Eating disorder severity and treatment outcome across race/ethnicity, sexual orientation, and socioeconomic status: Intersectional inequities in a clinical sample Authors: Catherine R Drury, Ariel L Beccia, Samar Shaqour, Megan Riddle, Sasha Gorrell, Simar Singh, Alan Duffy, Philip S Mehler, Kianna Zucker, Daniel Le Grange, Renee D Rienecke, Erin E Reilly Published: 2026 ABSTRACT AbstractMarginalized populations experience increased eating disorder (ED) risk and encounter significant barriers to treatment. Intersectionality provides a framework for understanding how systemic oppression contributes to inequities in EDs; however, intersectional approaches have yet to be applied to a clinical ED sample. The current study examined inequities in ED severity and treatment outcome across the intersections of race/ethnicity, sexual orientation, and socioeconomic status (SES). Adult women (N=3016; M = 27.2 years) with transdiagnostic EDs presenting to affiliated treatment sites across the United States completed the Eating Disorder Examination-Questionnaire (EDE-Q) at admission and discharge. Race/ethnicity and sexual orientation were self-reported; SES was measured using the area deprivation index of participants’ neighborhoods. Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) was used to estimate baseline EDE-Q global score; change in EDE-Q global score and binge eating, self-induced vomiting, laxative use, and driven exercise frequency from admission to discharge; and reason for discharge (routine or non-routine) across intersectional subgroups. In this sample of women with access to treatment, MAIHDA models predicted higher baseline levels of overall ED pathology among sexual minorities (predicted M = 4.10). Few differences in ED symptom improvement were observed across intersectional subgroups, with some small yet potentially meaningful inequities. Racially/ethnically minoritized subgroups appeared slightly less likely to complete treatment (predicted percent non-routine discharge = 41.50%). Future research should build on these findings by analyzing other dimensions of inequity (e.g., gender, weight status, disability status) to further characterize and address intersecting systems of oppression that disparately influence ED outcomes.Keywords: Eating disorders, Treatment, Race/ethnicity, Sexual orientation, Socioeconomic status, Disparity, Intersectionality Introduction Eating disorders (EDs) cause significant suffering and impairment via their impact on physical and psychological health and quality of life (van Hoeken and Hoek, 2020). Although EDs occur across age, race, ethnicity, gender, sexual orientation, and socioeconomic status (SES; Diemer et al., 2018; Grammer et al., 2021; Huryk et al., 2021; Rodgers et al., 2018), ED research has historically overrepresented a particular patient demographic (i.e., White, middle-to upper-class, cisgender women), due in part to longstanding, erroneous stereotypes about who is affected (Halbeisen et al., 2022). This imbalance has created a circular, self-perpetuating problem, whereby individuals who fit this demographic are more likely to be identified as having an ED (Gordon et al., 2006; Sala et al., 2024) and therefore more likely to receive ED treatment and participate in research (Sonneville and Lipson, 2018; Penwell et al., 2024). Thus, while the field continues to uncover heterogeneous presentations and develop tailored treatments (Anderson et al., 2023), ED pathology and treatment response among historically marginalized populations remain poorly characterized (Burnette et al., 2022). To encourage the development of inclusive assessment and treatment approaches, additional research on how EDs present and progress among marginalized populations is urgently needed (Goel et al., 2022; Halbeisen et al., 2022). In response to calls to dismantle ED stereotypes (Halbeisen et al., 2022), an increasing number of community-based studies have examined the distribution of ED pathology within and between social groups. Critically, this emerging research has found that marginalized populations may be disproportionately affected by EDs. For example, forms of adversity/disadvantage such as discrimination and food insecurity increase ED risk (Hazzard et al., 2020; Mason et al., 2021). Additionally, sexual and gender minorities exhibit higher rates of EDs and ED pathology than heterosexual and cisgender people (Nagata et al., 2024), and several studies suggest a heightened prevalence of EDs and ED-related behaviors among socially minoritized racial, ethnic, and multiracial groups (Burke et al., 2021; Gordon et al., 2024; Qeadan et al., 2023; Simone et al., 2022). Intersectionality provides a more nuanced framework for examining health inequities, including in ED presentation and treatment outcomes. Based in Black feminist scholarship (Crenshaw, 1993), intersectionality recognizes that systems of power and oppression embedded in society (e.g., structural racism, structural sexism) intersect, or mutually constitute one another, differentially shaping lived experiences according to one’s unique constellation of social identities (e.g., race/ethnicity, gender). Thus, intersectionality recognizes the variability in risk and resilience within heterogeneous populations such as “people of color” or “women” (Bowleg, 2012). Regarding health inequities, it is increasingly understood that unjust societal systems and structures directly influence disease risk and access to care (Bailey et al., 2017; Homan, 2019). Examining health inequities along a single axis (e.g., by gender) or only among multiply marginalized populations (i.e., not including people experiencing multiple forms of privilege or both privilege and disadvantage) is therefore insufficient (Cole, 2009; Evans et al., 2018). Instead, an intersectional perspective requires consideration of the ways by which multiple inequities (as proxies for experience with systems of power and oppression) interact to shape risk for a given outcome across intersectional social