Background: Sobal and Stunkard's review (1989) of 34 studies from developed countries published after 1941, found inconsistent relationships between socioeconomic status (SES) and childhood adiposity. Inverse associations (36%), no associations (38%), and positive associations (26%) were found in similar proportions. In view of the trends in pediatric obesity, the relationship between SES and adiposity may have changed.
Objective: To describe the cross-sectional association between SES and adiposity in school-age children from western developed countries in epidemiological studies since 1989.
Methods and Procedures: PubMed database was searched to identify potentially relevant publications. Epidemiological studies from western developed countries presenting cross-sectional data on the bivariate association between an SES indicator and objectively measured adiposity in childhood (5–18 years), carried out after 1989 were included. SES indicators included parental education, parental occupation, family income, composite SES, and neighborhood SES.
Results: Forty-five studies satisfied the review criteria. SES was inversely associated with adiposity in 19 studies (42%), there was no association in 12 studies (27%), and in 14 studies (31%) there was a mixture of no associations and inverse associations across subgroups. No positive SES-adiposity associations were seen in unadjusted analyses. With parental education as the SES indicator, inverse associations with adiposity were found in 15 of 20 studies (75%).
Discussion: Research carried out within the past 15 years finds that associations between SES and adiposity in children are predominately inverse, and positive associations have all but disappeared. Research is needed to understand the mechanisms through which parental social class influences childhood adiposity.
Sobal and Stunkard's review of the literature on the relationship between socioeconomic status (SES) and obesity published in 1989 (1) has been cited in over 500 publications to date (2). Their review included 144 studies published before 1989 with data on cross-sectional associations between SES and obesity in women, men, and children from developed and developing countries. In women from developed countries, SES was inversely associated with obesity prevalence (i.e., higher SES groups had lower rates of obesity) in 85% of studies, not associated in 13%, and positively associated in one study (2%). In men from developed countries, SES was inversely associated with obesity in 51% of studies, not associated in 17%, positively associated in 30%, and in one study (2%) the association was U-shaped. For girls and boys respectively, SES was inversely associated with obesity in 40 and 32% of studies, not associated in 35 and 41%, and positively associated in 25 and 26% of studies. In contrast, in studies from developing countries, the majority of associations were positive across all age/gender groups, with no inverse associations.
Three reviews of the literature on SES and weight status in adults have extended Sobal and Stunkard's work (1). Two focused on the association between SES and weight change in longitudinal studies (3,4). The first concluded that low SES in early life was not associated with adiposity in childhood, but was associated with higher adiposity in adulthood (3). The second concluded that weight gain was associated with a lower occupational status in non-black adults (4). A third review updated the evidence on cross-sectional associations between SES and obesity since 1989 in adults from developing countries (5). The proportion of positive SES-obesity associations had decreased to 14 and 50%, respectively in studies of women and men from developing countries, with inverse associations appearing for the first time and featuring in 71% of the studies involving women (5).
Sobal and Stunkard's literature review (1) has not been updated for children. Given that the prevalence of overweight and obesity has reached epidemic proportions in developed countries and shows no sign of stabilizing (6,7), understanding socioeconomic influences on adiposity in contemporary society is important for planning and implementing effective prevention initiatives. The aim of this systematic review is to describe the cross-sectional association between SES and adiposity in school-age children in epidemiological studies from western developed countries since 1989.
Methods and procedures
The National Library of Medicine's PubMed database was searched to identify potentially relevant publications. Exposure search terms included “social class,” “occupation(s),” “employment,” “education,” and “income(s)” as both Medical Subject Headings (MeSH) and text words, and “socioeconomic factors” and “educational status” as MeSH, and socio(-)economic as a text word. Outcome search terms were “obesity” and “body mass index” (BMI) as MeSH and text words, as well as “overweight,” “adiposity,” and “fatness” as text words. Methodology search terms included “epidemiologic studies,” “cohort studies,” and “cross-sectional studies” as MeSH, as well as “epidemiolog*,” “cross-sectional,” “cohort,” and “survey” as text words. The “OR” Boolean operator was used to combine search terms within each category (exposure, outcome, and methodology), and the three sets of results were combined using the “AND” Boolean operator. The PubMed limits function was applied to limit the search to studies involving humans, children 0–18 years, and published in English between 1 January 1990 and 20 December 2005. Publications assigned MeSH after the latter date were not included.
