Chapter Social environment: definitions, measures and associations with weight status a systematic review

Transcription

1 Chapter.3. Social environment: definitions, measures and associations with weight status a systematic review K. Glonti, J.D. Mackenbach, J. Ng, J. Lakerveld, J.-M. Oppert, H. Bárdos, M. McKee, H. Rutter Based on Glonti et al. Psychosocial environment: definitions, measures and associations with weight status a systematic review. Obes Rev 7(S), 8-95 Abstract Socio-ecological models suggest that many elements of the social environment act as upstream determinants of obesity. This systematic review examined definitions, measures and strength of associations between the social environment and adult weight status. Studies were included if they were: conducted on adults, the outcome was weight status, carried out in any developed country, and investigated at least one social environmental construct. Six databases for primary studies were searched: EMBASE, MEDLINE, PsycINFO, Scopus, Web of Science and the Cochrane Library. We restricted our search to studies published in English between January 995 and February 05. An adapted Quality Assessment Tool for Quantitative Studies was used to evaluate risk of bias of included studies. Out of 4,784 screened records, 4 articles were assessed using full-text. A total of 9 studies were included. The strongest associations with weight status were found for social capital and collective efficacy, although few studies found significant associations. There was heterogeneity in the definitions and metrics of social environmental constructs. There is limited evidence that greater social capital and collective efficacy are associated with healthier weight status. The research conducted to date has not robustly identified relations. We highlight challenges to undertaking research and establishing causality in this field and provide recommendations for further research. 3.3

2 Introduction.3 Obesity is a worldwide public health concern, with increasing obesity rates, particularly in high income countries.[] Poor diets, physical inactivity and sedentary behaviour are major drivers of obesity, and frameworks such as the ANalysis Grid for Environments Linked to Obesity (ANGELO) have been designed in an effort to conceptualise and understand factors operating at various levels that drive the obesity pandemic through these behaviours.[] These suggest that there is a dynamic relation between individuals and their broader social environment. Although there is no universally agreed definition, social environment has been broadly defined as the environment that influences an individual s behaviour through promoting a sense of social control through the creation of social norms.[3] It encompasses social determinants of health that can potentially be altered.[4] Social environment can be grouped into five dimensions - social support and social networks, socioeconomic status and income inequality, racial discrimination, social cohesion and social capital, and neighbourhood deprivation.[5] A 0 systematic review of the influence of built environments on cardio-metabolic risk factors, including obesity, found the strongest associations with community socioeconomic factors, while aspects of social interactions were rarely investigated.[6] Two systematic reviews linking obesity to socioeconomic status have been published[6, 7] but the remaining social constructs that gained prominence in epidemiological research on weight status and obesity have not yet been summarised.[9-] The remaining constructs are of a psychosocial nature and include social cohesion[3]; collective efficacy,[4] social capital,[6] social support, and social networks.[9] The social environment includes the communities to which individuals belong and the policies created to provide order to their lives[5] and varies across contexts and settings.[9] Our aim was systematically to review and evaluate evidence on associations or lack of them between the social environment and weight status in adults. We also summarise the definitions and measurement approaches used to explore relations between the social environment and weight status, and provide recommendations for future research. By understanding the role of the social environment in weight status, potential targets for policies to prevent increasing obesity may be identified.[] This review was conducted as part of a European-funded project (SPOTLIGHT) addressing the social and environmental determinants of obesity.[3] Methods The review included published primary research that investigated the association between social environmental constructs and weight status in adults (aged 8 years). Six electronic databases: EMBASE, Global Health, MEDLINE, PsycINFO, Scopus, Web of Science, and the Cochrane Library were searched in February 05. Reference lists of all included papers were searched for articles that had not been identified in the electronic searches. We restricted our search to studies published in English between January 995 and 5th February 05. Studies were included if they were: conducted on adult populations; the outcome was weight status; investigated at least one 4

