SEARCH

SEARCH BY CITATION

Keywords:

  • BMI;
  • education;
  • sex;
  • socioeconomic position;
  • Switzerland

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

Objective: To assess whether the rapid increase in obesity prevalence among persons with higher education levels observed in one U.S. study is also observed in a European adult population.

Research Methods and Procedures: This study involved continuous surveillance of the adult population of Geneva, Switzerland (1993 to 2004), with annual random, independent, cross-sectional, representative samples (6635 men and 6558 women, ages 35 to 74 years) and analysis of 12-year trends in obesity prevalence across educational level subgroups.

Results: Obesity prevalence in men had an upward trend in the medium education subgroup (p < 0.02), a borderline upward trend in the high education subgroup (p < 0.08), but no trend in the low education subgroup. There was a borderline trend interaction between the male low and medium education subgroups (p < 0.09). Obesity prevalence in women had a borderline increase in the low education subgroup (p < 0.08), an almost borderline increase in the high education subgroup (p = 0.11), but no significant trend in the medium education subgroup. There was no evidence of trend interaction between the female education groups.

Discussion: In Geneva, as in the United States, the inverse association between education level and obesity rates has weakened over time among men, but, inconsistent with the U.S. findings, has persisted for women. Explanations may include more physically demanding occupations for men with low education levels and different attitudes toward body image between the sexes.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

The World Health Organization estimates that, worldwide, more than 1 billion individuals are overweight (BMI ≥25 kg/m2) and 315 million are obese (BMI ≥30 kg/m2). This obesity epidemic is escalating rapidly; in 1995, there were 200 million obese individuals (1, 2). Today, two-thirds of the U.S. adult population are overweight or obese, and obesity is currently the second leading cause of preventable disease and death, after smoking (3, 4). The health care costs related to obesity are steadily increasing, particularly for the most obese (5). Obesity increases the risk of fatal and non-fatal—but highly debilitating—non-communicable diseases, particularly cardiovascular diseases, type 2 diabetes mellitus and other endocrine and metabolic disturbances, sleep apnea, osteoarthritis, certain types of cancer, and several psychological problems, as well as poor health-related quality of life (6, 7).

Several interacting factors have been found to be involved in the complex process that determines body composition and, specifically, adiposity: genetic predisposition, biological characteristics, and factors specific to culture, environment, and behavior (1, 2, 3). Socioeconomic position (SEP)1 or status determines an individual's access to food and exercise facilities. Moreover, it is inversely associated with healthy working and living conditions (8, 9). SEP is also associated with energy intake and expenditure and, as a result, with adiposity (10).

In developed countries, a consistent inverse association between SEP and obesity has been observed among women, but less consistently among men (9,11–14). This association is influenced by ethnicity: for African Americans and Mexican Americans, the association between SEP and obesity varies by sex and age. The most severe inequality in obesity was found within the 40- to 49-year age range in both sexes, and, in contrast to white men and women, a positive association between obesity and SEP was found among African-American and Mexican-American men only (11). Among white women and men, those in lower-status occupations tend to gain more weight than those employed in higher-status occupations. When using education level as an indicator of SEP, this inverse association remains but is weaker (13).

Recent findings in the United States using education level as a proxy for SEP suggest that obesity increases more rapidly among more educated persons. It has been shown that the prevalence of obesity increased between 1970 and 2000, particularly in the most highly educated men and women, and the inverse association of obesity and education level became smaller over time (15). However, these findings are not supported by the results of another study, which used education and income to classify SEP and found a similar increase of BMI in all SEP groups (16).

It is currently uncertain whether trends by education level such as those found in the United States (15) pertain to European populations. Whether or not obesity affects education subgroups differentially is of high public health relevance. We, therefore, analyzed obesity trends across educational level subgroups in Geneva, Switzerland.

