Effectiveness of a Digital Cognitive Behavior Therapy–Guided Self-Help Intervention for Eating Disorders in College Women

Key Points Question Does a coached, digital, cognitive behavior therapy (CBT) intervention result in improved outcomes among college women with eating disorders (EDs) compared with referral to usual care? Findings In this cluster randomized clinical trial that included 690 women with binge-purge EDs from 27 US universities, the digital CBT intervention was superior to referral to usual care in decreasing ED psychopathology, compensatory behaviors, depression, and clinical impairment through long-term follow-up, as well as in realized treatment access. There was no difference in abstinence from all ED behaviors or academic impairment between groups. Meaning These results support the efficacy of a coached, digital, CBT intervention for college women with EDs, which has the potential to bridge the treatment gap for this problem.

EDs, which has the potential to bridge the treatment gap for this problem.

+ Visual Abstract + Invited Commentary + Supplemental content Introduction
Eating disorders (EDs) are severe psychiatric disorders associated with high morbidity and mortality, marked impairment, and poor quality of life. 1,2 College campuses are faced with an elevated prevalence of EDs, with 13.5% of US college women and 3.6% of US college men affected. 3 Notably, 95% of first-time cases occur by age 25 years, 4 highlighting the importance of intervention with this group. However, fewer than 20% of students with EDs report receiving treatment. 3,5 Inadequacies in care delivery are associated with prolonged illness, poorer prognosis, and greater relapse, highlighting the need for improved modalities for screening and intervention, particularly in the at-risk group of college students. 6,7 Current treatment delivery efforts for EDs on college campuses are hindered by factors such as limited counseling center capacity and access to evidence-based treatments. [6][7][8][9] College students report additional barriers, including lack of time and stigma. 10 Digital technologies, highlighted as the future of psychiatry, 11 have the potential to improve mental health care on college campuses by overcoming barriers. 12 Furthermore, online screens have been developed that can identify individuals with a possible ED, who can then be offered services. 13,14 However, to date, there have been no large-scale studies in college populations that have evaluated the effects of a digital intervention for treating EDs, linked with an online screen.
The aim of the current study was to test the hypothesis that a digital cognitive behavioral therapy (CBT)-guided self-help program, Student Bodies-Eating Disorders (SB-ED), would significantly reduce ED psychopathology in college women screening positive for an ED (excluding anorexia nervosa), compared with referral to usual care. We focused on women because of the higher prevalence of these problems in women vs men. 3 Secondary aims were to test the hypotheses that SB-ED, vs referral to usual care, would increase abstinence from all ED behaviors, reduce ED behaviors (ie, binge eating and compensatory behaviors), depression, anxiety, ED-associated clinical impairment, and academic impairment, and increase realized treatment access.

Participants and Procedure
We recruited US universities for participation in this cluster randomized clinical trial. Participants were female students at participating universities, aged 18 years or older, who completed an online EDs screen and screened positive for a Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) 2 ED (except anorexia nervosa, which requires more intensive medical monitoring) using the Stanford-Washington University ED Screen 14 by endorsing 6 or more episodes of binge eating, vomiting, and/or laxative or diuretic use in the past 3 months. The Stanford-Washington University ED Screen has high sensitivity and specificity for ED cases vs face-to-face interview. 14 Participants also indicated their race (selecting all that applied) and ethnicity on the screen using prespecified options (although an other race option could be selected and defined).

JAMA Network Open | Psychiatry
Students were recruited using campus-specific recruitment strategies, developed collaboratively with campus stakeholders, including use of email, flyers, presentations, social media, and counseling or health center staff offering the study to individuals in need. For more information, see Fitzsimmons-Craft et al. 13 All participating universities either required their own institutional review board approval or deferred to the institutional review board of record, which approved all study procedures. Informed consent was obtained online before screening. Upon screen completion, eligible participants who agreed to participate in the trial were asked to complete online assessments at baseline, at a postintervention assessment (occurring 8 months after baseline), and at 1-year and 2-year follow-up assessments. Participants were remunerated with a $10 gift card for completion of the baseline,

Study Conditions
Participants were randomized at the university level to either intervention or control. Participants were provided information on how to access their assigned condition immediately upon screen completion and were reminded of this information 1 week later via email.

Intervention
SB-ED is a digital, guided self-help, CBT intervention. SB-ED is part of the Healthy Body Image Program, an online platform for screening and tailored intervention for college students at risk for or with EDs. [15][16][17] The intervention covers the core components of CBT for EDs, 18 including reducing ED behaviors (eg, via self-monitoring and regular eating), improving body image, regulating emotions, addressing shape checking and avoidance, challenging negative thoughts, and preventing relapse.
The program includes psychoeducational content, as well as meal planning and tracking tools, selfmonitoring logs, and other interactive tools (eg, texting platform facilitating coach-user communication). Users were provided access for 8 months, and each user was assigned a personal coach (see eTable 1 in Supplement 2 for an outline of final version of the intervention).
The program was hosted and maintained by a private company (Lantern). In year 1 of the trial, the program was offered in a traditional web-based format, using longer, weekly sessions. Because of the lower engagement in year 1 than expected and user-experience designers' impressions of changes that could improve engagement, in years 2 and 3 of the trial, the program was redesigned to comprise 40 shorter, core sessions requiring approximately 10 minutes each. This version covered the same content in the year 1 version of the program and was offered via iPhone (Apple) mobile telephone application (app) in addition to web access. These changes were designed to facilitate user experience improvements, but the core intervention principles remained the same. 19,20 Coaches were psychology doctoral students, social work masters students, study staff, or postdoctoral fellows, working at 1 of the universities overseeing the trial and were under the supervision of a clinical psychologist. Coaches used a clinical management dashboard to efficiently monitor multiple users. The dashboard provided information on users' goals and intervention use, as well as the ability to message users. Coaches underwent extensive training, including in CBT for EDs, 18 motivational interviewing, key tenets of effective digital coaching, and technical training.
Coaches were responsible for providing timely messages to users, supporting them in making changes, and for providing ongoing feedback on progress and symptom changes. In addition, in years 2 and 3, in an effort to further increase usability, coaches offered up to 2 optional 20-minute telephone calls at the beginning and end of users' time in the program. The first call was intended to build rapport, establish goals, and address barriers to use. The second call was intended to review progress and relapse prevention strategies. Clinical supervisors audited coaches' messages regularly and each week reviewed the correspondence between 2 coaches and all of their active users, providing feedback as needed. All coaches participated in weekly group supervision.