groups. Intersectional research approaches hold promise for illuminating and addressing persistent inequities in EDs (Burke et al., 2020). To date, only a handful of studies (Beccia et al., 2019, 2021; Billman Miller et al., 2024; Burke et al., 2023; Egbert et al., 2024; Gordon et al., 2024) have considered ED pathology from an intersectional framework – even fewer have examined the interaction of more than two axes of inequity. Consistent with intersectionality, this burgeoning research area has found particularly high ED prevalence among multiply marginalized populations. For example, Burke and colleagues (2023) studied EDs at the intersection of SES, gender identity, sexual orientation, and race/ethnicity, and found disproportionately high prevalence among Hispanic/Latine, sexual minority men, and lower-SES women, with considerable heterogeneity across all groups. Likewise, Beccia and colleagues (2021) examined ED pathology and diagnosis across the intersections of gender identity and expression, sexual orientation, and weight status and identified several multiply marginalized populations with heightened ED prevalence, including gender non-conforming, sexual minority, and larger-bodied girls and women. Notably, these inequities occurred over and above individual-level differences, underscoring the importance of adopting an intersectional perspective that explicitly considers structural influences. All existing intersectional ED research has been conducted in non-clinical, community samples, leaving unstudied the ways in which structural and systemic factors that contribute to inequities in ED onset may also influence variability in illness course. For example, given that marginalized individuals are more likely to encounter barriers to specialized ED care (Moreno et al., 2023; Penwell et al., 2024), including delays in ED identification (Gordon et al., 2006; Sala et al., 2024; Sonneville and Lipson, 2018), there may be disparities in ED severity at initial presentation to treatment. Moreover, considering certain social groups have been underrepresented in the development of ED treatments and treatment programs (Burnette et al., 2022; Halbeisen et al., 2022), there may be inequities in treatment response across intersectional social groups as well. The current study sought to fill these gaps by investigating intersectional inequities within a transdiagnostic sample of adult women treated for EDs at intensive ED treatment centers. Specifically, we examined the intersection of race/ethnicity, sexual orientation, and SES relative to (a) baseline level of ED pathology and (b) treatment outcome, as measured by change in ED pathology from admission to discharge and reason for discharge from treatment. Based on prior work, we hypothesized that multiply marginalized patients would exhibit higher levels of ED pathology at baseline and less improvement in symptoms than multiply privileged patients, and be less likely to complete a full course of treatment. Method Participants and procedure Our sample (N=3016) was a naturalistic sample of adult cisgender women with transdiagnostic EDs admitted to a multisite ED treatment facility in the United States (US) October 2020-September 2023. Treatment sites spanned 24 locations and included inpatient units, non-hospital-based residential treatment centers, partial hospitalization programs (PHP), and intensive outpatient programs (IOP). All sites utilized a singular treatment approach, detailed in the Supplementary Material. Sociodemographic information and ED diagnoses were determined during semi-structured interviews administered by licensed clinicians and informed by DSM-5 (American Psychiatric Association, 2013) criteria. Upon admission, patients were approached regarding study participation and provided informed consent before completing self-report questionnaires, which were readministered at treatment discharge. Study procedures were approved by the Salus Institutional Review Board. Of the 3798 patients who provided informed consent and completed outcome measures at baseline, 82.2% were cisgender women. As examining the intersection of four social identities (i.e., race/ethnicity, sexual orientation, SES, and gender) would have required a much larger sample size, patients who identified as cisgender men (7.0%), transgender (1.8%), non-binary (6.0%), or did not disclose a gender identity (3.1%) were excluded from the current study in order to remove any confounding heterogeneity caused by the influence of gender on ED pathology. Among the remaining patients, 3.4% (n = 106) were excluded due to missing both race/ethnicity and sexual orientation data, resulting in a final sample size of 3016 cisgender women with EDs. Measures Social identities Consistent with intersectionality, we emphasize that the social identities and labels used in this study stand as proxies for experiences with systems of oppression and other social processes (e.g., structural racism, classism) that may differentially influence ED severity and treatment response. For each of the social identities examined, decisions were made to categorize participants to allow sufficient sample sizes for intersectional analysis, while preserving nuance and variability within the data as much as possible. Although similar classifications have been used in prior intersectionality-based studies (e.g., Beccia et al., 2019; Forrest et al., 2023), this is a notable limitation addressed further in the Discussion. Race/Ethnicity. During initial clinical interview, clinicians asked participants an open question regarding their race/ethnicity, selected one of seven options that combined race/ethnicity into a single variable, and confirmed their selection with participants: American Indian/Native American or Alaska Native; Asian; Black/African American; Hispanic/Latine; Native Hawaiian/Pacific Islander; White; or Other, which included bi/multiracial patients. Some participants reported a more specific race/ethnicity (e.g., Indian), which was noted in a comments section. Specific