Research Methods and Procedures

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

Twelve-Year Surveillance and Study Subjects

The ongoing, community-based surveillance project Bus Santé (“Health Bus”) has been continuously monitoring chronic disease risk factors in the resident adult population of the canton of Geneva, Switzerland, since 1993 (17). The canton of Geneva (246 km2) has a population of about 438,500 primarily French-speaking inhabitants, of whom 61% are of Swiss ancestry and 39% are of foreign origin. Of the latter, approximately three-quarters are of European origin, and just over half come from Mediterranean countries, namely Spain, France, Italy, and Portugal. Survey participants were selected independently and uniformly throughout each year, starting in 1993, to represent the approximately 218,000 non-institutionalized men and women ages 35 to 74 years residing in the canton.

Eligible subjects are identified using a standardized procedure from an annual residential database established by the cantonal government. All Swiss and foreign citizens living in the canton with an official residence permit are registered. The only specific information from the list used in the survey (sex, age, and Swiss or foreign origin) is highly accurate. Stratified random sampling by sex within 10-year age strata is proportional to the corresponding population distributions. Selected subjects are mailed an invitation to participate, and, if they do not respond, up to seven telephone attempts at different times on various days of the week are made. If telephone contact is unsuccessful, two more letters are mailed. Each subject's recruitment lasts from 2 weeks to 2 months. Subjects who are not reached (15% of men, 19% of women) are replaced using the same selection protocol. Previous results have shown that subjects who cannot be contacted usually no longer reside in the canton and, therefore, are not eligible for the study. Subjects who refuse to participate are not replaced. Participating subjects are not eligible in future surveys. Annual participation rates have ranged from 57% to 65%.

For assessment of socioeconomic and health-behavioral, clinical, and biological chronic disease risk factors, each participant completes three self-administered questionnaires, general health, physical activity (18), and diet (19), that are specifically developed for the target population. During a scheduled appointment at a mobile epidemiology clinic, the questionnaires are checked by trained survey personnel for correct completion, and body weight and height are measured.

Assessment of Relative Weight (BMI)

Quetelet's BMI is a recognized simple measure for assessment of the general degree of overweight and obesity. It is calculated by dividing body weight in kg by height in squared meters (kg/m2). Body weight and standing height are measured after participants remove all outerwear and empty their pockets.

Weight is measured using a regularly calibrated mechanical beam balance (maximum weight, 150 kg; precision, 0.1 kg). Participants are asked to step on the center of the balance and to remain in a relaxed vertical position. From the measurement reading, 1 kg is subtracted to account for the mean value of the weight of garments worn across seasons.

Standing height is measured using a stadiometer (maximum height, 220 cm; precision, 0.5 cm). The level of the top of the head, which is in the Frankfort plane, is determined with a movable right-angled block on the measuring rod. Participants are told to breathe in before the measurement to assure that they are standing as tall as they can.

The following BMI categories were used (6): normal weight: 18.5 ≤ BMI < 25 kg/m2; overweight: 25 kg/m2 ≤ BMI < 30 kg/m2; and obesity: BMI ≥ 30 kg/m2.

Classification into Education Subgroups

SEP was defined by the highest attained education level (20, 21, 22). Low, medium, and high education levels were coded as follows: low education: mandatory school only, i.e., 9th grade or less; medium education: high school, i.e., 10th to 12th grade; and high education: university or higher.

Statistical Analysis

Age-adjusted multiple linear regression models stratified by sex were used to evaluate trends and differences in the prevalence of obesity from 1993 to 2004 across the three education subgroups. In most of the analyses, the 12 years from 1993 to 2004 were categorized into the four time periods 1993–1995 (reference), 1996–1998, 1999–2001, and 2002–2004 to provide large enough sample sizes at each time-point. A p value of <0.05 was considered statistically significant, and a level of 0.05 ≤ p ≤ 0.10 was considered borderline significant.

