Introduction

Eating disorders (ED) with higher prevalence rates are anorexia nervosa (AN), bulimia nervosa (BN) and binge eating disorder (BED) according to the 5th version of the Diagnostic and Statistical Manual of Mental Disorders ([DSM-5] ^{APA, 2013}). In the last two decades studies about epidemiology on ED have increased significantly, however it is worth to point out that the three basic frequency measures in this kind of studies are incidence, prevalence and mortality. Incidence expresses the volume and acceleration of new cases -disease or disorder-over a specific population and period, usually one year (^{Striegel-Moore, Franko, & Ach, 2006}); prevalence rates refer to the number of individuals in relation to the total population that suffer a disease or disorder in a specific time (^{Moreno, López, & Hernández, 2007}); mortality rates point out the number of deaths caused by a specific disease. This measure is often used as an indicator of illness severity (^{Rothman, 2002}). All these measures yield important information that helps us to characterize ED in terms of risk, occurrence and trends over time; however this study will focus exclusively on prevalence, because this measure is essential in planning health services, designation of economical resources and administration of medical care facilities (^{Hoek & van Hoeken, 2003}; ^{Kleinbaum, Kupper, & Morgenstern, 1982}). According to epidemiological literature there are different types of prevalence, (a) point prevalence is a particular assessment in certain point in time; (b) period prevalence is the percentage of cases established within a period of time (usually 1-year period); and (c) lifetime prevalence is defined as the number of individuals that at any time have experienced a disorder (^{Hoek & van Hoeken, 2003}; ^{Hunter & Risebro, 2011}).

Lifetime prevalence rates for AN, BN and BED are 0.9%, 1.5% and 3.5% among women and 0.3%, 0.5% and 2.0% for men respectively (^{, Hudson, Hiripi, Pope, & Kessler, 2007}). Also the category called "Other Specified Eating Disorders (OSED)" included in DSM-5, which is applied when the individual does not meet the full criteria for any of the ED, has a lifetime prevalence of 4-5% (^{Le Grange, Swanson, Crow, & Merikangas, 2012}). Nevertheless it has been documented that disordered eating behaviors (DEB) are more common among community sample, such as, restrictive dieting, fasting, self-induced vomiting, abuse of laxatives and/or diuretics and binge eating (^{Garner, 2008}; ^{Tam, Ng, Man, & Young, 2007}). These behaviors are important risk factors because they have physiological complications, for example, delayed linear growth and delayed puberty (^{Daee et al., 2002}); dental erosion, mouth and esophagus ulcers and in severe cases the onset of esophagus cancer (^{Matsha et al., 2006}; ^{Mitchell, Pomeroy, & Adson, 1997}); or digestive and urinary abnormalities (^{Mitchell et al., 1997}). However the psychological consequences are as dangerous or even more than physiological complications since individuals who present them at early and late adolescence not only are more likely to develop an ED in adulthood, but also because these individuals are more susceptible to engage on depression, low self-esteem, anxiety, substance abuse or suicide attempts (^{Garner & Keiper, 2010}; ^{Kotler, Cohen, Davies, Pine, & Walsh, 2001}; ^{Nunes, Barros, Anselmo, Camey, & Mari, 2003}; ^{Preti, Rocchi, Sisti, Camboni, & Miotto, 2011}; ^{Tylka & Mezydlo, 2004}).

Based on epidemiological research, in South Australia, ^{Hay, Mond, Buttner, and Darby (2008)} assessed the prevalence of DEB with women in two moments, the first in 1995 (M = 43.4, SD = 19.2 years) and the second in 2005 (M = 45.1, SD = 24.5 years) finding a point prevalence of 3.2% and 7.5% in binge eating; 1.3% and 2.1% in purging behaviors; 2.5% and 5.2% in strict dieting, respectively. These data give evidence of an increase in prevalence of DEB over time.

