Eating disorder measures in a sample of military veterans: A focus on gender, age, and race/ethnicity

Abstract

Early detection of eating disorders (EDs) is crucial for both prevention and treatment; however, few ED measures have been validated among older adults, men, and racially/ethnically diverse individuals, who may have varying symptom presentations. We examined the psychometric quality of three self-report ED measures within a diverse sample of U.S. military veterans, a population that may have elevated rates of EDs. Participants (N = 1,187) completed the Eating Disorder Diagnostic Scale-5 (EDDS-5), the Eating Disorder Examination-Questionnaire (EDE-Q), the SCOFF (Sick, Control, One, Fat and Food) questionnaire and measures of associated mental health symptoms. We examined proportions of probable EDs and reliability estimates, associations among ED measures, and their relationship with mental health measures for the sample as a whole and based on age, gender, and race/ethnicity. Proportions of probable EDs ranged from 9.9–27.7% and were comparable for White, Black, and Latinx participants. Participants aged 40–49 had significantly higher proportions of EDs compared to other age groups, whereas participants aged ≥ 60 had significantly lower proportions of EDs. Participants with obesity had significantly higher proportions of probable EDs compared to participants with healthy weight or overweight. There was fair to moderate agreement between the ED measures, with varying evidence for psychometric quality across demographic subsamples. Overall, the EDDS-5 performed best in this sample and yielded estimates of probable EDs consistent with expectations. These data add to the growing body of literature on the assessment of EDs and provide insight into measures that may be most useful in research and intervention efforts.

Eating disorders (EDs) are complex psychiatric conditions with both medical and social-cognitive symptoms and consequences. Early detection is crucial for both prevention and treatment; however, nearly all ED measures were developed among primarily White, female adolescent and young adult samples, and few have been validated among diverse samples, including older adults, men, or racially and ethnically diverse individuals (Forbush et al., 2017) who may have varying symptom presentations. Further, many extant self-report ED measures were developed primarily based on symptoms of anorexia nervosa (AN) and bulimia nervosa (BN) and do not capture more common presentations, including binge eating disorder (BED; Garner, 2004; Garner et al., 1982; Thelen et al., 1991).

Interviews are required to make a definitive ED diagnosis. However, self-report questionnaires are often used to determine “probable” diagnoses. For many ED measures, this is done using clinical cutoffs based on the total score, or by creating algorithms that align the questionnaire items as closely as possible with Diagnostic and Statistical Manual of Mental Disorders (DSM) criteria. Although there is value in assessing transdiagnostic ED constructs rather than solely focusing on DSM criteria (Fairburn et al., 2003), use of these measures to derive probable diagnoses may be problematic when the questionnaire does not assess all DSM criteria or uses a time frame that differs from the three-month time frame used by the DSM. There are also specific challenges to assessing DSM ED criteria such as binge eating, fasting, or excessive exercise. Self-report measures likely over-estimate the prevalence of binge eating, a core symptom of both BN and BED, partly due to the difficulty of determining what exactly qualifies as an eating “binge” (Fairburn & Beglin, 1994; Grilo et al., 2001a, 2001b; Lydecker et al., 2016). Further, there is limited evidence regarding the reliability and validity of items that assess non-purging compensatory behaviors such as excessive exercise and fasting (Berg et al., 2012).

Evidence for the psychometric properties of ED measures is particularly limited among racially/ethnically diverse individuals, men, and older adults (Forbush et al., 2017). It has long been assumed that the majority of ED cases occur only in White women, in part due to the theory that White beauty standards prioritize thinness, while other cultural beauty standards seem more accepting of larger bodies (Shaw et al., 2004). However, studies of nationally representative U.S. samples have found no racial/ethnic differences in the prevalence of EDs assessed via diagnostic interviews (Hudson et al., 2007) or that Hispanic/Latinx individuals have higher rates of some EDs (Swanson et al., 2011). Because most self-report ED measures were developed in samples of White women, and few have subsequently been validated among racially and ethnically diverse samples, the extent to which they are equally applicable for diverse populations is unclear (Forbush et al., 2017).

Approximately one-fourth to one-third of ED cases are male (Hudson et al., 2007). Yet, most ED measures have been developed using female samples (Forbush et al., 2017). Although some of these ED measures have been validated subsequently among men (Forbush et al., 2017; Lavender et al., 2010), most use language to describe and assess EDs that may be more appropriate for women. For example, many measures use language that focuses on the pursuit of thinness and desiring one’s stomach or thighs to be smaller rather the desire to be larger or more muscular, which may be more common for men (Darcy, 2011; Hildebrandt & Craigen, 2016; Jones & Morgan, 2010). Thus, it is important to evaluate the psychometric characteristics of measures separately for men.

EDs also continue to be under-addressed in middle-aged and older adults, and few studies have investigated the psychometric properties of ED measures in these populations (Forbush et al., 2017). Given that the average age of onset for most EDs is in adolescence or the 20s (Hudson et al., 2007), extant ED research and measure development has primarily focused on samples of adolescent and young adults. Yet, a growing body of research suggests that EDs may also be prevalent among older adults (Mitchell & Bulik, 2014), as many EDs, even when treated, never fully remit (National Practice Guideline Number CG69, 2004). Findings from clinical samples suggest that there also may be de novo ED cases that onset in middle or later adulthood (Beck et al., 1996; Keith & Midlarsky, 2004). Very few epidemiological studies have investigated EDs in middle and later adulthood. However, one 20-year study of BN and BN-related eating disorder not otherwise specified found that 4.5% of women in their sample met criteria for an ED when participants were 40 years old, on average. Of these cases, one-third had onset when participants were ≥ 30 years old (Keel et al., 2010). Thus, there is a need for additional attention to EDs and their measurement in racially/ethnically diverse groups, men, and older populations.

In the current study, we administered the Eating Disorder Diagnostic Scale-5 (EDDS-5), the Eating Disorder Examination-Questionnaire (EDE-Q), and the SCOFF (Sick, Control, One, Fat and Food) questionnaire to a nationally representative sample of male and female U.S. military veterans. Veterans are a population that has been traditionally understudied in the field of EDs but that may have elevated rates of EDs, in part due to high rates of overweight and obesity as well as a greater prevalence of trauma exposure in this population (Bartlett & Mitchell, 2015; Cuthbert et al., 2020). Yet, there are few prevalence estimates from nationally representative samples of veterans (Huston et al., 2019; Mitchell & Wolf, 2016), and little data regarding validity of ED screening measures in this population. We chose these three measures for several reasons. First, the EDE-Q and SCOFF are very widely used and well-known in the field of EDs. However, neither instrument fully assesses DSM-5 criteria for EDs. Thus, we also included the EDDS-5, which has been less well-studied but is one of the few self-report questionnaires that assesses DSM-5 criteria for AN, BN, and BED. The sections below review the EDE-Q, EDDS, and SCOFF (see also Table 1) with a focus on their validity in diverse samples.

