点击显示 收起
1 From the School of Physical and Health Education (IJ), the Departments of Community Health and Epidemiology (IJ, WFB, KS, and WP) and Emergency Medicine (KS and WP), the Social Program Evaluation Group, Faculty of Education (WFB), and the Centre for Health Services and Policy Research (WFB), Queen's University, Kingston, Ontario, Canada.
2 Supported by Health Canada, which funds the Canadian version of the World Health Organization-Health Behaviour in School-Aged Children Survey (WHO-HBSC). The WHO-HBSC is a WHO/Euro collaborative study. International Coordinator of the 2001-2002 study: Candace Currie, University of Edinburgh, Scotland; Data Bank Manager: Oddrun Samdal, University of Bergen, Norway. This publication reports data solely from Canada (Principal Investigator: William Boyce) and was funded in part through a grant from the Canadian Population Health Initiative (CPHI). CPHI is a program of the Canadian Institutes of Health Information (CIHI). The CIHI supports research advancing knowledge of the determinants of the health of the Canadian population and develops policy options to improve population health and reduce health inequalities. CIHI is a national, not-for-profit organization responsible for developing and maintaining Canada's comprehensive health information system. 3 Address reprint requests to I Janssen, School of Physical and Health Education, Queen's University, 69 Union Street, Kingston, Ontario, Canada, K7L 3N6. E-mail: janssen{at}post.queensu.ca.
ABSTRACT
Background: Low socioeconomic status (SES) is a risk factor for obesity. However, few studies have used a multilevel analysis to determine the influence of both individual- and area-level determinants of SES on obesity, and these studies have been limited to adults.
Objective: The primary objective was to examine associations between individual- and area-level measures of SES and obesity among adolescents by using a multilevel analytic approach. A secondary objective was to examine associations between individual- and area-level measures of SES with unhealthy eating and physical inactivity.
Design: The study sample consisted of 6684 youth in grades 6–10 from 169 schools across Canada. Individual-level SES exposures included material wealth and perceived family wealth. Area-level SES exposures included unemployment rate, percentage of adult residents with less than a high school education, and average employment income from head of household. Associations between SES and the outcome measures were examined by using multilevel logistic regression procedures that modeled students (individual level) nested within schools (area level).
Results: Both individual-level and all 3 area-level SES measures were inversely associated with obesity. The odds for unhealthy eating were increased for those living in an area with a low percentage of residents with a high school education. The odds of being physically inactive increased with decreasing levels of material wealth and perception of family wealth.
Conclusions: Individual- and area-level SES measures were independently related to obesity, which suggests that both individual and environmental approaches may be required to curtail adolescent obesity.
Key Words: Health surveys • social class • body mass index • food habits • motor activity
INTRODUCTION
Identification and targeting of groups at high risk for obesity is of critical importance given that obesity is a global epidemic (1). Strong evidence from population-based studies in industrialized countries indicates that a low socioeconomic status (SES) is a risk factor for obesity (1–3). Two different themes of research have linked SES and obesity. The first theme focuses on how the individual characteristics of the poor, such as lack of education and low income, contribute to the development and maintenance of obesity. For example, overweight individuals with a low SES are less likely to perceive themselves as having a weight problem compared with overweight individuals with a high SES (4), and poor individuals may not be able to afford healthy foods (eg, fruit and vegetables) that would help them maintain a normal body weight (5). The second theme linking SES and obesity focuses on the obeseogenic environments in which the poor and disadvantaged live. Obeseogenic environments are those that promote the consumption of energy dense foods and dissuade physical activity participation. Poorer neighborhoods tend to have a high concentration of fast-food restaurants (6) and a paucity of parks and other recreational facilities (7, 8).
In the past, associations between SES and obesity were observed for either individual (2, 4) or environmental or area (9–11) level characteristics. However, simultaneous examination of both individual- and area-level SES determinants by using a multilevel analysis has rarely been performed. Using a multilevel analysis is meaningful because this is a novel means of disentangling the effects of individual- and area-level predictors. The few multilevel analytic studies that examined the association between SES and obesity are limited to adult populations (12–14). Each of those studies identified gradients in risk of obesity with adults in lower SES classes being the most vulnerable. Individual and area SES characteristics were independently associated with obesity. To date, none of the existing multilevel analyses have examined adolescents, a population group in which the prevalence of obesity is increasing at an alarming rate (15).
