Article Text
PDF
Abstract
Background By 2020, it is predicted that 60 million children worldwide will be overweight. Maternal smoking in pregnancy has been suggested as a contributing factor. Our objective was to systematically review studies on this, thereby expanding the evidence base for this association.
Methods Systematic review with meta-analysis, Prospero Registration number CRD42012002859. We searched PubMed, Embase, Global Health, Web of Science and the Grey literature. We included prevalence, cohort and cross-sectional studies involving full-term, singleton pregnancies. Published and unpublished studies through to 1 January 2015 in all languages, demonstrating an objective overweight outcome up until 18 years of age and data presented as an OR, were included. Quality assessment was undertaken using an adaption of the Newcastle-Ottawa scale. Statistical analysis was performed using Review Manager V.5.3.
Findings The meta-analysis included 39 studies of 236 687 children from Europe, Australia, North America and South America and Asia. Maternal smoking in pregnancy ranged from 5.5% to 38.7%, with the prevalence of overweight from 6.3% to 32.1% and obesity from 2.6% to 17%. Pooled adjusted ORs demonstrated an elevated odds of maternal smoking in pregnancy for childhood overweight (OR 1.37, 95% CI 1.28 to 1.46, I2 45%) and childhood obesity (OR 1.55, 95% CI 1.40 to 1.73, I2 24%).
Interpretation Our results demonstrate an association between maternal prenatal smoking and childhood overweight. This contributes to the growing evidence for the aetiology of childhood overweight, providing important information for policymakers and health professionals alike in planning cessation programmes or antismoking interventions for pregnant female smokers.
- CHILD HEALTH
- OBESITY
- PREGNANCY
Statistics from Altmetric.com
Introduction
Childhood obesity is a major public health challenge of the 21st Century. The global prevalence of childhood overweight and obesity rose from 5% in 1990 to 7% in 20121 with 42 million children worldwide under the age of 5 classified as overweight.2 By 2020, almost 60 million children worldwide are predicted to be overweight or obese.3 Obese children suffer mental, psychological and physical ill health and are at an increased risk of developing non-communicable diseases (NCDs) later in life. NCDs accounted for 68% of global deaths in 20124 and are linked by risk factors including tobacco usage and overweight. Evidence suggests that once obesity in childhood is established it is difficult to reverse5 and likely to continue to adulthood.6 Tackling this is a global health priority.7
The Barker hypothesis describes adverse early life environmental factors disrupting normal growth and development, resulting in a susceptible adult phenotype prone to cardiovascular disease. Identifying early development factors, such as maternal smoking in pregnancy, which may result in subsequent obesity, enables them to be addressed effectively. Children with a high birth weight have an increased risk of later obesity, but evidence also shows that low birth weight can predispose to obesity. Maternal prenatal smoking is recognised to result in low birth weight, and this could be the link between maternal prenatal smoking and childhood obesity. One longitudinal study8 found different patterns of weight gain among children born to smokers versus non-smokers. Overweight children aged 4.5 years had a birth weight approximate to the population mean, but during the first 5 months of life those born to smoking mothers gained more weight than those born to non-smoking mothers. Individuals with a lower birth weight who later develop obesity are at the highest risk for cardiovascular morbidities in the long term.9 However, one study10 concluded that the link between maternal prenatal smoking and childhood obesity was independent of birth weight. Confounding factors are therefore likely to be important in this association, in particular socioeconomic circumstances and lifestyle factors.
A systematic review of published data up until June 200611 suggested that maternal smoking was positively associated with childhood overweight. However, the meta-analysis comprised only 14 studies, 5 of which included children born prematurely (<37 completed weeks gestation). Preterm infants may have different growth patterns from term infants, at least up to 5 years of age,12 ,13 and so it may not be appropriate to combine them when examining causes of childhood obesity. There has also been a marked increase in the number of studies published in this area since 2006. The 2010 UK Infant Feeding Survey found that one in eight mothers (12%) continued to smoke throughout pregnancy.14 In many low-income and middle-income countries, rates of smoking are rising, particularly among female adolescents.15
Objective
The objective of this study was to assess the association between maternal smoking in pregnancy and the risk of developing overweight and obesity in childhood, through a systematic review of published and unpublished work through to 2015.
