The built environment as determinant of childhood obesity: A systematic literature review

We evaluated the epidemiological evidence on the built environment and its link to childhood obesity, focusing on environmental factors such as traffic noise and air pollution, as well as physical factors potentially driving obesity‐related behaviors, such as neighborhood walkability and availability and accessibility of parks and playgrounds. Eligible studies were (i) conducted on human children below the age of 18 years, (ii) focused on body size measurements in childhood, (iii) examined at least one built environment characteristic, (iv) reported effect sizes and associated confidence intervals, and (v) were published in English language. A z test, as alternative to the meta‐analysis, was used to quantify associations due to heterogeneity in exposure and outcome definition. We found strong evidence for an association of traffic‐related air pollution (nitrogen dioxide and nitrogen oxides exposure, p < 0.001) and built environment characteristics supportive of walking (street intersection density, p < 0.01 and access to parks, p < 0.001) with childhood obesity. We identified a lack of studies that account for interactions between different built environment exposures or verify the role and mechanism of important effect modifiers such as age.

developing cardiovascular disease, type 2 diabetes, and certain cancers, as well as diminished mental health. [2][3][4][5] Obesity is preventable and reversible. Restricting energy intake and increasing energy expenditure have previously been the focus of prevention and treatment strategies. Most efforts and initiatives have, however, so far been unsuccessful at a population level, and a broadened approach is warranted. 6 The causes of obesity are multifactorial ranging from individual, household, to policy settings. In this context, place-based obesogenic factors are increasingly being recognized as important determinants of obesity, including the social context, the environment individuals live in, and behaviors linked to modern, urban living. 7 In order to target place-based mitigation approaches, interventions, and policy implementations, a clear understanding of the spatial context in which obesity determinants act is needed. 8 The place we live in has increasingly been recognized as a strong determinant of health, including obesity. 9 In this context, the term "built environment" has been coined to describe the physical and built infrastructure in which people live, learn, work, play, socialize, and travel. 10 Within urban settings, the natural infrastructure is an integral part of the wider concept of the built environment. The built environment has strong influences on residents' behaviors, with physical activity and sedentary lifestyles being the most widely studied. 11 Additionally, environmental pollution linked to the built environment such as air pollution and traffic also has strong impacts on urban health. 12 This systematic review synthesizes the empirical evidence on the built environment as determinant of childhood obesity. We focused on environmental factors including traffic noise and air pollution, as well as physical factors potentially driving obesity-related behaviors, including neighborhood walkability, and availability and accessibility of parks and playgrounds. Supported by a rigorous quality assessment and a focus on objectively measured built environment characteristics, we provide a quantitative synthesis of the updated evidence base with an emphasis on conceptual and methodological aspects and public health implications.

| Search strategy
We followed the Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) guidelines and registered the protocol with the International Prospective Register of Systematic Reviews (PROSPERO) database (registration number CRD42020170337). We used a comprehensive and reproducible search strategy to identify peer-reviewed journal articles in the English language, published from inception until February 2020, focusing on three databases: EMBASE, MEDLINE, and Web of Science. A preliminary search identified relevant keywords and MeSH terms at the intersection of two concept clusters: "childhood obesity" and "built environment" (Table S1).

| Eligibility criteria
Studies were eligible for inclusion if they met the following criteria: (1) Population: Children and/or adolescents under the age of 18 years; (2) Exposure: Objectively measured environmental and physical features of the built environment potentially linked to the onset of obesity; (3) Outcomes: Objectively measured and self-reported body mass index (BMI) or BMI standardized for age and sex (BMI z score); (4) Study design: Observational studies (cross-sectional and longitudinal) quantitatively assessing associations of outcome and exposure.
We excluded studies that assessed the built environment as confounder only, those that used self-reported perceived features of the built environment, and studies using controlled experiments in manipulated settings (Table S2). We also excluded studies with an explicit focus on the food environment as this was outside the scope of the review. After the removal of duplicates, articles were screened independently by two reviewers (D.M. and D.F.) against the eligibility criteria, using the online tool Covidence. 13

| Data extraction
Data extraction was performed independently by two reviewers (D.M. and E.H.), and discrepancies were mediated by D.F. Information was extracted on study characteristics (first author, year, study design, study area, sample size), participant characteristics (age, sex), exposure (built environment characteristic, data collection method), outcome measures (outcome, data collection methods, measure of association), individual-and area-level confounders, and main findings (direction and magnitude of association, statistical significance).

