How obesity relates to socio-economic status: identification of eating behavior mediators

Abstract

Background:

Socio-economic status (SES) is a strong determinant of eating behavior and the obesity risk.

Objective:

To determine which eating and lifestyle behaviors mediate the association between SES and obesity.

Methods:

We performed a case–control study of 318 obese people and 371 non-obese people in northern France. Ten eating behavior traits were assessed using the Three-Factor Eating Questionnaire Revised 21-Item and an eating attitude questionnaire (on plate size, the number of servings, reasons for stopping eating and the frequency of eating standing up, eating in front of the television set (TV) and eating at night). The SES score (in three categories) was based on occupation, education and income categories. Mediation analysis was performed using the test of joint significance and the difference of coefficients test.

Results:

The age- and gender-adjusted obesity risk was higher for individuals in the low-SES groups (odds ratio (OR) (95% confidence interval (CI)=1.82 (1.48–2.24), P<0.0001). Additional servings were associated with a higher obesity risk (OR=3.43, P<0.0001). Cognitive restraint (P<0.0001) and emotional eating (P<0.0001) scores were higher in obese participants than in non-obese participants but did not depend on SES. Of the 10 potential factors tested, eating off a large plate (P=0.01), eating at night (P=0.04) and uncontrolled eating (P=0.03) significantly mediated the relationship between SES and obesity.

Conclusion:

Our results highlighted a number of obesogenic behaviors among socially disadvantaged participants: large plate size, uncontrolled eating and eating at night were significant mediators of the relationship between SES and the obesity risk.

Introduction

Obesity (defined as an excess body fat that can be harmful to health) has been recognized as a disease by the World Health Organization since 1997.¹ The prevalence of obesity is increasing worldwide (that is, in both developed and developing countries) and has reached epidemic proportions in some areas. Obesity results from an imbalance between food intake and energy expenditure. It is nevertheless a complex multifactorial disease influenced by both genetic and environmental factors and gene-environment interactions. The environmental factors notably include socio-economic and behavioral parameters. In 1989, Sobal and Stunkard² established that obesity was inversely correlated with socio-economic status (SES), especially in women.³ With a view of slowing down the progression of obesity, it is essential to determine which factors may explain the most the association with SES. Potential explanatory factors include economic difficulties, educational level, lifestyle behaviors, the lower cost of energy-dense foods (and its consumption in greater quantities), individual beliefs about the relationship between food and health, a lack of motivation for health because of immediate life difficulties, lower levels of physical activity, and greater exposure and sensitivity to the advertising of high-fat food or sodas (as extensively described elsewhere).^{4, 5, 6, 7}

Eating behavior has a major role in the physiological regulation of energy balance and (when dysregulated) in obesity. This regulation involves (i) the homeostatic pathway that regulates (through hunger and satiety) the amount of food as a function of the body’s energy needs,⁸ and (ii) the hedonic pathway that depends on environmental (such as the abundance and palatability of food⁹) and psychological factors. It has been suggested that three psychological dimensions can describe eating behavior, cognitive restraint (CR), uncontrolled eating (UE) and emotional eating (EE). CR corresponds to a person’s tendency to restrict food intake to achieve weight loss or to prevent weight gain.¹⁰ However, excessive CR may have a counterproductive effect and may ultimately be followed by weight gain.^{11, 12} UE is a tendency to eat more than usual because of a loss of control over intake. Finally, EE corresponds to overeating in response to negative emotions, for example, when feeling lonely, depressed or anxious. Moreover, associations between eating behaviors and SES have been already reported.^{7, 13}

Although the relationships between obesity and SES, and between obesity and eating behaviors have been described separately, few studies have looked at whether or not eating behaviors can explain (mediate) the relationship between obesity and SES. It has only been shown that fruit and vegetable intake mediates the relationship between educational level and high adiposity in women.¹⁴

Hence, the primary objective of the present case–control study was to establish which eating behaviors mediated the relationship between SES and the obesity risk in a population in northern France.

Materials and Methods

Study design

Subjects were recruited between 2008 and 2011 as part of the Gene and Environment Case Control Obesity Study (GECCOS), which included 605 obese patients recruited by Lille University Medical Center (Lille, France) and 403 non-obese participants recruited in the Centre de Prévention et d'Education pour la Santé at the Institut Pasteur de Lille (Lille, France). The main inclusion criteria for obese participants were as follows: age 18 years or over, and body mass index (BMI)>30 kg m⁻². The inclusion criteria for non-obese participants were as follows: age 18 years or over, BMI<30 kg m⁻², no personal history of diabetes or obesity, stable body weight (no more than a 3 kg variation in the previous 6 months) and no weight gains or losses of >10 kg in adulthood (other than for pregnancies). Non-obese participants were secondly chosen in order to constitute a similar group to cases in terms of proportion of women and age (in 5-year classes) using frequency matching. The protocol was approved by the local investigational review board (CPP Nord Ouest IV, CP05/63). All participants provided their written, informed consent.

