Search strategy

The literature was reviewed in February 2017, using the PubMed search below. The screening process was conducted according to the criteria outlined by the PRISMA guidelines, and is presented in Fig. 1.

(((((((anorexia[All Fields] AND nervosa[All Fields]) AND DSM-5[All Fields] AND prevalence[All Fields]) OR ((bulimia[All Fields] AND nervosa[All Fields]) AND DSM-5[All Fields] AND prevalence[All Fields])) OR ((binge[All Fields] AND eating[All Fields] AND disorder[All Fields]) AND DSM-5[All Fields] AND prevalence[All Fields])) OR ((eating[All Fields] AND disorder[All Fields]) AND DSM-5[All Fields] AND prevalence[All Fields])) OR (OSFED[All Fields] AND DSM-5[All Fields] AND prevalence[All Fields])) OR (PICA[All Fields] AND DSM-5[All Fields] AND prevalence[All Fields])) OR (avoidant[All Fields] AND restrictive[All Fields] AND food[All Fields] AND intake[All Fields] AND disorder[All Fields]) AND DSM-5[All Fields] AND prevalence[All Fields]) OR (rumination[All Fields] AND disorder[All Fields]) AND DSM-5[All Fields] AND prevalence[All Fields])

Eligibility criteria

Exclusion criteria

All papers investigating ED prevalence in clinical samples were excluded. Editorials, commentaries, [systematic] reviews, as well as articles reporting prevalence of disordered eating without utilizing DSM-5 ED diagnoses (e.g. ED features or syndromes) were omitted. Exclusions were tracked and registered (see Fig. 1). Omitted review articles were screened for additional references. No exclusions for age, gender or geographic areas were applied.

Title, abstract and article selection

Publication titles in studies obtained through the search strategy outlined earlier were reviewed by the authors CLD and LW. All titles clearly irrelevant to the aim and scope of the paper (e.g. “Food addiction – substance use disorder or behavioral addiction”) were excluded during this initial screening. Publications with ambiguous titles (e.g. “Eating disorders in older women”) and publication with titles clearly relevant to the aim of the paper (e.g. “Prevalence and severity of DSM-5 eating disorders in a community cohort of adolescents”) were included at this stage of the screening process. Abstracts were then screened by CLD and LW using a similar strategy (i.e. screening for irrelevance, ambiguity and certainty), after which full texts were retrieved and assessed for eligibility by CLD and LW using the outlined eligibility and exclusion criteria. Agreement was reached for all included and excluded publications. None of the reviewers were blind to the authors of the texts, nor their affiliations.

Search summary

The PubMed search strategy produced 94 titles, which after the initial title screen was reduced to 43. Of these 43, 21 were omitted in the abstract screening process. A total of 22 papers met eligibility criteria and were retained. Subsequently, three full-text studies were excluded based on lack of relevance, leading to a total inclusion of 19 studies. No additional studies were found through the screening of review articles. A more detailed overview of the screening process is presented in Fig. 1 .

The majority of included studies presented data from US samples (N = 5), closely followed by Australia (N = 3), Germany (N = 2), Finland (N = 2) and Canada (N = 2). Portugal, Sweden, United Kingdom, Switzerland and the Netherlands all represented one individual study. Ages ranged from 11 to 96, with eight studies reporting data from adults only (>18), nine studies reporting rates from mixed age samples (i.e. children, adolescents and adults) and two studies presented data collected in samples below the age of 15. Sample sizes ranged from 496 to 22,397, and approximately 25% (N = 5) of the included studies followed a two-stage design approach in estimating ED prevalence. Six studies used interviews alone to establish ED diagnoses, and seven studies employed self-reports. Trace et al. [6] did not specify the design of the study. Disregarding prevalence reports using data from overlapping samples [7, 10, 13, 14], all studies employing 2-stage designs had used different screening instruments.

Prevalence rates

In 2-stage design studies, lifetime AN prevalence rates in females ranged from 1.7% [15] to 3.6% [9], and point prevalence ranged from 0.67% [9] to 1.2% [15]. Only one study reported male AN prevalence and they found both point and lifetime to be 0.1% [15]. BN was most commonly assessed using point prevalence, but was only assessed in two out the five 2-stage design studies [9, 15]. These studies reported nearly identical point BN prevalence rates of 0.59% and 0.6% respectively in females. Point prevalence rates for BED were reported in three studies, and ranged from 0.62% [9] (females only) to 3.6% [16] (males and females combined). Lifetime prevalence rates for OSFED ranged from 0.3% in males [15] to 0.6% in females [10] and point prevalence from 0.0% in males [15] to 2.4% in a sample combining male and females rates. Lifetime prevalence rates of UFED in females were 0.2% [15] and 0.9% [10] respectively.