In a first series of models, obesity prevalence (%) was regressed on age (years), time period (1 = 1993–1995 … 4 = 2002–2004), education (1 = low, 2 = medium, 3 = high), and a time period × education product interaction term. For analyzing trends, the time period and education variables were assumed to be continuous, while, for estimating prevalence, the age and time period × education variables were considered to be categorical. For illustrative purposes, annual trends are shown in the figures.

Based on the categorical versions of the above models, all pairwise differences in obesity prevalences between the education subgroups were calculated within each time period. Next, a second series of regression models that included (continuous) dummy variables (1 = yes, 0 = no) for each of the three possible sets of education subgroup pairs (low and medium; medium and high; low and high), together with the relevant time period × dummy variable product interaction terms, was fitted. We then performed the corresponding F tests of the null hypotheses that the interactions and trends were equal to 0 to assess whether the obesity prevalences evolved differently among the education subgroups across successive survey periods and whether there were significant trends.

In addition, differences in obesity prevalence among each of the education subgroups and time periods relative to those of the 1993–1995 low education reference subgroup were calculated together with simultaneous 95% Dunnett-type confidence intervals for all comparisons with a control.

Finally, we investigated the possibility of differential age effects on the results by re-running all of the analyses after stratifying the data by the age subgroups <50 and ≥50 years.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

Education and Anthropometric Measures

Table 1 shows the education levels and anthropometric measures of the study participants by sex. Overall, men had higher BMI as well as higher obesity and overweight prevalences compared with women (all p < 0.0001).

Table 1.  Education level and anthropometric measures of adults 35 to 74 years of age, Geneva, Switzerland, 1993 to 2004
Characteristic or measurementWomen (n = 6558)Men (n = 6635)Sex difference p*
  • SD, standard deviation.

  • *

    χ2 test for characteristics, Student's t test for measurements.

  • Overweight, 25 ≤ BMI < 30 kg/m2; obese, BMI ≥ 30 kg/m2.

Education level (%)   
 Low24.815.2<0.0001
 Medium46.555.0<0.0001
 High28.729.80.15
Mean age in years51.151.60.013
 (SD)(10.6)(10.8) 
Mean weight in kg64.179.9<0.0001
 (SD)(11.6)(12.5) 
Mean height in cm162.8175.0<0.0001
 (SD)(6.6)(7.3) 
Mean BMI24.226.1<0.0001
 (SD)(4.4)(3.7) 
Overweight (%)24.043.9<0.0001
Obese (%)10.013.4<0.0001

Trends in Obesity by Education Groups

The results of the analyses of overall and education level-specific trends in obesity prevalence over the four grouped time periods are shown in Tables 2and 3. Annual trends are illustrated in Figures 1 and 2.

Table 2.  Trends and differences in age-adjusted obesity prevalence by education level, Geneva, Switzerland, 1993 to 2004: women
Sample size1993 to 1995 (n = 1140)1996 to 1998 (n = 1815)1999 to 2001 (n = 1804)2002 to 2004 (n = 1799)Trend p*
  • CI, confidence interval.

  • *

    Prevalence and trend p obtained from multiple linear regression models of obesity (%) on age (years), education level, time period, and education level × time period interaction term. Education level and time period were considered to be categorical (continuous) for analyzing prevalence (trend).

  • For example, during 1996 to 1998, the difference between women with low vs. high education was (16.0% − 5.2%) = 10.8%.

  • Overall interaction, p = 0.91.

  • §

    Difference relative to the single reference group, low education in 1993 to 1995, with simultaneous Dunnett-type 95% CIs. A difference is statistically significant at overall p = 0.05 if the corresponding CI excludes 0. For example, in 2002 to 2004, high education women had (12.7% − 7.1%) = 5.6% lower obesity prevalence than low education women in 1993 to 1995, and this difference was statistically significant.

