Maternal and early‐life area‐level characteristics and childhood adiposity: A systematic review

Summary There is a cross‐sectional evidence that physical and social environments are linked to childhood adiposity. Evidence is scarce for the role of preconception, pregnancy, and early‐life area‐level characteristics in shaping childhood adiposity. We aimed to systematically review evidence for associations between physical and social environmental conditions experienced in these periods and childhood adiposity. Published literature was identified from the CINAHL, Embase, MEDLINE, and PsycINFO databases. Longitudinal studies linking an area‐level environmental exposure in the preconception, pregnancy, or early‐life (less than 1 year) periods and a measure of adiposity between the ages of 2 and 12 years were examined. Eight studies in the United States, Denmark, South Korea, United Kingdom, and Canada satisfied the inclusion criteria. Storm‐induced maternal stress, nitrogen oxides exposure, traffic noise, and proximity were associated with greater childhood adiposity. Frequent neighbourhood disturbances were associated with lower adiposity, while particulate matter exposure was associated with both higher and lower adiposity in childhood. Area‐level characteristics may play a role in the ongoing obesity epidemic. There is a limited evidence of longitudinal associations between preconception, pregnancy, and early‐life area‐level characteristics with childhood adiposity. Numerous factors that appear important in cross‐sectional research have yet to be assessed longitudinally, both individually and in combination.

socio-economically deprived areas are more than twice as likely to be affected by obesity than children in the most affluent areas. 7 Over the last 30 years or so, the Developmental Origins of Health and Disease (DOHaD) paradigm has identified the preconception, antenatal, and early-life periods as key to shaping future susceptibility to non-communicable diseases (NCDs). 8 The circumstances experienced in these key phases of life enact epigenetic and behavioural adaptations among offspring, which have implications for their development and later health. 9 The environment that mothers experience in the preconception period and those that their children are exposed to in-utero and in their first year of life are likely to be important dimensions influencing later adiposity growth in childhood through numerous plausible mechanisms.
The characteristics of the physical environment influence dietary and physical activity habits that affect overall health and risk of mothers being affected by overweight or obesity at conception, and during pregnancy. Proximity to fast food outlets may encourage consumption of food that is of poor nutritional value. 10 Mothers living within a half-mile of a fast food restaurant are more likely to gain over 20 kg during pregnancy, 11 and high gestational weight gain is a known risk factor for offspring being affected by obesity. 12 Conversely, the lack of accessible healthy food (in so called "food deserts") may also affect maternal diet in the preconception and pregnancy periods, with gestational undernutrition (as indicated by premature births and low birth weight) being linked with the risk for children to be affected by obesity through offspring compensatory growth post-birth and increased leptin resistance. 13,14 Attractive open and green environments encourage women to walk in the prenatal and perinatal periods, enhancing physical activity and offering opportunities for social interaction that may alleviate stress. 15 Stress during pregnancy has been linked with alterations in placental endocrine and immune processes, resulting in higher risk of infants being born premature and small for gestational age, which is associated with compensatory growth in early infancy and subsequent adiposity in childhood. 16 Some environmental factors that affect childhood adiposity may be specific to the pregnancy period. Mothers exchange ingested and inhaled pollutants with offspring via placental transfer, which affects fetal and infant development. 17 Gestational exposure to organic pollutants from indoor and outdoor sources has been shown to lead to elevated insulin and leptin levels, in addition to impaired glucose tolerance in rats, factors that affect the storage and expenditure of energy and therefore the risk of becoming affected by obesity. 18 The diversity of maternal gut bacteria (the "microbiome") affects nutrition exchange and the composition of the offspring microbiome at birth. 19 The composition of the antenatal microbiome is influenced by maternal exposure to environmental pollutants, with particular combinations being associated with susceptibility to NCDs. 20 For example, high counts of Lactobacillius bacteria in the antenatal microbiome are associated with high risk of offspring being affected by overweight and obesity during infancy and childhood. 21 The environment and diet children are exposed to in their first year of life may also affect their weight during childhood. The diversity of gut microbiota is formed in the first few hours of human life as a response to the antenatal microbiome 22 and rapidly evolves as a result of exposure to environmental pollutants, 20 and the microbiome is associated with weight in childhood. 23 The proximity of supermarkets and fast food outlets to the home and workplace is associated with dietary patterns among adults and hence families. 24 Through breast milk, mothers exchange nutrients from food with infants 25 ; this exchange develops infant familiarity and preference for the foods which mothers eat. 26 In this way, the food environment at this stage can affect later childhood diet through post-natal diet. Exposure to environmental allergens in the first year of life has been linked with lower risk of recurrent wheezing, 27 which may make physical activity more feasible during childhood.
Exposure to common air pollutants (sulphur dioxide, nitrogen dioxide, and particulate matter) during the first year of life hamper lung function, leading to respiratory problems in childhood, 28 which may also affect physical activity patterns during childhood.
The creation of "health-promoting environments" is one of the World Health Organization's objectives for preventing NCDs such as obesity. 29 What exactly constitutes a health-promoting environment is contested, and applications at the regional and national levels have been mixed. Preschools, schools, and deprived neighbourhoods are identified as environments that are conducive to obesity within the European Union 2014 to 2020 Action Plan, with no attention paid to the early-life neighbourhood environment. 30 The U.K. Government's action plan also identifies schools and early-year settings as environments where children are exposed to obesity-related risks, with no reference to the neighbourhood environment. 31 In addition, neither framework acknowledges the role that the preconception environment plays in subsequent offspring health.
In previous systematic reviews, neighbourhood socio-economic deprivation, 6 parental perception of neighbourhood safety, 32 fast food availability, 5 access to open natural (green) spaces, and physical activity facilities 33 were associated with childhood adiposity. The evidence base mostly consists of cross-sectional studies; therefore, the extent to which the environment is causally associated with childhood adiposity is difficult to establish, as there is no information on the length of exposure to environmental influences. 34 In the context of conflicting definitions of "health promoting environments," and to inform policies that can target high-risk neighbourhoods with preventive interventions, a comprehensive review is needed to collate the evidence on longitudinal associations between specific area-level characteristics and childhood adiposity. Hence, the aim of this study is to systematically identify research which characterizes area-level environmental exposures experienced in the preconception and antenatal periods as well as the first year of life and test their association with later childhood adiposity. The search strategy is detailed in Table S1. The final search was conducted on the 28th of August 2018, after consulting with a specialist librarian. Studies were limited to those published in English, and from January 1, 1990, to ensure that up-to-date literature was assessed. The reference list of all full-texts that were included was searched. The protocol for this review was published on the PROS-PERO international prospective register of systematic reviews (CRD42017082020), and this review is reported in line with the PRISMA guidelines. 35  where the cut-off is clearly defined and justified. The outcome must be measured in childhood (between 2 and 12 years old). Characteristics of the residential or workplace environment must be assessed through geo-referencing, or be self-reported. Environmental characteristics must be measured during the preconception, pregnancy, or early-life (younger than 1 year old) periods. Studies where the sole outcome was change in adiposity were excluded, as a change in growth velocity may not result in a difference in adiposity when there are differences in birth and early-life weight. Studies that used personal devices to monitor environmental features were also excluded, as measurements would have been affected by in-home, neighbourhood, and out-of-neighbourhood features. Self-reported measures were only eligible if they explicitly mention the residential or workplace neighbourhood, the surrounding area or the "local area."

