Background On the population level, the association of age and other sociodemographic factors with eating disorders (ED) is unclear.
Methods We used cross-sectional data from the French general population-based NutriNet-Santé e-cohort (n=49 603 adults; 76.3% women; mean age=50.4±14.6 years). ED were evaluated in 2014 with the 5-item SCOFF screening questionnaire and the Expali algorithmic tool. Likely cases of restrictive, bulimic, hyperphagic and other ED were the dependent variables. Age, marital status, education, occupation, physical activity and smoking were the independent variables. Associations were estimated via gender-stratified multivariable polytomous logistic regression.
Results Among women, age displayed inverse linear associations with both restrictive and bulimic ED, underscoring 18–25 years as the most vulnerable period (adjusted OR=3.37, 95% CI: 2.24 to 5.08 for restrictive ED; adjusted OR=2.98, 95% CI: 2.37 to 3.74 for bulimic ED, respectively). A similar association was observed in men regarding bulimic ED. In women, age was not associated with hyperphagic ED for which living alone, low education, low physical activity, being a homemaker/disabled/unemployed/retired, a manual worker or a former/current smoker had increased importance. In men, 18–39 years emerged as the least vulnerable period regarding hyperphagic ED (adjusted OR=0.74, 95% CI: 0.56 to 0.99). Across gender, having postsecondary education had significant inverse associations with all except restrictive ED, whereas being a student had a significant positive association with restrictive ED.
Conclusions The findings support gender-specific associations of age with four ED subtypes and could inform future prevention initiatives targeting specific ED among specific age groups.
Trial registration number NCT03335644; Pre-results.
- mental health
- public health
- social factors In
Statistics from Altmetric.com
A group of mental illnesses known as eating disorders (ED) is considered insufficiently recognised and underdiagnosed in western countries,1 yet is very important from a public health viewpoint because of high rates of comorbidity with many behavioural, physical and mental conditions (anxiety disorders, depression, alcohol use disorders, obesity, sleep disorders, gastrointestinal disorders, hypertension, etc)1–4 and high rates of premature mortality.5 From a healthcare utilisation viewpoint, the effects of ED might persist over the lifespan.6 ED have distinct risk factors and onset patterns.7–9 According to the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5), the principal ED include binge eating disorder (BED), anorexia nervosa (AN) and bulimia nervosa (BN).7 BED is defined as recurring episodes (≥once a week over ≥3 months) of eating a substantially larger amount of food over a short time compared with the amount most people would eat under similar circumstances.7 BED, in comparison with other ED, might be more common among males and among older individuals.10 Unlike BED, AN predominantly affects adolescent girls and young women and is characterised by distorted body image, excessive dieting, dangerously low calorie intake and severe underweight.7 A recent meta-review highlighted AN as the mental illness associated with the highest premature mortality risk.11 In turn, BN is characterised by weekly binge eating episodes followed by inappropriate behaviours to avoid weight gain, including self-induced vomiting.7
Regarding the prevalence of ED among adults in western countries, and in France in particular, there is notable paucity of data, coupled with heterogeneous assessment methods. Research conducted during 2001–2003 in six European countries reported that ED prevalence (any subtype, based on diagnostic interviews) among adults aged 18+ years ranged from 1.3% (Germany) to 4.2% (France).12 More recent evidence from a representative German sample (aged 14–95 years) indicated that the point prevalence of ED (any subtype) was 10% based on self-reported data.13 Next, a narrative review reported AN, BN and BED prevalence in the range of 1%–4%, 1%–2% and 1%–4%, respectively, among European women, and an overall 0.3%–0.7% ED prevalence among European men.1 Finally, a meta-analysis of community surveys published through 2013 reported lifetime prevalence estimates for AN, BN and BED of 0.2%, 0.8% and 2.2%, respectively.14 These authors, however, highlighted epidemiological data limitations, such as diagnostic criteria modifications and heterogeneity of analysed ED subtypes.14 Recent evidence from a large sample of French university students suggested that the prevalence of ED (any subtype, based on self-reports) was 20.5%, with the disorders being nearly three times as prevalent among females as among males.15 Several studies have indeed suggested that younger cohorts have higher odds of ED than do older ones4 16; however, the association of age with different ED subtypes is not well elucidated, and non-US research is particularly scant. With this study, we strove to expand knowledge about ED by analysing their presence and sociodemographic correlates in a very large sample of European adults recruited from the general population. We focused particularly on the role of age, which we expected to be ED-specific and gender-specific.
