A Pilot Randomized Controlled Trial of a Cognitive-Behavioral Therapy-Guided Self-Help Mobile App for the Post-Acute Treatment of Anorexia Nervosa: A Registered Report

Abstract

Introduction:

Relapse following acute treatment for anorexia nervosa (AN) is common. Evidence suggests cognitive-behavioral therapy (CBT) may be useful in the post-acute period, but few patients have access to trained providers. mHealth technologies have potential to increase access to high-quality care for AN, including in the post-acute period. The aim of this study is to estimate the preliminary feasibility and effectiveness of a CBT-based mobile intervention plus treatment as usual (TAU), offered with and without an accompanying social networking feature.

Method:

In the current pilot randomized controlled trial, women with AN who have been discharged from acute treatment in the past two months (N=90) will be randomly assigned to a CBT-based mobile intervention plus treatment as usual (TAU), a CBT-based mobile intervention including social networking plus TAU, or TAU alone. We will examine feasibility, acceptability, and preliminary effectiveness of the three conditions in terms of reducing eating disorder psychopathology, reducing frequency of eating disorder behaviors, achieving weight maintenance, reducing depression and suicidal ideation, and reducing clinical impairment. We will examine rehospitalization and full recovery rates in an exploratory fashion. We will also examine whether the mobile intervention and social networking feature change the proposed targets and whether changes in targets are associated with benefit, as well as conduct exploratory analyses to identify within-mobile intervention predictors and moderators of outcome.

Discussion:

Ultimately, this research may lead to increased access to evidence-based treatment for individuals with AN and prevention of the extreme negative consequences that can result from this serious disorder.

Anorexia nervosa (AN) is a costly, life-threatening illness affecting 1-4% of women in their lifetimes (Smink, van Hoeken, & Hoek, 2013; Udo & Grilo, 2019). Patients with severe AN are often treated in the acute setting, but 31-52% of patients relapse after treatment (Berends, Boonstra, & Van Elburg, 2018; Berends et al., 2016; Khalsa, Portnoff, McCurdy-McKinnon, & Feusner, 2017). Further, the discharge criterion in acute settings is typically weight restoration (Marzola, Nasser, Hashim, Shih, & Kaye, 2013). However, research has indicated weight-based recovery is not “enough,” and that full recovery involves physical, behavioral, and psychological components (Bardone-Cone et al., 2010). Cognitive-behavioral therapy (CBT) is useful with this population, including following discharge, and may aid in addressing psychological symptoms, achieving full recovery, and decreasing relapse (Byrne et al., 2017; Fairburn et al., 2013; Jansingh, Danner, Hoek, & van Elburg, 2020; Pike, Walsh, Vitousek, Wilson, & Bauer, 2003). However, few patients have access to providers trained in these techniques (Cooper & Bailey-Straebler, 2015; Fairburn & Wilson, 2013). This factor highlights the critical need for a new model of treatment delivery to support individuals with AN in the post-acute period.

mHealth technologies have potential to increase access to high-quality services for post-acute treatment of AN by addressing treatment barriers (e.g., lack of providers, logistical barriers), but to date, research on these approaches, as well as use in the real-world, have been limited. However, prior work provides indication that a CBT-based mobile app for the post-acute intervention of AN could be useful. First, research supports the use of in-person CBT for post-acute care for AN. Among 33 patients with AN randomly assigned to one year of outpatient CBT or nutritional counseling following hospitalization, CBT was significantly more effective (Pike et al., 2003). Those receiving nutritional counseling relapsed significantly earlier (p<.004) and at a higher rate than the CBT group (53% vs. 22%). Second, digital interventions have demonstrated some initial success in supporting aftercare for AN. Fichter et al. (2012) evaluated a digital, CBT-based relapse prevention program over nine months after inpatient treatment vs TAU in 258 women with AN in Germany. In addition to online psychoeducational content, participants had access to an electronic message board, monthly moderated one-hour chat sessions, and asynchronous text-based support from a therapist. Results indicated the intervention group gained weight while the TAU group had weight loss (d=.22); intervention completers gained significantly more weight than participants in TAU (p<.05) (Fichter et al., 2012). Further indicating the promise of digital interventions, Neumayr, Voderholzer, Tregarthen, and Schlegl (2019) randomized 45 women with AN in Germany to an 8-week mobile intervention (German version of “Recovery Record,” an app for self-monitoring of eating as well as ED thoughts, feelings, and behaviors) with therapist feedback as an adjunct to TAU or to TAU alone. Patients reported high acceptance of the app, and non-significant small to moderate between-group effect sizes regarding body mass index (BMI) (d=−.24) and overall ED psychopathology (d=.56) were found favoring the intervention at post-intervention; however, there were no differences at 6-month follow-up (Neumayr et al., 2019). The intervention is currently being evaluated in a larger trial (Schlegl, Neumayr, & Voderholzer, 2020). Other ongoing work is currently examining the use of telehealth interventions for aftercare support for this population (e.g., Giel et al., 2021; Steinglass et al., 2022), with one of those efforts also including an online platform for facilitating between-session practice and monitoring (Steinglass et al., 2022).

