The pilot study described above informed the design of this study as follows: (1) training clinicians in private practice—clinicians who treat a high proportion of youth with eating disorders but were not well represented in the pilot study and often do not receive training in EBTs, (2) refining materials on externalization and agnosticism in 1 training arm (ET-FBT) to confirm whether the preliminary data suggested that fidelity would improve with additional training in these key components of FBT, (3) incorporating strategies to improve retention by adding continuing medical education (CME) incentives (especially important for those in private practice who do not have institutional support for training), and (4) improving methods for capturing patient weight outcomes using therapist-deidentified reports because obtaining patient consent to collect these data, as attempted in the pilot study, was not feasible (too few patients consented) or effective (insufficient data were collected on outcomes to evaluate the impact of fidelity). This online training study was a longitudinal study with 3 primary assessment time points (baseline, end of online training, and end of CCC). A detailed description of the study protocol has been previously published [35].

This study was conducted entirely virtually with no physical site locations. Private practice clinicians from across the United States were invited to participate. Recruitment and data collection occurred from November 2020 to October 2024.

Therapists were required to be licensed (Masters, PhD, PsyD, and MD [psychiatrist]) in the United States and confirm that they regularly treated adolescents with AN in the context of a private practice and that they had not previously been trained in FBT. Methods of recruitment included—social media posts, professional organizations, and listservs (Twitter [subsequently rebranded X], Academy of Eating Disorders, Eating Disorder Referral and Information Center, and licensing organizations). Interested therapists were directed to an online questionnaire for further study information and screened for eligibility. Prior to online training, eligible therapists reported weight change from sessions 1 to 4 of an adolescent with AN they had recently treated using any treatment other than FBT. This weight change would be compared to weight change by session 4 of FBT post–online training during CCC to explore whether training affected early weight gain. Baseline assessments were conducted prior to training allocation and were completed online. After completing assessments, therapists were randomized to either ET-FBT or WT using block randomization. The training period for each group lasted approximately 3 months and was followed by 3 months of weekly CCC.

Detailed descriptions of the measures and related procedures used in this study can be found in a previously published protocol paper related to this study’s design [35]. Therapists completed the measures detailed below at baseline, at the end of online training, and at the end of CCC. In addition, the consultants who conducted CCC completed the Supervisor Report of Fidelity to Key Components of FBT at the end of 12 hours of CCC [29,30]. Consultants were faculty members at one of the participating sites in this study. Both consultants were trained in FBT and certified by the Training Institute for Child and Adolescent Eating Disorders as consultant supervisors. The consultants were masked to the participant training condition.

The list of measures used in this study is as follows:

The sample size for this study was determined to facilitate our investigation focusing on improvements in therapist outcomes, including self-efficacy, working alliance, and fidelity. These outcomes were not primary in this study, although they were critical for planning and enriching the next step of the confirmatory efficacy trial. We estimated power based on the proposed linear mixed effects (LME) modeling with repeatedly measured outcomes (baseline [pretraining], end-of-training [EOT], and follow-up [end of CCC]). We assumed a modest intraclass correlation of 0.5 for the repeated measures, with 50% attrition by follow-up. We also assumed a medium effect size (Cohen d=0.4-0.5) as a clinically meaningful minimum effect size. Under this scenario, with the expected combined sample of 140 clinicians at pretraining, the estimated power to detect significant improvement in therapist competency from baseline to follow-up is 0.98‐1.0 (2-tailed; α=.05). In terms of retention, our primary outcome, we set at least 50% retention by EOT and at least 50% retention by follow-up among those who start CCC as a success. Assuming about 65% retention at EOT, using a 1-sample chi-square test, the estimated power to achieve retention significantly better than 50% is 0.95. For our exploratory investigation, including group comparisons (ET-FBT vs WT) in therapist and patient outcomes and detecting baseline moderators of group effects, we planned to monitor both clinical significance (ie, effect sizes) and statistical significance (P value), with more emphasis on identifying clinical significance.

The 2 online training strategies, as noted in the “Study Size” section, are described more fully in a study by Citron et al [35], but a brief description here provides a summary of the main features of the 2 trainings. The WT arm includes an online webinar lecture series. In this arm, participants watched a weekly video lecture (approximately 1 h) created by authors JL and DLG, based on the Treatment Manual for Anorexia Nervosa: A Family-Based Approach [39]. Participants had 1 week to view the lecture, and the materials were not available for review. The lectures discuss the scientific evidence supporting FBT, how therapists set up treatment for FBT, the main interventions used in FBT during each phase, and recorded role-plays illustrating interventions throughout the 3 phases. The Enhanced Training (ET-FBT) arm uses an online training platform that releases 1 of 7 lecture bundles every 12 days. Once released, participants had access to the training material throughout their participation in the study. The ET-FBT lecture bundles consisted of 5 to 8 short didactic videos that discuss the treatment model and provide mock therapy session video clips (modeling FBT with a typical adolescent with AN case), supplementary readings (such as the Treatment Manual for Anorexia Nervosa: A Family-Based Approach [39]), clinical vignettes and session roleplays, and training assignments. Participant assignments included questions on the medical management of adolescent patients with AN, reflection on clinical vignettes and the therapist’s approach, and identification of treatment moderators. ET-FBT included additional focused training related to the key components of FBT—agnosticism and externalization. Brief online didactic lectures defined the 2 principles and discussed their uses, challenges, and application in clinical practice. Participants were provided with clinical vignettes on the application of agnosticism and externalization. Both trainings lasted approximately 12 total hours.