Table 1.

Summary of EDD-5, EDE-Q, and SCOFF

	Content	Development Sample(s)	Validation Sample(s)	Pros	Cons
EDDS-5	DSM-5 criteria for anorexia nervosa, bulimia nervosa, binge eating disorder, purging disorder, night eating syndrome, atypical anorexia nervosa, and low frequency bulimia nervosa and binge eating disorder	Women aged 13–65 (Stice et al., 2000)	Adolescent girls and young women, female undergraduate students, and adolescent girls (Stice et al., 2004)	Assesses DSM-5 criteria; time frame is past three months	Has not been validated in diverse samples; no psychometric data available for current version (EDDS-5); did not evidence measurement invariance for White and Black female undergraduates (Kelly, Mitchell, et al., 2012)
EDE-Q	Cognitions and behaviors related to bulimia nervosa and anorexia nervosa; some binge eating disorder symptoms	Community women aged 16–35; female patients with bulimia nervosa and anorexia nervosa (Fairburn & Beglin, 1994)	Undergraduate women (Luce et al., 2008); community women (Mond et al., 2006); clinical samples of men and women (Schaefer et al., 2018; Smith et al.,; Black undergraduate women (Kelly, Cotter, et al., 2012); older adults (Hilbert et al., 2012); Latinx undergraduates (McEntee et al., 2020; Serier et al., 2018)	Very widely used Norms available Has been translated into other languages Screening cutoffs available for multiple groups	Does not assess all DSM-5 criteria for anorexia nervosa, bulimia nervosa, and binge eating disorder; only asks about symptoms in past month; theoretical factor structure not supported by data
SCOFF	Five items assessing features of anorexia nervosa and bulimia nervosa	Women with anorexia nervosa or bulimia nervosa (Morgan et al., 2000)	Community sample of primarily White women (Solmi et al., 2015); Finnish young adults (Lahteenmaki et al., 2009); female primary care patients (Luck et al., 2002); female U.S. military veterans (Maguen et al., 2018); Chinese psychiatric outpatients (Liu et al., 2015)	Brief; best at detecting anorexia nervosa or bulimia nervosa among women	May be less effective at detecting BED; few studies have investigated its effectiveness among men or among older adults

Note: EDDS-5=Eating Disorder Diagnostic Scale-5, EDE-Q=Eating Disorder Examination-Questionnaire. The SCOFF acronym corresponds to the five questions included in this measure: 1) “Have you made yourself sick…” 2) Do you worry that you have lost control…” 3) Have you recently lost more than one stone (14 lbs).” 4) “Do you believe yourself to be fat.” and 5) Would you say that food.”.

EDE-Q

The EDE-Q is one of the most widely used self-report measures of ED symptoms and is based on the EDE, a semi-structured interview used to assess EDs and associated symptoms (Fairburn & Beglin, 1994). It was originally designed as an outcome measure for trials of cognitive behavior therapy for BN; thus, the items are more relevant for BN compared to other disorders, and the content reflects traditionally female body image concerns (Thomas et al., 2014). The EDE-Q assesses specific ED behaviors, including binge eating and compensatory behaviors (self-induced vomiting, laxative use, over-exercise), though it does not specifically assess fasting as a compensatory behavior or binge characteristics required for a BED diagnosis. The EDE-Q uses a 28-day timeframe rather than the three-month time frame included in the DSM-5. In addition, few studies have evaluated the psychometric properties of the item assessing excessive exercise (Berg et al., 2012). Thus, although many published articles have used the EDE-Q to generate probable ED diagnoses, the limitations of the EDE-Q preclude doing so with confidence.

EDE-Q scores appear to differentiate ED cases from non-cases and are correlated with other measures of ED symptoms, although the majority of validity studies have been conducted in samples of young, White women. Norms are available for undergraduate women (Luce et al., 2008), community samples of women up to age 42 (Mond et al., 2006), undergraduate men (Lavender et al., 2010; Schaefer et al., 2018), male and female patients with EDs (Schaefer et al., 2018; Smith et al., 2017), and Black undergraduate women (Kelly, Cotter, et al., 2012). Studies have also compared mean scores and the factor structure of the EDE-Q in samples of Latinx and non-Latinx adults (McEntee et al., 2020; Serier et al., 2018).

EDDS-5

The EDDS-5 is a self-report measure that was recently updated to address DSM-5 criteria for AN, BN, and BED and that can be used to assess purging disorder (PhenX Toolkit: Protocols). In the development and follow-up validation studies, the original EDDS, based on DSM-IV criteria, evidenced good agreement with interview diagnoses of EDs (Stice et al., 2000, 2004). There are currently no published investigations of the psychometric properties of the EDDS-5, however. No previous studies have evaluated the reliability and validity of the EDDS among samples of men. However, a study of differential item functioning in men compared to women did not find evidence of clinically significant differences for EDDS items or EDE-Q items (Schaefer et al., 2019). An investigation of the construct validity of several ED measures among White and Black undergraduate women found that factor loadings and intercepts for the EDDS differed significantly between these two groups (Kelly, Mitchell, et al., 2012).

SCOFF

The SCOFF questionnaire is a widely used, brief screening measure for EDs. A recent review found that it performs best in samples women with AN or BN (Kutz et al., 2020). Previous studies have validated the SCOFF for use in clinical and community samples (Lähteenmäki et al., 2009; Luck et al., 2002; Morgan et al., 1999). However, the SCOFF showed low sensitivity in a racially diverse sample of men and women (Solmi et al., 2015). Further, it may be less effective at detecting BED, and few studies have investigated its effectiveness among men or among older adults (Solmi et al., 2015). In a Chinese study of psychiatric outpatients, internal consistency estimates were 0.45 among women and 0.61 among men. Although the SCOFF evidenced good sensitivity and specificity, using cutoff scores of ≥ 2 among men and ≥ 3 among women, it under-performed in the subsample of men with obesity (Liu et al., 2015). A recent study investigated the SCOFF among a sample of primarily middle-aged, female VA patients. The researchers found that the SCOFF correctly identified 66.1% of participants with AN, BN, or BED and 79.9% of cases without EDs (Maguen et al., 2018). However, this study compared the SCOFF to probable EDs based on the EDE-Q, which as noted above, is a limitation.