Methods
Selection criteria
The evidence was systematically reviewed with results reported in accordance with PRISMA guidelines.16 The review protocol is accessible through PROSPERO (registration number CRD42012002859). Eligibility criteria defined inclusion criteria as: children born at full term (>37 completed weeks gestation), outcome of overweight measured between 2 and 18 years, studies required a control group with maternal prenatal smoking status to be specified for all groups and objective and quantitative measurement of obesity with results presented as an OR with 95% CIs. Studies in all languages were included, and data published through to 1 January 2015. Published and unpublished studies were included. Studies were excluded if they included premature infants (<37 weeks completed gestation), multiple pregnancies, animal studies and studies with no comparison group or no exposure of maternal smoking in pregnancy. Studies were also excluded if all mothers had medical conditions potentially impacting offspring growth, for example gestational diabetes. Discussions, policy documents, reviews and case reports were excluded. If studies used data from the same cohort population, the study measuring outcomes at the older age was the one included.
Search strategy
With the assistance of a professional librarian, the following online databases were systematically searched: PubMed, EMBASE, Global Health and Cinahl and Web of Science. The search terms were (pregnancy OR maternal OR mother OR pregnant) AND (smoking OR cigarette OR nicotine OR smoker* OR tobacco) AND (childhood OR offspring OR infant OR children OR adolescent OR teen) AND (obesity OR overweight OR ‘body mass index’ OR weight OR height OR percentile OR fatness OR adiposity). A total of 17 247 studies were identified.
The Grey Literature search consisted of databases ‘Dissertations & Theses’ and ‘Index to Theses’; OPENGREY; Google search of organisational websites (org.uk, nhs.uk, gov.uk, .gov and .org) and ongoing clinical trials identified through clinicaltrials.gov and the WHO trials registry. A simplified version of the above search terms identified 408 studies and trials.
Following removal of duplicate references, titles and abstracts were reviewed for potential inclusion. Full text articles were obtained for included studies. For conference abstracts, the author was contacted to clarify if the data were subsequently published and if not, a copy of the study data was requested. If a grey literature article was suitable for inclusion, the author was contacted for a full copy of the study and data.
Data collection
Both authors (SR and EP) independently reviewed full text articles, resulting in 39 studies meeting the inclusion criteria and so suitable for the meta-analysis. Disagreements were resolved through discussion. The references for each included study were reviewed to identify possible further studies. The PRISMA flow diagram details reasons for study exclusion at each stage (figure 1).
PRISMA flow diagram—the association between maternal smoking during pregnancy and childhood overweight.
Definition of outcomes
The primary outcome was childhood overweight and obesity in relation to maternal prenatal smoking with data extracted as unadjusted and adjusted OR with 95% CIs. If the unadjusted OR was not available, it was calculated from published data or authors contacted for raw data. Extracted data included study location, decade of birth of participants, age outcomes that were measured and specific definitions of overweight and obesity.
All studies, except one,17 used body mass index (calculated as weight (kg)/height2 (m2)) as the objective outcome; however, definitions of overweight and obesity varied. Nineteen studies18–36 used the International Obesity Task Force definitions: ‘overweight’ defined as age-specific and sex-specific cut-offs corresponding to an adult BMI of ≥25 and BMI ≥30 for ‘obesity’.37 Other studies defined overweight as BMI ≥85th or ≥90th centile for age and sex; with obesity as BMI ≥95th or ≥97th centile for age and sex. The study17 not using BMI defined overweight as ‘weight for height more than 2 SDs’.