Inclusion and exclusion criteria
In order to select studies systematically, the inclusion and exclusion criteria in Box 1 were specified prior to searching the PubMed database and applied in two steps. In the first step, studies were eliminated if it was evident from the abstract that at least one criterion was not satisfied. In the second step, full text articles of the remaining studies were obtained and the study methods were evaluated against the criteria. Those satisfying all criteria were included in this review. If data from the same study were reported in multiple publications, we included only the last publication remaining after applying the following criteria in order: (i) results stratified by gender, (ii) greater number of participants included in the analyses, and (iii) the paper with the earlier publication date.
In this review, the term “adiposity” is used to refer to all levels of excess weight in children because definitions of overweight and obesity are inconsistent between studies.
Box 1: Study inclusion and exclusion criteria
Cross-sectional data from birth cohort, population, or school-based studies. Intervention studies, case–control studies, and studies that recruited subjects from clinical settings or groups at high risk for developing obesity and related diseases were excluded. Studies including only one socioeconomic status (SES) or ethnic minority group were excluded.
Study time frame
Studies presenting data collected after 1 January 1990. This cut-point was selected because we judged that an adequate number of studies would have been published thereafter to meaningfully examine the association between SES and adiposity and still be relevant to informing obesity prevention initiatives in “contemporary society.”
Age: Five to eighteen years old. Studies in which the majority of participants were younger or older than this age range were excluded.
Country of origin: Studies from western countries classified by the World Bank as high income economies (Gross National Income per capita > US$10,066) (8).
SES indicators: Assessment of at least one household or neighborhood level SES indicator reported either by a parent or a child aged >12 years old, including parental education, parental occupation, family income, home postcode, or a composite measure derived from at least one of the preceding indicators. Living space, parental unemployment status, parental marital status, children's educational level, school type, and school postcode were not considered appropriate measures of SES for this review.
Adiposity: Height- and weight-based indices such as BMI. Studies with a combination of measured and self-reported heights and weights, or where some of the data were collected prior to 1990, were included if the majority of data satisfied the inclusion criteria. Only studies using age- and sex-specific categorical (e.g., overweight, obese, or combined overweight and obese) or continuous adiposity variables (BMI z -scores or BMI stratified by age and sex) were included.
Result presentation: Bivariate association between SES and adiposity.
Identification of studies
The search strategy identified 2,393 potentially relevant publications. A summary of the 45 studies satisfying the criteria for this review is presented in the appendices (Supplementary Tables online) according to the format of adiposity variable (Appendix 1 (Studies 1–35): categorical; Appendix 2 (36–45): continuous) and treatment of gender (Appendix 1a (1–20) and 2a (36–40): combined; Appendix 1b (21–35) and 2b (41–45): stratified) in the data analysis. At least one of the criteria for this review was not satisfied by 2,348 studies that were eliminated in the two-step procedure described in the Methods and Procedures section.
Overview of included studies
Country of origin. Data from the 45 studies included in this review originate from the United Kingdom (n = 11), Germany (n = 7), United States (n = 7), Australia (n = 6), Italy (n = 4), France (n = 2), The Netherlands (n = 2), Belgium (n = 1), Canada (n = 1), Republic of Ireland (n = 1), Spain (n = 1), Sweden (n = 1), and Switzerland (n = 1).