These results suggest a digital, guided, CBT-based aftercare intervention for AN may have potential, but there are limitations to the existing work, including: 1) to date, no primarily app-based aftercare interventions have been developed in English; 2) use of older technology (e.g., long web-based chapters; Fichter et al., 2012) vs app-based intervention with ability to access the program in shorter bursts; and 3) even when new technology has been used (i.e., Recovery Record app), it has not included specific features for this population, such as helping foster motivation or support (Neumayr et al., 2019).

Indeed, social support has been found to be critical to recovery from AN (Dawson, Rhodes, & Touyz, 2014; Wetzler et al., 2020), with qualitative work revealing that motivation for ED recovery is especially fueled by supportive relationships (Linville, Brown, Sturm, & McDougal, 2012). However, many individuals working toward recovery from AN report feeling misunderstood by others, including health professionals, friends, and family (Dawson et al., 2014), and as such, peer support is critical (Wetzler et al., 2020). One scalable way in which individuals with AN could access support from peers is via social media (Aardoom, Dingemans, Boogaard, & Van Furth, 2014). Development of positive, recovery-focused online communities is particularly important in EDs given the existence of many pro-ED online communities that serve to connect people experiencing these problems and encourage disordered behaviors (Rouleau & von Ranson, 2011). Although pro-ED communities contain many harmful aspects, users perceive social support as a key function (Bohrer, Foye, & Jewell, 2020; Rouleau & von Ranson, 2011). In addition, other work has begun to report on the specific benefit of recovery communities on social media (e.g., Au & Cosh, 2022; Herrick, Hallward, & Duncan, 2021). These findings point to the importance of social support, particularly from peers, in the process of recovery from AN, including the need for a positive social networking outlet.

Current Study

In the current pilot randomized controlled trial (RCT), women with AN who have been discharged from acute treatment in the past two months (N=90) will be randomly assigned to a CBT-based mobile intervention plus TAU, a CBT-based mobile intervention including a social networking feature plus TAU, or TAU alone. Here, when we refer to TAU, we wish to clarify that participants will not be required to engage any type of care, aside from endorsement of having a physician who is monitoring their medical safety (eligibility criterion). We will examine feasibility metrics and collect data on acceptability of the mobile intervention and the social networking feature. We will also estimate preliminary effectiveness of the mobile intervention vs the mobile intervention plus social networking vs TAU. We hypothesize that the mobile intervention plus social networking will be most successful in reducing ED psychopathology, reducing frequency of ED behaviors, achieving weight maintenance, reducing depression and suicidal ideation, and reducing clinical impairment. We will examine rehospitalization and full recovery rates in an exploratory fashion. Next, we will examine whether the mobile intervention changes the theorized targets of the intervention (i.e., reduced dietary restraint and weight/shape concerns and increased motivation to recover), as well as whether the social networking feature increases social support and changes motivations for social media use, and whether changes in targets are associated with clinical benefit. We hypothesize that the mobile intervention will significantly change the targets, more so than TAU, resulting in greater clinical benefit. Finally, we will conduct exploratory analyses to identify within-app predictors (e.g., sessions completed) and moderators of outcome (e.g., length of illness, age, psychiatric co-occurrence).

Method

Participants

We will recruit 90 English-speaking women within two months of discharge from an acute setting (i.e., inpatient, residential, partial hospitalization, intensive outpatient) for treatment of AN from a number of ED treatment centers, including for-profit and academic medical centers, across the U.S. Tailored recruitment strategies will be developed and implemented with treatment centers and may include use of fliers (e.g., included in discharge paperwork, hanging in treatment setting), social media, and email. Inclusion criteria will also include participant endorsement of having a physician who is monitoring medical safety, identifying as a cisgender woman, being 18+ years old, owning a mobile phone, and having a BMI ≥ 17 kg/m². This trial is registered at NCT05499676. Institutional review board approval was obtained from Washington University.