| Quality assessment
The quality of the eligible studies was assessed independently by two reviewers (D.M. and E.H.), and discrepancies were mediated by D.F. We used a modified Newcastle-Ottawa scale for quality assessment, 14 which we adapted for the assessment of observational studies. The elements used for the assessment include (1) representativeness of the exposed population, (2) selection of the nonexposed population, (3) objective ascertainment of the exposure, (4) sample size, (5) appropriateness of considered confounding factors, (6) assessment of the outcome, and (7) statistical test used for analysis (Table S3). Stars were assigned for each criterion with a maximum of 12 stars. A score of 0-4 was defined as poor quality, 5-8 as fair quality, and 9-12 as good quality. Publication bias was assessed using a funnel plot.

| Data synthesis
Due to the heterogeneity in exposure metrics and methodologies used across eligible studies, a meta-analysis was not possible. Instead, we used an alternative methodology to assess and synthesize the strength of associations, the weighted z test. 15 This approach has previously been used for systematic reviews on the built environment and health 16,17 and is based on the numb1er of studies with findings in the expected direction and their level of significance. For each study, we assigned a z value based on the level of statistical significance (α) and direction of association (expected direction of association based on research hypothesis vs. unexpected direction of association). If associations were in the expected direction, then z = 1.96 for α = 0.05, and z = 1.64 for α = 0.10; if associations were in the unexpected direction, then z = À1.96 for α = 0.05, and z = À1.64 for α = 0.10; z = 0.00 was assigned to null (statistically not significant) associations with p > 0.10. We summed the z value for each reported finding and weighted these by the quality assessment score for each study, divided by the square root of the sum of squared quality assessment scores. To determine the strength of association for each built-environmentoutcome combination, a two-tailed p value was computed for each weighted z value with interpretation of weak evidence if p < 0.05, strong evidence if p < 0.01, and very strong evidence if p < 0.001. 16 To avoid overrepresentation of individual studies reporting built environment-outcome associations by different subgroups (e.g., boys/ girls, geographic area, and age group), we applied fractional weights to each finding so that the sum of the weights across all reported associations was 1. 17 For example, if a study reported a positive association of fine particulate matter with childhood obesity, but that association was significative (α = 0.05) only in boys (z = 1.96) and not in girls (z = 0.00), the z value assigned to the study was 1.96 * 0.5 + 0 * 0.5 = 0.98. Following the standard set for meta-analysis, associations for each built environment feature-outcome combination were only synthesized if five or more studies reported such associations. We did not find evidence for publication bias ( Figure S5).

| Study characteristics
The four studies investigating effects of traffic noise on childhood obesity were recent (2016-2019) longitudinal studies from Northern Europe ( Table 1). [18][19][20][21] Two studies used national birth cohorts, 19,20 the others longitudinal studies with national coverage. Sample sizes ranged from 3963 to 40,974 participants. All studies assessed exposure to noise through standard modeling methods, linked to the home addresses of the subjects. Three studies used an implementation of the Nordic prediction method for road traffic noise, one study a national noise standard. 18 Methodologies between studies were generally comparable. The Swedish study 20 obtained height and weight from school and health records and, in part, measurements, whereas the three other studies used height and weight from questionnaires.
The Norwegian study 21 accounted for age and sex in the model via interaction terms to explore the effect of noise on BMI trajectory, whereas all others studies either used a age/sex standardization of BMI (BMI z score) and/or categorized BMI based on sex and agespecific cut-offs for overweight and obese from the International Obesity Task Force (IOTF). All studies accounted for age, sex, and maternal education in analysis, in addition to other study-specific confounders including maternal BMI prior pregnancy, 19-21 parental smoking, 18-20 neighborhood socioeconomic status, 18 and physical activity. 20 One study controlled further for urbanization and nitrogen oxides (NO x ). 19 Studies used either linear mixed models, 18,21 multiple regression, 19 or quantile regression 20 with increasing levels of adjustment. All studies were of high quality with scores of 9-10 out of the maximum 12 stars (see Table S5).

| Summary of findings
Due to the small number of studies, meta-analysis was not applied and findings descriptive. Impacts of traffic noise on childhood obesity were observed in three studies, but overall results were mixed and varied by life stage (see Table S4a). Positive associations of roadtraffic noise exposure during pregnancy and the risk of being with overweight in school-age children (7/8 years) were observed in Denmark and Norway 19,21 but not Sweden. 20 For the same age group, no impact of childhood noise exposure on weight was found. [18][19][20][21] Wallas et al., however, studied the effect of traffic noise exposure during adolescence and found a strong association with adolescence BMI between the ages of 8 and 16 years, which was slightly stronger for girls. 20 3.2 | Childhood obesity and air pollution
All studies assessed air pollution exposure at the home address, one study also at school. 26 Five studies modeled air pollution exposure using dispersion models, 25 29 and Sweden. 28 The majority of studies adjusted for age and sex, one study used Tanner stage, 24 and one only studied 4-year-old children. 28 Three studies did not adjust for age but used age and sex standardized BMI measures. 26,29,35 Covariates varied widely across studies and included parental socioeconomic status, maternal BMI, birth weight, parental smoking, and passive smoking exposure. All studies had a quality rating of good, ranging from 9 to 11 stars (see Table S5).