A set of studies carried out by Keel and colleagues evaluated, in a longitudinal study, the point prevalence of DEB (^{Heatherton, Mahamedi, Striepe, Field, & Keel, 1997}), BN symptoms (^{Keel, Heatherton, Dorer, Joiner, & Zalta, 2006}) and BN and OSED of BN (^{Keel, Gravener, Joiner, & Haedt, 2010}). They reported that purging behaviors-defined as the use of vomiting, laxatives or diuretics to control weight-prevalence were 5.1% in 1982, 3.5% in 1992 and 4.3% in 2002, concluding that these behaviors did not change significantly across cohorts; however point prevalence in binge eating (29.2% in 1982, 20% in 1992 and 14.8% in 2002) and fasting (19.6% in 1982, 12.7% in 1992 and 11.1% in 2002) decreased significantly from 1982 to 2002 (^{Keel et al., 2006}).

When there is a marked uncertainty in a specific topic, such as the prevalence of DEB, it is recommended to carry out a systematic review since this kind of studies gather relevant, valid and reliable information selected under rigorous methodological criteria that allow to discuss inconsistencies among studies to redesign and improve future research (^{Beltrán, 2005}). In this sense, some literature reviews have tried to explain the epidemiology of ED (from 1981 to 2002; ^{Hoek & van Hoeken, 2003}), of OSED (among 1980-2003; ^{Chamay-Weber, Narring, & Michaud, 2005}), and combining ED and OSED but only with Spanish population (from 2000 to 2010; ^{Peláez, Raich, & Labrador, 2010}). Finally to our knowledge there are two extent studies that comprised ED, OSED and DEB that analyzed studies of the last three decades (^{Jacobi, Abascal, & Taylor, 2004}) and one of them only reviewed Japanese publications (^{Chisuwa & O'Dea, 2010}).

Based on these background, we can summarize that (1) DEB are more common than full entities; (2) empirical studies yield inconsistent prevalence rates and (3) literature reviews limit their analysis to specific population or full entities, therefore the aim of this paper is to systematically review empirical studies that have provided estimates of prevalence of the DEB in women, specifically on restrictive dieting, fasting, laxatives or diuretics abuse, self-induced vomiting and/or binge eating. Particular attention will be paid to methodological differences across studies since these may be linked to discrepancies in the results reported. Analysis of strengths and limitations of these studies will be followed with recommendations for future studies of DEB.

Several hypotheses emerged for this review: (a) different type of prevalence (point, period and lifetime) will yield diverse rates; (b) methodological differences will yield different rates of prevalence and (c) restrictive dieting and binge eating will be the most prevalent DEB.

Method

According to PRISMA statement (Preferred Reporting Items for Systematic reviews and Meta-Analyses; ^{Moher, Liberati, Tetzlaff, & Altman, 2009}) on February 2013, a search of articles was carried out through MEDLINE and SCIENCE DIRECT databases, using different combinations of the following key words contained in the title, abstract and/or within the article's keywords: eating disorders, eating disorders not otherwise specified (term known as OSED in DSM-5), anorexia nervosa, bulimia nervosa, prevalence and women. Considering the dates of previous reviews for the current study were eligible studies published between January 2000 and January 2013.

To choose the studies for this review, the first two authors determined the relevance and adequacy of each eligible paper according the following selection criteria:

Results

Discussion

The main objective of the present paper was to systematically review empirical studies that have provided estimates of prevalence of the DEB in women, as well as make some recommendations for future epidemiological studies of DEB.

This review found inconsistent results among studies analyzed, however this is relevant since systematic reviews provide a general view of how have been investigated the prevalence of DEB. There are several confounding factors that may be responsible for these conflicting findings. One explanation is that different types of prevalence were reported, supporting the first hypothesis of this review which was: different type of prevalence (point, period and lifetime) will yield diverse rates. It is common that prevalence rates are consider just as one general epidemiological measure, however this may have different types and therefore different objectives. In this literature review 80% of the studies reported the point prevalence of DEB, this is reasonable since in ED field the temporality and frequency in these behaviors (at least during the past three months and twice a week according with DSM-5) are crucial data to determine the clinical relevance, therefore the point prevalence considers this aspect and gives a general view about the presence or absence of these behaviors among the population, allowing health services to provide facilities more attached to the necessities of the society. Hence it is recommended that both reader and researcher identify which kind of prevalence will be studied, since the unclearness in this aspect may affect the methodology, the results and main conclusions of the research, yielding prevalence rates over or sub-estimated not only of DEB but of any disorder.