In interview based studies, lifetime prevalence of AN in females ranged from 0.8% [17] to 1.9% [4]. Lifetime BN prevalence in females of was only reported in one of these studies, with a rate of 2.6% [17]. In two out of the six interview-based studies, lifetime rates of BED were reported. BED in females were 3.0% [17] and 3.6% [18] respectively, whereas the corresponding rates for males were only assessed in one study [18] reporting a rate of 2.1%. In females prevalence of OSFED was reported in the interview based prevalence studies, with overall OSFED point prevalence rates ranging from 5% [19] to lifetime prevalence rate of 11% [17].

Point prevalence was most commonly reported in studies using the self-report design. Here, AN rates ranged from 0.06% [14] to 1.2% in females [15]. BN prevalence rates ranged from 0.45% [11] to 8.7% in females [20]. BED rates ranged from 0.0% (in males) [20] to 4.1% (in females). Two studies [21, 22] set out to investigate the occurrence of night eating syndrome (NES). In these studies, point prevalence was established using the Night Eating Questionnaire (NEQ) [23] and a brief version of the Eating Disorder Examination Questionnaire, the EDE-Q8 [24]. Rates reported for both genders were 1.1% and 4.2% respectively.

Impact of the DSM-5 on prevalence estimates

It is of interest to review whether the transition from DSM-IV to DSM-5 has yielded the intended alterations in reported prevalence, i.e. increased prevalence of AN and BN, and decreased prevalence of the residual category EDNOS/OSFED. Several of the included studies in the present review have estimated prevalence of EDs as defined by both DSM-IV and DSM-5 in the same sample. Such recoding of diagnostic categories according to DSM-IV versus DSM-5 informs us on the impact of the DSM-5 on different ED prevalence rates. One study [7] assessed prevalence of AN by first adopting DSM-IV criteria. Subsequent to recoding according to the DSM-5 criteria, the authors observed a 60% increase in lifetime prevalence of AN among the 2825 female participants (mean age 24 years), from 2.2 to 3.6%. Another study recoding from DSM-IV to DSM-5 diagnoses investigated 1383 males (49%) and females at ages 14, 17, and 20 [20]. Significantly greater ED prevalence rates were reported when using DSM-5 criteria at all ages for females, and at age 17 only for males. These findings are in line with the study of Flament et al. [13] reporting that prevalence of full-threshold EDs increased from 1.8 to 3.7% after recoding from DSM-IV to DSM-5 criteria. Decreased prevalence of the residual ED categories (EDNOS/OSFED) when adopting DSM-5 criteria have also been reported across studies having recoded ED diagnoses [9, 13, 20]. It is worth noting that although research recoding ED prevalence from DSM-IV to DSM-5 has been important in the initial evaluation of the proposed DSM-5 criteria, these studies may have missed patients who initially did not qualify for an ED diagnoses based on DSM-IV, but who would have met criteria for a DSM-5 ED. This may imply a risk of inaccurate prevalence estimates in these studies.

A main intention of the DSM-IV revisions was to minimize the use of catch-all diagnoses such as Eating Disorders Not Otherwise Specified (EDNOS), the most frequently, reported ED diagnosis in the DSM-IV. However, the revisions made may not only contribute to altered rates of AN, BN and OSFED diagnoses, but may also increase the likelihood of reaching the threshold for a formal ED diagnosis. As the new OSFED category includes disorders which lack strict diagnostic criteria (e.g. PD and NES which lacks frequency criteria), clinicians and researchers should be vigilant with regards to the characteristics, especially significant distress and impairments, separating eating disturbances from eating disorders.