3 social environmental construct; were not focused solely on socioeconomic status; presented original research and were carried out in any developed country, i.e. classified as very high human development by the Human Development Index of the United Nations Development Programme. We adhered to an adapted version of the Preferred Reporting Items for Systematic Reviews and MetaAnalyses (PRISMA),[8] to conduct this review and report its findings. Study screening and selection Following removal of duplicates, two reviewers (JDM and KG) independently screened search results by title and abstract. Full texts of potentially suitable articles were obtained and reviewed for inclusion by both reviewers. Disagreements were resolved by discussion and consensus. Data extraction and analysis Data from each study on country, study design, definitions used, analyses performed, and key findings were extracted for a narrative synthesis. Substantial differences in variables, methods, data and context limited direct comparison and rendered a statistical meta-analysis of the findings difficult. For this reason, we did not extract the effect sizes and p-values presented in the original studies. Risk of bias assessment We used an adapted version of the Quality Assessment Tool for Quantitative Studies (as developed by the Effective Public Health Practice Project) [9] that has been used in this way before [0] to assess the risk of bias of the included studies. The tool contains 9 items within eight key domains. As some of these relate only to specific study designs, each paper received a score accordingly. Studies can have between six and eight ratings, with each component score ranging from (low risk-of bias; high methodological quality) to 3 (high risk-of-bias; low methodological quality). An overall rating for each study was determined based on its component rating. The study assessments were completed by two independent reviewers (JDM and JN). The ratings for each of the eight domains, as well as the total rating, were compared and consensus was reached on a final rating for each included article. Results The results of the screening process are described using an adapted Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flowchart (Figure ).[8] In total, 4,784 articles were screened by title and abstract for inclusion. The full text of 4 articles was obtained and assessed for eligibility. Although a total of 0 studies met the eligibility criteria for this review, only 9 were included because two papers referred to the same study with overlapping results.[, ] We included the 0 study by Prince et al.[] rather than the later one by the same group because of its more comprehensive description of methods and results. The studies were conducted in Australia,[3, 4] Canada,[, 5, 6] Portugal,[7] UK,[8] and.[9-30] Eighteen studies used a cross-sectional and one a longitudinal design. Full details are provided in Table

4 Country Australia Source Ball and Crawford, 006 [9] Bjornstrom, 0 [4] Burdette et 006[8] Study design general linear models One way anova and Logistic regression models Bivariate regression and multiple linear regression Analysis Collective efficacy as neighbourhoodlevel measure Collective efficacy as neighbourhoodlevel measure Social support as individual-level measure Social environmental construct and level 6 Census tracts Census tracts National level (urban, rural and remote areas) Spatial scale Obesity definition ( 30 kg/m) from measured and/or selfreported weight and height Obesity definition ( 30 kg/m) from self-reported Obesity definition ( 30 kg/m) from self-reported Weight measure Table.3... Description of the studies on associations between social environment and weight status No significant association between collective efficacy and either BMI or obesity Collective efficacy does not mediate the relationship between inequality and obesity. It is associated with a decrease in the odds of obesity Social support for healthy diets from family, and sabotage to healthy diets and physical activity from friends, significantly predicted women's current BMI, with higher levels of support and lower levels of sabotage associated with higher BMI Associations with weight status.3 Moderate () Strong () Moderate () Quality rating

5 Australia Canada Christian et 0[8] Greiner, et 004[5] Holtgrave and Crosby, 006 [33] Kaplan, et 003[] Keller et 03[3] Longitudinal Bivariate correlations Multivariate logistic regression Linear regression model Multivariable logistic regression Multiple linear regression Postcode sectors State level State level Participants were asked about their community New housing development 7 Social cohesion and social support as individual-level measure Social support as individual-level measure Social capital as state-level measure Community social capital as neighbourhoodlevel measure Social capital and social cohesion as neighbourhoodlevel measure Waist-to-hip ratio body composition from measured Obesity definition ( 30 kg/m) from self-reported Obesity definition ( 30 kg/m) from self-reported Obesity definition ( 30 kg/m) from self-reported No definition of obesity from self-reported Moderate () Strong () Strong () Strong () Strong () No significant association between BMI and social capital or social cohesion No statistically significant association between obesity and community ratings or community involvement Social capital correlated moderately with obesity, shown to have a protective effect against obesity Social support was associated with a lower likelihood of obesity for women and higher likelihood of obesity for men No statistically significant associations between BMI and social cohesion or social support.3

6 Kendzor et 03[34] Kim et 006 [35] Moore et 03[9] Mujahid et 008[6] Marginal maximum likelihood estimation Paired t-tests, Spearman rank correlations, anova and two-level mixed regression Two-level analyses and three-level analyses Latent variable modelling Social cohesion as neighbourhoodlevel measure Social cohesion as neighbourhoodlevel measure Social capital as state-level and county-level measure Social support as individual-level measure 8 mile area around home for nonmesa participants Census tracts State and county Houston metropolitan area No definition of from measured No definition of obesity from self-reported Obesity definition ( 30 kg/m) from self-reported Obesity definition ( 30 kg/m) from measured No definition of obesity No statistically significant association between BMI and social cohesion in women. Statistically significant associations were present for men who had higher social Point estimates of associations revealed that higher social cohesion was associated with lower BMI for both the home and workplace environments, although associations were not statistically significant for home environment and social cohesion Social capital had modest protective effects on obesity Social support was inversely associated with negative effect and positively associated with BMI.3 Strong () Strong () Strong () Strong ()