Adiposity. Adiposity was determined using BMI in 41 studies. Two studies each used relative weight (# 11, 13) and weight-for-height (# 1, 23). At least one measure of body fat distribution was examined in association with SES in six studies; five used skinfolds (# 10, 13, 41, 43, 44), two also included waist girth and/or waist-to-hip ratio (# 13, 44), and one used fat mass (# 45). Adiposity was analyzed as a categorical variable in 35 studies and a continuous variable in 10 studies. Twelve different growth references were used across the studies for classifying weight status, but the International Obesity TaskForce (9) classification of overweight and obesity was used in 16 studies making it the most consistent definition.
SES indicators. A single SES indicator was reported in 31 studies; the remaining 14 studies reported two (n = 11), three (n = 2), or four (n = 1) indicators. Overall, 63 analyses of associations between an SES indicator and adiposity were identified from the 45 studies included in this review. The majority of SES indicators were measured at the household level including: parental education (n = 26), parental occupation (n = 14), indices of family income (n = 11), and composite SES measures (n = 5). Neighborhood level SES indicators were used in seven analyses.
Bivariate association between SES and adiposity
All SES indicators. A summary of the type of SES-adiposity association in the 45 studies included in this review is shown in Table 1. Complex-inverse associations were found in nine studies (# 11, 14, 15, 23, 26, 28, 30, 32, 38) but for simplicity of presentation herein, this type of association will be described as an inverse association. Inverse (including complex-inverse) SES-adiposity associations were found in 19 studies, there was no association in 12 studies, and in 14 studies the association varied by subgroup with both inverse and no associations. No positive SES-adiposity associations were seen in any of the bivariate analyses in the studies included in this review. In one study, a statistically significant complex-positive association emerged between SES and obesity, but not SES and overweight, in girls but not boys, and only after adjusting for ethnicity and age (# 29). In 9 of the 14 studies (# 4, 6, 11, 14, 19, 20, 34, 38, 40) that included more than one SES indicator, the association was inverse for all indicators. In each of these studies, a measure of parental education featured as one of the SES indicators. In one study there was no association for either indicator (# 10) and in the other four studies (# 7, 9, 12, 35) the type of association varied by SES indicator/level of adiposity/ethnicity, with a mixture of negative and no association.
Table 1. . Summary of the bivariate SES-adiposity association stratified by SES indicator type in 45 studies that satisfied the criteria for this review
|Education||2 (parental)||38 (maternal)||10 (maternal and paternal)|
| ||4 (maternal)|| ||17 (maternal: overweight and obesity combined, both age groups)|
| ||6 (maternal)|| ||35a (parental: white boys, black boys and girls)|
| ||8 (parental)|| ||43a (parental: boys)|
| ||11 (maternal and paternal)|| ||45 (maternal: boys)|
| ||12a (maternal)|| || |
| ||14a (maternal)|| || |
| ||16 (parental)|| || |
| ||18 (parental)|| || |
| ||19a (maternal and paternal)|| || |
| ||20 (parental)|| || |
| ||34 (maternal and paternal)|| || |
| ||38 (paternal)|| || |
| ||39 (parental)|| || |
| ||40a (maternal and paternal)|| || |
| ||17 (maternal: obesity, 10–12 year olds)|| || |
| ||35a (white girls)|| || |
| ||43a (parental: girls)|| || |
| ||45 (maternal: girls)|| || |
|Occupation||3 (parental)||11(paternal)||7 (parental)|
| ||4 (paternal)||15 (parental)||13 (parental)|
| ||9 (parental)|| ||29a,b (parental)|
| ||11 (paternal)|| ||36 (parental)|
| ||24a (parental: girls)|| ||41 (parental)|
| ||25(paternal: boys)|| ||42 (parental)|
| || || ||24a (parental: boys)|
| || || ||25 (paternal: girls)|
|Indices of family income||5a||14a||12a|
| ||20|| ||27|
| ||34|| ||33a|
| ||7(obesity)|| ||7 (overweight)|
| ||9 (overweight and obesity combined)|| ||9 (obesity)|
| ||21a (white 8–16 year olds)|| ||21a (white 2–7 year olds, all black and Mexican|
| ||35a (white girls)|| ||American 2–16 year olds)|
| || || ||35a (white boys, black boys and girls)|
|Composite measure||6||28a (girls)||31a (study of girls only)|
| ||19a||32 (boys)||28a (boys: overweight)|
| ||28a (boys: at risk of overweight)|| ||32 (girls)|
|Neighborhood||44 (girls: BMI, waist girth, WHR)||23||1a|
| || ||26||22|
| || ||30a (boys)||37|
| || || ||30a (girls)|
| || || ||44 (boys: all adiposity measures, girls: skinfolds)|
Parental education. Parental education (parental (n = 8), maternal (n = 12), or paternal (n = 6)) was used as the SES marker in 20 studies, with 26 analyses presented. Inverse associations between parental education and adiposity were found in 15 of the 20 studies and in one study there was no association (# 10). In four studies the association was mixed, in two the association was inverse for girls but not boys (# 43, 45), in one study the association was inverse for white girls but there was no association for boys or black girls (# 35), and in the other, there was no association with combined overweight and obesity but there was an inverse association for obesity in 10–12 year olds (# 17).