The purpose of this study was to address limitations of the existing literature and expand knowledge about EDs among military veterans. We had two aims for this paper. The first was to report proportions of probable EDs based on the EDDS-5, EDE-Q, and SCOFF in a nationally representative sample of veterans, with a focus on similarities and differences across genders, racial/ethnic groups, body mass index (BMI) categories, and ages. Although these probable diagnoses were not definitive due to potential limitations with the measures’ applicability to different subgroups, as well as their self-report nature, these data are helpful for characterizing our sample, illustrating the usefulness and limitations of these measures, and providing a basis for comparison with other studies of EDs among veterans. We had no hypotheses regarding differences in proportions of EDs across measures. However, we expected that female veterans, younger veterans, and veterans with overweight or obesity would screen positive for EDs at the highest rates, but that there would be no differences by race or ethnicity. The second aim was to examine other psychometric characteristics of the measures in the full sample and key subgroups, including their internal consistency reliability, their associations with one another, and their relationships with associated mental health symptoms. We had no specific hypotheses regarding differences in reliability or associations with mental health symptoms. However, we expected that there would be at least moderate agreement among them, and that agreement would be higher for women relative to men, White, non-Latinx participants compared to Latinx participants and participants of other races, and younger participants compared to older participants.

Method

Participants and Procedure

We recruited participants from a nationally representative sampling frame of U.S. military veterans to investigate proportions of EDs as well as the psychometric properties of ED screening measures. Potential participants (N = 4,126) were randomly selected from the population of U.S. military veterans by the VA/DoD Identity Repository (VADIR), with the goal of recruiting a diverse sample of veterans in terms of age and gender. Women were oversampled relative to their representation in this population (1:1 ratio).

We contracted with a survey research firm to send recruitment letters and administer the survey. The study was launched in February 2019. Of the 4,126 veterans, 4,072 had locatable mailing addresses. We received completed surveys from 1,187 participants (29% response rate). This response rate is consistent with other surveys of U.S. military personnel and veterans, which often vary between 20% and 30% (Coughlin et al., 2011; Defense Manpower Data Center (DMDC), Research, Surveys, and & Statistics Center (RSSC), 2016). An additional 87 potential participants completed less than half of the survey. Notably, just over half of the sample (n = 662) completed the survey prior to March 15, 2020, when many schools and businesses began to shut down due to the COVID-19 pandemic. Proportions of probable EDs according to the SCOFF questionnaire (F = 2.15, p = 0.14), EDE-Q (F = 2.00, p = 0.16), and EDDS-5 (F = 0.09, p = 0.76) were not significantly different between participants who completed the survey prior to March 15, 2020 compared to those who completed the survey after the pandemic began.

The cover letter sent by the survey research firm described the study and provided a URL to access the online survey. Potential participants were informed that the survey would focus on eating behaviors, military experiences, risk factors, and healthcare use. A unique access code was provided to each participant in the letter; they entered the code to access the survey, and this code was used as their study ID. A study fact sheet was included in lieu of an informed consent form. An opt-out form was included for participants who did not wish to be contacted again. We included a $2 bill with the initial mailing, as providing upfront payments to all potential participants improves response rates (Singer et al., 2000). Potential participants were informed that they may keep the money regardless of their participation. Responders received an additional $20 cash. The survey took an average of 53.67 minutes to complete.

After the initial invitation letter, we sent three reminder and thank you postcards to potential participants who had not yet responded. We sent a paper survey to potential participants who had not responded to the initial invitation or first two reminders to complete the web survey, followed by the fourth and final reminder postcard. Dillman and colleagues recommend use of mixed-mode (e.g., web-based survey + paper) survey designs in order to achieve high-quality data results and reduce non-response bias (Dillman et al., 2009). The majority of participants (n = 899) completed the web version of the survey, and the remaining 288 completed the paper version. The local Institutional Review Board approved all study procedures.

Measures

The EDDS-5 is a self-report measure that can be used to determine probable ED diagnoses and also generates a symptom composite score (Stice et al., 2000). The original EDDS was based on DSM-IV criteria for AN and BN and proposed criteria for BED. As in the DSM, questions reference the past three months. The EDDS has been updated recently to reflect DSM-5 criteria for AN, BN, and BED and can also be used to assess possible purging disorder, night eating syndrome, atypical AN, and subthreshold (low frequency) BN and BED. The EDDS-5 also assesses self-reported height and weight, which were used to calculate BMI. For examination of proportions of probable EDs by BMI category, participants were grouped as having underweight (BMI < 18.5), healthy weight (BMI = 18.5–24.9), overweight (BMI = 25.0–29.9), and obesity (BMI ≥ 30).

We used the diagnostic scoring algorithms provided by the measure’s developers (PhenX Toolkit: Protocols) to assess probable AN, atypical AN, BN, BED, subthreshold BN, subthreshold BED, purging disorder, and night eating syndrome. Notably, the EDDS-5 does not specifically assess broad restriction of caloric intake related to AN and atypical AN. Rather, a low BMI is required for AN, while weight loss is required for atypical AN. As described below, the original scoring algorithm resulted in a higher prevalence of BN than BED, which is inconsistent with results from other nationally representative studies. We considered that possible over-endorsement of the fasting and excessive exercise items may be causing participants to be incorrectly diagnosed with BN instead of BED, given the lack of reliability or validity information for these types of items (Berg et al., 2012). Therefore, we also estimated probable diagnoses based on the EDDS-5 with the fasting and exercise items excluded from the algorithms for BN and BED. Notably, a skip pattern was introduced to reduce participant burden, where participants who did not endorse binge eating skipped items assessing compensatory behaviors and night eating syndrome; thus, we were unable to estimate the prevalence of probable night eating syndrome or purging disorder. We calculated the EDDS-5 composite score using standardized scores for items 1–17.

The EDE-Q is a self-report measure of ED symptoms (Fairburn & Beglin, 1994). It has four subscales: dietary restraint, eating concerns, weight concern, and shape concern and can also be used to generate a global severity score. The proposed four-factor structure has not been supported by research, as these scales were derived rationally rather than empirically (Thomas et al., 2014). For this reason, it is recommended that researchers focus on the global scale rather than the subscales (Berg et al., 2012). We calculated the continuous EDE-Q global score and also dichotomized these scores based on cutoffs of 2.8 among women and 1.68 among men (Mond et al., 2006; Mond et al., 2008; Schaefer et al., 2018).