geographic nationalities (e.g. Japanese) were then converted to the relevant classification (e.g. Asian); where such a conclusion could not be made, Other was used to avoid assumptions about racial/ethnic identity. Due to sample size limitations, American Indian/Native American or Alaska Native and Native Hawaiian/Pacific Islander were collapsed into a single category, termed Indigenous. Sexual Orientation. Sexual orientation was also assessed during initial clinical interview as part of a dual, open-ended question regarding both sexual orientation and relationship status. Because clinicians often recorded patients’ relationship status without noting their sexual orientation, there was a high degree of missingness in sexual orientation data (54.5%). Responses comprised 11 categories: asexual, bisexual, gay, lesbian, mostly heterosexual, pansexual, queer, questioning, heterosexual, multiple identities, and other. Based on documented differences in mental health outcomes between gay or lesbian and bisexual people (Chan et al., 2020), the following collapsed categories were used: heterosexual; gay or lesbian; bisexual, pansexual, or queer; and “other,” which included patients who identified as asexual, mostly heterosexual, multiple identity, other, and questioning. Socioeconomic Status. We operationalized SES as the area deprivation index (ADI; Kind and Buckingham, 2018) of the home neighborhoods in which participants resided at the time of admission. Participants’ zip codes were used to calculate ADI based on 5-year (2018-2022) American Community Survey data regarding the income, education, employment, and housing quality of each area. We mapped zip codes to zip code tabulation areas (ZCTA) using a crosswalk table (Human Resources and Service Administration, 2020); the ‘sociome’ package (Krieger et al., 2023) for R was used to calculate ADI. Because our sample represented all 50 states plus the District of Columbia, the reference area was set to all ZCTA in the US such that ADI scores were assigned relative to all other US ZCTA. By definition, the ADI has an M of 100 and an SD of 20, with higher ADI scores representing greater socioeconomic disadvantage. For the current study, ADI scores were converted to a binary variable (0 = higher SES, 1 = lower SES) divided along the median score (ADI = 82.75), as in prior work (e.g., Deng et al., 2022; Hufnagel et al., 2021). Eating disorder pathology The Eating Disorder Examination-Questionnaire (EDE-Q; Fairburn and Beglin, 1994, 2008) is a 28-item self-report measure of ED pathology over the past 28 days. The measure generates four subscales, the average of which yields a global score. Items not included in the global score assess the frequency of binge eating episodes, self-induced vomiting, laxative use, and driven exercise. In the current study, internal consistency for the EDE-Q global score was Cronbach’s α = .91 at admission and α = .92 at discharge. Discharge status Treatment center staff were trained to identify and record which of the following categories best captured patients’ reason for discharge: routine (i.e., deemed appropriate by patient’s care team based on symptom improvement or need for a different care setting), administrative (i.e., requested by care team due to unacceptable behavior or poor program fit), resource constraint (i.e., insurance denying/ending coverage, patient unable to pay for treatment), patient request (i.e., without medical risk but prior to the care team recommending discharge), against medical advice (i.e., patient leaving treatment when not advised due to medical risk), and COVID-19 (i.e., due to contracting COVID or fear of contracting COVID). Similar to prior work (Gorrell et al., 2022), discharge status was operationalized as a binary construct (0 = routine, 1 = non-routine) to examine intersectional inequities in likelihood of completing a full course of treatment. Statistical analyses Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA) was used to examine inequities in ED outcomes at the intersection of race/ethnicity, sexual orientation, and SES. Considered best practice in quantitative intersectionality research (Evans et al., 2018; Merlo, 2018), MAIHDA is a theoretically driven, descriptive approach that involves fitting multilevel models in which individuals at level 1 are nested within intersecting subgroups or “social strata” at level 2. In so doing, this approach examines both heterogeneity at the stratum level (or average differences between stratum) and the proportion of total outcome heterogeneity that can be attributed to stratum-level (versus individual-level) differences (Merlo, 2018). MAIHDA therefore aligns with intersectionality in that each social stratum is uniquely considered to hold advantages and disadvantages relative to the outcome. A published tutorial for MAIHDA (Evans et al., 2024) informed our analyses and interpretation. We first examined sample sizes across social strata (Table S1). Although MAIHDA accommodates small strata sample sizes by providing more conservative estimates via shrinkage, current guidelines recommend ≥20 participants in at least 80% of all strata (Evans et al., 2018, 2024). Thus, due to the large amount of missing sexual orientation data constraining strata sample sizes, we chose to conduct two separate analyses across the intersection of: 1) race/ethnicity and SES, and 2) sexual orientation and SES. Outcomes were EDE-Q global score at admission; change in EDE-Q global score, change in binge eating episode frequency, change in self-induced vomiting frequency, change in laxative use frequency, and change in driven exercise episode frequency from admission to discharge; and reason for discharge. For each of these 14 outcomes, we fit a two-level multilevel simple intersectional model with random intercepts for social strata. Model results are reported in terms of variance partition coefficients (VPCs), which represent the proportion of total sample variance in the outcome that can be attributed to between-stratum