| Inclusion and exclusion criteria
Research published in non-peer-reviewed or "grey" literature (including books, book chapters, conference proceedings, working papers, and theses) were also excluded due to the scale of peer-reviewed papers retrieved in preliminary searches and the lack of quality control afforded by the exclusion of peer-review in these outputs. 36

| Screening process
A 10% randomly selected sample of titles was screened for eligibility independently by two reviewers (S.W. and N.Z.) using Rayyan, a screening management software. 37 The 10% threshold was used, as a simulation study has shown that there is no decrease in study selection bias if the sampling fraction is increased 38 past 10%. The percentage agreement between the two reviewers was 94% at the title stage.
Discrepant decisions for inclusion/exclusion were arbitrated by a third reviewer (N.A.A.), and then one author (S.W.) screened the remaining titles for inclusion. The titles screened for inclusion followed the same process for abstracts, with the agreement between reviewers standing at 100%. All full-texts were screened independently by S.W. and N.Z., with disagreements resolved in a meeting with the two reviewers and N.A.A. At this stage, study authors were contacted for details of subgroup analyses if their age intervals for the exposure or outcome included ineligible ages, or for further clarity on exposure assessment.
Two authors replied with no further data gained, and one author did not reply.

| Data extraction
Data extraction was conducted for all final included articles by S.W.
using a modified version of the Cochrane Collaboration's data extraction form. 39 The fully adjusted association estimates between each eligible environmental indicator and outcome were extracted, including for all subgroup analyses. In cases where there were multiple time points, all age-eligible associations were extracted. Significant associations were identified through confidence intervals that did not overlap the null, or P values < .05 if confidence intervals were not presented.

