Sexual and gender minority individuals report higher rates of abuse and more severe eating disorder symptoms than cisgender heterosexual individuals at admission to eating disorder treatment.

Abstract Eating disorders (EDs) occur at higher rates among sexual/gender minorities (SGMs). We currently know little about the risk factor profile of SGMs entering ED specialty care. Objective To (a) compare history of abuse‐related risk in SGMs to cisgender heterosexuals (CHs) when entering treatment, (b) determine if SGMs enter and exit treatment with more severe ED symptoms than CHs, and (c) determine if SGMs have different rates of improvement in ED symptoms during treatment compared to CHs. Method We analyzed data from 2,818 individuals treated at a large, US‐based, ED center, 471 (17%) of whom identified as SGM. Objective 1 was tested using logistic regression and Objectives 2 and 3 used mixed‐effects models. Results SGMs had higher prevalence of sexual abuse (OR = 2.10, 95% CI = 1.71, 2.58), other trauma (e.g., verbal/physical/emotional abuse; OR = 2.07, 95% CI = 1.68, 2.54), and bullying (OR = 2.13, 95% CI = 1.73, 2.62) histories. SGMs had higher global EDE‐Q scores than CHs at admission (γ = 0.42, SE = 0.08, p < .001) but improved faster early in treatment (γ = 0.316, SE = 0.12, p = .008). By discharge, EDE‐Q scores did not differ between SGMs and CHs. Discussion Our main hypothesis of greater abuse histories among SGMs was supported and could be one explanation of their more severe ED symptoms at treatment admission compared to CHs. In addition, elevated symptom severity in SGMs at admission coincides with greater delay between ED onset and treatment initiation among SGMs—possibly a consequence of difficulties with ED recognition in SGMs by healthcare providers. We recommend increased training for providers on identifying EDs in SGMs to reduce barriers to early intervention.

Minority Stress Theory posits that stigmatization and social exclusion are ongoing chronic stressors contributing to dysregulation of multiple organ systems in the body and ultimately causing the higher rates of chronic diseases and poorer health outcomes found in marginalized and oppressed populations (e.g., Hatzenbuehler & McLaughlin, 2014;Stuber, Meyer, & Link, 2008). The experience of persistent discrimination and micro-aggressions often results in a vulnerability that places individuals at higher risk for victimization in the forms of bullying, abuse (sexual, physical, and emotional), and other forms of violence (e.g., Balsam, Rothblum, & Beauchaine, 2005;Corliss, Cochran, & Mays, 2002;Kann et al., 2011). The minority stress literature has led to the conceptualization of stigma as a "fundamental cause" of inequalities in population health (Hatzenbuehler, Phelan, & Link, 2013). Furthermore, in their systematic review, Alencar Albuquerque et al. (2016) theorized that SGMs are less likely to seek treatment in part due to internalized anti-SGM bias and shame. For these reasons, investigations of treatment response in oppressed and marginalized groups-such as SGMs-must be a high priority for public health research.
Despite recent evidence showing that SGM individuals have higher rates of lifetime ED diagnoses as well as unhealthy weight control behaviors than their CH peers (Kamody, Grilo, & Udo, 2019;Meneguzzo et al., 2018), risk factors and pathways leading to the disorder are often different for these groups (Duffy, Henkel, & Earnshaw, 2016;Engeln-Maddox, Miller, & Doyle, 2011;Wang & Borders, 2017;Watson, Grotewiel, Farrell, Marshik, & Schneider, 2015). For instance, research has shown gender dysphoria contributes to EDs in transgender individuals (Duffy et al., 2016). In sexual minority men, Wiseman and Moradi (2010) found experiences of sexual objectification and childhood homophobic bullying were associated with disordered eating attitudes and behaviors. Similarly, Feldman and Meyer (2007) identified higher rates of EDs in SGM men who had experiences of childhood abuse compared to those without abuse histories. Thus, it is important to elucidate whether trauma-related risk factors (such as having experienced bullying and/or sexual, physical, or emotional abuse) are more prevalent in SGM compared to CH individuals who present for higher levels-of-care at an ED treatment center. A better understanding will aid in developing and tailoring SGM-specific screening tools and interventions, which, to our knowledge, are lacking.
In accordance with Minority Stress Theory, this study aims to determine if, compared to CH patients: (a) SGM patients have a greater prevalence of abuse history (including bullying, sexual abuse, and other forms of physical and emotional trauma) when they present to ED treatment; (b) SGM patients enter and exit treatment with more severe ED symptomatology scores; and (c) SGM patients have different trajectories of improvement in ED symptoms over the course of their treatment episode. Specifically, we will test the following Hypotheses: (1) SGMs will present to treatment with higher rates of abuse (bullying, sexual, physical/emotional/verbal) than CH peers, (2) SGMs will (a) enter and (b) exit treatment with more severe ED symptoms than CH peers, and (3) SGMs will respond to treatment (in terms of ED symptom reduction) more slowly than their CH peers.