7 UK Canada Portugal Poortinga, 006 [3] PowellWiley et 03 [7] Prince et 0[7] Santana et Multilevel logistic regression Multilevel multivariable logistic regression Multivariable logistic regression Multilevel logistic regression Social capital Social capital Social cohesion as neighbourhoodlevel measure Social capital and social support as neighbourhoodlevel measure 9 Administrative boundaries of Natural barriers (e.g. rivers), spatial scale measurements and individual-level data PHDCN-CS definitions (not specified), participants were asked about their neighbourhood Postcode sectors from self-reported Definition of obesity ( 5 kg/m) from self-reported Definition of obesity ( 30 kg/m) from measured Waist circumference Obesity definition ( 30 kg/m) from measured obesity Strong () Strong () Strong () Moderate () Trust is associated with a decrease in the odds of obesity. No statistically significant association between obesity and social cohesion In men, stronger neighbourhood sense of community belonging was significantly associated with being physically active, and lower neighbourhood SES and higher voting rates were significantly associated with lower odds for overweight and obesity. No statistically significant association between social capital cohesion; they had a higher mean BMI..3

8 Multilevel linear models Community social capital 88 counties Census tract level, sociodemographic data and GIS to identify neighbourhoods from self-reported. Definition of obesity ( 30 kg/m) overweight ( 7 kg/m) Definition of from self-reported Definition of obesity ( 30 kg/m) from self-reported Definition of overweight and obesity ( 5 kg/m) Greater CSC on average appears to increase the likelihood of eating fewer calories or less fat to reduce or maintain weight. Depth and breadth of involvement in voluntary associations significantly related to overweight status. More involvement in networks of associations corresponded with a lower likelihood of being overweight. The odds of obesity were significantly lower for adults who reported involvement in clubs, associations or help groups and overweight Strong () Moderate () Moderate () 0 BMI, body mass index; CSC, Community social capital; GIS, geographic information system; MESA, Multi-Ethnic Study of Atherosclerosis; PHDCN-CS, Project on Human Development in Chicago Neighborhoods Community Survey; SES, socioeconomic status. Multivariable logistic regression Social capital Yoon and Brown, 0 [3] Canada Participants were asked about their neighbourhood. Veenstra et 005[0] Logistic regression Social capital No definition Proc survey logistic procedures Sullivan et 04[30] wards 009[].3

9 Risk of bias assessment Thirteen studies had a low risk of bias, with the remaining six studies having a moderate risk of bias. Full details of all ratings are provided in Table S.3... Seven studies [, 5, 9, 0, 7, 8, 30] were rated as strong and twelve as moderate [3, 4, 6-8,,, 4-6, 9] for representativeness. All studies with the exception of four moderate ones [4, 3, 5, 30] received a strong rating for avoidance of confounding. Eight studies scored strong [6, 8, 0,, 5-8] on data collection, nine scored moderate [, 4, 6, 7, 9,, 3, 4, 30] and two scored weak [3, 5] as they did not provide any information about the validity and reliability of their measures. With the exception of one [6] all studies received a strong rating for the robustness of their data analysis. Seventeen studies showed high quality of reporting and received 'strong' ratings while the remaining two achieved moderate [8, 9] scores in reporting quality. Definitions and constructs of the social environment In the included studies the authors distinguished between five social environmental constructs; two studies examined collective efficacy, five social cohesion, ten social capital, and five social support. These definitions and applications of these constructs across the included studies are summarised below: Collective efficacy Bjornstrom investigated whether local income inequality was associated with an increased likelihood of obesity and whether collective efficacy mediated the relation.[9] The author refers to Sampson et al. s definition of collective efficacy as a community social resource that captures the extent to which residents share norms, trust one another and are willing to intervene for the common good.[3] Burdette et al. used the same definition and examine the hypothesis that mothers of young children would have a higher prevalence of obesity if they lived in a neighbourhood that they perceived as having a low level of collective efficacy. [3] Social cohesion Five studies examined social cohesion.[3,,, 4, 6] Two looked at associations between perception of neighbourhood physical and social environments and weight status[, ]; however neither specified an overall definition of social cohesion, instead referring to measures that have been used as indicators or proxies of social cohesion. Thus, both Mujahid et al. and Powell-Wiley et al. included willingness to help neighbours and people trusted as indicators of social cohesion. In addition, Mujahid et al. included sharing similar values, and getting along with each other,[] whereas Powell-Wiley et al. included reporting a close-knit neighbourhood as a constituent element of social cohesion.[] Christian et al. examined whether the individual, behavioural, social and built environments were correlated with body mass index (BMI). The authors did not define social cohesion but referred to The Neighbourhood Cohesion Scale.[3] Moore et al. evaluated how residential and work environments may separately and jointly relate to BMI, referring to the definition of social cohesion proposed by Sampson et al.[3] Finally, Keller et al. analyzed relations between social support, neighbourhood environment factors and body composition among postpartum Latinas in the. The authors do not explicitly define social cohesion.[6].3