We used data from the ongoing, population-based NutriNet-Santé e-cohort launched in France in 2009 (www.etude-nutrinet-sante.fr). Its design and objectives have been described elsewhere.17 Briefly, male and female volunteers aged ≥18 years with internet access are recruited via traditional strategies (eg, flyers) and periodic multimedia campaigns (television, newspapers, etc). Eligible participants provide informed consent and an electronic signature, and afterwards complete a set of baseline questionnaires on sociodemographics and lifestyle, physical activity (PA), anthropometrics, health status and diet. On a regular basis thereafter, they receive additional questionnaires on nutrition and/or health-related topics.
Eating disorder assessment
As part of the follow-up, the 5-item SCOFF questionnaire for ED screening18 was administered between June and December 2014. That instrument had been validated for use with French-speaking adults.19 A diagnostic threshold is fixed at two positive responses, with a sensitivity of 0.88 and a specificity of 0.93 using interviews as diagnostic reference.20 Data obtained with SCOFF are regarded as reflecting likely ED realistically approximating actual ED point prevalence.20 However, since SCOFF data do not permit differentiation among ED subtypes, the Expali validated algorithmic tool was also used.21 It takes into account each SCOFF response and the participant’s body mass index (BMI), and thus differentiates ED into four subtypes: restrictive disorders (which include AN, atypical AN and restrictive food intake disorder), bulimic disorders (which include BN and low-frequency/short-duration BN), hyperphagic disorders (which include BED and low-frequency/short-duration BED) and other ED.21 In total, 125 279 enrollees received the SCOFF questionnaire, of whom 51 073 completed it and were thus eligible for this analysis.
Self-reported sociodemographic information (gender, age, height and weight (calculated BMI), marital status (married/cohabiting or living alone), education (less than high school; high school diploma or equivalent; some college; undergraduate degree; graduate school), occupation (homemaker/disabled/unemployed; student; manual/blue collar; office work/administrative staff; professional/executive staff; retired)) and lifestyle (PA (low; moderate; vigorous), smoking (never; former; current smoker)) was collected ≤12 months before the SCOFF administration. Leisure-time PA was assessed with the International Physical Activity Questionnaire-Short Form, and scoring followed the established protocol.22
Descriptive characteristics (age category, BMI category, marital status, educational level, occupational status, PA, smoking status) were compared across ED subtypes via Χ2 test; mean age and mean BMI across ED subtypes were obtained via analysis of variance. Regarding age, linear trend tests were performed. Owing to the multiple comparisons across age and BMI, we also performed Tukey’s studentised range tests. The principal outcome (dependent) variable was the 4-category ED measure (restrictive, bulimic, hyperphagic and other ED), whereas the principal exposure (independent) measure was age. For interpretation purposes, age was modelled as a 7-level categorical variable: 18–25; 26–35; 36–45; 46–55 (reference); 56–65; 66–75 and ≥76 years. Owing to the lower rates of ED in men than in women and the relatively small number of cases in each age×ED cluster, the following age categories were used in the main analyses among men: 18–39; 40–59 (reference) and ≥60 years. Likewise, some categories of educational level and occupation were collapsed in the main analyses among men, while marital status, PA and smoking status were modelled with the same categories in men and women. Cross-sectional associations were estimated via multivariable polytomous logistic regression (no ED=reference category) providing adjusted ORs and 95% CIs. Thus, we were able to estimate the independent association of each sociodemographic correlate with each ED subtype. The choice of reference category for each variable in the analysis was guided by its value distribution, by evidence in the literature regarding modelling of the respective construct and by our interest in testing specific associations.