| Quality assessment
Quality assessment was conducted by two reviewers (S.W. and N.Z.).
All eligible articles were prospective cohort studies, and there is no agreed scoring criteria for such studies. As a result, we elucidated key strengths and weaknesses of each study using the National Institute of Health (NIH) Assessment Tool for Observational Cohort and Cross-Sectional Studies and the STROBE checklist. 40,41 The exclusion of sample members born preterm or low birth weight was considered a key weakness in studies which looked at in-utero exposure, because these outcomes may be on the causal pathway between the pregnancy environment and later childhood adiposity. This stance is informed by evidence that 13% to 24% of preterm births globally are attributable to PM 2.5 exposure in a logistic regression model 42 and that PM 2.5 exposure increases the risk of being born low birth weight. 43 Being born preterm or low birth weight subsequently affects childhood adiposity in turn through early-life compensatory growth. 44

| Analysis and synthesis
As we expected significant variation in study design and environmental measures, a narrative synthesis was planned a priori, rather than a meta-analysis approach. Environmental measures were grouped based on their similarity, and a summary of the effect sizes and precision is presented across each included study.

| RESULTS
A total of 11 783 records were identified in the search (Figure 1), of which 3821 were duplicates; 7962 titles were screened, of which 198 abstracts were further screened. A total of 23 full-texts were assessed for inclusion independently. Two duplicate studies conducted by the same lead authors using the same dataset were identified, 45,46 one study was retained 45 as environmental measures were included in the fully adjusted model, whereas they were not in the other study. In total, eight studies were included in the narrative synthesis. 45,[47][48][49][50][51][52][53] Four studies were based in the United States, one each in Canada, Denmark, England, and South Korea. Seven of these studies were reports of a prospective cohort, and one was a secondary analysis of prospective cohort data. 45 Generally, all studies were well reported and designed but had poor recruitment rates (less than 50%) or poor follow-up rates (less than 80%). Further study characteristics are collated in Table 1.

| Characteristics of included studies
All eight included studies used data from prospective cohorts. Four studies recruited women during pregnancy, 47,48,50,51 one study recruited shortly after birth, 53 one study used a combination of the two, 49 one study recruited 9 months after birth, 45 and one study recruited children through schools. 52 The recruitment rate varied between 12% 49 and 78% 47 (mean 51%) and was not presented in two studies. 51,53 The percentage of the recruited sample who participated at each outcome time-point varied between 22% 51 and 83% 49 (mean 60%).
The eight studies varied in terms of the timings of exposure and outcome measurement. Two studies had only one time-point for exposure, 45,48 and four had only one time-point for the outcome. 45,[47][48][49] Five studies assessed the average exposure over the entire pregnancy, 47,48,51-53 four across the first year of life, 45,[50][51][52] and two in the preconception period, defined as 3 months prior to conception. 49,53 Three studies investigated trimester-specific measures, including two studies that assessed the exposure in each trimester, 49,53 and one study that assessed exposure in the third trimester only. 50 There were a total of eight anthropometric outcomes examined across the eight studies (BMI z-score, weight-for-age z-score,  Quality assessed through an adapted version of the National Heart, Lung, and Blood Institute Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies and the STROBE checklist. 39,40 overweight or obesity based on BMI cut-off, obesity based on BMI cut-off, fat mass, waist-to-hip-ratio, skinfold thickness, and waist circumference), with four studies examining more than one outcome. [47][48][49][50] All outcomes were measured by research assistants, school nurses, or GPs, with the exception of one study, where the measurer of height and weight was "undefined" for 35% of the sample. 48 BMI was the most common outcome and was present in five studies (four used age-and sex-adjusted z-scores, 47-50 one did not 52 ). An age-and sexadjusted cut-off for overweight/obesity was used in four studies, 45,48,49,53 and fat mass was assessed in two studies. 47,50 The following outcomes were assessed in only one study: waist-to-hip ratio 47 ; waist circumference and skinfold thickness 50 ; weight-for-age (ie, no height adjustment). 51 A total of 16 environmental measures were tested across the eight included studies (10 measured through geo-referencing and six selfreported), of which seven were significantly associated with childhood adiposity in one or more studies ( Table 2). Five groups of environmental indicators emerged from the review, which will be discussed in turn.