variance, as well as predicted outcomes for each social stratum. For context, VPCs in epidemiological studies are often less than 5% and rarely exceed 10% (Evans et al., 2024). As a descriptive analysis, MAIHDA does not focus on significance testing, effect sizes, or sensitivity analyses; instead, results were interpreted in the context of the VPCs and between-stratum differences in predicted outcomes. Data were analyzed in R version 4.4.1 (R Core Team, 2023). The ‘cmdstanr’ package (Gabry et al., 2024) was used to fit the MAIHDA models using Bayesian Markov chain Monte Carlo estimation methods, with vague priors, a burn-in period of 1000 iterations, and a total of 8000 iterations for each model (Evans et al., 2024). Posterior samples were checked using R-hat and effective sample size, and visual assessments of trace plots and autocorrelation plots. We adjusted for baseline scores in the change score analyses but did not include any other adjustment variables in our models that might obscure the magnitude of inequities across social strata (Fox et al., 2022; Lesko et al., 2022). For example, prior intersectional studies of ED risk have adjusted for age to account for variability in ED risk across the age spectrum; however, given that the current study used a clinical sample, differences in outcomes across age groups may reflect barriers to care that lead to marginalized groups accessing treatment at a later age. Thus, adjusting for age would adjust away the inequitable systems that influence treatment access and clinical outcomes. While we report results for age-adjusted models in the Supplementary Material, these findings may underestimate the magnitude of inequities and should be interpreted with caution. Results Sociodemographic and clinical characteristics for the entire sample (N = 3016) are shown in Table 1. Most participants were White (84.2%) and heterosexual (66.8%), with an average ADI score less than the average for all US ZCTAs (i.e., less socioeconomic disadvantage). Most participants had a diagnosis of anorexia nervosa (45.9%) or other specified feeding or eating disorder (33.7%). Table 1Participant demographic and clinical characteristics (N = 3016). n % Level of care at admission Inpatient61020.2 Residential100733.4 PHP75725.1 IOP168 5.6 Virtual IOP47315.7ED diagnosis Anorexia nervosa, restricting type90830.1 Anorexia nervosa, binge-purge type47815.8 Bulimia nervosa234 7.8 Binge eating disorder196 6.5 Avoidant/restrictive food intake disorder176 5.8 Other specified feeding or eating disorder101733.7 Unspecified feeding or eating disorder7 0.2Race/Ethnicity Indigenousa13 0.4 Asian95 3.2 Black/African American96 3.2 Hispanic/Latine163 5.5 White251784.2 Bi/multiracial105 3.5Sexual Orientation Heterosexual91766.8 Gay or lesbian86 6.3 Bisexual, pansexual, or queer28220.5 Otherb88 6.4Reason for discharge Routine184461.1 Non-routine117238.9 M SD Range Age (years) 27.2110.4915.00-68.00Area deprivation index (ADI)c 82.7218.6621.17-140.58Total length of stay (days) 66.7348.731.00-738.00% expected body weight at admission103.6438.8238.13-425.47ED symptoms at admission EDE-Q global score 3.73 1.570.00-6.00 Binge eating episodes 5.5015.080.00-500.00 Self-induced vomiting 6.3721.560.00-500.00 Laxative use 2.1011.100.00-500.00 Compulsive exercise 5.6910.820.00-300.00Note. Percents are valid percents. PHP = partial hospitalization program; IOP = intensive outpatient program; ED = eating disorder; EDE-Q = Eating Disorder Examination-Questionnaire; SES = socioeconomic status.aIndigenous includes American Indian/Native American or Alaska Native and Native Hawaiian/Pacific Islander participants.bOther sexual orientation includes participants who identified as asexual, mostly straight, multiple identity, other, and questioning.cADI scores were assigned relative to all US zip code tabulation areas (M = 100, SD = 20). Higher ADI scores represent greater socioeconomic disadvantage. Participant demographic and clinical characteristics (N = 3016). Note. Percents are valid percents. PHP = partial hospitalization program; IOP = intensive outpatient program; ED = eating disorder; EDE-Q = Eating Disorder Examination-Questionnaire; SES = socioeconomic status. Indigenous includes American Indian/Native American or Alaska Native and Native Hawaiian/Pacific Islander participants. Other sexual orientation includes participants who identified as asexual, mostly straight, multiple identity, other, and questioning. ADI scores were assigned relative to all US zip code tabulation areas (M = 100, SD = 20). Higher ADI scores represent greater socioeconomic disadvantage. Similar to other studies of ED higher levels of care (e.g., Brown et al., 2018; Reilly et al., 2020), a large number of participants (46.0%) were missing EDE-Q data at discharge and therefore excluded from change score analyses. Participants with missing data were older, p=.034, had a shorter length of stay, p < .001, lower EDE-Q global scores, p=.014, and were at a higher percent of expected body weight at admission, p < .001. Indigenous, Hispanic/Latine, and bi/multiracial participants had a larger proportion of missingness than White participants, p = .009. There was also a larger proportion of missingness among participants admitted to PHP or IOP, with a non-routine discharge, and with an ED diagnosis other than anorexia nervosa (all ps < 0.001). Participants with and without missing EDE-Q data did not significantly differ in sexual orientation or ADI. Table 2 provides a summary of strata sample sizes, which were sufficient for both models, with ≥20 participants in 83% and 100% of strata. When the sample was reduced to include only participants with EDE-Q data at discharge for the change score analyses, there were ≥20 participants in 75% of strata for the race/ethnicity and SES model and 100% of strata for the sexual orientation and SES model. Table 2Sample sizes of intersectional social strata (N = 3016).Strata defined by race/ethnicity and SES (n = 12)Sample size per stratumNumber of strata% of strata 100 