There is a considerable debate around the prevalence of DEB, since it is difficult to determine if these have rise, decrease or remained stable over time, this dispute is not only related to type of prevalence issues, but also to differences in methodologies among studies such as sample features, research design and instruments, therefore the second hypothesis that was methodological differences will yield different rates of prevalence, was accepted. For instance, large samples are important in epidemiological studies to be able to generalize findings; in this review 35% of the papers reported less than 1000 participants. Although there is not a consensus about what does it mean "large samples", ^{Jacobi, Hayward, de Zwaan, Kraemer, and Agras (2004)} have suggested for better estimations of prevalence rates, a sample size of at least 3000 subjects within a community-based study. However the size is not enough to assure the representativeness of the sample, also it is necessary to consider the method of sample selection and the response rate. The gold standard procedure in epidemiology for sample selection is the randomized methods, and ^{Punch (2003)} established that good response rates in face to face surveys goes from 80% to 85%, questionnaires sent by mail starting from 60%, by online of 30% or more and in classroom by paper starting from 50%. According to these assumptions it is worth to highlight that in this review most of studies accomplished with these two latter criteria, however these criteria must be achieved in all epidemiological researches, since these are key features that will reflect more real and certain prevalence rates among the population studied.

Regarding to research design, both cross-sectional and longitudinal studies yield relevant information, for instance, cross-sectional studies allow to predict strategies for prevention and intervention programs; however longitudinal studies may suggest a trend of these DEB and as an additional goodness, to acknowledge if the strategies carried out by health services are impacting positively over time in population. Therefore it is suggested for future investigations to consider perform longitudinal studies even the enormous time and costs that this design requires, coinciding with ^{Jacobi, Hayward, et al. (2004)} who mention that the first step in identifying risk factors should be through longitudinal studies.

The last methodological confounding factor in this review has to do with instruments, highlighting two aspects: (a) type of question and (b) answer option. The first point arises when were analyzed the different questions utilized to estimate the point prevalence of DEB. In this analysis were identified questions such as "Do you currently diet/binge/vomit twice a week...? ", "during the past six months or in the past 1 year do you... ?" and "have you ever ...?", in which was observed temporal differences thus different types of prevalence. For instance two studies carried out in the same country with similar sample sizes and similar participants' features, found prevalence rates extremely discrepant. ^{Huon et al. (2002)} assessed the prevalence rate of restrictive dieting through questions that refer to the current moment finding 0.6%, while ^{Tam et al. (2007)} evaluated the same behavior with questions that enquired participants to think in the past year, yielding 51.7% suggesting that the temporality expressed implicitly in questions have a significant influence in the prevalence rates. Moreover it is common to use screening instruments to assess prevalence rates; however it is not enough to count with a wide recognized instrument, it is necessary to carefully select items that reflect the presence of the behavior and not the attitude toward the behavior. For example, it is better to select items such as "I vomit after eating " instead of "I would like to vomit after eating ".

The second point to discuss concerning instruments is the use of Likert scales to identify the DEB, since criteria were not specified to consider presence or absence of the behaviors, which is crucial factor in epidemiological research. For example, in a Likert scale with five answer options (always, usually, sometimes, rarely and never), rigorous criteria may limit presence as "always" and absence as "never", but less exigent criteria may consider more than one answer option. This aspect was undervalued since almost any study specified meticulously how the presence or absence of DEB was established.