10 Social capital.3 A further ten studies focused on social capital.[, 3, 5, 7, 8, 0, 5, 7, 8, 30] The aim of the exploratory study by Holtgrave and Crosby was to examine the relation between social capital and obesity and diabetes in the adult population in the. The authors conceptualised social capital as a combination of trust, reciprocity, and cooperation among members of a social network, referencing work by Kawachi et al. and Putnam.[8] Prince et al. examined the relation between variables related to the built and social environments and physical activity (PA) and overweight/obesity. Civic participation and trust, as well as sense of belonging, were used as indicators of social capital.[] The authors referred to Kawachi et al. s construct of social capital,[33] described as elements of community organization such as civic participation and sense of trust between citizens that contribute to the mutual benefit of the community. Poortinga examined the associations of perceived social capital with people s physical activity, BMI and general health,[8] also drawing on Kawachi et al. s construct of social capital.[33] Greiner et al. examined the relation between community social capital and health outcomes that included obesity. The authors broadly subscribe to Putnam s [35] definition of social capital as the resources available to individuals and communities through their social connections.[0] Yoon et al. also refer to Putman s definition, focusing on the structural component of community social capital, manifest as community-level resources such as density of social networks that facilitate cooperative actions for mutual benefit among members of a community.[7] Kim et al. explored the relations between social capital measured at US state and county levels and individual leisure-time physical inactivity.[30] The authors broadly refer to social capital as the resources embedded in social structures which are accessed and/or mobilised in purposive action [34] and also refer to Putnam s state-level social capital index.[35] Christian et al. also explored, but did not define, social capital. Elements of social capital included satisfaction with number of friends, number of people known and whether the participants neighbourhood was perceived as a good place to live and raise children.[3] Veenstra et al. explored the association between involvement in voluntary associations and body mass index, describing social capital in terms of the depth and breadth of involvement in voluntary organizations, with reference to Berkman et al.[5, 36] Santana et al. examined the contribution of the local environment (both physical and social) and personal attributes to the risk of weight gain and obesity.[7] Although the authors referred to the concept of bonding social capital they did not explicitly define it, although in the literature on social capital it is normally used to describe links between individuals within the same group, as distinct from bridging social capital, which describes links between individuals of different groups.[37] Bonding social capital was assessed at neighbourhood level using variables such as the number of recreational or sport clubs, number of local newspapers, and the local newspaper circulation per inhabitant.[7] Sullivan et al. examined possible associations between perceived neighbourhood environments and obesity. The authors provided no definition of social capital nor a reference to a social capital construct but used participation in clubs, associations, and groups in neighbourhood as a proxy indicator for social capital.[5] Poortinga s study investigated the links between perceived aspects of the social and physical environment,

11 physical activity obesity, and self-rated health. The author provided no particular definition for social capital but defined it through social trust and civic participation.[8].3 Social support Five studies examined social support.[4, 6, 6, 9] Ball and Crawford explored the biological, psychological, social and environmental correlates of young women s current weight and retrospective -year weight change. The authors made no reference to any particular definition of social support but refer to it as social support from family/friends for physical activity and healthy eating behaviours.[4] Kaplan et al. examined the factors associated with overweight and obesity among older men and women. They provide no definition of social support but reference previous work,[38] stating that their measure encompasses major distinct dimensions of social support such as instrumental, informational, appraisal and emotional support.[6] Kendzor et al. evaluated a previously tested model of the psychosocial pathways through which socioeconomic status might influence BMI. One of the hypothesised psychosocial mediators is social support. The authors do not provide a definition of social support or any reference to a particular concept.[9] Keller et al. analysed relations between social support, neighbourhood environmental factors and body composition among postpartum Latinas in the. The authors do not define social support or social cohesion.[6] Finally, Poortinga provided no particular definition for social support but used the social support scale and grouping from the 987 Health and Lifestyle Survey.[8, 39] Measures of the social environment There was great heterogeneity across studies with regard to the constructs and measures of the social environmental factors. Full details are provided in Table.3... Collective efficacy In accordance with the construct of Sampson et Bjornstrom [9] and Burdette et al. [3] examined collective efficacy as a neighbourhood-level measure comprised of aggregated responses to items that capture trust, social cohesion, and the willingness to intervene for the common good among residents. Social cohesion The five studies that examined social cohesion all differed in the measures used. Mujahid et al. assessed social cohesion as neighbourhood-level measure assessed by a social environment score that combined items in four dimensions (greater aesthetic quality, safety, less violent crime and social cohesion) in a factor analysis. Social cohesion was measured using four items based on research by Diez-Roux.[] Powell-Wiley et al. conceived social cohesion as a neighbourhood-level measure.[] It is assessed through three items abstracted from questions on neighbourhood perception used in the 994 Project on Human Development in Chicago Neighbourhoods community survey.[3] Christian et al. measured social cohesion at neighbourhoodlevel using an 8-item Neighbourhood Cohesion Scale.[3, 40] Moore et al. examined social cohesion as neighbourhood-level measure assessed with four items based on the 5-item Sampson Scale [3] but indicate in their article that they are using four items to assess social cohesion without further 3