Where studies presented different levels of adjustment for confounding factors, the maximally adjusted model was included (confounding factors described in table 1). If studies presented different levels of exposure, the lowest possible smoking dose and smoking early in pregnancy was used. When using raw data to calculate unadjusted OR, categories were combined as ‘non-smoking in pregnancy’ or ‘any smoking in pregnancy’ to maximise the number of participants included. One study25 had results on two cohorts: the 1958 birth cohort and data for offspring born to the original cohort. Only offspring data were included as another study38 provided outcomes at an older age for the 1958 cohort. Two studies38 ,39 presented separate OR for men and women, with no overall OR, so the meta-analysis included subgroup data from each of these two studies as separate studies.
Characteristics of included studies
Statistical analysis
Pooled ORs were calculated (with 95% CIs) for included studies using a random effect model and a generic inverse variance method in which the weight of each study is the inverse of the variance of the effect estimate. A random effect model was chosen due to likely variation in included studies and recognition that participants in each study could differ from each other in a number of ways including geographical location and social circumstances. The I2 calculation provided a percentage of the variation across study estimates due to heterogeneity rather than chance alone.40 Funnel plots were constructed to give a visual representation of possible publication bias. Statistical analysis was performed using Review Manager V.5.3.
Assessment of risk of bias in included studies
Each author (SR and EP) independently performed data extraction and quality assessment against purposefully designed forms. The quality assessment was adapted from the Newcastle-Ottawa Quality Assessment Scale,41 grading each paper a number of stars for selection, comparability, exposure and outcome. The adaptations were minor, including specifying the exact outcome (childhood overweight) and the criterion ‘Demonstration That Outcome of Interest Was Not Present at Start of Study’ was omitted as this was not relevant.
Role of the funding source
The funding sources had no direct role in study design, data collection, analysis, interpretation of data or writing of the report. Both authors had full access to the study data and had final responsibility for the decision to submit for publication.
Results
Thirty-nine studies were included in the meta-analysis (figure 1): 26 cohort studies,8 ,18 ,19 ,21–23 ,25 ,26 ,29 ,30 ,32–35 ,38 ,39 ,42–51 12 cross-sectional studies17 ,20 ,24 ,27 ,28 ,36 ,52–57 and one case–control study.31 Table 1 details characteristics for each included study. The combined number of participants was 236 687, with children born between 1958 and 2010. Study populations were from Europe, Australia, North America and South America and Asia. The prevalence of maternal smoking in pregnancy ranged from 5.5%21 to 51.4%39 (mean 19.1%). The age of outcome ranged from 234 ,51 to 1830 years of age. The prevalence of childhood overweight ranged from 6.3%17 to 42%28 (mean 17.2%), and obesity prevalence from 2.0%30 to 17%44 (mean 6.9%). Not all studies contained prevalence data for both overweight and obesity. Table 2 outlines the quality assessment of included studies.
Summary of quality assessment
The primary outcome was childhood overweight compared to maternal smoking in pregnancy (28 studies provided unadjusted OR). The pooled unadjusted OR of 1.43 (95% CI 1.32 to 1.56) (figure 2) demonstrated an elevated risk of overweight among children whose mothers smoked prenatally compared to those who did not. There was moderate heterogeneity with an I2 of 64%. The pooled adjusted OR for childhood overweight in relation to maternal smoking in pregnancy was 1.37 (95% CI 1.28 to 1.46) (figure 3), I2 of 45% (31 studies). This demonstrates an increased risk of childhood overweight in relation to maternal prenatal smoking even when adjusted for potential confounders. Funnel plots for unadjusted OR and adjusted OR for overweight are presented in online supplementary figures S1 and S2, respectively.
Pooled unadjusted OR for maternal prenatal smoking and childhood overweight.
Pooled adjusted OR for maternal prenatal smoking and childhood overweight.
Supplementary figure
Supplementary figure
Given the different definitions of childhood overweight, sensitivity analysis was performed restricting to studies using the international obesity task force definitions. This marginally reduced the pooled adjusted OR for overweight, but CIs were similar at OR 1.33 (1.23 to 1.46 95% CI). The heterogeneity was reduced with an I2 of 52%.