In light of this background, the primary purpose of this study was to examine associations between individual- and area-level measures of SES and obesity among adolescents by using a multilevel analytic approach. A secondary purpose was to examine associations between individual- and area-level measures of SES with unhealthy eating and physical inactivity, because the influence of SES on diet, physical activity, or both is the likely mechanism that links SES with obesity.
SUBJECTS AND METHODS
Survey methods
Individual-level data
The study sample included Canadian youth in grades 6–10 from the Health Behavior in School-aged Children Survey (HBSC). HBSC is a collaborative cross-national study of adolescents facilitated by the World Health Organization (16). A classroom-based health survey is conducted every 4 y in the participating countries. In this study we analyzed Canadian data from the 2001–2002 version of the HBSC. The Canadian sample was designed according to the international HBSC protocol (16), in that a cluster design was used with the school class being the basic cluster, the distribution of the students reflected the distribution of Canadians in grades 6–10 (representing youth with average ages of 11–15 y), and the sample was designed to be self-weighting (eg, sample weights were not used). Within each province, samples were selected to represent distributions of schools by size, location, language, and religion. Youth in private and special needs schools, street youth, and incarcerated youth were excluded. Although data are not available for race, 85% of the Canadian population is white and a similar proportion would be expected in the HBSC (17). The survey instrument contained questions about health behaviors, lifestyle factors, and demographic characteristics (16) and provided the individual measures of SES, obesity, unhealthy eating, physical inactivity, and the covariates that were used in this study. A total of 7235 students from 171 schools participated in the 2001–2002 Canadian HBSC survey. The analysis for this study was limited to 6684 students from 169 schools who had information on all of the SES exposure variables of interest. Ethics approval was obtained from the Queen's University General Research Ethics Board, and subject consent was obtained at the school board, parent, and student level.
Area-level data
Area-level data were obtained from the Canada Census of Population, which is conducted every 5 y to provide information on the population of Canada, its provinces, and smaller geographic areas. For each geographic unit, data are available that describe area demographic and economic circumstances. Through a linkage of school civic addresses, postal codes, or both, the 2001 census was used to provide SES measures for the areas in which HBSC sample schools were situated.
Socioeconomic status exposure variables
Individual-level variables
Two individual measures of SES were examined from the Canadian HBSC: a measure of material wealth (Family Affluence Scale) and a measure of perceived family wealth. The 3-point Family Affluence Scale (low, medium, or high) was developed on the basis of 4 measures of material family wealth, as reported by the students (car ownership, bedroom sharing, holiday travel, and computer ownership) (16). Perceived family wealth was designed to measure the students' perceptions of their family's socioeconomic circumstances. This variable was based on the question "How well off do you think your family is?" and the following response categories: very well off, quite well off, average, not very well off, not at all well off (16).
Area-level variables
PCENSUS (2001 Census of Canada Profile Data; version 2001; Tetrad Computer Applications Inc, Bellingham, WA) for MAPPOINT SOFTWARE (version 2002; Microsoft Corporation, Redmond, WA) was used to define the geographic area (5-km radius) surrounding each of the participating schools. Three measures of area-level SES were obtained for the Census subdivisions in these 5-km areas (Note: The area-level SES variables were also calculated by using a 1-km radius. No differences were observed between the SES estimates for the 5-km and 1-km radii.) These measures were used to infer the levels of SES in the areas served by the schools, a proxy for SES in the neighborhoods of individual participants. Area-level SES measures included 1) unemployment rate, 2) percentage of adult residents with less than a high school education, and 3) average employment income from head of household in the area. For analytic purposes, areas were assigned to SES quartiles for each of these SES measures.
Outcome variables
Obesity
Self-reported height and weight were obtained from the study participants. Body mass index (BMI; in kg/m2) was calculated. BMI cutoffs for children from the International Obesity Task Force were used to define adiposity status as obese or nonobese (18). These age- and sex-specific cutoffs were derived from a large international sample that used regression techniques by passing a line through the health-related adult cutoffs at 18 y. Youth with BMI values that corresponded to an adult BMI 30 were classified as obese, and all others were considered nonobese. A previous study in a large representative sample of American adolescents found that 94% were correctly classified as obese or nonobese on the basis of self-reported height and weight, with similar results in boys and girls (19).