| Summary of findings
To synthesize findings using the z test, we combined NO 2 and NO x results, and PM 2.5 and PM 10 were considered separately (  18,22,24,25,27 and four studies did not find significative results. 26,28,29,33 Two of the studies had mixed results, one found an effect only in boys 34 and in one study the effect dependent on the exposure period. 30 Overall, the association of NO 2 /NO x exposure on childhood obesity was strong with a two-tailed p value from the

| Summary of findings
There was limited evidence that the walkability index is linked to childhood obesity (p = 0.28), with only one out of 10 studies finding significant associations 40 (Table 2). Two further studies showed mixed results based on sex (effect on bodyweight status in girls, but not boys) 38 and geographic area (healthy BMI associated with higher levels of walkability in one of three studied cities). 42 The Walk Score was associated with decreased BMI z score in rural but not urban youths in one study 45 but did not show any significant association in another study. 37 The walkability index based on street element characteristics, however, did identify a significant association with childhood obesity. With regards to individual walkability indicators, street intersection density was the most widely used indicator (n = 7). Three studies found significant associations with childhood obesity, 36,43,51 one study found a weak positive association, 52 mixed results were found in two studies, with effects observed in girls but not boys, 38 and one out of three studied cities. 42 The z test revealed strong evidence to support a link between street intersection density and obesity measures (p = 0.005). Out of six studies analyzing associations with population density, only one study found an effect of lower residential density being linked to higher BMI z score, 36 and one study found an effect only in girls. Overall, the evidence did not suggest a  Land use mix was only analyzed in four studies, with one study finding a significant association.

| Study characteristics
The dominant study design of the 28 included studies was cross-sectional (n = 20), the others longitudinal (n = 8). 18,29,[53][54][55][56][57] One of the longitudinal studies conducted a quasi-experiment, which considered a pre-park and post-park time frame and dividing the children into those who live near the park (the exposure group) and those who live further from the park (the control group) to examine how exposure to a newly built park translates to changes in BMI z score over time. 58 Almost half of the studies were conducted in the United States The outcomes analyzed were BMI z score, BMI trajectories, BMI percentiles, and weight status. Anthropometric measures were rarely used: waist circumference (n = 3), 55,76,77 waist-to-height ratio (n = 1), 55 sum of skinfold (n = 1), 48 and percentage body fat (n = 2). 47,77 The quality of the studies was either fair (n = 6) or good (n = 22).
The main reasons for fair quality were small sample sizes, selfreported outcomes (height and weight), or study population scarcely representative of the population (see Table S5).

| Summary of findings
Due to the great variability in exposure metrics, we synthesized findings across the following exposure categories: distance to the nearest park (n = 9), park area (n = 10), number of parks (n = 8), and presence/absence of parks (n = 5). Only three studies analyzed NDVI, which was insufficient for meta-analysis according to our criteria ( Table 2). The z test and related p value suggest that there was insufficient evidence to support an association of distance to park and childhood obesity (p = 0.170). Out of the nine studies, only one found a significant association. 67 Two studies concluded with mixed findings: One study found a significant association in boys of all ages and girls of high school age but not in younger girls, 52 and one study found an significant association in children living in urban areas but not those in rural areas. 18 The p value suggested weak evidence of an association with percentage of park area (p = 0.014). Three studies found significant associations, six studies found no statistically significant effects, and two studies had mixed results, with effects only found in boys and older children. The p value showed little evidence of an effect of number or density of parks on childhood obesity (p = 0.148). One study found a significant association, five studies did not find significant associations, and one study reported mixed results with effects only in girls. 69 The intervention study did, however, find an effect in the intervention group, which could not be replicated in the control group. 53 We identified strong evidence on the presence of a park within the sphere of influence and childhood obesity (p < 0.001). Out of the five studies, four studies found statistically significant effects.
Results from the three studies that explored the effect of greenness via the NDVI suggest a potential association in the more proximal environment of less than 250 m. 18,63,65 Three studies specifically focused on playgrounds, and none of them found statistically significative associations.