Anthropometric, clinical and biochemistry data and detailed information on SES, physical activity, eating behavior and feelings of hunger were collected using the same questionnaire at the first medical consultation (for obese participants) or at the study visit (for non-obese controls). Individuals with missing data on SES and eating behavior variables were excluded, leaving a total of 371 non-obese and 318 obese participants in the final analysis (Supplementary Figure 1). Age, gender and BMI did not differ significantly between subjects who were included and those who were not.

Socio-economic status

SES was scored (range: 0–40) on the basis of occupational social class, educational level and income.⁷ Occupational social class was categorized according to the French National Institute of Statistics and Economic Studies’ (INSEE) nomenclature. Participants were divided into eight groups: (1) Farmers (10 points in the SES score); (2) Craftspeople, tradespersons and general managers (10 points); (3) Senior managers and higher-intellectual professions (0 point); (4) Mid-level professions (5 points); (5) Clerical and service staff (5 points); (6) Manual workers (5 points); (7) Retirees (10 points); and (8) Other people with no professional activity (10 points). Ten points were attributed to inactive individuals and missing values.

The educational level was coded into five categories as follows: no formal education or primary-school education only (10 points in the SES score); junior high (10 points); high school (5 points); a 2-year college degree (5 points); and a 3-year college degree or higher (0 point).

Personal and household monthly incomes were coded separately into four classes as follows: <€800 (10 points in the SES score); €800–€1300 (8 points); €1300–€2700 (5.5 points); and>€2700 (0 point). Five points were attributed to missing personal and household income values.

Points for each category was summed up and individuals were then categorized into three SES groups. High-SES group coded as (0) (score (0–20)); intermediate-SES group coded as (1) (score (20–30)); and a low-SES group coded as (2) (score (30–40)).

Lifestyle variables

Time spent in front of the TV was reported in hours per week. The level of physical activity level was evaluated with the Ricci–Gagnon self-questionnaire on daily activity, and on sports and recreational activities; the total score ranged from 5 (inactive) to 40 (very active), as used by Duclos et al.¹⁵ In terms of smoking exposure, participants were categorized as (0) never-smokers or (1) former/current smokers (that is, participants reporting at least one cigarette per day in the past and/or at the time of the study).

Eating behaviors

Eating behaviors (CR, UE and EE) were assessed with the revised Three-Factor Eating Questionnaire R-21 (TFEQ-R21),¹⁶ and eating habits and circumstances with a questionnaire on the following: plate size (ordinary or large), number of servings per meal (1 or ⩾2), reasons for stopping eating (not hungry anymore, gastric fullness or an empty plate/dish) and the frequency with which the participant ate standing up, ate in front of television or ate during the night (never or sometimes/always). The Cronbach's coefficient-α (used to estimate the internal consistency of each behavior) were 0.77, 0.84 and 0.93 for the CR, UE and EE behaviors, respectively, in the GECCOS.

Clinical data

Anthropometric measurements were recorded for all participants. Weight was measured with an electronic set of scales. A specific set of scales was used for obese participants. Height was measured barefoot (with the head in the Frankfurt plane position) with a height gauge. BMI was calculated according to the Quetelet equation.¹⁷

Statistical analyses

Data were analyzed using SAS software (version 9.1, SAS Institute Inc., Cary, NC, USA). All analyses were adjusted for age and gender as potential confounding variables. Subjects were compared using multivariate linear (for quantitative variables) and logistic (for categorical variables) regression analyses. Interactions between obesity and gender, or between obesity and SES were successively tested by fitting an interaction term (obesity × SES) or (obesity × gender).

For the mediation analysis, we used the model depicted in Figure 1. Path ‘a’ in this model refers to the associations between SES and each potential eating behavior mediator, as assessed by individual logistic regression analyses. Path ‘b’ refers to associations between potential mediators and the obesity risk, as assessed by individual logistic regression analyses and adjusted for SES levels (because SES may be a confounding factor in this association). Mediation between SES and the obesity risk was established using (i) the test of joint significance,¹⁸ which indicates that a variable is a mediator when both paths a and b are significant, and (ii) the Sobel test. The total effect of SES on the obesity risk (path c) was examined in a logistic regression analysis. The direct effect of SES on the obesity risk (path c’) adjusted for one significant mediator at a time (defined in the test of joint significance) was examined using individual logistic regression analyses. Then the mediation was quantified using the mediation ratio¹⁹ (1−β’/β × 100), where β is the initial coefficient for obesity (path c) and β’ is the coefficient for obesity when controlling for the significant mediators (path c’). β-Coefficients were standardized.