or more216.750 or more758.330 or more1083.320 or more1083.310 or more1191.7Less than 1018.3 Strata defined by sexual orientation and SES (n = 8)vSample size per stratumNumber of strata% of strata 100 or more450.050 or more450.030 or more8100.020 or more8100.010 or more8100.0Less than 1000.0Note. SES = socioeconomic status as measured by area deprivation index (ADI) divided along the median (ADI = 82.75). Sample sizes of intersectional social strata (N = 3016). Note. SES = socioeconomic status as measured by area deprivation index (ADI) divided along the median (ADI = 82.75). Results for all simple intersectional MAIHDA models are shown in Table 3; Figs. 1 and 2 show the stratum-specific predicted values for each outcome. Full summaries of strata sample sizes and observed and predicted values (including age-adjusted predicted values) are provided in the Supplementary Material (Table S2–S8). For ED severity at admission, VPCs indicated that only 0.2% of the total variation in ED pathology was attributable to stratum-level differences across race/ethnicity and SES, while 2.1% was attributable to stratum-level differences across sexual orientation and SES, over and above differences between individuals. Table S2 shows the stratum-specific predicted EDE-Q global scores at admission, which were similar across social strata in the race/ethnicity and SES model (range: 3.66-3.74). Strata including sexual minorities had slightly higher predicted levels of ED pathology at admission than those including heterosexual patients (range: 3.71-4.10), with lower-SES gay or lesbian patients having the highest predicted values. Table 3Parameter estimates for simple intersectional models (N = 3016).Race/ethnicity and SESSexual orientation and SES Estimate[95% CI]Estimate[95% CI] EDE-Q global score at admission Intercept3.69[3.56, 3.78]3.89[3.70, 4.12] Stratum-level variance0.08[0.00, 0.23]0.23[0.03, 0.57] Individual-level variance1.57[1.53, 1.61]1.59[1.53, 1.65] VPC0.2%2.1%Change in EDE-Q global score Intercept−0.17[−0.36, 0.06]−0.16[−0.40, 0.10] Baseline EDE-Q global score0.47[0.43, 0.50]0.46[0.40, 0.51] Stratum-level variance0.14[0.01, 0.40]0.10[0.00, 0.34]] Individual-level variance1.19[1.15, 1.23]1.19[1.13, 1.25] VPC1.4%0.7%Change in binge eating episodea Intercept−0.73[−1.13, −0.31]−0.58[−1.37, 0.31] Baseline binge eating0.92[0.90, 0.94]0.87[0.83, 0.91] Stratum-level variance0.26[0.01, 0.83]0.62[0.04, 1.64] Individual-level variance4.67[4.50, 4.84]6.16[5.86, 6.50] VPC0.3%1.0%Change in self-induced vomitinga Intercept−0.37[−0.66, −0.08]−0.28[−0.73, 0.18] Baseline self-induced vomiting0.94[0.94, 0.95]0.93[0.92, 0.94] Stratum-level variance0.19[0.01, 0.58]0.29[0.01, 0.95] Individual-level variance3.11[3.01, 3.22]3.79[3.59, 4.00] VPC0.4%0.6%Change in laxative usea Intercept−0.23[−0.38, −0.06]−0.17[−0.37, 0.00] Baseline laxative use0.99[0.98, 0.99]0.96[0.94, 0.98 Stratum-level variance0.12[0.01, 0.38]0.13[0.00, 0.41] Individual-level variance1.68[1.62, 1.74]1.69[1.61, 1.79] VPC0.5%0.5%Change in driven exercisea Intercept−0.43[−0.89, 0.20]−0.54[−1.40, 0.20] Baseline driven exercise0.78[0.75, 0.80]0.79[0.75, 0.83] Stratum-level variance0.34[0.01, 1.09]0.60[0.03, 1.87] Individual-level variance4.94[4.77, 5.12]5.11[4.85, 5.38] VPC0.5%1.4% OR[95% CI]OR[95% CI] Reason for dischargeb Intercept0.65[0.58, 0.81]0.64[0.54, 0.75] Stratum-level variance1.12[1.00, 1.43]1.11[1.00, 1.40] VPC0.4%0.3%Note. SES = socioeconomic status; EDE-Q = Eating Disorder Examination-Questionnaire; VPC = variance partition coefficient.aChange in frequency over the past 28 days from admission to discharge.b0 = routine discharge and 1 = non-routine discharge. Parameter estimates for simple intersectional models (N = 3016). Note. SES = socioeconomic status; EDE-Q = Eating Disorder Examination-Questionnaire; VPC = variance partition coefficient. Change in frequency over the past 28 days from admission to discharge. 0 = routine discharge and 1 = non-routine discharge. Predicted values for strata defined by race/ethnicity and socioeconomic status (N = 3016). EDE-Q = Eating Disorder Examination-Questionnaire. Spikes indicate approximate 95% confidence intervals. Two-digit stratum ID numbers reflect coding as follows: Digit 1 = race/ethnicity [1 = Indigenous; 2 = Asian; 3 = Black/African American; 4 = Hispanic/Latine; 6 = White; 7 = Bi/multiracial]; Digit 2 = socioeconomic status [0 = Higher; 1 = Lower]. Predicted values for strata defined by sexual orientation and socioeconomic status (N = 3016). EDE-Q = Eating Disorder Examination-Questionnaire. Spikes indicate approximate 95% confidence intervals. Two-digit stratum ID numbers reflect coding as follows: Digit 1 = sexual orientation [1 = Heterosexual; 2 = Gay or lesbian; 3 = Bisexual, pansexual, or queer; 4 = Other]; Digit 2 = socioeconomic status [0 = Higher; 1 = Lower]. Regarding change in ED symptoms from admission to discharge, heterogeneity across social strata for all outcomes was low, with small differences in symptom reductions. All strata seemed to benefit similarly from treatment in terms of overall ED pathology (predicted change score range: 1.49-1.75; VPCs = 0.7-1.4%). Across both the race/ethnicity and SES model and the sexual orientation and SES model, strata with lower SES tended to experience slightly greater reductions in ED pathology than strata with higher SES, with lower-SES Hispanic/Latine and gay or lesbian patients demonstrating the greatest improvements (Table S3). Change in overall ED pathology was lowest among strata including higher-SES Asian patients and higher-SES patients whose sexual orientation was categorized as “other.” While reductions in ED behaviors were largely similar across social strata, small inequities were found for binge eating (predicted change score range: 3.20-4.02; VPCs = 0.3-1.0%) and driven exercise frequency (predicted change score range: 4.06-5.45; VPCs = 0.5-1.4%), with a complex patterning across both models (Table S4 and S7). Specifically, in the sexual orientation and SES model, strata including lower-SES