Overall (n = 6558) (n: obese/overweight/normal)8.4 (100/250/790)10.0 (173/445/1,197)10.7 (189/437/1,178)11.6 (195/440/1,164)0.12
Low education (n = 1008) (n: obese/overweight/normal)12.7 (42/86/186)16.0 (75/1583/221)16.2 (756/150/217)17.0 (73/153/188)0.077
Medium education (n = 3646) (n: obese/overweight/normal)8.9 (49/119/381)8.8 (76/201/608)10.8 (91/196/552)10.4 (81/175/522)0.18
High education (n = 1981) (n: obese/overweight/normal)3.6 (9/45/223)5.2 (22/86/368)5.0 (23/91/409)7.1 (41/112/454)0.11
Difference in obesity prevalence within time period    Interaction p
 Low minus medium3.87.25.46.60.53
 Medium minus high5.33.65.83.30.67
 Low minus high9.110.811.29.90.84
Difference in obesity prevalence relative to low education in 1993 to 1995 (95% CI)§     
 (Low 93 to 95) minus low (95% CI)0 (reference)−3.3 (−9.1; 2.6)−3.5 (−9.3; 2.4)−4.3 (−10.2; 1.7) 
 (Low 93 to 95) minus medium (95% CI)3.8 (−1.7; 9.5)3.9 (−1.3; 9.2)1.9 (−3.3; 7.2)2.3 (−3.0; 7.7) 
 (Low 93 to 95) minus high (95% CI)9.1 (2.5; 15.7)7.5 (1.7; 13.4)7.7 (2.0; 13.5)5.6 (0.1; 11.2) 
Table 3.  Trends and differences in age-adjusted obesity prevalence (%) by education level, Geneva, Switzerland, 1993 to 2004: men
Sample size1993 to 1995 (n = 1202)1996 to 1998 (n = 1784)1999 to 2001 (n = 1848)2002 to 2004 (n = 1801)Trend p*
  • CI, confidence interval.

  • *

    Prevalence and trend p obtained from multiple linear regression models of obesity (%) on age (years), education level, time period, and education level × time period interaction term. Education level and time period were considered to be categorical (continuous) for analyzing prevalence (trend).

  • For example, during 1996 to 1998, the difference between men with low vs. high education was (21.8% − 8.6%) = 13.3%.

  • Overall interaction, p = 0.21.

  • §

    Difference relative to the single reference group, low education in 1993 to 1995, with simultaneous Dunnett-type 95% CIs. A difference is statistically significant at overall p = 0.05 if the corresponding CI excludes 0. For example, in 2002 to 2004, high education men had (19.2% − 10.3%) = 8.9% lower obesity prevalence than low education men in 1993 to 1995, and this difference was statistically significant.

Overall (n = 6635) (n: obese/overweight/normal)11.1 (135/476/591)14.0 (249/795/740)13.1 (244/828/776)14.6 (261/811/729)0.011
Low education (n = 1883) (n: obese/overweight/normal)19.2 (38/82/75)21.8 (63/133/86)15.7 (43/135/87)19.4 (52/134/80)0.53
Medium education (n = 3051) (n: obese/overweight/normal)11.4 (78/295/308)14.5 (144/453/406)14.6 (152/467/411)16.0 (150/428/354)0.012
High education (n = 1624) (n: obese/overweight/normal)6.0 (19/99/208)8.6 (42/209/248)9.1 (49/226/278)10.3 (59/249/295)0.078
Difference in obesity prevalence within time period    Interaction p
 Low minus medium7.87.31.13.40.081
 Medium minus high5.48.35.55.70.93
 Low minus high13.213.26.69.10.12
Difference in obesity prevalence relative to low education in 1993 to 1995 (95% CI)§     
 (Low 93 to 95) minus low (95% CI)0 m (reference)−2.6 (−10.9; 5.6)3.5 (−4.9; 11.8)−0.2 (−8.6; 8.1) 
 (Low 93 to 95) minus medium (95% CI)7.8 (5.8; 14.92)4.7 (−2.2; 11.6)4.6 (−2.4; 11.5)3.2 (−3.8; 10.1) 
 (Low 93 to 95) minus high (95% CI)13.2 (5.2; 21.2)10.6 (3.1; 18.0)10.1 (2.7; 17.4)8.9 (1.6; 16.2) 
image

Figure 1. Annual trends in obesity prevalence by education level in Geneva, Switzerland, 1993 to 2004: women 35 to 74 years of age.