| Participants and setting
This retrospective, longitudinal cohort study utilized a de-identified dataset of 2,818 participants with diagnosed EDs entering higher levels-of-care in a large, US-based, ED treatment center. The majority (n = 2,049) entered at one of the residential treatment centers, 459 entered a partial hospital program, and 310 entered an intensive outpatient program. Eligibility for higher level-of-care treatment was determined by the American Psychiatric Association's Practice Guidelines for the Treatment of Patients with Eating Disorders (Yager et al., 2014) and medical necessity criteria established by third-party payors.
Ninety percent of the Center's population has their treatment covered by private insurance, 8% is self-pay, and 2% is scholarship.

| Materials and procedure
Participants were eligible for study inclusion if they: (a) admitted and discharged between 2015 and 2018, (b) consented for their data to be used for research purposes, (c) were discharged in accordance with the treatment plan or by insurance determination (i.e., discharged into a lower level-of-care), and (d) had an ED diagnosis other than avoidant/ restrictive food intake disorder (ARFID) (see Figure 1 for the STROBE flow diagram). Patients with ARFID (n = 83) were excluded due to research showing that our main outcome measure (i.e., global scores on the Eating Disorder Examination-Questionnaire, EDE-Q) does not sufficiently capture ARFID symptomology (Cooney, Lieberman, Guimond, & Katzman, 2018). During treatment, participants were followed through the program as they moved through higher levels-of-care in a descending trajectory (e.g., residential treatment to partial hospitalization to intensive outpatient, or partial hospitalization to intensive outpatient). Data were recorded at admission and discharge from each levelof-care. To avoid duplication of individuals who may have had more than one treatment episode in the allotted timeframe, only participants' most complete treatment episode was retained to facilitate the longitudinal modeling process. If two or more episodes contained admission and discharge data from only one level-of-care, only the first treatment episode was retained. The final dataset contained 7,456 observations, with between two and six data points for each patient. See Figure 2 for details of the sample's movement through the higher level-of-care sys-
Specifically, the EDE-Q measures cognitive and behavioral aspects of ED symptomatology over the prior 28 days using a scale from 0 (no days) to 6 (every day). Patients were asked to complete the EDE-Q at admission and discharge from each level-of-care.
Our main exposure variable-SGM status-was created using information gathered during structured clinical intake interviews via licensed clinicians (or clinicians overseen by licensed clinicians). Clinicians asked patients their sexual orientation-defined as straight/heterosexual, lesbian/gay, bisexual, unsure, or other-and, whether they identified as transgender.
Patients were also asked their gender-identity (as opposed to sex assigned at birth). The electronic medical record (EMR) allowed for answers of female, male, or blank, with a text field that could be filled in with nonbinary genders. Gender-identity and sexual orientation were combined to form a variable with three categories for SGM status: (a) those who identified as cisgender and heterosexual; (b) SGMs, who identified as transgender and/or as lesbian/gay, bisexual, or other; and (c) unsure individuals, who were cisgender but identified their sexual orientation as "unsure." Table 1 shows the breakdown of gender identification by SGM status. Notably, all patients identified themselves as either female or male, as opposed to nonbinary; thus, gender-identity is presented in only two categories.
History of abuse and bullying were also assessed during the Two a priori specified covariates were also drawn from the structured clinical intake interviews: months-since-ED-onset and prior history of ED treatment in a higher level-of-care (yes/no). To characterize the study sample, we also extracted the following variables from the EMR: patients' current ED diagnoses, weight and height (from which body mass index-BMI-was calculated as weight in kilograms divided by height in meters 2 ), race/ethnicity, and age. To account for time (the primary predictor variable for testing Hypothesis 3) and model the trajectory of change in EDE-Q scores over the course of care, we measured "days in treatment" as the number of days since admission-beginning with 0 to mark the initial admission day-and the exact day of discharge and admission into the next lower level-of-care to numerically mark each following timepoint. All other variables adequately met distribution assumptions.