All tests were two-sided and p<0.05 was considered statistically significant (except for Tukey’s studentised range tests for which the significance level was set at p<0.01). The analyses were conducted with SAS (V.9.4, SAS Institute, Cary, North Carolina, USA).
Of the 125 279 enrollees who received the SCOFF, a total of 51 073 enrollees (40.8%) returned a completed questionnaire. Non-responders were significantly more likely than responders to be younger, to be current smokers, to be living alone and to hold manual/blue-collar jobs, and less likely to have postsecondary education (all p<0.0001; data not tabulated). As shown in figure 1, from the sample of 51 073 individuals, we excluded 2.9% owing to missing/aberrant anthropometric or sociodemographic data. Thus, we retained n=49 603 for analysis (37 865 (76.3%) women; 11 738 (23.7%) men). Mean age was 50.4±14.6 years and mean BMI was 24.0±4.5 kg/m2. Overall, 5986 (15.8%) women and 939 (8.0%) men were identified as likely cases of ED. The sociodemographic characteristics of women and men by ED subtype are summarised in tables 1 and 2, respectively (p<0.0001 from age×gender interaction tests). Gender-specific percentage distributions of restrictive, bulimic, hyperphagic and other ED were 1.3%, 4.4%, 8.1% and 2.0% among women, and 0.2%, 1.1%, 5.0% and 1.7% among men (p<0.0001). The cumulative point prevalence of likely ED by age and subtype among women and men is shown in online supplementary figures 1 and 2, respectively.
Supplementary file 1
Illustrating the heterogeneity of the sample, the raw percentage distributions suggested that in women, restrictive and bulimic ED might be less common after age 35, whereas hyperphagic and other ED might peak in the 60s. The patterns in men were less clear-cut, partly because of a small number of cases. Only hyperphagic ED appeared to display a pattern similar to that seen in women. Bulimic ED remained somewhat elevated between the mid-30s and the mid-70s.
Tukey’s studentised range tests in women revealed that all pairwise comparisons across age were significant except for no ED–hyperphagic ED, other ED–hyperphagic ED and no ED–other ED. In turn, all pairwise comparisons across BMI were significant in women. Among men, the following significant pairwise comparisons across age were found: no ED–bulimic ED, hyperphagic ED–bulimic ED, other ED–bulimic ED and other ED–restrictive ED. Finally, all pairwise comparisons across BMI were significant in men, except for no ED–bulimic ED, restrictive ED–bulimic ED and hyperphagic ED–other ED.
Age and other sociodemographic correlates of ED among women
Since all sociodemographic variables displayed significant associations with ED in bivariate analyses, they were included in the logistic regression models. The only exception was BMI, which was not included in the adjusted analysis in order to prevent problems of model fit or multicollinearity. In fact, BMI was taken into account by the Expali algorithm differentiating the four ED subtypes. In the multivariable polytomous logistic regression model in women (table 3), age (reference=46–55 years) displayed an inverse linear association with restrictive ED (reference=no ED), which was significant through age 75 years (p<0.0001). Likewise, age showed an inverse linear association with bulimic ED, significant for all age categories. For both restrictive and bulimic ED, age 18–25 years appeared to be the most vulnerable period (adjusted OR=3.37, 95% CI: 2.24 to 5.08 for restrictive ED; adjusted OR=2.98, 95% CI: 2.37 to 3.74 for bulimic ED, respectively). For both disorders, age appeared to be the most important sociodemographic correlate; other significant associations were found with living alone (reference=married/cohabiting), vigorous PA (reference=moderate PA) and being a current smoker (reference=never smoker).