The SCOFF questionnaire is a widely used, brief screening measure for EDs (Morgan et al., 1999). The acronym corresponds to the five questions included in this measure: 1) “Have you made yourself sick…” 2) Do you worry that you have lost control…” 3) Have you recently lost more than one stone (14 lbs)…” 4) “Do you believe yourself to be fat…” and 5) Would you say that food…”. Items are scored as yes/no. Endorsing at least two items was found to have a 100% sensitivity for AN and BN and 87.5% specificity for controls (Morgan et al., 2000). As has been recommended in studies of U.S. samples, we used a cutoff score of ≥ 2 to categorize participants as having a probable ED or not. Because of low internal consistency values (see below), and because the SCOFF typically is analyzed as a dichotomous variable, we did not include the SCOFF total score in correlation analyses.

We included measures of depression, stress, and anxiety in order to examine associations between ED measures and related mental health symptoms. Depression symptoms were assessed using the Patient Health Questionnaire-9 (PHQ-9; Kroenke et al., 2001) as well as the Depression Anxiety Stress Scales-21 (DASS-21; Antony et al., 1998) depression subscale. Anxiety and stress were assessed using the DASS-21 anxiety and stress subscales, respectively. Cronbach’s alphas for the PHQ-9 in the full sample and in each gender, racial/ethnic and age group were acceptable (range: 0.83–0.93). Ranges of Cronbach’s alpha values were 0.86–0.96 for the DASS-21 depression subscale, 0.85–0.93 for the DASS-21 stress subscale, and 0.84–0.92 for the DASS-21 anxiety subscale for all groups except the participants aged ≥ 70. Cronbach’s alpha for the DASS-21 anxiety subscale was 0.65 in this subsample.

Participants self-reported demographic information, including age, gender, race, and ethnicity.

Statistical Analyses

Prevalence estimates and chi-square tests were calculated using the “survey” package in R 4.0.0. We used a multi-step process to calculate sample weights in order to obtain more nationally representative prevalence estimates and more precise standard errors. All weighting steps were done separately by sex. The first step calculated a base weight for each sampled case, which is the reciprocal of the case’s selection probability. The second step adjusted the base weights for nonresponse by computing nonresponse adjustment factors within weighting cells. Weighting cells were defined by data variables known for both respondents and nonrespondents. Variables used to create nonresponse adjustments cells were those available for every record in VADIR, including sex, service component, personnel category, pay grade, race, ethnicity, marital status, education level, region, state, presence of address on file, and presence of address on Lexus Nexus database. Nonrespondents included veterans who refused to participate in the survey and veterans who were unable to be contacted, including those without locatable addresses, those mailed surveys and returned with postal non-deliverable status, and veterans who never returned the survey. The adjusted base weights for respondents are base weights multiplied by the corresponding adjustment factor, whereas the adjusted base weight for nonrespondents is zero.

The final step computes adjustments to ensure the weighted respondents correspond to expected proportions from the population. Ineligible records are dropped prior to this step. The raking procedure was used to adjust the weighted sample of respondents by age, race, and ethnicity based on population counts from the VADIR database, referred to as control totals. The purpose of raking to control totals from the frame is to adjust the weighted respondent sample such that the proportions of the selected variables are close to the to the proportions of the same variables in the population. The final weight used in analyses was the base weight multiplied by the nonresponse and raking adjustment factors. Because weights were developed by sex within each study population, the sum of the weighted respondents for males corresponds to the count of males in the population and the sum of the weighted respondents for females corresponds to the count of females in the population. Concatenating male and female files together properly represents the proportions of males and females in the population. Thus, the final weight is appropriate for either analyzing by sex or across sex.

Using SPSS version 26 (IBM Corp, 2019), we computed unweighted two-by-two cross-tabulations to compare probable EDs based on the EDDS-5 vs. EDE-Q, EDDS-5 vs. SCOFF, and EDE-Q vs. SCOFF. We calculated Cronbach’s alpha and Kuder-Richardson 20 values to examine internal consistency of the EDDS-5, EDE-Q, and SCOFF. We then used the DAGstat software (Mackinnon, 2000) to calculate Cohen’s kappa as a measure of agreement between each pair of measures. Kappa values from 0.01–0.20 indicate slight agreement, values from 0.21–0.40 indicate fair agreement, values from 0.41–0.60 indicate moderate agreement, values from 0.61–0.80 indicate substantial agreement, and values 0.81–1.00 indicate nearly perfect agreement. We estimated a weighted correlation for the EDDS-5 and EDE-Q to investigate their association with one another. We estimated weighted correlations between the EDDS-5 and EDE-Q, respectively, and other study measures (PHQ-9, DASS-21) to investigate their association with related mental health symptoms.

Analyses were conducted in the full sample, among subsamples identifying their gender as male or female, respectively, and among subsamples by age (19–29, 30–39, 40–49, 50–59, 60–69, and ≥ 70). Proportions of probable EDs were examined among participants categorized as having healthy weight, overweight, or obesity. As only six participants were categorized as having underweight, we excluded this subgroup from analyses based on BMI categories. We also conducted analyses in race- or ethnicity-stratified subsamples among participants identifying as White, Black, and Latinx, as these were the largest groups. The Latinx subsample included multiple races. We compared rates of probable EDs among men and women, racial/ethnic groups, and age groups using Wald chi-square tests. As the tests for the age groups were six (19–29, 30–39, 40–49, 50–59, 60–69, and ≥ 70) × two (ED or no ED), we created dummy variables for the age categories to investigate differences that were significant in the overall models. We used a Bonferroni-corrected p-value threshold of 0.008, calculated as 0.05/6, for the post-hoc tests. Similarly, we used a Bonferroni-corrected p-value of 0.05/3 for post-hoc tests comparing proportions of probable EDs across the three BMI categories. Cramer’s V was calculated as a measure of effect size.

This study was not pre-registered. Study data and analysis code are not publicly available. Analysis code is available upon request.

Results

Descriptives

Of the 1,187 survey responders, 541 identified as male, 594 identified as female, one participant identified as female-to-male transgendered, and five participants identified as other gender categories. Forty-six participants did not respond to this question. Weighted and unweighted means and standard deviations for age and BMI as well as proportions of racial/ethnic groups and education categories are presented in Table 2. The average age of the sample was in the 50s, and their average BMI fell within the overweight range; 283 were categorized as heaving healthy weight, 447 as having overweight, and 420 as having obesity.