Figure 1

A conceptual model of potential mediators of the SES-obesity relationship. Path a: associations between SES and potential mediators; path b: associations between mediators and the obesity risk, adjusted for SES; path c: the association between SES and the obesity risk; path c’: the association between SES and the obesity risk, adjusted for significant mediators.

The threshold for statistical significance was set to P⩽0.05.

Results

Lifestyle and eating behaviors and obesity

The socio-demographic and lifestyle characteristics of 318 obese participants (mean BMI: 41.3 kg m⁻²) and 371 non-obese participants (mean BMI: 22.4 kg m⁻²) are summarized in Table 1. On average, the obese participants were slightly older than the non-obese controls (41.3 vs 39.3 years, respectively; P=0.03) and had a lower SES, according to the socio-professional category (with fewer managers and mid-level occupations), a lower educational level and a lower income. The age- and gender-adjusted obesity risk was higher for individuals in the low-SES groups (odds ratio (95% confidence interval)=1.82 (1.48–2.24), P<0.0001). Compared with non-obese participants, obese participants had a lower physical activity score (19.1 vs 10.4, respectively; P<0.0001) and spent more time watching TV (16 vs 28 h per week, respectively; P<0.0001). The obese and non-obese groups did not differ in their smoking habits.

Table 1 Descriptive, socio-economic and lifestyle variables by obesity status in the GECCOS

Concerning eating behaviors and the obesity risk, we first looked for significant interactions with gender but did not find any (P>0.09; data not shown). Hence, men and women were pooled in all subsequent analyses (Table 2). During meals, obese participants were more likely than non-obese participants to eat off a large plate (26 vs 14%, respectively). The age- and gender-adjusted obesity risk was 2.22 (P<0.0001) for eating off a large plate. Furthermore, obese participants were more likely than non-obese participants to have at least two servings (30 vs 12%, respectively); the associated adjusted obesity risk was 3.27 (P<0.0001). Obese and non-obese participants differed in terms of the reasons for stopping eating (P=0.007); gastric fullness or an empty plate/dish were more frequently mentioned by obese participants than by non-obese participants (18 vs 12% for gastric fullness and 43 vs 37% for an empty plate/dish, respectively). The associated adjusted risks of obesity were 1.97 (P=0.004) and 1.45 (P=0.03), respectively, compared with not being hungry anymore.

Table 2 Eating behavior phenotypes by obesity status in the GECCOS

When considering the circumstances, we found that obese participants were more likely to eat in front of the TV (73 vs 65%, P=0.025) or at night (22 vs 6%, P<0.0001) than non-obese participants; the adjusted risks of obesity were 1.53 (P=0.01) and 4.52 (P<0.0001), respectively, compared with individuals never displaying these behaviors. The three dimensions of eating behavior assessed by the TFEQ-R21 (CR, UE and EE) were all higher (P<0.0001) in obese participants than in non-obese participants.

All the above-mentioned associations persisted after adjustment for SES (Table 2).

Lifestyle/eating behaviors and SES

The characteristics of the participants by the SES group (high, intermediate and low) are summarized in Table 3. As expected, BMI was inversely proportional to SES (P<0.0001). Levels of physical activity (P<0.0001) were lower in the low-SES group than in the other two groups. In contrast, the mean time spent in front of the TV in the low-SES group was double that in the high-SES group (P<0.0001). There were no inter-group differences in smoking status.

Table 3 Descriptive, lifestyle and eating behavior variables by SES group in the GECCOS

In terms of eating behaviors, factors such as the number of servings, reasons for stopping eating, eating standing up, CR and EE did not differ as a function of the SES group. In contrast, eating off a large plate (P<0.0001), eating in front of the TV (P=0.001) and at night (P=0.03) were more prevalent among participants in the low-SES group. Furthermore, levels of UE were higher in the low- and intermediate-SES groups than in the high-SES group (P=0.03). We did not detect any significant interactions between eating behaviors, SES and obesity (Figure 2).

Figure 2

Eating behaviors as a function of obesity and SES levels. Eating behaviors were associated with (a) both obesity and SES, or (b) obesity but not SES. Data are quoted as the percentage or the mean±s.d. SES groups: high SES (white bars), intermediate SES (gray bars) and low SES (black bars). Analyses were adjusted for age and gender. P-interaction values for SES × obesity.

Mediation analyses

The results of the mediation analyses are summarized in Table 4. Of the 10 potential mediators tested (Figure 1), plate size, eating at night and UE were identified as mediators in the association between SES and obesity, because (i) both path ‘a’ and path ‘b’ were significant (in a test of joint significance) and (ii) Sobel tests were significant. The difference of coefficients was then calculated for these mediators. The total effect (path c) of SES on obesity was 0.60±0.11 (odds ratio (95% confidence interval)=1.82 (1.48–2.24), P<0.0001). The indirect effects (c–c’) of each mediator are shown in Table 4. The mediation ratios ranged from 3% (for eating at night and UE) to 8% (for eating off a large plate time) for the relationship between SES and obesity.