Eligible therapists were consented using a protocol approved by the institutional review board at Stanford University and the University of California, San Francisco (e-protocol 56548). All participants gave their informed consent for inclusion before participating in the study. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the 1975 Declaration of Helsinki, as revised in 2008. Therapists did not pay for training, but to enhance retention, they received CME credit for completing the training and CCC. Participants were compensated US $100 for each assessment time point (US $300 total). All reports are private and confidential per the institutional review board–approved protocols.

To estimate improvement in clinician outcomes, we used linear mixed effects (LME) modeling [40,41] with random intercepts and maximum likelihood estimation, which enabled us to fully use outcome data repeatedly measured for most clinicians (3 assessment time points). Given the 2 distinct phases in our study design (baseline to EOT and EOT or initiation of CCC to completion of CCC), we used piecewise growth models that allowed different slopes from baseline to EOT and from EOT to follow-up. We used the training group (ET-FBT vs WT) as the predictor of changes across the 2 phases. The piecewise LME model we used here is presented below.

We describe our general model for longitudinal LME modeling in 2 levels. The outcome Y for therapist i at time point t can be expressed as follows:

which is a function of 3 growth parameters:

where i=1, 2, . . . , N (N=123 in our study) and t=0, 1, 2 (baseline, EOT, and follow-up). Therapists were randomly assigned to 1 of the 2 conditions (Z_i=1 if ET-FBT, Z_i=0 if WT). We modeled the longitudinal trend using 3 growth parameters, the initial status (η_0i) and the changes from the baseline to EOT (η_1i) and from EOT to follow-up (η_2i). We used time scores of S_1t (=0, 1, 1) for the first and S_2t (=0, 0, 1) for the second slope. The fixed effects of interest are the mean initial status (α₀ under WT, α₀+γ₀ under ET-FBT), the changes in each treatment group (α₁, α₂ under WT, α₁+γ₁ , α₂+γ₂ under ET-FBT) and the group differences in these changes (γ₁and γ₂). We allow for individual variability at baseline (ie, random intercept), which is captured by the variance of level 2 residual (Σ_ζ=Var(ζ_0i)). We assumed the random effect to be normally distributed (ζ_0i~N(0,Σ_ζ)). We used the standard assumption for the level 1 residual, which is normally distributed and consistent across assessments (ε_i~N(0,σ²)).

In our LME modeling, data points that were missing due to clinician attrition or intermittent dropout were handled under the assumption that data are missing at random [42] conditional on observed information, which is a standard assumption in modern missing data approaches. In this procedure, all available cases, including those with missing information, were included in the analysis. For maximum likelihood estimation of mixed effect models, we used the lme4 package [42] in R software (R version 4.4.1; R Foundation for Statistical Computing). For the exploratory patient outcome, early weight gain by session 4 (restoring 2.4 kg or not), we planned to report simple proportions at baseline and follow-up as the information is not collected repeatedly from the same individuals. We also explored the number of years working in private practice and therapists’ age at baseline as possible treatment effect moderators and/or predictors of therapist outcomes and patient early weight response.

Participants were approximately 10% Hispanic, 9% non-White, and 96% female, with a mean age of 36.9 (SD 8.8) years (Table 1). Most were master’s level clinicians (77%), followed by psychologists (23%) and psychiatrists (1%). Recruitment and data collection occurred from November 2020 to October 2024.

We randomized 123 patients (88% of the expected number; Figure 1) over the course of 3 years. Participants were recruited from 33 states, with the largest percentage practicing in California (n=26, 21%), Virginia (n=11, 9%), New York (n=8, 6.5%), North Carolina (n=8, 6.5%), Michigan (n=8, 6.5%), Texas (n=7, 5.7%), and Florida (n=7, 5.7%). A total of 20 states (Alabama, Alaska, Arkansas, Colorado, Connecticut, Idaho, Indiana, Iowa, Maryland, Massachusetts, Minnesota, Missouri, Montana, Nevada, Ohio, Oregon, Rhode Island, South Carolina, Washington, and Wisconsin) had 3 or fewer participants. A total of 123 clinicians started training, and 117 (95%) completed it. Of those who completed online training, 89 (72%) planned to start CCC, but only 54 (44%) found an eligible patient and started CCC during the allotted timeline for study completion.