12 specification.[4] Finally, Keller et al. measured social cohesion as an individual-level measure, comprising of four items measuring willing to help neighbours, trust-worthiness, share same value, getting along with each other.[6].3 4

13 Social support as individual-level measure assessed by four subscales. These comprised two sets (family/friends) of 8 questions, adapted from Sallis et al. The four subscales were as follows () support for healthy eating (six items: e.g. how often have family, partner, children, parents complimented me on my eating habits); () sabotage of healthy eating (three items: e.g. offered me high fat or unhealthy foods); (3) support for physical activity (six items: e.g. participated in physical activity with me); and (4) sabotage of physical activity (three items: e.g. suggest things that are physically inactive). Collective efficacy as a neighbourhood-level measure composed aggregated responses to 0 items that capture social cohesion, trust and propensity for collective action: () This is a close-knit neighbourhood ; () Adults in this neighbourhood watch out for children ; (3) There are adults in this neighbourhood kids can look up to ; (4) People in this neighbourhood can be trusted ; (5) People generally don't get along with each other ; (6) Residents are willing to help their neighbours ; (7) Residents do not share the same values ; (8) Neighbourhood children were skipping school and hanging out on a street corner ; (9) Spray-painting graffiti on a local building ; and (0) Showing disrespect. Collective efficacy as a neighbourhood-level measure composed of aggregated responses to 0 items that capture social cohesion, trust and propensity for collective action: () This is a close-knit neighbourhood ; () Adults in this neighbourhood watch out for children ; (3) There are adults in this neighbourhood kids can look up to ; (4) People in this neighbourhood can be trusted ; (5) People generally don't get along with each other ; (6) Residents are willing to help their neighbours ; (7) Residents do not share the same values ; (8) Neighbourhood children were skipping school and hanging out on a street corner ; (9) Spray-painting graffiti on a local building ; and (0) Showing disrespect. Social capital as a neighbourhood-level measure composed of aggregated responses to five items measuring neighbourliness. Social cohesion as a neighbourhood-level measure composed of aggregated responses to 8-item neighbourhood cohesion scale () Overall, I am attracted to living in this neighbourhood; () I feel like I belong to this neighbourhood; (3) I visit my friends in their homes; (4) The friendships and associations I have with other people in my neighbourhood mean a lot to me; (5) Given the opportunity, I would like to move out of this neighbourhood; (6) If I need advice about something I could go to someone in my neighbourhood; (7) I believe my neighbours would help in an emergency; (8) I borrow things and exchange favours with my neighbours; (9) I would be willing to work together with others on something to improve my neighbourhood; (0) I plan to remain a resident of this neighbourhood for a number of years; () I like to think of myself as similar to the people who live in this neighbourhood; () I rarely have a neighbour over to my house to visit; (3) I regularly stop and talk with people in my neighbourhood; (4) Living in this neighbourhood gives me a sense of community; (5) Overall I think this is a good place to bring up children; (6) If the people in my neighbourhood were planning something, I'd think of it as something we were doing rather than they were doing; (7) I think I agree with most people in my neighbourhood about what is important in life; and (8) I feel loyal to the people in my neighbourhood. Social support: from family/friends for physical activity and healthy eating behaviours Collective efficacy: social cohesion, trust and propensity for collective action Collective efficacy: social cohesion and trust and informal social control Social capital: satisfaction with number of friends, number of people known, if neighbourhood is a good place to live and raise children Social cohesion: belonging, loyalty, helping each other, collective action and shared values Ball and Crawford, 006 [9] Bjornstorm, 0 [4] Burdette et 006[8] Christian et 0[8] 5 Measures Constructs Source Table.3... Social environment constructs and measures as presented in the included studies.3