Procedure

Following completion of a brief online eligibility screener, eligible participants will be invited for participation. If interested in moving forward, they will provide consent and complete the baseline assessment, after which they will be randomized. Combinations of the following conditions, conceived as potential moderators of outcome, will be used to create strata for randomization, based on prior research: length of illness +/− 3 years (Hay & Touyz, 2018), Clinical Impairment Assessment score +/− 32 (Reas et al., 2016; Welch et al., 2011), Patient Health Questionnaire-9 depression score +/− 10 (Kroenke et al., 2001; Kroenke & Spitzer, 2022), Generalized Anxiety Disorder-7 anxiety score +/− 10 (Spitzer et al., 2006), and age +/− 25 years (Wonderlich et al., 2020). Blocked randomization schedules will be created for each stratum to ensure balance among moderators. Assessments will occur online at baseline and at 6 weeks, 6 months, and 9 months post-randomization.

Control condition.

Participants assigned to the control condition, TAU, will be encouraged to follow the discharge plan provided to them by the ED program from which they were discharged. We will also encourage participants to follow up with their ED program and/or reach out to ED non-profit organizations in the U.S. for assistance with finding resources as needed. In particular, we will provide the websites for both the National Eating Disorders Association (NEDA) and the National Association of Anorexia Nervosa and Associated Disorders (ANAD), highlighting that both organizations offer helplines and online treatment provider databases.

Mobile intervention condition.

Participants randomized to the mobile intervention condition will receive access to our mobile app, hosted by SilverCloud Health. The intervention was developed using an extensive and iterative user-centered design process (Laboe et al., in preparation), with the initial version of the intervention based on two sets of existing, evidence-based materials. First, our team has successfully developed a guided self-help CBT-based mobile app, Student Bodies-EDs (SB-ED), which covers the core components of CBT for EDs (Fairburn, 2008), including targeting reducing dietary restraint and weight/shape concerns, and has been shown to be effective among college women with EDs (Fitzsimmons-Craft et al., 2020). Second, the app will be adapted to the unique needs of the target population by expanding and revising the content based on Pike et al.’s (2003) protocol, which includes a particular focus on enhancing motivation, developing skills, and supporting long-term recovery from AN. We will deliver the intervention as guided self-help given the robust literature to support the use of program guides in online CBT (Anderson & Cujipers, 2009; Kass et al., 2014; Spek et al., 2007), as well as given the severity of this population. The role of coaches will include supporting and enhancing user motivation, monitoring progress, facilitating goal setting and offering accountability, providing feedback on technique usage and encouraging practice, and monitoring for/responding to clinical risk. Communication with users will be delivered primarily within the app via two-way asynchronous messaging. Additionally, users will be offered optional, brief video chats at the beginning of the program and monthly thereafter, providing the coach an opportunity to facilitate greater personalization. Participants will have access to the app for 6 consecutive months and will additionally be encouraged to follow their discharge plan and engage with TAU.

Mobile intervention plus social networking condition.

Participants randomized to the mobile intervention plus social networking condition will receive access to our mobile app, as described above. They will also receive access to the social network, which will be delivered on Facebook and will provide a private community for users and coaches to interact through discussion threads. Like the app, the social networking component was also developed through an iterative user-centered design process (Laboe et al., in preparation). The group will be “secret,” meaning that no one outside of the group will be able to find it or know that someone is a member. To increase the likelihood that the social networking feature remains a safe space, users will be provided engagement guidelines, and the community will be moderated daily by coaches. The social networking feature will be designed to facilitate increased social support and also to facilitate changes in participants’ motivations for social media engagement (e.g., less appearance-focused). Participants will have access to the app and social network for 6 consecutive months and will additionally be encouraged to follow their discharge plan and engage with TAU.

Measures

Feasibility.

We will document recruitment rates (i.e., percent of those who are offered participation in the study and subsequently sign up) and study retention rates (i.e., percent follow-up completion) to assess appropriateness of follow-up procedures.

Mobile intervention and social networking engagement and acceptability.