Parental occupation. A total of 14 analyses of the association between parental (n = 10), maternal (n = 1) or paternal (n = 3) occupation and adiposity was presented across 13 studies. An inverse association between parental (including maternal and paternal) occupation and adiposity was found in 5 of the 13 studies and in 6 studies there was no association. A varied association was found in two studies, with an inverse association for one gender (girls (# 24), boys (# 25)) but not the other.
Indices of family income. Associations between indices of family income and adiposity were presented in 11 studies. An inverse association was found in 4 of the 11 studies, in 3 studies there was no association, and the association varied in 4 studies (# 7, 9, 21, 35). In two studies from the United Kingdom, the association varied according to the level of adiposity (International Obesity TaskForce definition), with income inversely associated with combined overweight and obesity but not obesity alone (# 9), and with obesity but not overweight (# 7). In two studies from the United States, the SES-adiposity association varied by ethnicity, and age or gender. In one of these studies, income was inversely associated with adiposity in white boys and girls aged 8–16 years but there was no association for white boys and girls aged 2–7 years, nor for black or Mexican-American children of any age (# 21). In the second study, income was inversely associated with adiposity in white girls but there was no association for boys or black girls, as seen for parental education in the same study (# 35).
Composite measure of SES. Five studies presented associations between a composite measure of SES and adiposity. An inverse association was found in two studies and there was no association in one study that included only girls. In the remaining two studies the association varied by subgroup; SES was inversely associated with adiposity in boys but not girls (# 32), and overweight in both boys and girls but “at risk of overweight” only in girls (# 28).
Neighborhood SES. The association between neighborhood SES and adiposity was presented in seven studies. An inverse association was found in two of them and there was no association in three. In the other two studies, the association varied according to the type of adiposity measure or gender. In one of these studies the association was inverse for boys but not girls (# 30) and in the other neighborhood SES was inversely associated with BMI, waist girth, and waist-to-hip ratio but not skinfolds in girls, and was not associated with any adiposity measure in boys (# 44).
Additional considerations for interpreting findings
Gender. In this review, 19 studies (Appendices 1b and 2b) presented data on the association between SES and adiposity stratified by gender and 1 study included only girls (# 31). In 10 studies the associations were consistent for boys and girls. In seven studies there was an inverse association for one gender (girls (# 24, 43, 44, 45), boys (# 25, 30, 32)), but not the other. In the remaining studies, there were inconsistent SES-adiposity associations between genders according to ethnicity (# 35) and level of adiposity (# 28). Overall, the SES-adiposity association was consistent for boys and girls in more than half of the reviewed studies that presented results stratified by gender. In the remaining studies there appeared to be no gender differences in the proportion of inverse associations.