14 Social capital as a state-level measure composed of aggregated responses to a combination of 4 variables that span the domains of community organizational life, involvement in public affairs, volunteerism, informal sociability and social trust. Social support as individual-level measure assessed with four items that reflected whether the respondents felt that they had someone they could confide in, count on, give them advice and make them feel loved and cared for. The total social support score was derived from the sum of all affirmative responses (yes vs. no) to the four items. The items tap on major distinct dimensions of social support such as instrumental, informational, appraisal and emotional. Social cohesion and social support as individual-level measure. Social cohesion (four items): () Willing to help neighbours, () Trustworthiness, (3) Share same values, and (4) Getting along with each other. Social support for exercise (PA): adapted 9-item version of Social Support and Exercise Survey assessing frequency of engagement and participation of family members and friends in PA Social capital: trust, reciprocity and cooperation among members of a social network Social support: instrumental, informational, appraisal and emotional Social cohesion: willing to help neighbours, trustworthiness, share same value and getting along with each other Social support: for exercise and general Social support: tangible support, belonging and appraisal Social capital Social cohesion: willing to help neighbours, trustworthiness, share same Holtgrave and Crosby, 006 [33] Kaplan et 003[] Keller et 03[3] Kendzor et 03[34] Kim et 006 [35] Mujahid et 008[6] 6 Social cohesion as neighbourhood-level measure assessed by social environment score that combined measures of aesthetic quality (five items), walking environment (seven items), availability of healthy foods (three items), safety (three items), violent crime (four items) and four items for social cohesion: () Willing to help neighbours () Trustworthiness, (3) Share same values, and (4) Getting along with each other. Social capital as a state and county level measure composed of aggregated responses to two state-level social capital scales from 0 indicators and two county-level scales from five indicators Social support as individual-level measure. The ISEL- is a -item self-report measure of perceived availability of social support that contains three subscales. The Tangible Support subscale measures the perceived availability of material aid (e.g. able to borrow money if needed), the Belonging subscale measures the perceived availability of others with whom one may engage in activities, and the Appraisal subscale measures the perceived availability of others with whom one can talk about problems. Items are rated on a 4-point scale, and scores may range from 4 to 6 on each subscale. Higher scores indicate greater social support. Authors refer to community social capital but divide the concept into two questions that measure community rating/trust (How would you rate your community as a place to live?) and social participation (During the past 5 years, have you been active in a coalition or civic group which attempted to address one or more community problems?), which are reviewed separately as county-level measures composed of aggregated responses. Community social capital: approximation to trust and social participation Greiner et 004[5] Measures Constructs Source.3

15 Social capital and sense of belonging to community as neighbourhood-level measure. Social capital: measured by proxy using 006 voting rates Sense of belonging to community: one item from CCHS questionnaire: How would you describe your sense of belonging to your community? Social capital as neighbourhood-level measure. Quantification by researchers observation Social capital: civic participation and trust Sense of belonging to community Social capital: number of recreational or sport clubs, number of local newspapers and local newspaper circulation per inhabitant Social capital: civic participation Social capital: depth and breadth of involvement in voluntary organizations PowellWiley et 03 [7] Prince et 0[7] Santana et 009[] Sullivan et 04[30] Veenstra et 005[0] 7 Social capital as neighbourhood-level measure. Three items in researcher-created questionnaire () Please tell me the name of any group or voluntary organization you belong to () How involved are you in the activities and affairs of this group or organization you belong to? (very involved, somewhat involved, not very involved, not at all involved) (3) Is there any other group or Social capital as neighbourhood-level measure. Social capital measured by proxy using participation in block clubs, associations and help groups in neighbourhood Social cohesion as neighbourhood-level measure assessed through three items from PHDCN-CS questionnaire: () Willingness to help neighbours, () Close-knit neighbourhood, and (3) People trusted Social cohesion: willingness to help neighbours, closeknit neighbourhood and people trusted Poortinga, 006 [3] Measures Social capital and social support as neighbourhood-level measure. Social capital: two items from HSE questionnaire Generally speaking, would you say that most people can be trusted? Or you can't be too careful in dealing with people. Civic participation was measured by asking respondents to indicate whether they regularly join in activities of 4 types of clubs or associations Social support: summation of responses to seven statements and subsequent categorization into three groups from the 987 Health and Lifestyle Survey value and getting along with each other Constructs Social capital: social trust and civic participation Social support: rating the physical and emotional support obtained from family and friends under various scenarios Source.3

16 Community social capital Yoon and Brown, 0 [3] Community social capital as community-level measure. Ratio of full-time employees in voluntary organizations to population (PSCI). Seventeen out of 8 categories of voluntary organizations were applied here. Only the category any group which meets over the internet could not be matched. voluntary organization you belong to? Measures 8 CCHS, Canadian Community Health Survey; HSE, Health Survey for England; ISEL-, Interpersonal Support Evaluation List; MOS-SS, Medical Outcomes Study: Social Support Scale; PA, physical activity; PHDCN-CS, Project on Human Development in Chicago Neighborhoods Community Survey; PSCI, Petris Social Capital Index. Constructs Source.3

