Medical Journal, 314, 547–552.

Davey Smith, G., & Phillips, A. N. (1992). Confounding in

epidemiological studies: Why ‘independent’ effects may not

be all they seem. British Medical Journal, 305, 757–759.

Department of Health. (1998). Independent inquiry into inequal-

ities in health (the Acheson report). London: The Stationery

Office.

Department of Health (2002). Tackling health inequalities: 2002

cross cutting review. Department of Health, London,

accessed at www.doh.gov.uk/healthineqalities/ccsrsummar-

yreport.htmon 3.6.2003.

Dolan-Mullen, P. (1999). Maternal smoking during pregnancy

and evidence based interventions to promote cessation.

Primary Care, 26, 577–589.

Elmen, H., Hoglund, D., Karlberg, P., Niklasson, A., &

Nilsson, W. (1996). Birth weight for gestational age as a

health indicator: Birth weight and mortality measures at a

local area level. European Journal of Public Health, 6,

137–141.

Ermisch, J., Francesconi, M., & Pevalin, D.J. (2001). Outcomes

for children in poverty. Research report No. 158, Depart-

ment of Work and Pensions, The Stationery Office, London.

ARTICLE IN PRESSN. Spencer / Social Science & Medicine 62 (2006) 1250–12591258

Ferri, E. (1993). Life at 33: The fifth follow-up of the National

Child Development Study. London: National Children’s

Bureau.

Graham, H. (1994). Gender and class as dimensions of smoking

behaviour in Britain—insights from a survey of mothers.

Social Science and Medicine, 38, 691–698.

Graham, H. (1996). Smoking prevalence among women in the

European Community 1950–90. Social Science & Medicine,

43, 243–254.

Graham, H., & Blackburn, C. (1998). The socio-economic

patterning of health and smoking behaviour among mothers

with young children on income support. Sociology of Health

and Illness, 20, 215–240.

Graham, H., & Owen, L. (2003). Are there socio-economic

differentials in under-reporting of smoking in pregnancy?

Tobacco Control, 12, 434.

Haglund, B., Cnattingius, S., & Nordstrom, M.-L. (1993).

Social differences in late fetal death and infant mortality in

Sweden, 1985–86. Paediatric and Perinatal Epidemiology, 7,

33–44.

Heath, A., & Payne, C. (1999). Twentieth century trend in

social mobility in Britain. Working paper 70, Centre for

Research into Elections and Social Trends, University of

Oxford.

Hertzman, C., Power, C., Matthews, S., & Manor, O. (2001).

Using an interactive framework of society and lifecourse to

explain self-rated health in early adulthood. Social Science

& Medicine, 53, 1575–1585.

Himmelberger, D., Brown, B., & Cohen, E. (1978). Cigarette

smoking during pregnancy and the occurrence of sponta-

neous abortion and congenital abnormality. American

Journal of Epidemiology, 108, 470–479.

Hobcraft, J. (2003). Continuity and change in pathways

to young adult disadvantage: Results from a British

Birth Cohort. CASE paper 66. London:Centre for

Analysis of Social Exclusion, London School of

Economics.

Hodnett, E.D., Fredericks, S. (2003). Support during pregnancy

for women at increased risk of low birth weight babies. The

Cochrane database of systematic reviews, issue 3, Art. No.

CD000198.

Hofvendahl, E., Hofvendahl, S., & Wallender, B. (1997). Is

smoking in pregnancy interrelated between generations.

Acta Paediatrica Scandinavica, 86, 1224–1228.

Jarvis, M. J., & Wardle, J. (1999). Social patterning of

individual health behaviours: the case of cigarette smoking.

In M. Marmot, & R. G. Wilkinson (Eds.), Social

determinants of health (pp. 240–255). Oxford and New

York: Oxford University Press.

Katz, M. H. (1999). Multivariable analysis: A practical guide for

clinicians. Cambridge: Cambridge University Press.

Kleinman, J. C., & Kopstein, A. (1987). Smoking during

pregnancy 1967–80. American Journal of Public Health,, 77,

823–825.

Kramer, M. S. (1987). Determinants of low birth weight:

Methodological assessment and meta-analysis. Bulletin of

the World Health Organization, 65, 663–737.

Kramer, M. S., Seguin, L., Lydon, J., & Goulet, L. (2000).

Socio-economic disparities in pregnancy outcome: Why do

the poor fare so poorly? Paediatric and Perinatal Epidemiol-

ogy, 14, 194–210.

Kuh, D., & Ben-Shlomo, Y. (Eds.). (1997). A life course

approach to chronic disease epidemiology. Oxford: Oxford

University Press.

Kuh, D., Power, C., Blane, D., & Bartley, M. (1997).

Social pathways between childhood and adult health. In

D. Kuh, & Y. Ben-Shlomo (Eds.), A life course approach to

chronic disease epidemiology. Oxford: Oxford University

Press.