heterosexual patients showed the least improvement in binge eating episode frequency relative to sexual minorities and higher-SES heterosexual patients. Regarding driven exercise episode frequency, patients whose sexual orientation was categorized as bisexual, pansexual, queer, or “other” experienced smaller improvements than gay or lesbian patients and higher-SES heterosexual patients. Lower-SES heterosexual, White, and bi/multiracial patients had smaller predicted change scores than their higher-SES counterparts. Low levels of heterogeneity were found for change in self-induced vomiting (predicted change score range: 5.21-7.30; VPCs = 0.4-0.6%) and laxative use frequency (predicted change score range: 1.65-1.80; VPCs = 0.5%) across models. Results from the reason for discharge models are shown in Table 3 and S8. A small proportion of the variation in reason for discharge was attributable to stratum-level differences (predicted percent non-routine discharge range: 38.12-41.50%; VPCs = 0.3-0.4%) and suggested a slightly greater likelihood of a non-routine discharge among strata that included racially/ethnically minoritized patients, with Indigenous patients having the highest predicted prevalence. Lower-SES White and heterosexual patients were least likely to have a non-routine discharge. Discussion The current study is the first to apply an intersectional approach to a clinical ED sample by examining and quantifying inequities in ED severity and treatment outcome across the intersections of race/ethnicity, sexual orientation, and SES. In this sample of women with access to specialized ED treatment, sexual minorities tended to report higher levels of ED pathology at admission than heterosexual patients, with lower-SES gay or lesbian patients endorsing the most severe symptoms. This finding is consistent with the disproportionate rates of ED pathology in community-based sexual minority samples (Calzo et al., 2017; Kamody et al., 2020) as a result of the discrimination and stigma experienced by these individuals (Meyer, 2003). The similar levels of pathology observed across racial/ethnic and SES groups further discredits the myth that EDs predominantly affect affluent, White populations. A growing body of work has found increased ED risk among racially/ethnically minoritized and lower-SES people (e.g., Burke et al., 2021; Gordon et al., 2024). The current study builds on this literature to suggest that diverse social groups who experience EDs and present to treatment may exhibit comparable levels of illness severity. Across ED symptoms, there was minimal heterogeneity in change from admission to discharge, suggesting that overall, patients may have benefited similarly from specialized treatment, consistent with prior studies of ED treatment outcomes across sexual and racial/ethnic identities (e.g., Donahue et al., 2020; Lydecker et al., 2019). However, it is important to interpret this finding within the context of the study’s methodological limitations. Specifically, the smaller sample sizes in social strata that included racially/ethnically minoritized and sexual minority patients resulted in more cautious estimates for these strata that may have led to the minimal heterogeneity observed, rather than a true absence of inequity. Additionally, the larger proportion of missing discharge data among patients with a non-routine discharge may have biased results toward those with longer, potentially more successful treatment and possibly failed to capture the full range of heterogeneity that might have been found had all patients been included in these analyses. Within the small treatment differences observed in the current study, the patterning across intersectional social groups was complex and seemingly inconsistent. Given the study’s naturalistic design, missing data, descriptive analytic approach, and small VPC values, the clinical significance of these differences should be interpreted with caution; however, several findings may highlight important directions for future investigation. First, lower-SES patients tended to experience slightly more improvement in overall ED pathology than higher-SES patients. However, some strata including lower-SES patients (e.g., White, bi/multiracial, heterosexual) appeared to experience small inequities in reduction of behavioral symptoms, particularly binge eating and driven exercise. Thus, while our results potentially align with qualitative accounts that experiences of oppression and deprivation impact ED treatment (Calzo et al., 2025; Frayn et al., 2022), more research is needed to better understand the treatment experiences of lower-SES patients, including in higher levels of care. Finally, the smaller improvements in driven exercise observed in some sexual minority groups (i.e., bisexual, pansexual, queer, and “other”) may be consistent with prior work that has found worse psychological treatment outcomes for bisexual patients (Rimes et al., 2019), who can experience even greater stigma and less community support than gay or lesbian people (Ross et al., 2018). In light of the methodological and measurement limitations addressed further below, it is also possible that reported differences between intersecting subgroups are reflective of regression to the mean, limitations in the use of self-report measures, or the suboptimal methods by which social identities were categorized. Although, descriptively, variability in discharge reason across intersecting social groups was also low, small differences were observed among strata including racially/ethnically minoritized patients, who were generally less likely to complete treatment than White patients. A non-routine discharge likely involved leaving treatment prior to symptom remission, potentially impeding full recovery and thus contributing to further inequities in outcomes. Critically, research on ED treatment among racially/ethnically minoritized populations remains limited. While a handful of studies have found that neither race nor ethnicity are significant