| Impact of built environment characteristics on childhood obesity
We systematically reviewed the ep1idemiological evidence on the influence of four built environment characteristics on obesity outcomes in children: traffic noise, air pollution, neighborhood walkability, and accessibility and availability of parks and playgrounds.
To our knowledge, this is the first systematic review on this topic that applied a systematic synthesize of findings to evaluate the strength of the available evidence.
Studies were generally of high quality, using objectively measured outcome and exposure measures and adjusting for relevant confounders. Some studies, however, had small sample sizes, which were not necessarily representative of the overall population. Overall, 42% of studies used longitudinal data; however, the small number of longitudinal studies investigating effects of neighborhood walkability and parks accessibility should be emphasized.
We found very strong evidence of association of BMI-derived obesity outcomes with NO 2 /NOx (p < 0.001) and presence/absence of parks in the neighborhood (p < 0.001), strong evidence with intersection density (p < 0.01), and some evidence with the amount of park area in the neighborhood (p < 0.05). There was little evidence of an effect on childhood obesity in relation to PM 2.5 , PM 10 , walkability index, residential density, distance to the nearest park, number of parks, and access to playgrounds.
Air pollution has been shown to decrease birth weight 78 and might independently affect weight in childhood through epigenetic and behavioral adaptation. Some hypotheses on the mechanism involved in the exposure both during pregnancy and childhood were highlighted in previous publications: Prenatal growth restrictions can lead to growth spurts in early childhood with implications on increased weight into later childhood and adolescence 79 ; heavy traffic roads, an important sources of air pollution, might deter active transport and reduce physical activity. 80 Our findings point towards this direction with traffic-related air pollutants NO 2 and NO x having a strong impact on increased weight in childhood, but not particulate matter (PM 2.5 and PM 10 ), which is driven to a lesser degree by local traffic. 18 Another explanation could be the biochemical mechanism that emphasizes the role of NO 2 as active oxidant involved in many physiological pathways in the human body, which might impact consequently the onset of obesity. 81 Despite evidence suggesting a link between walkability and physical activity, 82  We also found strong evidence for the presence (or accessibility) of parks with decreased prevalence of childhood obesity, whereas studies focusing on playgrounds did not find significative associations. This is supported by findings from Bird et al.
who concluded that parks that emphasize unstructured activities (i.e., with few team sport installations) were associated with lower percentage of truncal fat among children at risk of being with obesity. 83

| Methodological considerations
Some of the included studies investigated more than one built environment characteristics. Several studies explored walkability and parks. 29,47,48,52,70,84 Among the studies that considered walkability and greenspaces, walkability was not statistically significant, except intersection density in boys in one of the studies, 47 and greenspace was at least partially associated with weight outcomes in all studies.
No multi-exposure interactions were evaluated in these studies, except for a Pearson correlation coefficient between intersection density and park space, which did not show collinearity. 47 Overall, we found a lack of studies that explore the interaction between multiple exposures on childhood obesity. Bloemsma et al. 18 investigated the combined effect of noise, air pollution, and park accessibility. They found that the association of NO 2 with overweight remained after adjustment for noise and greenspace, but the associations between greenspace and overweight weakened substantially after adjustment for NO 2 , indicating that NO 2 is driving the relationship. To better understand the complex relationship of multiple built environment characteristics on childhood obesity, more evidence is required.
Our review highlighted a strong presence of effect modifiers. Sex was the most studied effect modifier, but there was no consistency across studies. Two studies reported an increased effect in boys for the association between air pollution exposure and BMI, 31,34 but one of the studies found also an opposite effect considering waist-to-hip ratio as anthropometric measure, which was statistically significant only in girls. 31 Walkability and intersection density were found to be associated with body weight status in girls but not in boys in one of the studies, 38 but in another study, a high level of street connectivity was related to lower percentage of body fat only in boys. 47 The association between park accessibility and obesity was gender-dependent in five studies, of which three showed more significant effects on boys 52,54,55 and two on girls. 47,69 Overall, sex affected the results in nine studies, concluding with an increased effect in boys in five studies, in girls in three studies, and with opposite effects depending on the considered anthropometric measure in one of the studies. Age was another common effect modifier, showing differential results in five studies. In one study, the exposure to road traffic noise was associated with increased BMI from school age to adolescence, but not at earlier ages, the relation increased in the older age groups. 20 Age also modified the association between greenspace exposure and BMI in four studies (two of them were based on the same sample), always with increased effects in older children. 29,52,54,55 Another effect modifier was urbanization, with one study finding a negative association between walk score and BMI z score for youths in rural settings and a positive association among urban youths, 45 whereas in another study, children living in a urban area had a negative association of the distance to the nearest park with weight status and no association for those living in rural areas. 18

| CONCLUSION
In summary, we found strong evidence for an association of traffic-related air pollution (nitrogen dioxide and nitrogen oxides exposure, p < 0.001) and built environment characteristics supportive of walking (street intersection density, p < 0.01 and access to parks,