Unhealthy eating
The subjects were asked how many times in a typical week they consumed each of the following food items: sweets (candy or chocolate), nondiet soft drinks, cake or pastries, potato chips, and french fries. The possible responses were "never," "less than once a week," "once a week," "2–4 days a week," "5–6 days a week," "once a day," and "more than once a day." Frequency of consumption of these 5 foods was used to compose a factor-analytically derived score of unhealthy eating. Factor loadings for the 5 items varied from 0.64 to 0.81. The factor itself had an eigenvalue of 2.6, accounted for 51% of the variance in the 5 variables, and was reliable (Cronbach's = 0.76). Subjects were classified as unhealthy eaters if they were in the top quartile of the factor-derived score.
Physical inactivity
After being given a definition and examples of common physical activities, participants were asked how many days in the past week and in a typical week they were physically active (cumulative activity) for 60 min (20). The average number of days from the past week and a typical week were used as an index of physical activity participation (20). The responses to these physical activity questions in youth are reliable for classifying subjects as meeting or not meeting physical activity guidelines (intraclass correlation = 0.79) (20). On the basis of the guidelines that recommend youth engage in 60 min of physical activity on 5 d/wk (21), subjects were categorized as physically inactive if they were active for <5 d/wk.
Statistical analysis
Associations among SES, obesity, unhealthy eating, and physical inactivity were examined by using multilevel logistic regression procedures that modeled students (individual level) nested within schools (area level). Odds ratios and associated 95% CIs were estimated. The highest level of SES was chosen as the referent category within each multilevel model. A three-step approach to the multilevel analysis was used. First, bivariate models were fit for SES exposures and specific outcomes. The second step involved fitting multilevel multivariate logistic regression models. Multivariate analyses were theory driven, and only those SES exposure variables that were significant (P < 0.05) in the bivariate models were included in multivariate analyses. The third analysis step consisted of path analysis, in which case a series of multivariate analyses were used to assess links among SES measures, unhealthy eating, physical inactivity, and obesity (Figure 1). Although the data are cross-sectional, the ordering of the variables in the figure from left to right illustrates a logical sequence. That is, one would expect SES to influence eating and activity patterns, which would in turn influence obesity. Age and sex were forced into all regression models according to a priori assumptions about potential confounding. During the multilevel analysis simple level 2 variation (random intercepts) was assumed for all outcomes. The multilevel analysis was conducted with the use of HLM software, version 5.05 (Scientific Software International, Lincolnwood, IL).
FIGURE 1.. Pathway analysis (conducted by using multiple logistic regression analysis) assessing the relations among socioeconomic status, unhealthy eating, physical inactivity, and obesity among 6684 Canadian youth. Solid lines indicate that a significant (P < 0.05) relation exists, and dashed lines indicate that the relation was not significant (P > 0.05).
RESULTS
Of the study population, 4% was classified as obese, 25% as unhealthy eaters, and 55% as physically inactive (Table 1). The study population is further described according to age, sex, and individual-level measures of SES in Table 1. The area-level SES measures indicated that there were mixed SES communities. The mean unemployment rate was 7% and ranged from 2% to 52%. The mean percentage of residents with less than a high school education was 28% and ranged from 12% to 61%. The mean average household income was C$30 000 and ranged from C$14 049 to C$53 745.
View this table:
TABLE 1. Distribution of the sample according to study variables1
Bivariate analyses that examined associations among SES measures and the 3 outcome measures are described in Table 2 (individual-level SES measures) and Table 3 (area-level SES measures). At the individual level, low material wealth and perception of low family wealth were associated with an increased likelihood of obesity and physical inactivity. None of the 2 individual-level SES measures were associated with unhealthy eating. At the area level, a high unemployment rate was associated with an increased likelihood of obesity, a high percentage of residents with less than a high school education was associated with an increased likelihood of obesity and unhealthy eating, and a low average income was associated with an increased likelihood of obesity. None of the 3 area-level SES measures were associated with physical inactivity.
