In a population-based survey of American households—the first nationally representative study of eating disorders in the United States—we found estimates of lifetime prevalence for eating disorders that are broadly consistent with earlier data. However, we found a surprisingly high proportion of men with anorexia nervosa and bulimia nervosa (representing approximately one-fourth of cases of each of these disorders). By contrast, clinical and case registry studies (Fairburn and Beglin 1990
; Hoek and van Hoeken 2003
) report that fewer than 10% men among cases of these disorders, and population-based studies report a 15% proportion of men for anorexia nervosa (Garfinkel et al 1996
) and 8%–10% of men for bulimia nervosa (Bushnell et al 1990
; Garfinkel et al 1995
). Note, however, that estimates from population-based studies, including ours, are unstable because they involve small numbers of men with eating disorders (no more than 5 men with either disorder in any study).
Our findings provide unique data regarding the lifetime duration of eating disorders, and the onset and duration of binge eating disorder, together with extensive information on sociodemographic features of individuals with all 5 disorders. Also, our study provides support for the common impression that the incidence of bulimia nervosa has increased significantly in the second half of the twentieth century (Kendler et al 1991
; Hoek and van Hoek 2003
), and it provides the first data showing a similar trend for binge eating disorder. Nevertheless, there are some data suggesting that the incidence of bulimia nervosa may be leveling off in recent years (Currin et al 2005
). Whether the incidence of anorexia nervosa has increased over time is unclear and subject to debate. We failed to find a significant increase, but had little power to detect such a trend; case registry study data have yielded conflicting findings and interpretations (Fombonne, 1995
; Lucas et al 1999
; Hoek and van Hoeken 2003
; Currin et al 2005
We found that lifetime anorexia nervosa is associated with a low current BMI, a finding consistent with follow-up studies of clinical samples of individuals with anorexia nervosa showing that low weight often persists after resolution of the disorder (Steinhausen 2002
). By contrast, binge eating disorder was found to be strongly associated with current severe obesity (BMI _ 40)—a finding also consistent with earlier reports (de Zwaan 2001
; Streigel-Moore and Franko 2003
; Hudson et al 2006
). Although the causal pathways responsible for this latter association are unclear, shared familial factors (such as shared genes or shared family environmental exposures) are likely at least partly responsible (Hudson et al 2006
We also assessed role impairment in all disorders except anorexia nervosa, where analysis was precluded because no 12-month cases were identified. While the majority of respondents with bulimia nervosa, binge eating disorder, or any binge eating reported at least some role impairment in at least 1 role domain, only 21.8% of respondents with subthreshold binge eating disorder reported any role impairment. Severe role impairment was uncommon in all conditions. It is important to note, though, that participants may possibly have under-reported role impairment due to factors such as minimization, shame, secrecy, or lack of insight stemming from the ego-syntonicity of symptoms.
Less than half of individuals with bulimia nervosa or binge eating disorder had ever sought treatment for their eating disorder (a measure not assessed for anorexia nervosa), although the majority of individuals with all 3 disorders had received treatment at some point for some emotional problem. This finding, coupled with the observation that physicians infrequently assess patients for binge eating (Crow et al 2004
) and often fail to recognize bulimia nervosa and binge eating disorder (Johnson et al 2001
), highlights the importance of querying patients about eating problems even when they do not include such problems among their presenting complaints.
Several findings in this study are particularly noteworthy. First, we found that anorexia nervosa displayed a significantly shorter lifetime duration and lower 12-month persistence, as well as lower overall levels of comorbidity, than either bulimia nervosa or binge eating disorder. These findings contrast with previous studies (Steinhausen 2002
) that have conceptualized anorexia nervosa as a chronic and malignant condition. This discrepancy may be due to the fact that our population-based method identified individuals with milder cases of anorexia nervosa who might have been missed in previous follow-up studies, which were based largely on clinical samples. Alternatively, our population-based method might have missed more severe cases of anorexia nervosa, either because they were unavailable, unreachable, hospitalized, or unwilling to participate in an interview about emotional problems. Parenthetically, we would note that while we found no cases of current anorexia nervosa in our study, 15.6% of the individuals with a lifetime diagnosis of anorexia nervosa still had a current BMI of less than 18.5 at the time of interview. Indeed, these individuals (3 cases) were all below 85% of ideal body weight, thus meeting our operationalization for DSM-IV criterion A for anorexia nervosa. However, all of these individuals failed to meet at least one of the other criteria for anorexia nervosa currently—although our data did not permit an analysis of which specific criteria were lacking in individual cases. Nevertheless, these data suggest that a minority of individuals with past anorexia nervosa may continue to maintain an abnormally low body weight, even though they no longer meet full criteria for anorexia nervosa.
Our findings also provide further evidence for the clinical and public health importance of binge eating disorder. In contrast to some earlier studies suggesting that binge eating disorder might be a relatively transient condition (Cachelin et al 1999
; Fairburn et al 2000
), the present findings, together with those from another recent study (Pope et al, in press
), suggest that this disorder is at least as chronic and stable as anorexia nervosa or bulimia nervosa. Binge eating disorder also appears more common than either of the other two eating disorders, exhibits substantial comorbidity with other psychiatric disorders, and is strongly associated with severe obesity. Collectively, these findings suggest that binge eating disorder represents a public health problem at least equal to that of the other 2 better-established eating disorders, adding support to the case for elevating binge eating disorder from a provisional entity to an official diagnosis in DSM-V.
Subthreshold binge eating disorder, by contrast, was found to be associated with such low impairment and comorbidity that it likely does not merit consideration for inclusion as a DSM disorder. It should be recalled, in this connection, that the main difference between subthreshold binge eating disorder and binge eating disorder is that the former lacks the criterion of distress (see in Supplement 1). These findings suggest that the criterion of distress may be important for defining clinically meaningful forms of binge eating.