Download figure to PowerPoint

image

Figure 2. Annual trends in obesity prevalence by education level in Geneva, Switzerland, 1993 to 2004: men 35 to 74 years of age.

Download figure to PowerPoint

Among women, an apparently steady increase in overall obesity prevalence was not statistically significant (Table 2). Likewise, there appeared to be similar increases in all of the education subgroups (Figure 1). In fact, the increase in the low education subgroup was borderline significant (p = 0.077), and the increase in the high education subgroup just missed being borderline significant (p = 0.11) (Table 2). There was no evidence of trend differences between the education groups (all interactions p ≥ 0.53) (Table 2). When the data were stratified by the age groups <50 vs. ≥50 years, there was no significant indication of trend interaction in either age group (all interactions p ≥ 0.12) (age-stratified results not shown).

Among men, there was a significant steady increase in overall obesity prevalence (p = 0.011) (Table 3). There appeared to be a declining trend in the low education subgroup and similar upward trends in the medium and high education subgroups (Figure 2). In fact, there was no significant trend in the low education subgroup (p = 0.53), a borderline upward trend in the high education subgroup (p = 0.078), and a significant upward trend in the medium education subgroup (p = 0.012) (Table 3). There was also a borderline trend difference between the low and medium education subgroups (interaction p = 0.081) (Table 3). When the data were stratified by the age groups <50 vs. ≥50 years, there were no significant trend interactions in either age group (all interactions p ≥ 0.13) (age-stratified results not shown).

Patterns of Obesity Prevalence Differences by Education Groups

Tables 2and 3 also focus on patterns of pairwise differences in obesity prevalence between the education subgroups, either within time period or relative to the low education (reference) subgroup in 1993–1995.

Among women, there was no discernible pattern across time periods (Table 2). Among men, there was some indication that the low-medium and low-high within-time-period differences were decreasing with time (Table 3).

Among women, compared with the low education subgroup in 1993–1995, obesity prevalence was significantly lower only in the high education subgroup, with some indication that the difference tended to decline in the subsequent time periods (Table 2). Similarly, among men, compared with the low education subgroup in 1993–1995, obesity prevalence was significantly lower in the high education subgroup, and the difference declined in the subsequent time periods (Table 3). Obesity prevalence in 1993–1995 was also significantly lower in medium compared with low education subgroup men, also perhaps with some tendency to decline in the subsequent time periods (Table 3).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

The prevalence of obesity among adults in Geneva, Switzerland, tended to increase steadily over the 12-year period 1993 to 2004 in both sexes, but particularly so in men. Overall, men had higher prevalence of obesity (13.4% vs. 10.0%) and overweight (43.9% vs. 24.0%) than women. The association patterns between education level and obesity trends differed by sex. In men, obesity prevalence tended to increase in the medium and high education subgroups, as opposed to little change in the low education subgroup. In women, obesity prevalence tended to increase in all education groups, but particularly in the low education subgroup. Thus, the inverse association between obesity and education level may have decreased for men but persisted for women.

In the United States, obesity has increased in both sexes, particularly among persons with higher SEP. The differences in obesity prevalence between education groups decreased from 1970 to 2000 (15). However, when analyzing the same National Health and Nutrition Examination Survey data using income as a proxy for SEP, Chang and Lauderdale (14) observed an unchanged inverse association between income and obesity for women. For men, the inverse association was weak for non-Hispanic whites in more recent waves of National Health and Nutrition Examination Surveys, but the association was even positive in African- and Mexican-American men (14). The association with obesity observed in our study depends somewhat on the choice of education as a proxy for SEP, but the results are consistent with those of Zhang and Wang (15) for men, although not for women.