For childhood obesity in relation to maternal smoking in pregnancy, the pooled unadjusted OR was 1.77 (95% CI 1.56 to 2.03), I2 58% (see online supplementary figure S3) and pooled adjusted OR was 1.55 (95% CI 1.40 to 1.73), I2 24% (see online supplementary figure S4). The pooled adjusted OR demonstrates a clear increased risk of childhood obesity following maternal prenatal smoking despite adjustment for confounders. Confounding factors varied between studies (table 1), often including a measure of parental education, breastfeeding and socioeconomic position.
Supplementary figure
Supplementary figure
The pooled results used the lowest exposure smoking dose and at the earliest available point in pregnancy, demonstrating the minimal effect of prenatal smoking. Most included studies (27 in total) classified maternal prenatal smoking as ‘yes’ or ‘no’, but the remaining studies categorised the level of cigarette exposure. Eight studies18 ,23 ,29 ,38 ,39 ,42 ,52 ,53 demonstrated a dose–response association of increasing odds of overweight with increasing numbers of cigarettes smoked per day, or persistent smoking throughout pregnancy. However, two studies19 ,30 found higher ORs for moderate rather than heavy smokers, whereas another study28 had the lowest OR for childhood overweight in moderate smokers. The final study46 demonstrated a higher OR for overweight in mothers who only smoked in the first trimester compared to mothers who continued to smoke throughout their pregnancy.
Discussion
This review provides robust data demonstrating maternal smoking during pregnancy is associated with an elevated odds of childhood overweight and obesity; if a child's mother smoked during pregnancy, the odds of the child becoming overweight are 37% greater and 55% greater for obesity than if the mother did not smoke. A number of plausible mechanisms have been put forward for this relationship. Smoking in pregnancy leads to low birth weight, likely through a vasoconstrictive action of nicotine and fetal hypoxia.58 This may affect postnatal growth patterns, leading to a higher likelihood of childhood obesity. There is also increasing evidence that tobacco smoke exposure at other periods in childhood increases the risk of childhood overweight.54 ,59 ,60
Confounding factors, such as socioeconomic status and parental education, are clearly important in the relationship between maternal smoking and childhood obesity. However, a clear relationship remained despite adjustment; implying measured confounders cannot fully explain this association. Nevertheless, residual confounding is likely to be relevant. Howe et al61 found that measured confounders did not impact the relationship between maternal prenatal smoking and adiposity changes; however, comparisons with partner smoking in pregnancy implied that unmeasured familial factors are important. Iliadou et al10 concluded that parental socioeconomic position and education partly explained the association between maternal prenatal smoking and childhood obesity, but unmeasured familial factors, including lifestyle factors and dietary habits, were also relevant. While included studies in this review did adjust for known confounders, there was no adjustment for potential (but unmeasured) confounders and so residual confounding is likely to affect the results.
Compared to a previous systematic review,11 these results demonstrate a lower pooled adjusted OR for overweight (1.37 vs 1.5) but similar pooled adjusted OR for obesity (1.55 vs 1.52). The strengths of this more recent meta-analysis are that it includes almost three times as many children as the previous study,11 using a wider range of databases, including published and unpublished studies. More confidence can be placed in these findings as premature infants were excluded, and outcomes were assessed only during childhood. This meta-analysis included populations from an extensive geographical area (Europe, Australia, North America and South America and Asia), demonstrating that the association exists across different regions. However, populations included were predominantly high-income western countries despite smokers in low-income and middle-income countries (LMICs) accounting for nearly 80% of global tobacco consumption.62 The increasing prevalence of overweight and obesity affects all countries, with close to 31 million overweight children living in LMICs, so this highlights an important gap in global health research. Publication bias may be a factor and also affect the results as a whole. Although we constructed and presented funnel plots, they have limitations for assessing publication bias. Visual interpretation of funnel plots may be too subjective to be useful,63 and some effect estimates can produce spurious asymmetry in a funnel plot.64
The quality of each included paper was assessed with most scoring well. However, some study populations were not entirely representative of the wider population, with some cohort studies losing large numbers to follow-up, potentially introducing bias. It is possible that more overweight or obese children were lost to follow-up leading to an underestimate of the effect size. All included studies were observational, so may all be subject to similar biases, including selection and recall bias. All studies used maternal self-reporting for the exposure of smoking during pregnancy. However, studies show self-reporting has high validity in early and late pregnancy.65 There is more propensity for people to misrepresent themselves as non-smokers while continuing to smoke, leading to an under estimation of the true association between maternal smoking and childhood overweight. This meta-analysis included data on the lowest smoking dose and smoking early in pregnancy, implying the results are the minimum effect that any smoking in pregnancy can have. While the design of included studies means that it is not possible to infer causality, this review contributes to the mounting body of evidence of a clear association between maternal smoking and childhood overweight.