Age. The studies could be broadly classified by the age range of participants: children (5–11 years, n = 18), adolescents (12–18 years, n = 9), and both children and adolescents (n = 18). In 10 of the 18 studies of children (# 6, 8, 11, 14, 15, 16, 18, 19, 20, 38), SES was inversely associated with adiposity, no association was found in 4 studies (# 1, 36, 37, 42), and the association varied by subgroup in 4 studies (# 7, 12, 17, 43). In five of the nine studies of adolescents there was an inconsistent SES-adiposity association between the genders (# 24, 25, 32, 35, 45) (as described in the gender section), an inverse association was shown in one (# 5) and in three studies there was no association (# 22, 31, 41). In the remaining 18 studies that included both children and adolescents, an inverse association was found in 8 (# 2, 3, 4, 23, 26, 34, 39, 40), there was no association in 5 (# 10, 13, 27, 29, 33), and the association varied by subgroup in 5 (# 9, 21, 28, 30, 44). It appears that inverse SES-adiposity associations are more common in studies of children compared with adolescents.
Ethnicity. The treatment of ethnicity in the studies included in this review can be classified in three categories: 12 studies included white children or children with a European ancestry (>95% of sample) or the data analysis was restricted to these groups, 15 included a variety of ethnic groups (>5% of sample) in the data analyses, and in 18 studies the ethnic composition of the sample was not specified. The studies including only white/European ancestry children originated from the United Kingdom (n = 5), Germany (n = 4), Italy (n = 1), and The Netherlands (n = 2). In eight of these studies, including those from Germany, Italy, The Netherlands, and one from the United Kingdom (# 26), SES was inversely associated with adiposity. In the remaining four studies from the United Kingdom, there was either no association (# 36, 42) or the association varied according to the level of adiposity (# 7, 9).
The 15 studies with a variety of ethnic groups originated from the United States (n = 7), United Kingdom (n = 4), Australia (n = 2), Belgium (n = 1), and Germany (n = 1). In five studies, including those from Australia, Belgium, Germany, and one from the United States (# 14), SES was inversely associated with adiposity. There was no association in four studies, one from the United States (# 1) and three from the United Kingdom (# 29, 31, 33), although as already reported, a complex-positive association emerged for obesity in girls in one of these studies after adjusting for age and ethnicity (# 29). In six studies the association varied by subgroup with differences between genders in two (United States (# 43), United Kingdom (# 30)), type of SES indicator in one (United States (# 12)), the level of adiposity in one (United States (# 28)), and in the other two US studies that stratified results by ethnicity, an inverse SES-adiposity association was shown only for white girls but not boys and black children (# 35) and white girls and boys aged 8–16 years but not younger white children or children from other ethnic groups of any age (# 21). Because the majority of studies did not specify the ethnicity of the sample, and in those that did, few stratified results by ethnicity, it was difficult to determine its influence on the association between SES and adiposity. However, in the studies that presented results separately by ethnicity and gender, there appeared to be no association between SES and adiposity in black children, while inverse associations were more common for white girls.
Sample size. We estimated that a sample size of 1,000 or more participants would be required to detect a twofold risk (odds ratio (OR) = 2.0) of adiposity in one SES group compared to another in a study with three, similarly sized SES groups. A sample size of 1,000 would provide 95% power, and a sample size of 500 (for gender-stratified analyses) would provide 75% power. Thirty-three studies included in this review had a sample size of at least 1,000, and in these, the proportion of inverse, no association, and mixed (inverse/none) associations was 52, 18, and 30%, respectively.
Magnitude of bivariate SES-adiposity associations. The magnitude of SES-adiposity associations between extreme SES groups was examined using an exploratory analysis of the ORs for adiposity in the lowest SES group with the highest SES group as the reference group. When the original report presented ORs calculated using the lowest SES group as the reference, these were recalculated. Similarly, when the original report presented results as proportions, these were converted to ORs if sample size data were available. Results from 24 studies (70 analyses) are presented in Table 2 stratified by the total number of SES groups and the definition and level of adiposity. Within a study, separate analyses were sometimes performed for different levels of adiposity, gender, ethnicity, age, and type of SES indicators.