17 Social capital Different approaches to measuring social capital were used in ten studies. Holtgrave and Crosby used state-level correlational analyses. The measure of social capital was obtained from Putnam s public use data set.[35] The measure is a combination of 4 variables that span the domains of community organizational life, involvement in public affairs, volunteerism, informal sociability, and social trust.[8] Prince et al. assessed social capital as neighbourhood-level measure by proxy, using voter turnout in the 006 Ottawa municipal election and by aggregating selfreported neighbourhood values using a single item repeated over four cycles of the Canadian Community Health Survey (CCHS) to generate a measure of sense of community belonging.[] Poortinga examined social capital as a neighbourhood-level measure. Social trust and civic participation were used as indicators of social capital,[4] measured using two questions from the 003 Health Survey for England.[8] Greiner et al. used two constructs; community rating and social participation to approximate the trust and social participation items employed in a prior study by Goodman et al.[37] as structural indicators of social capital, although they are measured separately, not as a score.[0] Yoon et al. s measure of community social capital is the Petris Social Capital Index (PSCI). The PSCI is a geographically based proxy for community social capital and is calculated from the County Business Patterns and population data from the U.S. Census Bureau, and represents a ratio of total employees in voluntary organizations. The PSCI is based on membership of voluntary organizations, one of the social capital measures collected by the Social Capital Community Benchmark Survey (SCCBS). In the SCCBS there are 8 different categories of voluntary organizations. The authors were able to match all of them with the exception of any group which meets over the internet.[7] Kim et al. measured social capital at US state and county level. Two state-level social capital scales were derived from 0 indicators and two county-level scales from five indicators.[30] Christian et al. assessed social capital as a neighbourhood-level measure comprising aggregated responses to five items measuring neighbourliness.[40] Veenstra et al. measured social capital as a neighbourhood measure using two items on involvement in voluntary associations/organizations.[5] Santana et al. assessed social capital as a neighbourhood-level measure. The authors measured bonding social capital by the number of local recreational/sports associations and number of local newspapers as well as the local newspaper circulation per inhabitant.[7] Sullivan et al. assessed social capital as neighbourhood-level measure by proxy using participation in block clubs, associations, and help groups in neighbourhood.[5] Social support Ball and Crawford measured social support as an individual-level measure for weight-related behaviour with eight subscales. These comprised two sets (family/friends) of 8 questions, adapted from Sallis et al.[4] Kaplan et al. measured social support as an individual-level measure assessed using four items that captured whether the respondents felt that they had someone they could confide in, count on, give them advice, or make them feel loved and cared for. The total social support score was derived from the sum of all affirmative responses to the four items.[6] Kendzor et al. assessed social support as an individual-level measure. The authors used The Interpersonal Support Evaluation List (ISEL-) to assess social support on three different levels: () Tangible Support measuring the perceived 9.3

18 .3 availability of material aid; () Belonging measuring the perceived availability of others to engage in activities; (3) Appraisal perceived availability of others with whom one can talk about problems.[9] Keller et al. measured social support as an individual-level measure. They distinguish between social support for exercise (adapted 9-item version of Social Support and Exercise Survey), assessing frequency of engagement and participation of family members and friends in physical activity, and general social support (9-item Medical Outcomes Study) using the Social Support Scale (MOSSS).[6] Poortinga examined social support as a neighbourhood-level measure, a summation of responses to seven statements and subsequent categorization into three groups from the 987 Health and Lifestyle Survey.[39] Associations between the social environment and obesity All studies used BMI to define weight status, with Powell-Wiley et al.[] and Moore et al.[4] using waist-circumference (WC) as an additional measure. However discrepancies in the definition of weight status were noted. Four studies [3,, 4, 6] did not report their definition of obesity, despite using change in BMI as an outcome and drawing links between obesity and social determinants in their findings. Eleven studies [4, 6, 8-0,, 3, 5, 7-9] defined obesity as a BMI 30kg/m, consistent with WHO guidelines.[] The remaining three studies either combined both obese and overweight BMI status in their reporting [] or focused primarily on overweight individuals,[5, 7] using a BMI cut-off of 5kg/m as per WHO guidelines. One study [5] defined overweight as having a BMI 7kg/m. A number of associations between constructs of the social environment and weight status were identified as detailed below. Collective efficacy From the two studies on collective efficacy only Bjornstrom found that increased levels are associated with a decrease in the odds of obesity. The analysis was interpreted as showing that it exerted an independent and beneficial effect on obesity but did not mediate the relation between inequality and obesity.[9] Social cohesion From the five studies on social cohesion, three found significant associations with obesity. Mujahid et al. found no significant association between neighbourhood social environment and BMI in women, although men with higher social environment scores had a higher mean BMI. This was also the case when social cohesion was investigated separately men with higher social cohesion had a higher mean BMI.[] Prince et al. found that the social environment had no impact on female physical activity or overweight/obesity status. However in males, a stronger neighbourhood sense of community belonging was significantly associated with being physically active, whereas lower neighbourhood socioeconomic status and higher voting rates were significantly associated with lower odds of overweight and obesity.[] Finally, Moore et al. found that in general, point estimates of associations revealed that higher social cohesion was associated with lower BMI for both the home and workplace environments, although associations were not statistically significant for home environment and social cohesion.[4] 30