predictors of treatment effectiveness in ED intervention trials (Thompson-Brenner et al., 2013; Linardon et al., 2017), racially/ethnically minoritized groups were more likely to drop out of treatment (Thompson-Brenner et al., 2013). Additional work in this area is needed to ensure accessible, inclusive treatment that makes recovery possible for patients of all racial, ethnic, and cultural identities. This study has several limitations that highlight additional important takeaways and areas for further study. First, although participants represented a diverse range of racial/ethnic, sexual, and socioeconomic identities, the majority were White, heterosexual, cisgender women, and lived in less socioeconomically disadvantaged areas, which limits generalizability and conclusions about populations most affected by structural influences. As EDs occur across races/ethnicities, sexual orientations, genders, and socioeconomic backgrounds and disproportionately impact marginalized groups (Burke et al., 2023; Diemer et al., 2018; Grammer et al., 2021; Rodgers et al., 2018), the limited representation of minoritized and marginalized patients may be attributed to this being a sample with access to ED treatment primarily covered by private health insurance. Due to structural racism and other systems of power and oppression, individuals whose identities do not align with historical ED stereotypes face substantial barriers (e.g., financial, stigma, delayed diagnosis) to accessing specialized care (Penwell et al., 2024) and thus are underrepresented in treatment research such as the current study. For example, racially/ethnically minoritized patients are less likely to be identified as having an ED or referred to ED treatment due to racist medical curricula and trainings (Gordon et al., 2006; Sonneville and Lipson, 2018) and more likely to be uninsured or have public insurance due to inequities in employment and earnings (Lee et al., 2021), while most specialty ED providers only accept private insurance or out-of-pocket payment (Accurso et al., 2021). Regarding the impact on our findings, social strata including marginalized patients had smaller sample sizes and, therefore, potentially less accurate predicted values that may have underestimated inequities (Evans et al., 2018, 2024). Moreover, the limited gender diversity among the patients seen at the treatment center combined with the relatively small overall sample for intersectional analyses precluded including gender as an intersecting social identity and reduces the study’s generalizability across genders. Future research can recruit clinical samples more representative of the diverse population affected by EDs (e.g., Beccia et al., 2021; Burke et al., 2023), including gender-diverse patients and those with public health insurance (Accurso et al., 2021). Several assessment-based limitations may have impacted our results and can inform data collection and assessment decisions within ED programs seeking to examine treatment outcomes through an intersectional lens. Perhaps most notably, the amount of missingness in outcome data, though similar to other research in ED higher levels of care (e.g., Reilly et al., 2020), may have inflated estimates of treatment effectiveness and limits confidence in the intersectional comparisons, particularly because missingness was associated with demographic and clinical characteristics. Specifically, missingness further constrained the sample sizes of strata that included racially/ethnically minoritized patients in change score analyses and limited generalizability primarily to more severe ED presentations and younger patients with a routine discharge and longer length of stay. Moreover, the large amount of missing sexual orientation data caused by the manner in which sexual orientation was assessed (i.e., as part of a dual, open-ended question regarding both sexual orientation and relationship status) weakened the robustness of analyses and prevented simultaneous examination of race/ethnicity, sexual orientation, and SES, which would provide a clearer picture of the patterning of inequities across all intersecting subgroups. Examining only two identity axes at a time deviates from the premise of inter-sectionality and may have led to incomplete interpretations. Other assessment-based limitations include collapsing racial/ethnic and sexual orientation categories to allow sufficient subgroup sample sizes for intersectional analyses, which compromises nuance within these social identities and may have obscured meaningful heterogeneity and inequities within subgroups. As race and ethnicity are separate and multi-dimensional constructs (Burton et al., 2010; Helms, 2007), future research can incorporate multiple elements of participants’ cultural backgrounds and racial/ethnic identities (e.g., country of birth, religion, national identity; Burton et al., 2010) to more sensitively and flexibly capture the different experiences of racially/ethnically minoritized groups. Additionally, calculating ADI using zip code provides a less precise measure of SES than using census block group (Krieger et al., 2003) and may introduce misclassification by not accurately capturing individual SES experiences. Dichotomizing ADI at the median may have further obscured heterogeneity across smaller ADI ranges, thereby weakening conclusions about SES-related inequities. Despite precedent for using these methods to quantify health inequities (e.g., Badreddine et al., 2024; Deng et al., 2022; Hufnagel et al., 2021; Lehr et al., 2023), future research can examine ED outcomes across more individualized and incremental measures of disadvantage, such as income-to-needs ratio, educational attainment, and measures of material hardship and deprivation (e.g., food insecurity; Diemer et al., 2013). Although all treatment sites utilized a singular treatment approach, participants may have experienced variations in treatment delivery and program intensity, given the naturalistic study design. While the decision to exclude