Data collection is one of the methodological procedures more important in any research, because this stage allows responding research questions proposed (^{Singh, 2006}), therefore, the selection of the instrument must be a decision carefully taken since this incises in the results of research. To this respect, paper-and-pencil instruments are useful tools in epidemiological research to collect large amounts of information at a low cost per respondent, notwithstanding the instrument selected need to cover two important conditions: validity and reliability in specific populations (^{Bhattacherjee, 2012}).

According to the results of this review restrictive dieting and binge eating were within the more prevalent DEB, supporting the last hypothesis formulated for this review. This finding could be influenced by the last confounding factor that has to do with cultural issues. To illustrate this, we can observe that ^{Westenhoefer (2001)} in a longitudinal study found a decrease in most of DEB, one possible explanation that he proposes is that Germany is considered a country where people is more aware about health and wellness issues; supporting this proposal, the ^{International Markets Bureau (2010)} mention that experts and media have worked together to warn population about health risks caused by the practice of DEB, making the wellbeing as a lifestyle and marketing concept. In contrast, ^{Nunes et al. (2003)} reported one of the highest prevalence rates in laxatives (8.5%) in Brazilian population, according to the authors this may be due to Brazilians women have more access to diet pills than women in other countries, these pills are each time more accessible to adolescents since medical prescription is no needed to obtain them, and the fact that media promotes them as "soft or natural weight control methods" increase the desire to consume these products. Besides, Brazil is considered one of the countries with the highest beauty standards, hence one of the countries where more cosmetic surgeries are performed per year in the world (^{International Society of Aesthetic Plastic Surgery, 2013}) this may cause a strong social pressure in women, pushing them to consume products that will help them to reach "the perfect body". Other study that reflects cultural aspects in prevalence rates is the one carried out by ^{Kiziltan et al. (2006)} in Turkish population; they reported high prevalence rates in fasting (10%) and binge eating (16%). One possible explanation is that fasting is a behavior performed by Muslims girls for religious reasons; it is known that long periods without eating may lead an increased amount of food intake and this may be misunderstood as a binge eating, yielding high prevalence rates (^{Peláez et al., 2005}). This underlines the importance to adapt culturally the instruments to detect when the practice of DEB is truly pathological and not performed for religious, cultural or health purposes.

The analysis carried out in this literature review add knowledge to understand the differences among studies about prevalence rates of DEB, however some limitations should be consider: (1) male prevalence rates were not analyzed, it is suggested that future research address this aspect since there is evidence that the practice of these behaviors are becoming more popular among adolescent boys and young men (^{Fortes et al., 2013}; ^{Petrie et al., 2008}), (2) the search review was performed only in two data bases (MEDLINE and SCIENCE DIRECT ), it is suggested that further research included information from other sources such as books, dissertations, articles in different languages or indexed in different data bases to enrich the knowledge in epidemiological field of ED, (3) excessive exercise is also a relevant DEB but was not included in this review, so further research should study it since there is evidence that it is associated with muscle dysmorphia (^{Hale et al., 2013}) and (4) although this study was design as a systematic review it is possible to also consider it as a first approximation of a meta-analysis, since according with ^{Crombie and Davies (2005)} the validity of a meta-analysis depends on the quality of the systematic review on which it is based, therefore it is suggested that future researches give continuity to this work to provide more detailed and accurate information on DEB, following criteria proposed by PRISMA statement (Preferred Reporting Items for Systematic reviews and Meta-Analyses; ^{Moher et al., 2009}).

The main strength of this paper is the meticulous analysis performed in each article; this allowed deriving the following methodological considerations for epidemiological research in ED, which will contribute to describe more accurately the real state of the population study, as well as to have a greater scientific rigor:

Sponsors

This research was sponsored by CONACyT No. 131865-H granted to Dr. Juan Manuel Mancilla-Díaz. The first author was supported by a doctoral scholarship from CONACyT Scholar No. 245864

Ethical disclosures

Conflict of interest

The authors declare no conflict of interest.