Age was not associated with hyperphagic ED, except for a significant inverse association after age 65 years. For these ED, living alone, having low education (reference=high school diploma or equivalent), being a homemaker/disabled/unemployed/retired (reference=office work/administrative staff), a manual worker, a former/current smoker or having low PA emerged as significant correlates. Regarding other ED, age displayed a significant positive association only through the mid-30s, while other sociodemographic correlates with significant associations included manual work, low PA and being a former smoker. Having postsecondary education displayed significant inverse associations with all ED subtypes except restrictive ED, while being a student displayed a significant association only regarding restrictive ED (adjusted OR=1.58, 95% CI: 1.04 to 2.40).
In the overall multivariable polytomous logistic regression model in women, all sociodemographic covariates were statistically significant (all p<0.0001).
Age and other sociodemographic correlates of ED among men
Regarding the sociodemographic variables included in the logistic regression model, the same considerations applied in men as in women. Results of the multivariable polytomous logistic regression analysis in men (table 4) were markedly different from those obtained in women, even though the test for linear trend regarding age and each ED subtype was also significant (p<0.0001). There was a non-significant association of age (reference=40–59 years) with restrictive ED, for which living alone (reference=married/cohabiting), being a current student or a former smoker (reference=never smoker) displayed significant associations. Regarding bulimic ED, age showed a significant inverse association, with age 18–39 years appearing as the most vulnerable period (adjusted OR=1.63, 95% CI: 1.02 to 2.62). In contrast, that age category was the least vulnerable period regarding hyperphagic ED (adjusted OR=0.74, 95% CI: 0.56 to 0.99), for which living alone and being a former smoker displayed significant associations. There was also a significant inverse association between vigorous PA (reference=moderate PA) and hyperphagic ED. Next, age ≥60 years appeared to have a strong positive association with other ED (adjusted OR=1.85, 95% CI: 1.09 to 3.13). Finally, similar to the findings in women, having postsecondary education (reference=up to high school) displayed significant inverse associations with all ED subtypes except restrictive ED.
In the overall multivariable polytomous logistic regression model in men, all sociodemographic covariates except occupation were statistically significant (p value range between 0.009 and <0.0001).
This cross-sectional study, carried out in a large sample of French adults, provided epidemiological evidence of gender-specific and subtype-specific associations of age with ED. Among women, age displayed comparable inverse linear associations with both restrictive and bulimic ED, underscoring 18–25 years as the most vulnerable period. A similar finding was observed in men regarding bulimic ED, although the magnitude of the effect size was smaller. Age was generally not associated with hyperphagic ED in women. In men, age 18–39 years was the least vulnerable period regarding hyperphagic ED and the most vulnerable period regarding bulimic ED, highlighting the heterogeneity among ED risk factors.
Our findings are largely consistent with prior evidence suggesting that younger cohorts have higher odds of presenting with ED than do older ones.4 16 However, in this study, women aged up to 45 years appeared to be at increased risk of both restrictive and bulimic ED, whereas men appeared to be at increased risk of bulimic ED through the age of 40 years. Prior research with middle-aged Europeans has also reported non-negligible rates of ED after age 40 years. For example, in an Austrian sample of women aged 40–60 years, 4.6% had reported symptoms meeting full DSM-4 criteria for ED and another 4.8% were categorised as subthreshold ED cases.23 In turn, in a large sample of middle-aged women in the UK, 15.3% met the criteria for lifetime ED, while the 12-month ED prevalence was 3.6%.24
Regarding the other sociodemographic correlates of ED in women, and largely consistent with prior data,25–27 low education, being a homemaker/disabled/unemployed/retired, being employed in manual work, being a former/current smoker or having low PA emerged as important, independent determinants of hyperphagic ED. Living alone was also independently correlated with this ED type in women; however, in a clinical sample of women with ED, those living with a partner reported a higher frequency of purging behaviour and generally presented greater ED symptomatology and psychopathology compared with their peers living alone.28 The correlation between marital status and ED has been suggested to depend on overweight status.29 Next, in adjusted models in both men and women, having postsecondary education displayed significant inverse associations with all ED subtypes except restrictive ED, whereas being a student displayed a significant positive association only with restrictive ED. The latter is consistent with evidence highlighting the college years as a particularly vulnerable period regarding ED owing to the convergence of factors such as individuation, emancipation and intimacy.30
A common risk behaviour among individuals with ED is smoking, which has been reported as an appetite/weight control strategy.31 32 A case-control study documented that women with ED had higher rates of smoking and greater nicotine dependence than did controls, with women with binge/purge ED reporting the highest smoking rates.31 A recent meta-analysis also revealed that the highest rates of smoking were found among individuals with BED (lifetime prevalence of 47.7%), followed by BN (lifetime prevalence of 39.4%) and AN (lifetime prevalence of 30.8%).33 In our sample, current smoking was positively associated with all ED subtypes except other ED in women, whereas former smoking was positively associated with all ED subtypes except restrictive ED in women, and with restrictive and hyperphagic ED in men.