View this table:
TABLE 2. Bivariate analyses of associations among individual-level (level 1) socioeconomic status (SES) variables, obesity, unhealthy eating, and physical inactivity in Canadian youth1
View this table:
TABLE 3. Bivariate analyses of associations among area-level (level 2) socioeconomic status (SES) variables, obesity, unhealthy eating, and physical inactivity in Canadian youth1
Multivariate analyses that examined associations among the individual- and area-level SES measures and the occurrence of obesity, unhealthy eating, and physical inactivity are described in Table 4. Note that only the SES exposure variables that were significant (P < 0.05) predictors of the outcomes in the bivariate analyses were included in the multivariate analyses. As with the bivariate analyses, in the multivariate analyses individual- and area-level SES measures were related to obesity, whereas individual-level SES measures alone were associated with physical inactivity and area-level SES measures alone were associated with unhealthy eating. Additional analyses indicated that there was no interaction effect of the material wealth and perceived family wealth exposure variables in any of the multivariate models (P > 0.1).
View this table:
TABLE 4. Multivariate analyses of associations among socioeconomic status (SES) variables, obesity, unhealthy eating, and physical inactivity in Canadian youth1
The prevalence of participants with the 3 study outcomes was vastly different (4% obese, 25% unhealthy diet, 55% physically inactive), which may have affected the relations observed given that the likelihood of finding significant odds ratios is greater for more extreme outcomes. Consequently, all logistic regression analyses were repeated after forcing the outcomes to represent comparably sized groups for adiposity status [BMI above age- and sex-specific overweight cutoff (18): 19.2%], unhealthy eating (lowest quintile of unhealthy eating factor derived score: 20.0%), and physical inactivity (<3 d/wk active for 60 min: 22.2%). Risk estimates derived from these analyses were comparable to those detailed in Tables 2–4 (data not shown).
Stratified analyses were performed to consider the potential interactive effect of sex and grade (elementary school, grades 6–8 compared with secondary school, grades 9–10) on the observed relations. Although comparable levels of statistical significance were not always obtained because of smaller sample sizes, 1) similar patterns of associations were observed in girls and boys compared with those observed for the entire study sample, and 2) similar patterns of associations were observed within elementary and secondary school students compared with those observed for the entire study sample (data not shown).
The results of the path analysis are shown in Figure 1. The only SES variable significantly related to unhealthy eating was a high percentage of residents with less than a high school education. Material wealth and perceived family wealth were related to physical inactivity. Surprisingly, obesity was not significantly associated with unhealthy eating. However, obesity was directly associated with physical inactivity and 2 SES measures (material wealth, unemployment rate).
DISCUSSION
The primary purpose of this study was to examine associations between individual- and area-level measures of SES and obesity in adolescents by using a multilevel analytic approach. The principal finding was that individual- and area-level SES measures were independently related to obesity.
Our findings suggest that obesity prevention and intervention strategies in adolescents may need to focus on both the individual characteristics of those with a low SES (eg, providing more recreational opportunities that are affordable to those with a low SES) and the areas in which economically disadvantaged people tend to live (eg, increasing the number of supermarkets and decreasing the number of fast-food restaurants in poorer neighborhoods). With that being said, we recognize that this was a cross-sectional population-based study and not a randomized, controlled trial. A randomized trial would provide a superior level of evidence; however, conduction of a randomized trial in which both individual- and area-level SES variables are treated is a daunting and perhaps impossible task. Thus, a recommendation to incorporate both individual and environmental approaches to treat adolescent obesity at the population level will likely never be based on the strongest form of scientific evidence.
Many earlier reports have examined the relation between SES and obesity in adolescent populations from industrialized countries as previously reviewed (22). Although not all of these studies found significant associations, many found an inverse relation between individual- or area-level measures of SES and obesity. The inconsistent findings among studies may be explained by differences in the study populations, analytic methods used, or SES indicators examined. The present study, which was conducted in a large representative sample of Canadian youth, found modest relations between both individual- and area-level determinants of SES and obesity. Thus, it appears that adolescents belonging to deprived families are more likely to be obese, independent of neighborhood SES, and that adolescents living in poorer neighborhoods are more likely to be obese, regardless of family wealth.