Logan, S., & Spencer, N. J. (1996). Smoking and other health-

related behaviour, social and environmental context and the

young. Archives of Disease in Childhood, 74, 176–179.

Lumley, J., Oliver, S.S., Chamberlain, C., & Oakley, L. (2001).

Interventions for promoting smoking cessation during

pregnancy (Cochrane review). In The Cochrane database

of systematic reviews, issue 3. Art No.: CD001055.pub2.

Lynch, J. W., Kaplan, G. A., & Salonen, J. T. (1997). Why do

poor people behave poorly? Variations in adult health

behaviour and psychosocial characteristics, by stage of the

socioeconomic lifecourse. Social Science & Medicine, 44,

809–820.

Lynch, J. W., Kaplan, G. A., & Shema, S. J. (1997). Cumulative

impact of sustained economic hardship on physical,

cognitive, psychological and social functioning. New Eng-

land Journal of Medicine, 337, 1889–1895.

Marsh, A., & McKay, S. (1994). Poor smokers. London: Policy

Studies Institute.

Meis, P. J., Michielutte, R., Peters, T. J., Bradley Wells, H.,

Evan Sands, R., Coles, E. C., et al. (1997). Factors

associated with term low birthweight in Cardiff, Wales.

Paediatric and Perinatal Epidemiology, 11, 287–297.

Meyer, M. B., Jonas, B. S., & Tonascia, J. A. (1976). Perinatal

events associated with maternal smoking during pregnancy.

American Journal of Epidemiology, 103, 464–476.

Mitchell, E. A., Ford, R. P. K., Stewart, A. W., Taylor, B. J.,

Becroft, D. M. O., et al. (1993). Smoking and the sudden

infant death death syndrome. Pediatrics, 91, 393–396.

Notkola, V., Punsar, S., Karvonen, M. J., & Haapakoski, J.

(1985). Socio-economic conditions in childhood and mor-

tality and morbidity caused by coronary heart disease in

adulthood in rural Finland. Social Science & Medicine, 21,

517–523.

Power, C., & Hertzman, C. (1999). Health, well-being, and

coping skills. In D. P. Keating, & C. Hertzman (Eds.),

Developmental health and the wealth of nations: Social,

biological and educational dynamics. New York and

London: The Guilford Press.

Power, C., Manor, O., & Fox, J. (1991). Health and class: The

early years. London: Chapman & Hall.

Reid, I. (1998). Class in Britain. London: Polity Press.

Rose, G. (1992). The strategy of preventive medicine. Oxford:

Oxford University Press.

Royal College of Physicians of London. (1992). Smoking and

the Young. London: Royal College of Physicians.

Salihu, H. M., Aliyu, M. H., Pierre-Louis, B. J., & Alexander,

G. R. (2003). Levels of excess infant deaths attributable to

maternal smoking during pregnancy in the United States.

Maternal and Child Health Journal, 7, 219–227.

Schiaffino, A., Fernandez, E., Borrel, C., Salto, E., Garcia, M.,

& Borras, J. M. (2003). Gender and educational differences

in smoking initiation rates in Spain from 1948 to 1992.

European Journal of Public Health, 13, 56–60.

ARTICLE IN PRESSN. Spencer / Social Science & Medicine 62 (2006) 1250–1259 1259

Sheehan, T. J. (1998). Stress and low birth weight: A structural

modeling approach using real life stressors. Social Science &

Medicine, 47, 1503–1512.

Spencer, N. J. (2003). Weighing the evidence: How is birthweight

determined?. Oxford: Radcliffe Medical Press.

SPSS. (1999). SPSS for Windows, version 10. Chicago: SPSS Inc.

Sweeting, H., & West, P. (2001). Social class and smoking at

age 15: The effect of different definitions of smoking.

Addiction, 96, 1357–1359.

Vagero, D., & Lundberg, O. (1989). Health inequalities in

Britain and Sweden. Lancet, 2, 35–36.

Victora, C. G., Huttly, S. R., Fuchs, S. C., & Olinto, M. T. A.

(1997). The role of conceptual frameworks in epidemiolo-

gical analysis: A hierarchical approach. International

Journal of Epidemiology, 26, 224–227.

Wong, P. P., & Baumann, A. (1997). How well does

epidemiological evidence hold for the relationship between

smoking and adverse obstetric outcomes in New South

Wales? Australian and New Zealand Journal of Obstetrics

and Gynaecology, 37, 168–173.

World Health Organisation (1999). International consultation on

environmental tobacco smoke and child health: Consultation

report. WHO, Division of Non-Communicable Diseases,

Tobacco Free Initiative, Geneva.

Zhang, J., & Yu, K. F. (1998). What’s the relative risk? A

method of correcting the odds ratio in cohort studies of

common outcome. Journal of American Medical Associa-

tion, 280, 1690–1691.

Further reading

Logan, S., Spencer, N. J., & Blackburn, C. (1996). Smoking is

part of a causal chain. British Medical Journal, 313,

1332–1333.