Table 2. . Summary of the OR (95% CI) for adiposity in the lowest SES group when the highest SES group is the reference group in bivariate analysis, listed according to the total number of SES groups in analyses
From these 24 studies, 21 (42 analyses) found a significant inverse SES-adiposity association in at least one analysis. Regardless of the total number of SES groups or the definition or level of adiposity, the OR for adiposity in the lowest compared with the highest SES children, ranged from 1.30 (# 8) to 6.70 (# 24) with a median OR of 2.04. In 11 of these studies, ORs for two levels of adiposity were available (e.g., BMI > 90th percentile and BMI > 97th percentile) and in all of these, ORs were larger for the higher level of adiposity.
There was a non-significant inverse association (OR > 1, 95% confidence interval (CI): <1 to >1) in at least one analysis in 10 studies ((# 7, 15, 17, 21, 24, 26, 27, 29, 30, 31); 20 analyses). The ORs for adiposity in the lowest SES group compared with the highest SES group ranged from 1.04 (#7, 31) to 2.92 (# 15) with a median OR of 1.28. In three studies (# 12, 21, 29; 7 analyses) there was a non-significant positive association (OR < 1, 95% CI: <1 to >1). The OR for adiposity in the lowest SES group compared with the highest SES group ranged from 0.64 (# 12) to 0.97 (# 29) with a median OR of 0.81. In one study a significant positive association (OR < 1, 95% CI <1) emerged for black girls aged 8–16 years, the OR (95% CI) for adiposity in the lowest SES group compared with the highest SES group was 0.60 (0.36–0.99) (# 21), although in the original report there was not a statistically significant difference between SES groups.
Multivariate association between SES and adiposity
Multivariate analyses were reported in 16 of the 45 studies included in this review. In addition to alternate SES indicators, the following covariates were included in one or more of the analyses: age, gender, parental adiposity, parental marital status, parental area of birth, maternal age, maternal employment status, maternal smoking during pregnancy, birth weight, birth length, birth order, infant feeding, size of flat, own bedroom, dietary intake, parental control over intake, physical activity, sedentary activity, family car ownership, behavior problems, cognitive stimulation score, academic grade retention, medication, depressive symptoms, geographical region, parent road safety concern, and school/district factors. Different combinations of covariates were used across different studies making it difficult to draw general conclusions.
Because parental adiposity is often the strongest predictor of adiposity in children (e.g., # 3, 16), we examined the effect of including it as a covariate in the multivariate analyses. Nine of the sixteen studies reporting multivariate analyses included parental adiposity as a covariate (# 3, 6, 12, 16, 19, 34, 40, 43, 45). In six of them, the inverse SES-adiposity association remained for maternal education (# 45), parental education (# 43), overweight but not obesity (# 6, 16), maternal education but not paternal education (# 19, 34) and income (# 34). In three, the inverse association between SES and adiposity disappeared after including parental adiposity as a covariate (# 3, 12, 40). In the seven studies that included other covariates but not parental adiposity, the inverse SES-adiposity association did not change in four studies (# 8, 9, 17, 39), and remained for paternal occupation (# 11), maternal but not paternal education (# 11), parental education but not income (# 20), and paternal occupation but not maternal education (# 4).
In this review of 45 studies from a diversity of western developed countries, results of bivariate analyses showed that SES was inversely associated with adiposity in children in 19 studies (42%), there was no association in 12 studies (27%), and in the remaining 14 studies (31%) both inverse and no associations were reported according to the type of SES indicator, the level of adiposity, the type of adiposity indicator, or the participants' gender, age, or ethnicity. No statistically significant positive association was found in any of the bivariate analyses, although one significant positive association emerged in one study after adjustment for age and ethnicity.