Appendix table 1
Lifetime prevalence estimates of DSM-IV eating disorders and related behavior by age and sex
Note that subthreshold binge eating disorder may be defined in different ways. For example, relaxing the frequency criteria to less than the average of 2 days per week for 6 months required by DSM-IV identifies groups with characteristics similar to the full disorder (Striegel-Moore et al 2000
; Crow et al 2002
). We were unable, however, to evaluate these definitions due the nature of the CIDI questions, and instead defined subthreshold binge eating disorder by relaxing criteria other than frequency of binges. Thus, while our definition of subthreshold binge eating disorder does not appear to identify a clinically meaningful entity, other definitions may well do so.
Unlike subthreshold binge eating disorder, the entity “any binge eating” is associated with severe obesity, modest levels of impairment, and high levels of comorbidity with other mental disorders. These features appear to be accounted for cases of bulimia nervosa or binge eating disorder within the “any binge eating” group, given that such features are not shared by those with subthreshold binge eating disorder, and individuals with anorexia nervosa contribute only a small number of cases. The findings for any binge eating are interesting to consider in the light of findings from twin studies of binge eating. These studies have suggested that there are genetic influences on binge eating (Bulik et al 1998
) and on binge eating without compensatory behaviors (Reichborn-Kjennerud et al 2004a
). On the basis of our findings here, it is tempting to speculate that the heritability of binge eating behavior may be attributable primarily to cases of bulimia nervosa and binge eating disorder—both of which have been shown to be familial (Strober et al 2000
; Hudson et al 2006
)—rather than to cases of subthreshold binge eating disorder within the group.
Several limitations of the study should be considered. First, some CIDI questions did not precisely mirror the DSM-IV criteria for the various eating disorders, as illustrated by in the diagnostic algorithms discussed in our methods section. Perhaps the most important inconsistency is that, in order to have parallel duration requirements for bulimia nervosa and for binge eating disorder, we required only 3 months of illness for a diagnosis of binge eating disorder, in contrast to the 6 months required by DSM-IV. Thus, it is possible that we may have overestimated the prevalence of binge eating disorder by including some cases with a duration of only 3 to 5 months.
Second, diagnoses were based on unvalidated, fully structured lay interviews where lifetime information was assessed retrospectively. These may be important considerations, given that an earlier version of the CIDI was found to underdiagnose eating disorders (Thornton et al 1998
), possibly because some individuals minimized or denied symptoms. Version 3.0 of the CIDI was designed to reduce this sort of under-reporting by using a number of techniques developed by survey methodologists to reduce embarrassment and other psychological barriers to reporting (Kessler and Ustun 2004
)—but these changes necessitated indirect assessments of loss of control and distress, as noted above. In any event, pending validation studies, it would seem prudent to think of the NCS-R estimates as lower bounds on the true prevalence of eating disorders.
Third, in our analyses of the associations between eating disorders and body weight, we possessed only current BMI, rather than maximum or minimum adult BMI, or BMI at the time of the disorder. Thus, we likely underestimated the magnitude of these associations.
Fourth, because recall of earlier experiences may diminish with age, our retrospective assessments may have overestimated the magnitude of cohort effects (Giuffra and Risch 1994
). Since cohort effects and age effects are confounded, and no prospective studies have been performed over the period under study, it is not possible to assess the magnitude of this potential bias. Prospective studies will be useful to track possible cohort effects in the future.
Fifth, our results are based on the assumption that any exiting from the population available for sampling was non-informative and that there was no selection bias (in the form of non-response bias) due to sampling from available subjects; these limitations are discussed elsewhere (Hudson et al 2005
). For example, the validity of our results would be threatened if the development of eating disorders rendered individuals less likely to be available for sampling, which might occur if there were a high mortality due to eating disorders, or a significant proportion of cases hospitalized at the time of sampling. Although some clinical follow-up studies have suggested substantial mortality for anorexia nervosa (Sullivan 1995
; Steinhausen 2002
; Keel et al 2003
), data from a community case registry study (Iacovino 2004
) did not find excess mortality.
Another possible threat to validity would be bias in sampling of available individuals, in that individuals with eating disorders might be more or less likely to participate. However, we carried out a non-response survey to deal with this problem, which offered a larger financial incentive ($100) to main survey nonrespondents for a short (15-min) telephone interview that assessed diagnostic stem questions. Very little evidence was found that survey respondents and non-respondents differed on stem question endorsement for the NCS-R core anxiety, mood, impulsecontrol, or substance use disorders (Kessler et al 2004b
). Thus, it is likely that non-response bias for eating disorders was minimal.
Sixth, while we examined 2 provisional entities in addition to those for which criteria were provided in DSM-IV, we did not examine many other possible entities that lie within the category of Eating Disorder Not Otherwise Specified (Fairburn and Bohn, 2005
)—such as subthreshold forms of anorexia nervosa and bulimia nervosa, alternative definitions for subthreshold binge eating disorder (discussed above), purging without either bulimia nervosa or anorexia nervosa (Keel et al 2005
), and night eating syndrome (Stunkard et al 2005
)— because the questions in the CIDI did not permit evaluation of these conditions.
In conclusion, the lifetime prevalence of the individual eating disorders ranged from 0.6–4.5%; these disorders displayed substantial comorbidity with other DSM-IV disorders and were frequently associated with role impairment. These patterns raise concerns that such a low proportion of individuals with these disorders obtain treatment for their eating problems. As it turns out, though, a high proportion of cases did receive treatment for comorbid conditions. Thus, detection and treatment of eating disorders might be increased substantially if treatment providers queried patients about possible eating problems, even if the patients did not include such problems among their presenting complaints.