Research evidence points to shared vulnerability between ED and substance use disorders, explained by shared genetic predisposition and common neurobiological pathways (primarily involving the dopamine reward system).34 35 A recent review of twin studies suggested that a moderate proportion of the shared variance between ED and substance use disorders could be attributed to overlapping genetic factors, particularly regarding ED characterised by binge eating and/or inappropriate compensatory behaviours.36 In turn, binge eating and substance use/addictions have been associated with reduced activity in the orbitofrontal and prefrontal cortex areas (which are linked to self-control), and with dopaminergic and opiate pathway dysregulation.34
It should be noted that ED could be conclusively diagnosed only in a clinical setting, which was not feasible in this epidemiological context. We used self-reported information obtained via the validated SCOFF questionnaire. Whereas it is not a diagnostic tool, it has been strongly recommended for ED screening given its excellent psychometric properties20 and its capacity to provide a realistic approximation of actual ED point prevalence. It is one of several instruments used in the assessment of likely ED, in addition to the DSM criteria. The latter were modified in 2013 with the most notable change being the addition of BED as a proper diagnosis.7 The update likely impacts overall prevalence and trend estimates,37 compounding the challenges regarding epidemiological data collection and ED burden quantification methodology.38 Next, the SCOFF is posited to identify likely ED cases at a given point in time. The questions are formulated in a way suggesting repeatability/chronicity of the behaviour. Hence, we could speculate that recently emerged ED might be less well-perceived in terms of symptomatology, and thus possibly underestimated by the SCOFF. Another limitation of the latter is the lack of information about ED subtypes. To overcome this limitation, we employed a clinical algorithm (Expali), which takes into account the 104 possible combinations of the four main BMI categories (underweight, normal weight, overweight, obese) and the responses to each of the five SCOFF questions, and thus differentiates ED into restrictive, bulimic, hyperphagic and other ED.21 The last category, which includes ED that are less prevalent and less well-known, yet could have equally adverse health and well-being consequences, include night-eating syndrome,39 orthorexia nervosa40 and male-specific muscularity-oriented disordered eating.8 A limitation of the Expali algorithm, however, is the division of ED subtypes into four broad categories. Yet, such categorisation might not be problematic in a primary care setting. For example, in the case of typical and atypical AN, the clinical evaluation and specialised referrals are virtually identical.
Overall, 15.8% of women and 8.0% of men were identified as likely ED cases. Even though ED point prevalence evaluated with SCOFF in a representative sample of German women and men was reported to be 10%,13 ED rates in our study are somewhat higher than those reported by most previous research with adults,4 12 14 16 which might be explained by the choice of the modelled ED subtypes and also by the respondent profiles. Here, caution is advised when generalising the findings which were derived from a select, possibly health-conscious and nutrition-conscious sample of volunteers with complete data. Whereas the sample for analysis was very large and heterogeneous (notable strengths of the study), only 41% of the solicited NutriNet-Santé participants returned a completed SCOFF questionnaire. This was not unexpected, given our prior experience and the large number of questionnaires administered in this e-cohort.41 42 It should be noted, however, that a comparison with national Census figures has revealed differences between the general French population and the NutriNet-Santé e-cohort, especially regarding marital status and the proportions of women and relatively well-educated individuals.43 Finally, the findings regarding restrictive ED in men should be interpreted with caution due to the small number of cases and the subsequent unstable estimates.