These results are clearly different from those reported in Sobal and Stunkard's review of 34 studies in children from developed countries published between 1941 and 1989, which showed a more equal distribution of inverse associations (36%), no associations (38%), and positive associations (26%) (1). It is difficult to comment on secular trends with certainty because definitions of adiposity in children have changed, and the representation of studies from different developed countries was not the same (e.g., half of the studies included in Sobal and Stunkard's review were from the United States). Furthermore, details regarding important characteristics of included studies such as study design, specific age range of participants, and type of results (bivariate or multivariate) were not reported in the earlier review. Nevertheless, it is rather striking that the protective effect of lower SES that used to be observed in a proportion of studies has all but disappeared in studies of school-age children from developed countries since 1990.
Examination of the association between SES and adiposity by the type of SES indicator revealed that parental education was more consistently inversely associated with adiposity (15 of 20 studies) than other indicators. With parental occupation or income as the SES indicator (in 13 and 11 studies, respectively), associations with adiposity were more evenly distributed between inverse and no association, particularly after taking account of associations that varied by subgroup. Comparatively few studies used a composite SES measure (n = 5) or neighborhood level SES (n = 7), limiting the scope to assess trends in the association with adiposity for these indicators. In the majority of studies (9 of 14) examining the association between adiposity and multiple SES indicators, the association was the same for all indicators and was inverse.
The stronger inverse association found between adiposity and parental education than between adiposity and parental occupation or income, may have several explanations. The results may reflect the relative stability and therefore greater validity of parental education as an indicator of SES, while parental occupation and income could be more liable to change (10). But also the theoretical framework proposed by Sobal (11) acknowledges that each SES indicator may operate via a different pathway to influence the development of adiposity. According to the framework, education influences knowledge and beliefs, occupation influences lifestyle and shared peer values, and income relates to access to resources (11). It is possible that knowledge and beliefs are more key to healthy lifestyles in the contemporary world than are access to resources or shared values (12).
Approximately a third of the studies (n = 16) in this review presented a multivariate model of the association between SES and adiposity. We were able to examine the effect of including parental adiposity as a covariate in a limited number of studies (n = 9). The SES-adiposity association totally disappeared in three studies once parental adiposity was included, but significant inverse associations with education remained in five studies (and composite SES in one study), indicating that education is more than a marker for parental overweight and probably exerts an independent effect on adiposity. Other covariates appeared to have a lesser effect on the SES-adiposity association. In the three of the four studies that included both maternal and paternal education in the analysis, the inverse association persisted for maternal education, suggesting it confers additional explanatory power beyond higher paternal education. This is consistent with the idea that mothers have more influence than fathers on children's behaviors.
Whilst Sobal and Stunkard did not investigate whether SES-adiposity associations between studies differed by SES indicator, gender was discussed as a possible explanation (1). In their review the SES-adiposity association was the same for boys and girls in 27 of 31 studies that included both genders. In contrast, in the studies that presented results stratified by gender in the present review, a consistent association for boys and girls was found in only half of the studies. However, among studies finding an inconsistent association by gender, the proportion of inverse associations was similar for boys and girls.
Age was another factor discussed by Sobal and Stunkard as a possible explanation for variation in SES-adiposity associations, but few studies in their review gave sufficient detail to examine it directly (1). Fortunately, all the studies included in the present review reported ages. The finding that SES was inversely associated with adiposity in 10 of the 18 studies in children (5–11 years) supports the idea that SES gradients in adiposity develop early in the life course. In contrast, the association between SES and adiposity was inverse in only one of nine studies in adolescents (12–18 years) although it varied by gender in five of them. This may be due to the difficulty of classifying adiposity during puberty, or it may reflect real changes in the SES distribution of adiposity, possibly mediated by more similar eating and physical activity behaviors across SES groups at this age. The studies in this review that included both children and adolescents but did not stratify results by age, had findings more similar to those seen in children.