Engagement data automatically captured within the app platform (e.g., sessions completed, number of messages to coach), along with engagement with the social networking component, will be documented and also examined as a predictor of outcomes. Participants in the mobile intervention conditions will also complete the Usefulness, Satisfaction, and Ease of Use (USE; Lund et al., 2001) questionnaire as a measure of usability. All participants randomized to one of the mobile intervention conditions will have additional items on their 6-month survey (end of intervention period) to assess their thoughts on acceptability of the mobile app and any suggestions on improvements for future versions. Similar questions about the social networking feature will be asked of those in that condition.

See Table 1 for clinical assessments and timepoints. Demographic information collected will include race, ethnicity, sexual orientation, income, insurance status, food security, and education level.

Table 1.

Clinical outcomes, targets, and other outcomes.

Outcome	Assessment	Time Points
		Baseline	6 weeks	6 months	9 months
Clinical Outcomes
ED psychopathology	Eating Disorder Examination-Questionnaire Global	X	X	X	X
ED behaviors	ED behavior frequencies from Eating Disorder Examination-Questionnaire	X	X	X	X
BMI/weight maintenance	BMI derived from height (self-report) and weight (measured via digital scale)	X	X	X	X
Depression and suicidal ideation	Patient Health Questionnaire-9	X	X	X	X
Clinical impairment	Clinical Impairment Assessment	X	X	X	X
Targets/Theorized Mechanisms
Dietary restraint	Dutch Restrained Eating Scale	X	X
Weight/shape concerns	Weight Concerns Scale	X	X
Motivation for recovery	Anorexia Nervosa Stages of Change Questionnaire	X	X
Social support	Online Social Support Scale + additional items	X	X
Social networking behavior	Motivations for Social Media Use Scale + additional items	X	X
Other Outcomes
Treatment utilization & rehospitalization rates	Treatment questionnaire	X	X	X	X
Full recovery rates	BMI, Stanford-Washington University ED Screen, Eating Disorder Examination-Questionnaire	X	X	X	X

Clinical outcomes.

For overall ED psychopathology (primary clinical outcome) and frequency of ED behaviors, we will use the Eating Disorder Examination-Questionnaire (EDE-Q; Fairburn & Beglin, 2008); overall ED psychopathology will be assessed using the Global score. For anxiety, we will use the Generalized Anxiety Disorder-7 (GAD-7; Spitzer et al., 2006). For depression and suicidal ideation, we will use the Patient Health Questionnaire-9 (PHQ-9; Kroenke et al., 2001). For clinical impairment, we will use the Clinical Impairment Assessment (CIA; Bohn & Fairburn, 2008).

We will calculate BMI at each timepoint as derived from self-reported height, which has shown to be accurate in adult ED populations (Meyer, Arcelus, & Wright, 2009), and weight assessed by a digital wireless scale, to determine weight maintenance. At each timepoint, a study coordinator will schedule a time with the participant to take their weight while they are on a video call. Participants will be instructed to take off their shoes, wear light clothing, and empty their pockets. Given the use of a wireless scale, the measurement will be automatically sent to the research team.

Targets.

We will assess target engagement (i.e., theorized mechanisms of the intervention) in all participants at baseline and 6 weeks. For change in dietary restraint, we will use the Dutch Restrained Eating Scale (DRES; van Strien et al., 1986). For change in weight/shape concerns, we will use the Weight Concerns Scale (WCS; Killen et al., 1984). For change in motivation, we will use the AN Stages of Change Questionnaire (ANSOCQ; Rieger et al., 2000; Rieger et al., 2002). For change in social support, we will use the Online Social Support Scale (OSSS; Nick et al., 2018), in addition to items developed for this study. The Motivations for Social Media Use Scale (MSMU; Rodgers et al., 2020), along with items developed for this study, will be used to assess whether the social networking component changes motivation for engagement in social media to be more adaptive.

Other outcomes.

We will assess treatment utilization at every timepoint including current treatment experiences, dates of treatment, primary type of treatment, and psychotropic medication use. At the baseline assessment, we will also collect data on all past treatment experiences and duration of illness. For determining recovery rates, at the 6- and 9-month assessments, we will apply criteria from Bardone-Cone et al. (2010) to determine participants who have achieved full recovery based on: 1) BMI > 18.5 kg/m²; 2) absence of binge eating, purging, and fasting in the past 3 months (assessed using the Stanford-Washington University ED Screen [SWED]; Graham et al. 2019); and 3) EDE-Q scores within 1 SD of age-matched community norms.