Table 2.

Demographic characteristics of the sample

Demographics	Unweighted	Weighted
Race
White	71.2%	70.2%
American Indian or Alaska Native	1.3%	1.1%
Asian	1.3%	2.5%
Black	13.2%	14.2%
Native Hawaiian or Pacific Islander	0.4%	0.2%
Race Unknown	4.5%	4.8%
Other	3.5%	3.4%
Multiple	4.5%	3.5%
Ethnicity
Hispanic/Latinx	7.8%	9.4%
Non-Hispanic/Latinx	87.9%	86.2%
Ethnicity Unknown	4.3%	4.3%
Education
Did not graduate high school	1.4%	2.1%
High school diploma or GED	15.8%	21.1%
Some college	39.3%	40.0%
Bachelor’s degree	22.5%	19.8%
Post-baccalaureate degree	17.3%	13.2%
Unknown	3.7%	3.8%
Body mass index category
Underweight	0.5%	1.0%
Healthy weight	23.8%	18.4%
Overweight	37.7%	41.9%
Obese	35.4%	35.8%
Unknown	2.6%	2.8%
	Mean (SD)	Mean (SD)
Age	53.9 (6.0)	54.5 (5.4)
Body mass index	29.1 (13.8)	29.1 (14.7)

The racial/ethnic composition of our sample was fairly consistent with estimates from the Veteran Population Projection Model, which uses data from the 2018 U.S. Census as well as from Veterans Affairs and the Department of Defense (Office of Policy and Planning, n.d.). However, our sample had fewer white veterans than the population as a whole. A total of 800 participants identified as White, non-Hispanic/Latinx only, and 152 identified as Black, non-Hispanic/Latinx only. In addition, 92 were Latinx. Subsample sizes by age were as follows: 19–29 (n = 55), 30–39 (n = 158), 40–49 (n =181), 50–59 (n = 322), 60–69 (n =280), and ≥ 70 (n = 141).

Proportions of Probable EDs and Internal Consistency Reliability Estimates

Internal consistency reliability (Cronbach’s alpha) for the EDDS-5 composite score was 0.93 in the full sample and 0.92 in the male and female subsamples. Cronbach’s alpha for the EDDS-5 ranged from 0.92–0.93 when examined by race/ethnicity and 0.91–0.93 across age groups. Cronbach’s alpha for the EDE-Q global score was 0.95 in the full sample and among women and 0.93 among men; values ranged from 0.94–0.95 across racial/ethnic groups and from 0.90–0.98 when calculated in each age group. The Kuder-Richardson Formula 20 (KR-20) coefficient of internal consistency for the SCOFF total score was 0.53 in the full sample, 0.49 among women, and 0.56 among men. Coefficients ranged from 0.41–0.56 when calculated by race/ethnicity and 0.29–0.69 when calculated within each age group.

See Table 3 for weighted proportions of probable EDs according to the EDDS-5, EDE-Q, and SCOFF in the full sample and among the male and female subsamples, as well as means and standard deviations for each measure. Using the diagnostic algorithms provided by the developer of the EDDS-5, we obtained weighted proportions of probable EDs as follows: 4.1% for BN, 0.4% for BED, 3.1% for atypical AN, and smaller numbers of subthreshold BED (0.01%) and night eating syndrome (2.2%). No participants met criteria for probable AN or subthreshold BN. In total, 9.9% of participants, including 18.6% of women and 8.5% of men, met probable criteria for any ED according to the EDDS-5. Given that the discrepancy in the proportion of probable BN compared to probable BED was in contrast to U.S. nationally representative studies, which have consistently reported a higher prevalence of BED than BN (Hudson et al., 2007; Udo & Grilo, 2018), we re-estimated ED diagnoses with the fasting and exercise items excluded from the algorithms for BN and BED, and obtained proportions that were more consistent with previous studies: 0.8% (BN) and 2.8% (BED) in the full sample. These proportions of probable EDs were 2.3% (BN) and 5.2% (BED) among women and 0.4% (BN) and 2.2% (BED) among men. Among the probable ED cases according to the EDDS-5, 23.6% were male. With the alternate scoring algorithm, 152 participants met criteria for a probable ED, compared to 154 participants with the developer’s scoring algorithm. Thus, we created an EDDS-5 “any ED” variable, representing whether participants had any probable ED diagnosis according to this measure, based on the alternate scoring algorithm to use in subsequent analyses.

Table 3.

Proportions of probable eating disorders and mean scores on the EDDS-5, EDE-Q, and SCOFF

	EDDS-5 Scoring			Alternate EDDS-5 scoring*			EDE-Q			SCOFF
	Total	Women	Men	Total	Women	Men	Total	Women	Men	Total	Women	Men
Bulimia Nervosa	4.1%	8.1%	3.1%	0.8%	2.3%	0.4%	--	--	--	--	--	--
Binge Eating Disorder	0.4%	1.2%	0.3%	2.8%	5.2%	2.2%	--	--	--	--	--	--
Atypical Anorexia Nervosa	3.1%	6.9%	2.7%	3.3%	6.9%	2.9%
Subthreshold BN	0	0	0	0.02%	0.2%	0	--	--	--	--	--	--
Subthreshold BED	0.01%	0.6%	0	0.1%	0.9%	0	--	--	--	--	--	--
Any eating disorder	9.9%	18.6%	8.5%	9.8%	18.3%	8.5%	25.0%	32.0%	25.2%	12.2%	21.6%	11.1%
Total score mean (SD)	−1.51 (10.29)	2.24 (13.21)	−2.15 (9.45)	--	--	--	1.24 (123)	2.09 (156)	1.13 (113)	0.47 (0.89)	0.79 (104)	0.43 (0.86)

Note: EDDS-5=Eating Disorder Diagnostic Scale-5, EDE-Q=Eating Disorder Examination-Questionnaire. The SCOFF acronym corresponds to the five questions included in this measure: 1) “Have you made yourself sick…” 2) Do you worry that you have lost control…” 3) Have you recently lost more than one stone (14 lbs).” 4) “Do you believe yourself to be fat.” and 5) Would you say that food.”. Cutoffs for the EDE-Q were 2.8 for women (Mond et al., 2006; Mond et al., 2008) and 1.68 for men (Schaefer et al., 2018). The cutoff for the SCOFF was ≥ 2 items endorsed (Morgan et al., 2000). The EDDS-5 total score was calculated as a sum of z-scores for items 1–17. Percentages, means, and standard deviations were calculated using sample weights.