| Statistical analysis
To reduce problems with multicollinearity when applying confounderadjustments to the model testing Hypotheses 2 and 3, we categorized the age variable (<19 or not), which prior to doing so was highly collinear with months-since-ED-onset (r > .80).
We used logistic regression analysis with Bonferroni adjustments to provide a conservative correction for multiple comparisons to test Mixed-modeling approaches invoke maximum likelihood estimation to handle missing data; this is robust to bias under the assumption of Missing at Random, meaning that data are missing as a function of the observed but not unobserved data (Little & Rubin, 1987).
Using mixed modeling approaches, all cases having at least one observation on the outcome (as did all cases in this dataset) are included in the analysis (Laird, 1988 Table 2 for EMMs by SGM status over time and The confounder-adjusted model (see Table 3, Model 4), which controlled for age (<19), gender-identity, intake BMI, months-since-ED onset, prior ED treatment, and all three abuse variables, found that gender-identity was a significant effect modifier of SGM status on both   (Dohrenwend, 2009), and they exist at both the provider level and more systemically within the US healthcare system (Buchmueller & Carpenter, 2010). Until we structurally address these matters with antidiscrimination policy that mandates healthcare provider education surrounding SGM health and expands cultural sensitivity training to include SGM allyship, health disparities like the differences shown here in ED symptom severity and abuse histories may be slow to change (Dohrenwend, 2009 T A B L E 3 Progression of mixed-effects models predicting global EDE-Q scores over treatment by sexual/gender minority status  was comprised largely of non-Hispanic White females who had private insurance. Therefore, it is not representative of the general population, or SGMs in particular, who are insured at lower rates than the general population (Charlton et al., 2018). The number of male-identifying SGMs in the sample was small (n = 28) and though there were significant effects identified, the low n may have limited our ability to detect other findings related to this group. For statistical modeling purposes, we also excluded a small number of patients (n = 39, 1.38%) who moved from partial hospital programming (or intensive outpatient) to residential treatment centers, which could slightly bias findings toward less treatment-resistant patients.
Lastly, there may have been misclassifications on important exposure variables due to the sensitive nature of the questions. For instance, nonbinary patients may have been misgendered as femaleor male-identifying. Similarly, there may have been underreporting in the abuse-related variables; research has long shown this is common with individuals who have been sexually abused (Kempe, 1978;Reitsema & Grietens, 2015).

| CONCLUDING REMARKS
There are critical research-practice gaps in understanding best care for patients with EDs (Kazdin, Fitzsimmons-Craft, & Wilfley, 2017).
Differences in how populations present and respond to treatment is understudied in the ED field (Cooper et al., 2016;Linardon, Brennan, & de la Piedad Garcia, 2016). Given the significant health disparities seen in SGMs (e.g., , which our findings have underscored, it is crucial to build upon recent data showing that EDs are especially prevalent among this population (Diemer et al., 2015;Kamody et al., 2019). The present study represents an essential step toward closing the research-practice gap by providing evidence of treatment effectiveness for this population in higher levels of ED specialty care. In future research, we suggest further advancing our knowledge by exploring potential differences in ED phenotypes that present in SGM versus CH samples. Knowing the most common ED symptom profiles among SGMs could help aid in earlier detection. Indeed, having population and clinical normative data across demographic characteristics-like SGM status-would offer a more optimal use of our existing assessment tools such as the EDE-Q for detection of EDs in diverse subpopulations (Avila, Golden, & Aye, 2019). Reducing barriers to early intervention in the SGM population is also an important goal, especially given the treatment delay shown in this sample. The present study provides a foundation to build upon for a better understanding of normative data in a clinical sample of SGMs, as well as noting that trauma exposures may play a role in symptom exacerbation and successful treatment of EDs in SGM individuals.