Methodological aspects

Methodological aspects are likely to influence both lifetime- and point prevalence rates including assessment measures adopted and samples investigated [3]. The majority of studies in this review adopted self-report to assign ED diagnoses, followed by interviews and 2-stage designs. Although self-report assessments have its obvious advantages in terms of being cost- and time effective, only diagnostic interviews can help determine the presence (or absence) of a formal ED diagnosis as defined by the DSM. Methodological issues such as these should be considered when interpreting prevalence rates across studies. In addition to assessment measure per se, recruitment strategy and design are important aspects to consider when evaluating quality of studies and consequently reliability of results. The 2-stage design, including stage one with screening followed by stage two with clinical diagnostic interview, have been considered the preferred approach to estimate prevalence rates. In the current study, aiming to review the prevalence of diagnosable EDs as defined by the diagnostic manual DSM-5, it may therefore be timely to consider the included 2-stage design studies [7, 9, 10, 15, 16] to be among the highest ranked studies in terms of quality. However, another marker of quality in epidemiological studies is sample size. The reviewed studies with the largest sample sizes include samples of N = 6041 [11], 13,295 [6], 10,038 [4], and 22,397 [27]. It is worth noting that none of these latter studies are among the mentioned 2-stage design studies, demonstrating the necessity to consider multiple methodological aspects when evaluating quality. In addition to the above-mentioned aspects, sample characteristics are likely to influence reported prevalence rates, and includes both age ranges and gender distributions. One of the intentions of the DSM-5 was to better capture EDs in males than during the DSM-IV era, and was a central rationale for removing the amenorrhea criteria for AN. However, as the DSM-5 revisions have contributed to higher prevalence of full-threshold AN and BN in general, it is difficult to detect whether the DSM-5 has led to better detection of male EDs per se. Also, although the amenorrhea criterion has been removed, the core ED psychopathology outlined in the DSM-5 is still biased towards females as it focuses on drive for thinness rather than muscularity. Furthermore, assessment measures used to detect ED psychopathology are often gender biased in that they have been developed to capture “female” psychopathology and symptoms, and also, as they most commonly are validated using female samples [28, 29]. Issues such as these may complicate the detection and description of ED pathology in males. Another methodological issue subsequent to the introduction of DSM-5 is the removal of a specific weight threshold for AN. This has clear advantages, maybe in particular in terms of individual and flexible evaluations in clinical settings, but it is worth noting that in research, the lack of an explicit weight criterion may lead to larger weight variations compared to earlier. In general, it is important to strive for consistent use of DSM-5 categories, which will aid the interpretation of prevalence rates across studies.

New diagnostic categories in the DSM-5 include ARFID (problem with eating not related to weight or shape concerns, leading to inability to take in adequate nutrition), pica (recurrent consumption of “nonnutritive, nonfood” items), and rumination disorder (RD; i.e. recurrent, effortless regurgitation of food). Although none of the included studies investigated these full-threshold diagnoses, three studies which were excluded due to their sub-threshold nature, addressed features of ARFID [30], pica and RD [31], reporting frequency numbers ranging from 0 to 3.2%. More research is needed to determine the prevalence of these diagnostic categories.

Finally, this review reports the prevalence of eating disorders according DSM-5 criteria. In a historical perspective there have been important changes to the diagnostic criteria since AN and BN were introduced in the DSM-III [32] in 1980. For example, in the DSM-III, the weight loss criteria for AN was “25% below original body weight”, which was then revised to “body weight less than 85% of that expected” in the DSM-IV [1], and subsequently, redefined in the DSM-5 [2] to “significantly low body weight”. For BN there were no criteria for frequency of binge eating and compensatory behavior in the DSM-III. With the DSM-III-R update, the frequency was specified to twice a week, and in DSM-5, it was reduced to once a week. These revisions will greatly influence prevalence rates in the years to come, and are crucial to address when comparing DSM-5 based prevalence to earlier epidemiological studies in EDs.

Strengths and limitations

This is the first systematic review of DSM-5 prevalence studies in EDs, and offers the reader an early snapshot of the extant prevalence literature. The core strength of the study is the thoroughness of the systematic literature review, and the detailed screening process conducted by two of the authors. Conversely, only one database was used to search the literature and only articles that were written in English (or had an available published English translation) were reviewed representing a limitation of the current study. Further, a meta-analysis was not performed which also represents a potential weakness. In addition, our search date (February 2017) dates our paper. In this interim, three publications relevant to the scope of our paper have been published. Hay et al. [33] investigated the prevalence and burden of ARFID and other DSM-5 EDs in an Australian population, Ernst et al. [34] explored how the DSM-5 revisions affected the prevalence, sex ratio and diagnostic distribution of EDNOS/OSFED in a student sample, and Micali et al. [35] investigated lifetime and 12-month prevalence of EDs amongst women in mid-life. Their relevance to the extant literature warrant a more detailed account in future epidemiological ED studies.