^*The alternate scoring algorithm for the EDDS-5 excluded the items assessing compensatory fasting and exercise from the criteria for bulimia nervosa and binge eating disorder

Using a cutoff of 2.8 among women and 1.68 among men for the EDE-Q, 27.7% of the full sample were categorized as having a probable ED, including 32.0% of women and 25.2% of men. Proportions were somewhat lower for the SCOFF: 12.3% of the full sample, and 21.6% of women and 11.1% of men, were characterized as having a probable ED. As hypothesized, women had significantly higher proportions of probable EDs according to the EDDS-5 (F = 15.86, p <0.001), EDE-Q (F = 4.89, p = 0.03), and SCOFF (F = 17.15, p < 0.001). Effect sizes were very small to small (Cramer’s V = 0.05–0.11).

Supplemental Table 1 includes proportions, means, and standard deviations for White, Black, and Latinx participants. As hypothesized, proportions of probable EDs identified by the three measures did not differ for White vs. Black participants or for Latinx vs. non-Latinx participants (all p values >0.05; full results available upon request). Notably, the proportions of probable EDs appeared higher for Latinx participants compared to non-Latinx White or Black participants. The lack of significant differences may have been due to lower power for this comparison. Cramer’s V values revealed a small effect size for the comparison of Latinx and non-Latinx participants (Cramer’s V = 0.13–0.14) and very small effect sizes comparing White to Black participants (Cramer’s V = <0.001–0.05).

Supplemental Table 2 includes proportions of probable EDs by age group, as well as means and standard deviations. Proportions of EDs for the EDDS-5 (F = 5.13, p < 0.001, Cramer’s V = 0.23), EDE-Q (F = 6.44, p < 0.001, Cramer’s V = 0.23), and SCOFF (F = 4.44, p < 0.001, Cramer’s V = 0.20) differed significantly by age. These effect sizes are medium-large. As hypothesized, participants aged 60–69 (F = 16.69, p <0.001) and ≥ 70 (F = 15.11, p < 0.001) had lower proportions of probable EDs compared to other groups according to the EDDS-5. Participants aged 40–49 had higher proportions of probable EDs according to the EDE-Q (F = 9.20, p = 0.002). Participants aged ≥ 70 had lower proportions of probable EDs according to the EDE-Q (F = 22.99, p < 0.001) and SCOFF (F = 13.10, p < 0.001), and participants aged 60–69 had lower rates of probable EDs according to the SCOFF (F = 12.36, p <0.001). Notably, consistent with hypotheses, participants aged 19–29 had the highest proportions of probable EDs; however, this difference was not statistically significant, likely due to low power, given the small subsample size of the youngest participants. Cramer’s V for tests comparing proportions of probable EDs for participants aged 19–29 to all other groups ranged from 0.06–0.12 (very small to small).

Supplemental Table 3 includes proportions of probable EDs by BMI categories. Proportions of probable EDs identified by the EDDS-5 (F = 8.09, p < 0.001, Cramer’s V = 0.20), EDE-Q (F = 26.78, p < 0.001, Cramer’s V = 0.30), and SCOFF (F = 6.09, p = 0.002, Cramer’s V = 0.16) differed significantly by BMI category. These effect sizes are generally medium-large. Post-hoc tests revealed that participants with obesity had higher proportions of probable EDs compared to participants with healthy weight on the EDDS-5 (F = 15.13, p < 0.001), EDE-Q (F = 50.53, p < 0.001), and SCOFF (F = 11.84, p < 0.001). Participants with obesity also had higher proportions of probable EDs compared to participants with overweight on the EDDS-5 (F = 11.26, p < 0.001), EDE-Q (F = 19.43, p < 0.001), and SCOFF (F = 6.13, p = 0.01). Proportions of probable EDs did not differ for participants with overweight compared to participants with healthy weight.

Agreement among ED Measures and with Mental Health Measures

We computed kappas to investigate the level of agreement among the three measures (see Table 4). Consistent with our hypotheses, there was “fair” agreement between the EDDS-5 and EDE-Q in the full sample and in the male subsample; there was “moderate” agreement between these two measures in the female subsample. We found “moderate” agreement between the EDDS-5 and SCOFF in the full sample and in both subsamples. Finally, agreement between the EDE-Q and SCOFF was “moderate” in the full sample and among women, but “fair” among men. Contrary to hypotheses, results were similar across race/ethnicity and age groups, with agreement among measures ranging from “fair” to “moderate” (see Supplemental Tables 4–5). There was one exception to this: agreement between the EDDS-5 and SCOFF was “substantial” for participants aged 19–29.

Table 4.

Agreement for probable ED diagnoses according to the EDDS-5, EDE-Q, and SCOFF

EDDS-5 and EDE-Q
	Total (95% CI)	Women (95% CI)	Men (95% CI)
Kappa	.39 (.33–.45) “fair”	.43 (.36– 51) “moderate”	.31 (.22–.39) “fair”
Positive agreement	.50 (.44–.55)	.57 (.50–.64)	.39 (.30–.48)
Negative agreement	.86 (.84–.88)	.85 (.82–.88)	.87 (.85–.90)
EDDS-5 and SCOFF
Kappa	.46 (.39–.53) “moderate”	.45 (.36–.54) “moderate”	.43 (.30–.56) “moderate”
Positive agreement	.54 (.47–.60)	.56 (.48–.64)	.48 (.35–.60)
Negative agreement	.92 (.91–93)	.89 (.87–.91)	.95 (.94–.96)
EDE-Q and SCOFF
Kappa	.47 (.41– 53) “moderate”	.54 (.47–.62) “moderate”	.35 (.26–.44) “fair”
Positive agreement	.58 (.53–.63)	.66 (.60–.73)	.44 (.35– 53)
Negative agreement	.87 (.86–.89)	.87 (.85–.90)	.88 (.85–.90)

Note: EDDS-5=Eating Disorder Diagnostic Scale-5, EDE-Q=Eating Disorder Examination- Questionnaire. The SCOFF acronym corresponds to the five questions included in this measure: 1) “Have you made yourself sick…” 2) Do you worry that you have lost control…” 3) Have you recently lost more than one stone (14 lbs).” 4) “Do you believe yourself to be fat.” and 5) Would you say that food.”. We used an alternate scoring algorithm for the EDDS-5 which excluded the items assessing compensatory fasting and exercise from the algorithms from bulimia nervosa and binge eating disorder. Cutoffs for the EDE-Q were 2.8 for women (Mond et al., 2006) and 1.68 for men (Schaefer et al., 2018). The cutoff for the SCOFF was ≥ 2 items endorsed (Morgan et al., 2000).