Given the diversity of included studies, the moderate heterogeneity between studies is not surprising. This can be attributed to the variability in study design, participants and definitions of exposures. Outcome definitions differed between studies and children were assessed at different ages in different studies. There may have been differences in the measurement of exposure variables such as socioeconomic position, and with participants coming from a number of different countries, this would have further introduced variability.
This systematic review highlights maternal smoking in pregnancy as an important factor contributing toward the childhood obesity pandemic. This reinforces the need for ongoing public health interventions regarding smoking cessation in all countries, particularly focusing on women of child bearing age. The pathway to childhood overweight is multifactorial, but also largely preventable. Estimates show that 40% of childhood obesity could be prevented through a combination of healthy eating promotion, active living and the cessation of smoking during pregnancy.23 Prevention of childhood obesity should start at the earliest possible stage of development and a smoke-free environment for all will be an important part of this.
In conclusion, this systematic review has confirmed evidence for an association between maternal smoking in pregnancy and childhood overweight and obesity. This knowledge is relevant to all from policymakers to those healthcare professionals in regular contact with women of child bearing age. Smoking cessation is of benefit to the long-term health of the mother, it benefits the growing fetus during pregnancy and now this study confirms evidence of benefit to her offspring's future health and well-being through removing this increased risk of overweight.
What is already known on this subject
Childhood obesity is a major public health problem with predictions that 60 million children worldwide will be overweight by 2020.
Maternal smoking in pregnancy is recognised to be associated with low birth weight, but increasing evidence shows an association with overweight and obesity in childhood as well.
What this study adds
This study provides robust evidence for an association between maternal smoking in pregnancy and an elevated risk of childhood overweight and obesity. The study identified papers from a wide range of databases and included published and unpublished studies with comprehensive inclusion and exclusion criteria. It includes populations from an extensive geographical area demonstrating the association to exist across different regions.
Acknowledgments
The authors acknowledge and thank Nia Roberts, outreach librarian at Bodleian Health Care libraries, for her assistance with the searching process of the review. They also acknowledge those authors who provided assistance with further information or data as necessary including Jonathan Gravel, Dr Joesph Braun, Dr Margaret Demment, Dr Plachata-Danielzik, Dr Seungmi Yang, Dr Nancy Sowan, Professor Emily Oken, Professor Xiaozhong Wen, Professor Kohta Suzuki, Professor Sarah Messiah, and Professor Christine Olson.
References
Footnotes
Twitter Follow Sarah Rayfield at @RayfieldSarah
Contributors SR devised the study, established inclusion criteria, did all searches of published and unpublished work, evaluated each article for possible inclusion, completed data extraction and quality assessment of each included study, contacted authors of relevant studies, was responsible for statistical analysis and interpretation of the data and drafted the manuscript. EP established inclusion criteria, provided supervision, evaluated each article for possible inclusion, completed data extraction and quality assessment of each included study and assisted with interpretation of the data and critical revision of the manuscript.
Funding SR is a Public Health trainee salaried by Oxford University Hospitals NHS Trust. EP is salaried by the Nuffield Department of Medicine, University of Oxford.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement This study is a systematic review, and all included studies are available.