19 Social capital Eight studies examined social capital, and five significant associations were found. Holtgrave and Crosby found that social capital correlated moderately with obesity. It was shown to have a protective effect against obesity.[8] Prince et al. found that higher social capital in males was significantly associated with lower odds of being overweight or obese.[] Sullivan et al. found that adults who reported that they were involved in neighbourhood clubs, associations, or support groups had significantly decreased odds of obesity. However, race/ethnicity appeared to modify these associations. After further stratification by race/ethnicity, report of involvement in neighbourhood clubs, associations, or support groups was only significantly associated with a decreased odds of obesity among non-hispanic white adults.[5] Yoon et al. found that greater community social capital on average appears to increase the likelihood of eating fewer calories or less fat to reduce or maintain weight.[7] Kim et al. found that social capital had modest protective effects on obesity.[30] In Poortinga s study, only trust appeared to be significantly associated with obesity; people with a high level of trust were 4% less likely to be obese than people with a low level of trust.[8] Social support From the five studies on social support, four found significant associations. Ball and Crawford found that perceived social support for healthy diets from family, and sabotage to healthy diets and physical activity from friends, significantly predicted women s current BMI, with higher levels of support and lower levels of sabotage associated with higher BMI.[4] Kaplan et al. found that social support was associated with a lower likelihood of obesity for women, and a higher likelihood of obesity for men.[6] Veenstra et al. found that depth and breadth of involvement in voluntary associations was significantly related to overweight. Greater involvement corresponded to a lower likelihood of being overweight.[5] Kendzor et al. found that social support was inversely associated with negative affect, and positively associated with BMI.[9] Discussion We systematically reviewed literature on associations between social environmental factors and weight status in adults. We found 9 relevant articles with few significant associations between the social environment and obesity, which indicates that the currently available evidence for such associations is weak. A total of associations on social environment and weight status were examined in 9 studies resulting in 3 significant associations. One was consistent with the hypothesis that collective efficacy is related to adult obesity. From the five studies on social cohesion, three found significant associations. Eight studies examined social capital, and five significant associations were found. From the five studies on social support, four found significant associations. There was a great heterogeneity in the definitions and measures used, which limited the comparability of study results. Social environment is a construct that encompasses many different dimensions so it is not surprising that attempts to simplify it to a single dimension fail to adequately capture the richness of the concept, and are unable to demonstrate the complex associations. We identified three main possible explanations for a lack of consistent associations: social environmental factors were not well-defined; the design of included studies did not allow for detection of meaningful associations; or there was absence of association between social environmental factors 3.3

20 .3 and weight status. The studies identified in this review lacked robust and consistent conceptual frameworks, clear definitions, and distinct constructs. Similar findings were identified in a recent systematic review on obesity interventions targeting social environmental factors,[43] which suggests that these constructs used in adult obesity research are still at an early stage of development. In the studies included in this review, the authors distinguished between four different constructs, but it is unclear how distinct they actually are. Social capital is the most popular construct, and most of the social capital definitions tended to draw on the work of Putnam[44] and Coleman.[45] Krishna et al. refer to social capital as being divided into structural and cognitive components; what people do (voting, attending church) and what people feel (trust, friendship).[46] This suggests that collective efficacy and social support should be part of the cognitive construct of social capital. On the other hand, Edwards et al. suggest that trust is not a part of social capital, but rather a factor that predisposes to social capital.[47] Social constructs are complex, so simple proxies are unlikely to capture their complexity.[48] Yet such proxies are often all that are available in the large surveys that have also considered overweight and/or obesity. For all social constructs, it is important to distinguish between individual and collective characteristics.[49] The mechanisms linking social capital or related constructs - to obesity may be dependent on the level at which the social factors are measured. In this review, we incorporated studies that measured social capital of individuals on different levels. However, none of the studies in this review studied the contextual effects of social environmental factors, which may be important given the evidence that the influence of micro-level measures of social capital on health vary according to characteristics of the country such as civil liberties and the scale of the informal economy.[50] Only a few studies related neighbourhood-level social measures (e.g. neighbourhoods with high and low social capital) to individual-level obesity. This kind of analysis would be needed to detect any neighbourhood-level (contextual) effect of social environmental factors. Social environmental constructs depend on specific physical contexts and settings [5] but the measurement approaches in all these studies are based on spatial aggregation of individual responses from surveys at neighbourhood level.[5] In contrast, more recent instruments designed to capture the influence of the physical environment on health combine both objective measures at community level and aggregate perceptions.[5] If the construct measured at neighbourhood level actually operates at a higher level of aggregation, such as city or state, it is not possible to identify meaningful links. As noted above, there is a distinction between bonding and bridging social constructs; bonding capital refers to social cohesion within a group, while bridging capital refers to social cohesion across different groups.[53] When studying social capital within neighbourhoods, it is important to consider the particular nature of neighbourhood social capital in order to avoid drawing false conclusions about the level at which this social construct acts. A better understanding of the interactions between social environmental factors and obesity related behaviours could provide useful support for environmental policy actions to tackle obesity. However the evidence base in this review is predominantly of North American origin, with resulting limitations in terms of its applicability to European countries. A particularly serious limitation is the unavoidable 3