Analytic Plan

To test feasibility, we will compare recruitment rates to other trials of digital ED interventions and other trials of psychological interventions for AN following acute treatment. We will also compare study retention rates to those trials and to the benchmark of >70% completion of follow-up at every timepoint.

To test acceptability, we will compare app engagement to other trials of digital ED interventions, particularly in AN. We will also document engagement with the social networking feature. We will examine usability and acceptability ratings using descriptive statistics, aiming for high usability and satisfaction.

To examine clinical outcomes, we will use intention-to-treat (ITT) analyses, using full-information maximum likelihood to handle missing data. The principal strategy to examine clinical outcomes over time, accounting for baseline levels, between the three groups will be the use of mixed random effect repeated measures models, with participant as a random effect and with treatment group, time, and treatment by time interaction as fixed effects. Use of other treatment will be assessed and modeled as a covariate to control for differences in the amount of other treatment between participants. Linear mixed models will be used to compare ED psychopathology (primary clinical outcome), BMI, depression, suicidal ideation, and clinical impairment. For frequencies of ED behaviors, we will use a generalized linear mixed model with a log link, and for rehospitalization and full recovery rates (binary outcomes), we will use a logit link. These models will allow us to test whether intervention effects in outcomes are observed at 6 months (i.e., end of intervention period) and whether the obtained effects are sustained during subsequent follow-up at 9 months (i.e., 3 months after the intervention period). Effect sizes and 95% confidence intervals will be reported. We note that even if the intervention is generally helpful overall for participants, it is possible it could be counterproductive in some individuals; our analyses will attempt to identify such cases.

To test whether the mobile intervention, and the mobile intervention plus social networking, will significantly change the targets (i.e., reduced dietary restraint and weight/shape concerns and increased motivation to recover), more so than TAU, resulting in greater clinical benefit, we will use moderated mediation analysis. This will allow us to examine whether clinical targets affect our outcomes of interest (particularly our primary clinical outcome, ED psychopathology), and whether these effects differ by intervention group. In the case of targets that have overlap with the assessment of ED psychopathology (i.e., dietary restraint and weight/shape concerns), we will use a modified EDE-Q Global score that removes the overlapping subscales (i.e., Restraint and Weight and Shape Concern, respectively). We will use a similar approach for testing the proposed social networking targets. We will test for mediation using structural equation models (SEM) with confidence intervals derived from bootstrapping of indirect and total effects.

Finally, we will conduct exploratory analyses to identify within-app predictors (e.g., sessions completed) and moderators of outcome (e.g., length of illness, age, psychiatric co-occurrence, baseline smartphone and social media usage). Potential moderators will be examined using interaction terms with intervention group assignment in models described above to identify subgroups that benefit most from the intervention (with or without the social networking feature) and others for whom additional tailoring may be necessary in future refinements. Assessment of within-app treatment predictors will use the models described above but will include each predictor represented as a covariate and predictor by time interaction.

Power

The primary purpose of this study is to examine the feasibility and acceptability of our intervention. Nevertheless, we have examined power for the clinical outcome of ED psychopathology, and a sample of n=30 per group will provide 28% power when the true EDE-Q Global between-group mean difference is 0.5 and 80% power when the difference is 1.05.

Conclusions

This pilot study will lay critical groundwork for our longer-term goals of developing and disseminating an effective mobile app to be used among individuals with AN in the post-acute intervention period. Information on feasibility, acceptability, and estimates of effect sizes for clinical outcomes will aid in the planning of a larger RCT. This study may also identify key moderators of effects, which can be addressed in future iterations. In our next-stage trial, we may also test effectiveness of the app among not only those with AN who have recently been discharged from intensive treatment but also among those who have not yet been in such a care setting but who do not have access to in-person providers. These aims and research questions will naturally extend this work and will bring us closer to achieving our goal of increasing access to evidence-based treatment for individuals with AN and ultimately preventing the extreme negative consequences that can result from this serious disorder.

Public Significance.

Relapse after acute treatment for anorexia nervosa is common, and few patients have access to trained providers to support them following acute care. This study will pilot a coached mobile app, including a social networking component, for this population. If ultimately successful, our approach could greatly increase access to evidence-based treatment for individuals with anorexia nervosa and ultimately prevent the extreme negative consequences that can result from this serious disorder.

Data Availability Statement

This article describes a study protocol, and thus, no data are available for this study at this time.