additional covariates from analyses (e.g., diagnosis, level of care, treatment duration) was made to avoid obscuring the magnitude of inequities across social strata (Fox et al., 2022; Lesko et al., 2022), observed differences/similarities in outcomes may have been confounded by clinical or demographic factors unrelated to the social identities included in analyses. Finally, despite precedent for the use of self-report measures in prior intersectionality ED research (e.g., Beccia et al., 2019, 2021; Burke et al., 2023; Gordon et al., 2024), the reliance on the EDE-Q as the primary outcome introduces limitations associated with self-report measures, including recall bias and limited sensitivity to culturally specific symptom presentations, which may systematically affect marginalized groups (Alexander et al., 2024; Parker et al., 2023). Although behavioral indicators and clinician report of patients’ ED symptoms were used at the treatment center to determine readiness for discharge, these metrics were not included in the study dataset and can be incorporated into future work to provide a more objective assessment of patient progress. In addition to ED pathology, assessment of clinically relevant co-occurring behaviors (e.g., substance use behaviors) would further situate this study’s findings within the broader literature on ED heterogeneity across intersecting social groups (Qeadan et al., 2023). In response to this work, treatment programs seeking to conduct intersectionality-based research should carefully consider how information about patients’ social identities is assessed and recorded, and collaborate with other treatment sites to standardize data collection. Collaboration and data sharing between multiple programs would generate clinical datasets large enough for more robust intersectional analyses. At the policy level, public insurance programs and healthcare systems in the US do not have sufficient resources to care for individuals with EDs (Mikhail et al., 2025), and specialty ED treatment programs often do not accept public insurance (Bailey-Straebler et al., 2024), preventing many marginalized individuals from accessing care and participating in clinical research. Addressing this problem will likely require intervention across the continuum of ED care, including outreach and education to reduce stigma among providers and the public about who is susceptible to EDs, development of ED screening tools that consider the experiences of diverse populations, advocacy for improved insurance coverage and reimbursement for ED treatment, and increased consideration of intersecting social identities in ED treatment models. Harmonizing and improving data collection in this space will also ultimately strengthen policy recommendations, including identifying best practices for measuring intersectional inequities across treatment settings. Historical ED stereotypes and systemic barriers to healthcare have long contributed to inequities in ED outcomes. By seeking to examine ED severity and treatment response across several intersecting social groups, the current study preliminarily demonstrates the complexity with which related systems of power and oppression differentially impact the lived experiences of individuals with EDs. Additional research is needed to further describe the illness and treatment trajectories of underrepresented patients, both through the quantitative methods used in this study and qualitative report. Consideration of the variability of patient experiences and perspectives, via an intersectionality framework, is an important step toward providing more equitable care to all those affected by EDs. Supplementary Material Funding statement This work was supported by the National Institute of Mental Health (DLG, grant number R34MH123596; SG, grant numbers K23MH126201, R21MH131787; EER, grant number K23MH131871; SS, grant number T32MH018261); the National Institute on Minority Health and Health Disparities (ALB, grant number 5F32MD017452); and the Brain and Behavior Research Foundation (SG, grant number 5F32MD017452). Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi.org/10.1016/j.socscimed.2026.119504 . Footnotes Ethics approval/patient consent statement All procedures were performed in compliance with relevant laws and institutional guidelines, and were approved by the Salus Institutional Review Board. All participants and their legal guardians provided informed assent and consent. CRediT authorship contribution statement Catherine R. Drury: Conceptualization, Data curation, Formal analysis, Methodology, Project administration, Visualization, Writing – original draft, Writing – review & editing. Ariel L. Beccia: Formal analysis, Methodology, Software, Writing – original draft, Writing – review & editing. Samar Shaqour: Data curation, Visualization, Writing – original draft, Writing – review & editing. Megan Riddle: Conceptualization, Methodology, Writing – review & editing. Sasha Gorrell: Methodology, Writing – review & editing. Simar Singh: Methodology, Writing – review & editing. Alan Duffy: Data curation, Resources, Writing – review & editing. Philip S. Mehler: Resources, Writing – review & editing. Kianna Zucker: Data curation, Writing – review & editing. Daniel Le Grange: Supervision, Writing – review & editing. Renee D. Rienecke: Resources, Supervision, Writing – review & editing. Erin E. Reilly: Conceptualization, Methodology, Supervision, Writing – review & editing. Declaration of competing interest Dr. Le Grange receives royalties from Guilford Press and Routledge. He is co-director of the Training Institute for Child and Adolescent Eating Disorders, LLC, and a member of the Clinical Advisory Board of Univa Health. Dr. Rienecke receives royalties from Routledge and consulting fees from the Training Institute for Child and Adolescent Eating Disorders, LLC. Data availability Data will be made available on request. References Associated Data Supplementary Materials Data Availability Statement Data will be made available on request.