The only two studies in this review to present results from a mixed ethnic sample stratified by ethnicity indicated inverse SES-adiposity associations for white children 8–16 years of age (# 21) or white girls (# 35), but not for non-white children. Similarly, a discrepant association between ethnic groups was found in two other US studies that were identified through the literature search but either did not fulfill the date (13) or results presentation criteria for this review (14). In the 1987 Growth and Health Study, a lower prevalence of obesity in girls with more educated mothers was found in white but not black girls (13). In the National Longitudinal Study of Adolescent Health (Wave 2, 1996), data presented graphically indicated a clear inverse association between adiposity and family income for white and Hispanic girls, and white boys but not children from other ethnic groups. With parental education as the SES indicator, the association was inverse for white, black, and Hispanic girls, and Asian boys, but not children from other ethnic groups (14). These results indicate that inverse SES-adiposity associations may be less consistent for children from other ethnic groups than white girls in the United States.
The variety of classifications of SES and definitions of adiposity in the studies included in this review prohibited the use of meta-analysis to determine the effect sizes. However, an insight into the magnitude of the association is provided by an exploratory analysis comparing the extreme SES groups, regardless of definition. In the majority of analyses a statistically significant inverse association was found between extreme SES groups, with the median odds of adiposity (at any level) in children from the lowest SES group approximately twice that of the highest SES group, with stronger effects for higher levels of adiposity (e.g., obesity vs. overweight). The inverse association was not statistically significant in a minority of studies, where effect sizes were generally lower.
To facilitate our understanding of how SES gradients in adiposity relate to the type of SES indicator, age, and ethnicity, some methodological recommendations for future research have emerged from this review. Two-thirds of studies reviewed used only a single SES indicator and few studies provided data on neighborhood SES, despite recommendations to do so (10,15). Future epidemiological studies of adiposity in children should incorporate SES indicators measured at both the household and neighborhood level (10,15) and consider Braveman et?al. 's recommendations to include more than one SES indicator where possible (16). It would also be useful if future studies and analyses of existing datasets report results stratified by age and ethnicity when sample size permits (16), following the example of Alaimo et al. (# 21).
The finding that parental education, compared with other SES markers, is more consistently protective against adiposity warrants further investigation. Since adiposity is the outcome of a positive balance between energy intake and expenditure, a greater understanding of SES gradients in dietary intake and physical activity are needed. Differences in child feeding (17,18) and physical activity (19) have been identified in some studies. However, in order to establish strategies for reducing health inequalities, research is needed to understand the mechanism by which parental education and other indicators of SES independently and collectively influence these behaviors.
A strength of this review is the comprehensive and easily replicable search strategy applied to the PubMed database. While additional databases are available, we limited our search to PubMed because we believed these citations would be representative of the literature in this field. We did not source additional studies from the reference lists of the publications included in the review because this could have introduced bias through selective citation by authors. Another strength of this review is the systematic selection of studies through the application of well-defined inclusion/exclusion criteria. Restricting the review to studies with objectively measured adiposity limits any impact of self-report bias on the findings.
It should be noted that many of the studies included in this review were not designed for the purpose of determining the association between SES and adiposity and therefore may be underpowered to detect such an association if one were to exist. This concern is supported by our finding that restricting analyses to studies with a sample size >1,000 produced a slightly higher proportion of inverse SES-adiposity associations, and it confirms the idea that this is a robust effect. The exclusion of studies not published in English is a limitation, but this was a consequence of practical considerations. Another limitation is that we cannot discount the potential influence of SES-related non-response bias because very few studies included in the review reported data on the characteristics of non-responders. However, this is likely to mean that families with lower SES and higher adiposity would be underrepresented, and therefore the true (inverse) effect may be higher than we found.
In conclusion, in the past 15 years the association between SES and adiposity in children from western developed countries was predominately inverse and very few positive associations were found. Whilst children from all SES groups are vulnerable to the development of adiposity, children whose parents (particularly mothers) have a low level of education appear to be at higher than average risk. The conclusions from this review have implications for both the prevention and treatment of childhood adiposity. Population-level strategies to prevent overweight and obesity in children are needed, but when supported by local data, prevention efforts targeting lower SES groups may be appropriate. The higher prevalence of overweight and obesity amongst lower SES children in many of the studies presented in this review reinforces the need for treatments that specifically address the needs of children and families from lower SES households.