In conclusion, given the deleterious impact of ED on quality of life, the high rates of comorbidity with many physical and mental disorders, and the fact that only a small percentage of individuals affected with ED seek treatment,1 4 12 16 24 future research and prevention efforts are needed. This study revealed gender-specific and subtype-specific associations of ED with age and other sociodemographic factors, and could inform prevention initiatives targeting specific ED among specific age groups. The results merit confirmation in similar and also in different contexts with objectively assessed ED.
What is already known on this subject
Eating disorders (ED) represent a heterogeneous group of mental illnesses that are insufficiently recognised and underdiagnosed in western countries.
The principal ED include anorexia nervosa (restrictive ED), bulimia nervosa and binge eating disorder (hyperphagic ED). ED are commonly associated with adolescent girls and young women.
What this study adds
This large cross-sectional study provided current population-level epidemiological data about the presence and correlates of different types of ED in France, thus augmenting knowledge about ED across age.
Overall, 15.8% of women and 8.0% of men were identified as likely ED cases, based on self-reported information provided via a validated screening tool.
Gender-specific percentage distributions of restrictive, bulimic, hyperphagic and other ED were 1.3%, 4.4%, 8.1% and 2.0% among women, and 0.2%, 1.1%, 5.0% and 1.7% among men.
Among women (mean age=48.8±14.3 years), age was inversely associated with both restrictive and bulimic ED, while being unrelated with hyperphagic ED.
Among men (mean age=55.8±14.2 years), age 18–39 years was the least vulnerable period regarding hyperphagic ED and the most vulnerable period regarding bulimic ED.
Across gender, having postsecondary education was inversely associated with all except restrictive ED, whereas being a student was positively associated with restrictive ED.
The findings could increase awareness about ED in western countries and could inform future prevention initiatives targeting specific ED among specific age groups.
The authors would like to thank Frédéric Coffinieres, Thi Hong Van Duong, Younes Esseddik (IT manager); Paul Flanzy, Régis Gatibelza, Jagatjit Mohinder and Maithyly Sivapalan (computer scientists); Julien Allegre, Nathalie Arnault, Laurent Bourhis, Véronique Gourlet, PhD, and Fabien Szabo de Edelenyi, PhD (data-managers/biostatisticians) for their technical contribution to the NutriNet-Santé study, and Nathalie Druesne-Pecollo, PhD (operational/logistics coordinator). The authors would also like to thank all volunteers in the NutriNet-Santé e-cohort.
Contributors VAA performed the literature review, the statistical analysis, led the writing and has primary responsibility for the final content; PG and SH designed the NutriNet-Santé study, directed its implementation and coordinated recruitment and data collection; SP coordinated the SCOFF questionnaire administration; M-PT, JL and PD provided methodological and theoretical guidance; VAA, M-PT, PD and CJ developed the study’s analytic strategy; all authors assisted with interpretation of data, read and edited each draft of the manuscript for important intellectual content. All authors read and approved the final manuscript.
Funding The NutriNet-Santé study is supported by the French Ministry of Solidarity and Health, the National Agency for Public Health (Santé Publique France), the National Institute for Health and Medical Research (INSERM), the National Institute for Agricultural Research (INRA), the National Conservatory of Arts and Crafts (CNAM), the Sorbonne-Paris-Cité Centre for Epidemiological Research and Statistics (CRESS), the NutriNet-SU.VI.MAX Association and the University of Paris 13.
Competing interests None declared.
Patient consent Not required.
Ethics approval This study was approved by the French Institute for Health and Medical Research (INSERM) and by the National Commission on Informatics and Liberty (CNIL).
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.