European trend studies using education as a proxy for SEP have mostly found a persistent inverse association (13, 23, 24, 25, 26, 27) or no association (28, 29) between education level and obesity prevalence. On the other hand, two Swedish studies have reported contradictory results: one observed an increase in obesity prevalence among middle-SEP men (30), whereas the other observed a decrease in BMI for men with low education (10). Both of the latter studies indicated a narrowing of obesity prevalence across SEP subgroups in men and are, therefore, consistent with our results. However, all of the above studies contain at least one of the following major limitations: the use of self-reported weight and height, which is known to be influenced by SEP and obesity (31, 32, 33); the study data were collected 10 or 20 years ago; and the assessment of trends was based on only two points in time. On the other hand, the present study results were based on measured weight and height; used annual random, independent, representative samples from 1993 to 2004; and evaluated trends across at least four time-points over a 12-year period.

How might the observed SEP and sex differences in obesity trends be explained? It has been shown in British adults that high-SEP subgroups have a higher frequency of weight monitoring and a lower threshold for defining themselves as overweight, and they make more deliberate efforts to control their weight (34). SEP influences an individual's social and economic environment, as well as access to food and health care services (8, 9). Low-SEP persons have been shown to have less healthy dietary patterns. They consume less fish, vegetables, and fruits; have lower iron, calcium, and vitamins A and C intake; and consume more energy-dense foods (10, 35, 36, 37).

The sex differences found in this study may have resulted from differing attitudes between men and women toward body weight and different practices for controlling body weight (9, 38). The faster rate of increase in obesity prevalence among low-SEP women may also have resulted from inverse selection owing to discriminatory hiring practices. Discrimination against obese job applicants, especially women, is documented in simulated job interviews (39, 40). In women, obesity is associated with unemployment and low income, whereas in men, obesity is not associated with economic or social disadvantage; among men, only thinness is found to be related to low income (41, 42, 43, 44).

Traditionally, women in couples perform the majority of routine household chores. This sex difference persists even if both partners work outside the household: women who work spend 3 times more energy doing household chores than men who work (18% of total energy expenditure in Geneva women vs. 6.5% in Geneva men, p < 0.0001, data not shown). Furthermore, low-SEP men are more likely to have physically demanding occupations than high-SEP men (45). These arguments may explain why obesity prevalence might be stable in low-SEP men and why women, and particularly high-SEP women, who are more concerned about body image and may practice better weight control than men, tend to have lower obesity prevalence (38, 43, 46, 47, 48).

Study limitations include the cross-sectional design, the length of the observation period, the sample size, and the use of education as a proxy for SEP. The cross-sectional design excludes establishing a causal relationship between education and obesity. All else being equal, a longer survey period or a larger sample size would have led to statistically significant trend interactions by education subgroups in men. The use of education as an indicator of SEP, while not ideal, at least allowed for more direct comparisons of the present study results with those of Zhang et al. (15).

Conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

To our knowledge, the present study is the first to show that obesity prevalence in men with medium or high education may be approaching that of men with low education in Europe. The inverse association between SEP and obesity has weakened over time for men but persists for women. Possible explanations include more physically demanding occupations in low-SEP men and different attitudes toward body image between the sexes.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References

This study was supported by grants from the Swiss National Fund for Scientific Research (32-31.326.91, 32-37,986.93, 32-46,142.95, 32-47,219.96, 32-49,847.96, 32-054,097.98, 32-57,104.99, 32-68,275.02). We thank Ryan Theis for his useful comments on earlier drafts of this article.

Footnotes
  • 1

    Nonstandard abbreviation: SEP, socioeconomic position.

  • The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must, therefore, be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Research Methods and Procedures
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. References