See Table 5 for weighted correlations among study measures in the full sample. The EDDS-5 and EDE-Q global scores were correlated at 0.68 in the full sample and among men and at 0.64 among women. Correlations between the EDDS-5 and EDE-Q were 0.67 among White participants, 0.62 among Black participants, and 0.71 among Latinx participants. Correlations between the ED measures and depression, anxiety, and stress symptoms ranged from 0.31–0.42 in the full sample, 0.39–0.49 among women, and 0.29–0.43 among men, suggesting that these measures are distinct from what is assessed in general mental health symptom measures. This pattern of correlations was similar among White participants (range: 0.31–0.47) and somewhat lower among Black participants (0.16–0.38). The ED measures were more highly correlated with other measures, particularly the PHQ-9, in Latinx participants (range: 0.39–0.64).

Table 5.

Correlations among continuous study variables in the full sample

	EDDS-5	EDE-Q	PHQ-9	DASS_21 depression	DASS_21 anxiety	DASS-21 stress
EDDS-5	1.0
EDE-Q	0.68	1.0
PHQ-9	0.42	0.39	1.0
DASS-21 depression	0.40	0.36	0.86	1.0
DASS-21 anxiety	0.38	0.31	0.74	0.76	1.0
DASS-21 stress	0.42	0.39	0.77	0.83	0.82	1.0

Note: EDDS-5=Eating Disorder Diagnostic Scale-5, EDE-Q=Eating Disorder Examination-Questionnaire; PHQ-9=Patient Health Questionnaire-9, DASS-21=Depression Anxiety Stress Scales-21.

Correlations between the EDDS-5 and EDE-Q by age group were as follows: r = 0.45 (19–29), r = 0.72 (30–39), r = 0.69 (40=49), r = 0.74 (50–59), r = 0.59 (60–69), and r = 0.51 (≥ 70). Correlations between the EDDS-5 and depression, anxiety, and stress symptoms varied by age: r = 0.40–0.66 (19–29), r = 0.14–0.29 (30–39), r = 0.36–0.53 (40–49), r = 0.39–0.49 (50–59), r = 0.14–0.24 (60–69), and r = 0.37–0.48 (≥ 70). Finally, correlations between the EDE-Q and depression, anxiety, and stress measures by age were as follows: r = 0.08–0.45 (19–29), r = 0.18–0.36 (30–39), r = 0.35–0.44 (40–49), r = 0.47–0.57 (50–59), r = 0.19–0.31 (60–69), and r = 0.12–0.31 (≥ 70).

Discussion

We examined proportions of probable EDs according to three widely used self-report measures of EDs, the EDDS-5, EDE-Q, and SCOFF, as well as the psychometric quality of these measures, within a nationally representative sample of U.S. military veterans. Proportions of probable EDs were relatively high among both male and female veterans. These estimates varied depending on the measure used, with proportions of any probable ED lowest based on the EDDS-5 (9.8%) and highest based on the EDE-Q (27.7%). Nonetheless, our estimates are fairly consistent with those reported in previous studies. Probable diagnoses based on self-report measures in samples of service members and veterans have ranged from 5–36% among women and 4–19% among men (Huston et al., 2018, 2019; Jacobson et al., 2009; Lauder et al., 1999; Masheb et al., in press; McNulty, 1997; Mitchell & Wolf, 2016; Warner et al., 2007).

Notably, although these proportions underscore the relevance of EDs and ED symptoms to veterans, they are likely somewhat inflated due to the low specificity of self-report screening measures. This may be due, in part, to the relative brevity of self-report questionnaires. In comparison, interview measures of EDs among non-nationally representative samples of veterans have reported lower estimates of EDs among women (4–5%) and among men (1–4%; Curry et al., 2014; Litwack et al., 2014). These findings are consistent with previous studies comparing prevalence estimates of EDs from interview vs. self-report measures, the majority of which have focused on the EDE and EDE-Q (e.g., Grilo et al., 2001a, 2001b; Lydecker et al., 2016). Specifically, a meta-analysis of studies comparing the EDE and EDE-Q found that participants consistently achieve higher subscale scores when they complete self-report measures compared to interviews. Further, scores on the two measures for the vomiting and laxative abuse items were highly correlated, but the items assessing binge eating were only correlated at 0.37–0.55 (Berg et al., 2011). Participants may over-report their symptoms on questionnaires, particularly if they have high levels of distress in general. Alternatively, participants may under-report their symptoms during interviews if they are ashamed or if they do not view their symptoms as problematic. Thus, although our results suggest that proportions of EDs in our sample are fairly high, it is possible that some participants over-reported their symptoms.

It is noteworthy that 8.5–25.2% of men in our sample screened positive for a probable ED. Consistent with U.S. population-based studies (Hudson et al., 2007), nearly one-fourth of probable ED cases (identified by the EDDS-5 in our study) were male. Stereotypes regarding EDs as women’s issues may compound the stigma of seeking treatment for an ED for men. Men are more likely to be misdiagnosed, and they are less likely to be referred for treatment by providers (Jones & Morgan, 2010; Strother et al., 2012). Similarly, proportions of probable EDs were comparable across groups characterized by race/ethnicity. Further, we found high proportions of probable EDs among participants aged 40–49, highlighting the potential vulnerability of middle-aged adults to EDs. Several factors may contribute to risk for EDs in this age group, including the effects of aging on one’s body in a society that places high value on a youthful appearance, as well as life stressors that may be more common in mid-life such as the “empty nest” (Mitchell & Bulik, 2014). More research is needed regarding differential risk for EDs across the lifecourse. Our results add to a growing body of literature suggesting that diverse groups are impacted by EDs, likely at higher rates than previously believed, and reinforce the importance of assessing EDs in military veterans, men, older adults, and racially/ethnically diverse men and women.