| Traffic
Five traffic-related measures were assessed in three studies. As

| Built environment
Hawkins et al 46   The findings of this systematic review clearly display that there are important environmental exposures for which the longitudinal evidence on association with childhood adiposity is lacking, including socio-economic deprivation, 6 neighbourhood safety, 32 and food access, 5 all found to be linked to childhood adiposity in cross-sectional research. Also, spaces for physical activity and green space were associated with childhood adiposity in a previous systematic review of cross-sectional research 33 but were not assessed in any studies in this review. The differences between the findings for longitudinal and cross-sectional studies may be explained by residential sorting, where families with risk factors predisposing their children to being affected by obesity (eg, low-income) may be more likely to move to disadvantaged neighbourhoods as their children grow, 56 although adjustment for early-life migration had no effect on estimates in Hawkins et al. 45 Given that the above factors were not assessed across multiple studies in this review, further longitudinal research may shed greater light on these discrepancies.
Across the eight studies, geo-referenced measures of the environment were more common than those which were self-reported ( Table 2). The results for self-reported measures were in contrast to cross-sectional research which has linked objective measures of the food environment 57 and spaces for physical activity 58 to childhood adiposity, although the evidence for these linkages is inconsistent. [59][60][61][62] Conversely, geo-referenced area-deprivation was not associated with childhood adiposity, despite cross-sectional evidence. 6 These findings suggest that there is some inconsistency in how these mea-  45 which is a limitation because paternal genetics, attitudes, and behaviours likely also affect childhood adiposity, and their exclusion from models overemphasizes the effect of maternal factors. 66 The exclusion of preterm and low birthweight babies in some studies 47,48,51 may have affected the estimates as these outcomes may be on the causal pathway between the environment and childhood adiposity, given evidence of links between the environment, birth outcomes, 42 and infancy catch-up growth for preterm and small babies. 44,67 Residual confounding may also occur at the area-level, where certain environments experience multiple forms of disadvantage in terms of suitability for healthy weight gain, for example, areas with limited park access tend to have fewer outlets selling healthy foods. 68 The results of these studies may have been All of the studies in this review used data from recruited prospective cohorts. Reliance on recruitment as opposed to routinely collected data may have led to sample bias affecting study estimates.
Differences between the target population and the sample were noted in several studies, 48-50 although Christensen et al noted that the effect of this bias was found to be insignificant in a comparison of cohort and register data. 69 The lack of studies drawing on routine or administrative data is likely related to difficulties in attaining datasets where parental residential information is linked to childhood data. As all of the included studies were observational in nature, it is unclear whether the associations were causal. Dancause et al 49 note, however, that the exposure in their study (exposure to a storm) was theoretically randomly distributed (with respect to socio-economic status), suggesting that there is potential for a causal mechanism between extreme weather conditions, maternal stress, and childhood adiposity.
In addition to the potential bias arising from low response rates from specific subgroups, five of the included studies 48-51,53 had attri- All of the objective measures used in studies within this review have been assigned through the mother or child's home address, which will have resulted in faulty assumptions regarding activity spaces and environmental exposures. In the preconception and early pregnancy timeframes, mothers may commute to work, and through these journeys, they may have been exposed to differing environments than those experienced in the areas surrounding their home.
For working mothers, this may lead to an underestimation of exposure, as areas surrounding workplaces have been found to be less socially advantaged and have higher densities of food outlets than residential neighbourhoods, for pre-retirement adults and women specifically. 72 papers, 75 especially given that all included papers presented at least one significant finding. This decision was made in the context of peer-review acting as a quality control process for journal articles.
There was no evidence that this exclusion led to a lack of papers with negative or "unexpected" findings, as there was disagreement between the findings in this review and the cross-sectional literature for PM 2.5 exposure, deprivation, and neighbourhood conditions. The lack of multiple studies for each environmental indicator limits the ability to summarize cumulative evidence. Few studies adjusted for area-confounders, so we cannot isolate each specific environmental characteristic's influence on later childhood adiposity from the general milieu (which likely differs across the range of each indicator). The studies were located entirely in high-income countries, so there is no evidence base to infer for middle-and low-income contexts. There was no consensus on the best tool to score or grade observational cohort research, so in the interest of being objective, we were limited to listing the strengths and weaknesses that were elucidated using two commonly used checklists. This approach may be more transparent in understanding how studies were assessed and allow readers to self-identify criteria which are important in their view.

| CONCLUSION
In summary, six area-level characteristics experienced during preconception, pregnancy, and early-life showed associations with childhood adiposity in this review. Worse air quality and greater exposure to traffic in the preconception, in-utero and early-life periods, were associated with greater adiposity in childhood. Other factors such as area deprivation and garden access, significant in cross-sectional research, were not associated with adiposity in longitudinal studies. This suggests that area factors may play a role in the ongoing obesity epidemic. However, numerous area-factors which appear important in cross-sectional research have yet to be assessed longitudinally. In addition, there is no evidence on the effects of multiple areadisadvantage. Further research to ascertain the role of area-level environment in the developmental origins of obesity is needed.