Table 4 Associations between SES and potential mediators (path a) and between potential mediators and the obesity risk (path b)

Discussion

The present case–control study is the first to have looked at the mediators of the relationship between severe obesity and SES. We found that the relationship between SES and the obesity risk is partly mediated by some eating behaviors (such as eating off a large plate, eating at night and UE). In contrast, other eating behaviors (the number of servings, reasons for stopping eating, CR and EE) were strong, but they are SES-independent determinants of the obesity risk.

It is also noteworthy that some eating behaviors were not at all associated with SES. One can therefore distinguish between behaviors linked to SES and the environment (that is, external cues and stimuli, such as eating off a large plate, and UE) and those linked to internal cues (disrupted regulation of eating) that do not differ between SES groups (two or more servings, stopping eating because of an empty plate or gastric fullness—signals unrelated to hunger per se—CR and EE).

Among the identified mediators, eating off a large plate is the one with the biggest effect (8%). Although large portion size is known to be associated with an increased prevalence of obesity,²⁰ the link between portion size and SES and the role of portion size as a mediator of the SES-obesity relationship had not been previously established. Large portion sizes may be owing to increased packaging size, larger servings in restaurants and use of a large plate. However, most studies have found that plate size has no effect on food intake,²¹ and a recent meta-analysis showed only a weak effect. However, the studies were very heterogeneous and were mainly conducted in non-natural settings and with non-obese participants.²²

Few studies have explored the relationship between the TFEQ score and SES. Only Dykes et al.¹³ showed that in a high-SES group, restraint and disinhibition scores were not significantly correlated with body weight; in contrast, the correlations were significant in the intermediate- and low-SES groups. In the present study, only levels of UE were higher in the low-SES group than in the high-SES group.

As expected, the TFEQ scores recorded were higher in obese participants than in non-obese participants. Consistently, earlier studies have found that the relationship between the TFEQ score and BMI depends on the study population in question. For example, CR was negatively correlated with BMI in a clinical sample but positively correlated with BMI in a web-based sample,²³ and positively correlated with BMI in normal-weight participants and did not correlate or negatively correlated with BMI in obese and overweight participants.^{24, 25} Similarly, an interaction of gender for the association between TFEQ scores and BMI has not been found consistently.^{23, 24, 26}

The underlying physiological mechanism regulating eating behavior has been described by Blundell et al.⁸ as the ‘satiety cascade’. Eating behavior is regulated by internal feelings (such as hunger and satiety) that vary from one individual to another.²⁷ It was first suggested over 50 years ago that obese individuals might be less sensitive to internal cues than to external cues.²⁸ In the present study, we found that the obese participants were more likely to finish a meal solely because of gastric fullness or because the plate/dish was empty. These results are in line with laboratory studies demonstrating that obese people are more responsive to stomach sensations²⁹ or external stimuli.³⁰ Hetherington³¹ reported that participants who rated gastric fullness as the most important reason for terminating a meal consumed more calories. Moreover, individuals prone to weight gain and obesity have altered neuronal responses to food cues in brain regions known to be important in energy-intake regulation.³² Physiologically, this can promote feelings of being unable to control the response to eating food, and may subsequently lead to increased UE and EE and decreased CR in some individuals.

One major strength of the present case–control study was its demonstration that several eating behaviors can mediate the relationship between SES and obesity. However, several obesogenic behaviors are independent of SES and appear to be more strongly linked to interoceptive differences in feelings of satiety. Our observations must be interpreted with a degree of caution because cases and controls were enrolled in different contexts (hospital recruitment and recruitment from the general population). However, the variety of participants also strengthen our data by offering a more comprehensive understanding of these phenomena across differing groups of individuals. A limitation of the study is that a large proportion of obese subjects were excluded from the analyses because of missing values, therefore reducing the statistical power of the study. However, age, gender and BMI did not differ significantly between subjects who were included and those who were not. Also, a limit of the mediation model is that it is a causal model, with a strong assumption that one variable has a causal effect on another variable. Even though the cross-sectional design does not allow us to establish a causal, one-way relationship between eating behaviors and obesity (or between eating attitudes and obesity), evidence of data from the literature on eating behaviors and SES supports the relevance of our model.^{7, 13, 33}

In conclusion, the present case–control study comprised a detailed analysis of the contribution of eating behaviors to the SES-obesity relationship. Eating behaviors are key factors in the management of obesity and energy balance disruption. More case–control studies and longitudinal are needed to better understand the relationships between obesity and its causal behavioral factors. Finally, better knowledge of these factors and their relationships with SES should enable the development of personalized strategies that help individuals with low SES to improve food choices, eating habits and circumstances, and increase levels of physical activity.