Our findings also suggest that there may need to be scoring modifications when these measures are used in military veterans. Specifically, it seems likely that EDE-Q global score thresholds of 2.8 for women and 1.68 for men may not be sensitive enough in this population. These findings are somewhat surprising, as younger adults, on which these cutoffs are based, tend to have higher rates of EDs (Hudson et al., 2007). However, there are several potential explanations. Of relevance is the finding that the incidence of EDs in the military has increased in recent years (Antczak & Brininger, 2008), suggesting that rates of EDs may be increasing among veterans as well. Veterans typically have high rates of trauma and psychiatric comorbidity, which also may contribute to elevated rates of EDs in this population (Hoerster et al., 2015; Maguen et al., 2012). Veterans also tend to have high rates of overweight and obesity (Almond et al., 2008; Koepsell et al., 2009, 2012), which are associated with EDs, particularly BED (Striegel et al., 2012). The average BMI in our sample fell within the overweight range, and participants with obesity had the highest proportion of probable EDs compared to participants with healthy weight or overweight. Few previous epidemiological studies have reported rates of EDs by BMI category (Duncan et al., 2017). Our results underscore the importance of screening for EDs among individuals with larger body sizes, who may be less likely to be diagnosed with an ED by a healthcare provider, despite high rates of disordered eating behaviors in this population (Nagata et al., 2018).

It should be noted that we could not assess purging disorder or night eating syndrome with the EDDS-5, due to the skip-out pattern, which could partly contribute to the somewhat lower proportions of probable EDs based on this measure. The EDDS-5 also does not assess general caloric restriction related to AN and atypical AN. Although the relatively lower number of probable ED cases identified by the EDDS-5 gives us more confidence that this measure will result in fewer false positives, the likely over-estimation of BN cases by the original scoring algorithm is a concern. Notably, ours is not the only study to report similar findings using the EDDS. For example, in a sample of Amazon MTurk workers, using the original EDDS but with DSM-5 scoring criteria, 7.3% of the sample met probable criteria for BED, while 15% met probable criteria for BN (Pollert et al., 2016). We hypothesize that these findings may be due to limitations of the items assessing compensatory behaviors, specifically fasting and excessive exercise. As noted above, there has been very little validation of items assessing these two behaviors (Berg et al., 2012). The EDDS items may not adequately assess truly excessive exercise or fasting. Although the fasting item asks how many times participants skipped at least two meals, this item does not distinguish between the pathological aspects of fasting as an ED behavior vs. intermittent fasting, a popular approach for weight management that is not necessarily pathological (de Cabo & Mattson, 2019). In a nationally representative sample of U.S. adults, 43% reporting dieting in the past year; of these individuals, 10% reported intermittent fasting, which was the most popular diet (2020 Food & Health Survey: International Food Information Council, 2020). There also is evidence suggesting that people may over-report exercise on questionnaires (Brenner & DeLamater, 2014). The EDDS item asks how many times participants “engaged in more intense exercise specifically to counteract the effects of overeating” but does not give examples of “more intense” exercise or provide specific benchmarks, e.g., duration. Thus, optimizing assessment of compensatory fasting and exercise is an important area of future research.

The EDDS-5 and EDE-Q evidenced strong internal consistency in the full sample and subsamples categorized by gender, race/ethnicity (White non-Latinx, Black non-Latinx, and Latinx), and age. The EDDS-5 and EDE-Q were strongly related with each other among most groups, although correlations were lower among participants aged 19–29 and ≥ 70. In addition, while both were correlated with measures of mental health symptoms, they were not correlated at levels that would suggest they are not distinct from these measures. However, it should be noted that although some of the correlations between these two measures and measures of depression, in particular, were nearly as high as the correlation between the EDDS-5 and EDE-Q in some groups. There was only fair to moderate agreement for probable EDs among the three measures. This is likely due to the different time frames assessed by each measure (past three months, past month, and unspecified) and also that the EDE-Q and SCOFF focus primarily on AN and BN symptoms, while the EDDS-5 includes BED.

Each of these measures has its own limitations; however, we found that the measures performed comparably across demographic subsamples, in general. Further, the EDDS-5 and EDE-Q evidenced similar patterns of correlations with other mental health symptoms. Because the EDDS-5 assesses DSM-5 criteria and yielded the lowest proportions of probable EDs of the three measures, which were most consistent with our expectations, we generally would recommend this questionnaire for assessing EDs in veteran samples. However, future validation work is needed to investigate the EDDS-5 diagnostic algorithms, and to determine what modifications may be necessary. The EDE-Q identified a very high proportion of participants as having probable EDs; we would not recommend use of this measure unless the goal was to capture everyone with a possible ED without regard to false positives. Further, the timeframe of the EDE-Q and lack of assessment of all DSM-5 criteria also preclude its use for deriving probable diagnoses. This questionnaire is likely more useful for examining specific symptoms or general severity of disordered eating. The brevity of the SCOFF questionnaire is attractive; however, the use of this measure for capturing BED is limited (Solmi et al., 2015), and this measure evidenced very low internal consistency in our study. For these reasons, we likely would not recommend use of the SCOFF.

Our findings should be considered in light of several limitations. First and foremost is the use of self-report data, which, as noted above, cannot be used to definitively diagnose EDs. Without interview-based diagnoses, we were unable to effectively investigate the sensitivity and specificity of the EDDS-5, EDE-Q, or SCOFF in our sample. In addition, it is possible that selection bias impacted our results, if potential participants with EDs were more likely to complete the study. However, an earlier study found no differences in eating symptoms between participants of a study advertised as “Disordered Eating in Young Women” and those of a study titled “Consumer Preferences” (Moss & von Ranson, 2006). Although proportions of probable EDs appeared higher among Latinx compared to non-Latinx participants, these differences were not statistically significant, possibly due to the relatively smaller sample size of Latinx participants. Further research is needed to investigate these measures in large samples of diverse groups, ideally using both interview and self-report measures, particularly in terms of determining whether they adequately capture diverse symptom presentations in men, older adults, and people of color.

Despite these limitations, our use of a nationally representative sample of male and female veterans was an important strength, as very few investigations of EDs have been conducted in such samples. Our findings suggest that veterans may have relatively high rates of EDs and underscore the importance of future research in this population. Further, our results suggest that several widely-used measures of EDs perform well in this population, although more work is needed to investigate their psychometric properties, particularly for identifying probable ED cases. Veterans Healthcare Administration (VHA) has recently begun offering ED treatment in VHA medical centers around the U.S.; however, EDs are not routinely screened for in VHA settings, partly due the lack of quality data on EDs and their measurement. Assessing EDs in populations that may have elevated rates of EDs, but which may go undetected if they are not screened for, is an important step toward increasing the reach of ED treatment.

Public Significance Statement:

Many eating disorder measures have not been tested in diverse groups, including U.S. military veterans, who are affected by eating disorders but are often not included in eating disorders research or treatment. We found that many different demographic groups of veterans were impacted by eating disorders, and there were some important differences in the performance of three different eating disorder measures in this sample.