Control
Upon completion of the baseline assessment, participants assigned to control received written feedback encouraging them to seek evaluation and/or treatment at their university's counseling center. Participants were provided with specific information on how to make an appointment at their respective center.

Randomization
The target university enrollment was 28, assuming that up to 2 universities would drop, leaving 26 participating. A blinded analyst performed randomization by randomly distributing pairs of universities as they were recruited using a random number generator. A priori randomization procedures specified that we assess for balance on the basis of students-to-therapist ratios at the counseling centers to ensure a balance in usual care across conditions. We used a biased coin technique 21 in our randomization so that intervention and control universities would be approximately balanced on this ratio by the end of randomization.

Measures
The primary outcome was reduction in overall ED psychopathology according to the Eating Disorder

Statistical Analysis
Power analysis was based on the primary hypothesis testing (ie, determine effects of intervention vs control on EDE-Q Global) at the postintervention assessment. The power calculation took into account that cluster randomization occurred at the university level and the assumption that there would be approximately 25 students eligible per university. We assumed an intraclass correlation coefficient of 0.05 to 0.15. With little prior information on the effect size of our primary hypothesis, we assumed a medium effect size (Cohen d, 0.5) at the postintervention assessment. The estimated power ranged from 0.85 to 0.99 with 26 universities (13 universities per condition, 25 students per university) for a total sample of 650 students. We used multilevel mixed effects modeling to conduct the analysis of primary and secondary study aims including data from all participants in line with the intent-to-treat principle. Random effects were specified to account for the nested data structure of multiple assessments (level 1) within individual participants (level 2) and multiple individuals within each university (level 3). Fixed effects were specified to contrast (ie, to estimate change) the postintervention assessment with baseline and to contrast overall follow-up assessment (combining 1-and 2-year follow-up assessments) with baseline. Level 3 fixed effects were specified for the intervention effects (ie, intervention vs control) accounting for randomization at the university level.

JAMA Network Open | Psychiatry
We specified a logit link for assessment of abstinence (binary outcome) and a log link for assessment of binge eating and compensatory behavior rate outcomes. For analyses of academic impairment,

Descriptive Statistics
We randomized 28 US universities, with 1 university failing to recruit participants. For those that did   Table 2 describes the outcome variables, and eTable 2 in Supplement 2 shows possible diagnoses for each assessment by condition. Table 3 summarizes the results of longitudinal mixed effects modeling of primary and secondary outcomes.

Primary Outcome
There was a significantly greater reduction in EDE-Q Global score in the intervention group vs the     t 1385 = −5.10; P < .001).

Subanalysis
Because the design of the intervention changed between year 1 (longer sessions, online only) and years 2 and 3 (shorter sessions, online, and mobile telephone app), the primary outcome analyses were rerun excluding year 1 participants. As in the full sample, there was a significantly greater    also associated with significant reductions in depression and ED-associated clinical impairment at the postintervention assessment and over follow-up compared with control, as well as with significant reductions in anxiety over follow-up. It is important to note that body mass indexes remained constant even as ED attitudes and pathology improved, demonstrating that these changes were not associated with changes in weight. Furthermore, although academic impairment outcomes did not differ between groups, the rates of course withdrawal and taking a leave of absence were quite low overall. Finally, with regard to realized treatment access, the intervention was far superior: 83% of students offered the intervention began it, whereas only 28% of students in the control group reported seeking treatment for their ED at any point over the 2-year follow-up. Overall, the intervention was associated with significantly greater improvements than control in the primary outcome, ED psychopathology, as well as in binge eating, compensatory behaviors, depression, and clinical impairment at postintervention assessment, with these gains sustained through longerterm follow-up for all outcomes except binge eating. These differences are notable given the control group demonstrated substantial improvement.

Strengths and Limitations
Strengths of this study include the large number of participants (a particular challenge in intervention trials for EDs), diversity of the population, broad inclusion criteria, delivery on a national scale, and long-term follow-up. In addition, we optimized the service over time, an important innovation for implementing psychosocial and digital interventions. 19,30 Regarding limitations, first, overall engagement with the intervention was 31%. Yet, even with this level of engagement, the intervention group demonstrated significant improvement vs the control group. Furthermore, this level of engagement is consistent with engagement with mental health apps in the real world; one review 31 indicated that median retention rates for mental health apps were 3.9% for 15 days and 3.3% for 30 days. Future research should address the issue of improving engagement with mental health apps, including SB-ED. Second, the population was recruited though an online screen; thus, ED status was determined on the basis of self-report rather than diagnostic interview. Importantly, however, this approach is consistent with what would be required to sustainably implement digital screening and intervention broadly on university campuses. 5 In addition, current results suggest the superiority of the intervention vs referral to usual care, but future research may wish to compare to other control conditions (eg, in-person CBT).

Conclusions
Overall, these findings support the use of a coached, digital CBT intervention, compared with referral to usual care, for college women with a wide range of EDs. Given its scalability, a digital CBT-guided self-help intervention for EDs has great potential to address the wide treatment gap for this problem.