To our knowledge, 3 previous studies have examined the relation between SES and adiposity level by using a multilevel analysis (12–14). Our findings in an adolescent population are consistent with those earlier studies in adults. Sundquist et al (12) found that education level and neighborhood deprivation had independent effects on obesity in 9240 Swedish men and women. Chaix and Chavin (14) reported that the risk of being overweight was increased with lower education and income at the individual level and lower gross domestic product at the area level in 12 948 French men and women. Finally, in 2259 American women Robert and Reither (13) found that individual and community SES disadvantages each explained in part the higher BMI values in blacks compared with nonblacks.
It is reasonable to assume that the mechanisms linking SES and obesity reflect the underlying effects of SES on dietary habits and physical activity status. In regard to dietary habits, energy-dense diets (eg, diets based primarily on foods with added sugars, fats, or both) are more affordable than prudent diets (eg, diet based primarily on healthy foods such as fresh fruit and vegetables) (5). Thus, it has been hypothesized that the association between deprivation and obesity is mediated in part by the low cost of energy-dense foods (5). This was not supported by our findings because individual-level measures of SES were not related to unhealthy eating and because the path analysis did not find an association between unhealthy eating and obesity. Poorer neighborhoods also tend to have more opportunities for unhealthy eating, as documented by a greater density of fast-food restaurants in neighborhoods with a low SES (6). This contention was supported by our findings because the area-level SES measure of education was related, albeit modestly, to unhealthy eating. In regard to physical activity, adolescents from families with lower SES have less opportunity to participate in sports and other physical activity pursuits because of cost or other access barriers (eg, poor parental support) (23, 24). Indeed, both individual-level measures of low SES were positively associated with physical inactivity in the present study, and the path analysis suggests that this is one of the pathways linking SES and obesity. Previous research also indicates that poorer neighborhoods provide less opportunities for physical activity (eg, fewer parks, unsafe streets and playgroups) (7, 8). Thus, we were surprised to find no associations between area-level SES determinants and physical inactivity in this study.
In general, the area-level SES determinants were weaker predictors of obesity, unhealthy eating, and physical inactivity by comparison to the individual-level SES determinants. It is possible that the area-level SES measures were not as relevant mechanistically. However, it is also possible that the weaker associations for area-level SES measures can be explained by less variability and more opportunity of nondifferential misclassification of exposure status between area-level measures than individual-level measures (25, 26). Hence, area-level measures are more likely to underestimate any true effects. Also note that the area-level SES measures used in this study were limited to census variables that focused on aggregated individual-level variables. A need exists for further multilevel studies that consider the influence of additional area-level characteristics, such as the density of parks and fast-food restaurants, on obesity.
Because the HBSC is representative of 6th–10th grade Canadian youth, these findings are valid for many Canadians and likely to youth of a comparable age from several other industrialized countries. However, it should be noted that 85% of the Canadian population is white. Although racial and ethnicity data were not available in this study, a similar racial composition would be expected in the HBSC sample. Previous research has indicated that race may influence the relation between SES and obesity. In a large biracial sample of American girls, the rates of obesity varied by SES level in whites but not in African Americans (27, 28). This finding suggests that the results of this study may not be applicable to all racial groups.
Several notable strengths to this study include the large and representative sample of Canadian adolescents, the consideration of both individual- and area-level SES measures, the use of a multilevel analysis, and the consideration of the dietary and physical activity pathways that link SES and obesity. However, some limitations also warrant recognition. First, because this was a cross-sectional study, the temporal directions of the associations under study cannot be identified. Misclassification bias is another limitation. Most notably, the 5-km radii surrounding the schools were used to estimate area-level SES for individual students, which would be inappropriate if the students lived further than 5 km from their schools. Nonetheless, the probabilities of SES misclassification were likely unrelated to obesity, unhealthy eating, or physical inactivity.
In summary, this study examined associations among SES with obesity, unhealthy eating, and physical inactivity among Canadian adolescents. Our findings indicate the importance of considering both individual- and area-level measures of SES. Because this study is the first multilevel analysis that examined these associations in adolescent populations, replication of these findings in different settings or contexts is warranted.
ACKNOWLEDGMENTS
IJ was responsible for the study design, performed the data analysis, and wrote the drafts and the final article. WFB, KS, and WP aided in the presentation and interpretation of the results and statistical analysis. None of the authors had any duality or conflicts of interest.
REFERENCES