There was evidence of considerable improvement in the training outcome measures within both treatment arms (Table 2). In TVAN and fidelity self-ratings, both arms showed similarly large effect sizes (Cohen d of 1.0-1.9) in terms of changes from baseline to EOT and follow-up (end of CCC), leading to insignificant group differences. In WAI, both groups showed similar modest but significant improvements from baseline to EOT. However, by follow-up, the ET-FBT condition showed significantly more improvement (Cohen d=0.53; P=.04) in working alliance (WAI). Expert fidelity ratings at the end of CCC, provided by the consultants who were blind to the training condition, did not suggest a difference between the groups (WT=21.3 (SD 4.21) and ET-FBT=19.3 (SD 6.93); Cohen d=−0.35, 95% CI −0.94 to 0.24; P=.25). These expert ratings of fidelity were somewhat lower than self-reported fidelity but were significantly correlated with each other (r=0.33; P=.03).

The primary patient-related outcome for this study was early weight gain of 2.4 kg by session 4, used as a proxy predictor (surrogate marker) for end-of-treatment weight restoration [32-34]. The percentage of early weight gain of 2.4 kg by session 4 for the sample at baseline (prior to online training) was 15.4% (19/123). At follow-up, the percentage of patients who achieved early weight gain of 2.4 kg by session 4 more than doubled in both training approaches to a rate of 34% (33% under ET-FBT; 35% under WT) compared to the rate at baseline. This rate of early response is almost identical to that achieved by trained and supervised FBT therapists participating in RCTs [43].

We examined the number of years working in private practice and therapists’ age at baseline as possible treatment effect moderators and/or predictors of self-reported fidelity. The number of years working in private practice did not show a significant association with self-reported fidelity either at EOT or follow-up. Therapist age showed little or negative association with self-reported fidelity in the WT arm (EOT: r=0.13, P=.36; follow-up: r=–0.39, P=.048), whereas it showed a positive association with fidelity in the ET-FBT arm (EOT: r=0.32, P=.03; follow-up: r=0.26, P=.25). As a result, at follow-up, the moderating effect of age was significant (r difference=0.65; P=.04), which can be interpreted as older therapists benefiting more from the ET-FBT condition. We did not find any significant associations between baseline therapist characteristics and patient early weight response.

The results of this study found that both online training strategies were acceptable and feasible to conduct with clinicians in private practice. They led to improvements in self and expert reported fidelity to the key components, self-efficacy related to the delivery of FBT, and working alliance. Furthermore, both training strategies also appeared to lead to improved patient outcomes, more than doubling the early response rates compared to the rates of those treated prior to FBT training. This is a rate that approximates what is achieved in RCTs with expert FBT clinicians [43]. The only difference between the two training approaches was that the WAI showed significantly greater increases in the ET-FBT group, which may be attributable to baseline differences in this measure. Thus, overall, our hypothesis that ET-FBT would outperform WT was not supported. While this result was not as expected, it is important that both training programs led to fidelity to FBT key components and patient clinical improvements. This supports the promise that online training of therapists in FBT is likely a viable way to increase the supply of therapists who can deliver the treatment with reasonable fidelity, with an expectation that their patients will experience substantial benefit. These results are in line with previous nonrandomized studies of online training for clinicians treating eating disorders, where significant improvements in skills and confidence were noted for clinicians learning CBT or interpersonal psychotherapy [8,9]. Taken together, these studies support the conclusion that online training is likely a feasible and practical strategy for training clinicians in the main evidence-based approaches for eating disorders.

The recruitment process was hampered in part by the COVID-19 pandemic, resulting in only 123 participants (88% of our original aim of 140) enrolling. This was because many therapists reported being overwhelmed by referrals and therefore had no time for training [44-46]. Nonetheless, interest in receiving online training in FBT was high, with 1143 expressing interest, 426 completing the online screening, 251 being deemed eligible after screening, 123 starting the training, and 117 completing it (95%), which was much higher than the 50% we had anticipated based on our pilot studies. However, in this study, of those who completed the online training, 89 (72%) expressed interest in receiving CCC, but only 54 (44%) of these clinicians were able to find an eligible patient and start CCC within the allotted timeline for study completion. This was lower than the 50% we had anticipated. At the same time, the majority of those who found an eligible patient and started CCC completed the 12 hours offered (n=47, 87%) and provided patient outcome data (ie, patient weight change from sessions 1‐4). This level of obtaining patient outcome data was much greater than the 10% we were able to collect in the pilot study, where obtaining patient consent was a major barrier [28]. Overall, the acceptability of study procedures was high, with rates of completion of study measures at the EOT reaching 79% (48/61) for ET-FBT and 82% (51/62) for WT, and with an overall rate of assessment completion of 80% (99/123), with no statistical differences between randomized groups.

As noted above, it was somewhat surprising and contrary to our hypothesis that additional training in agnosticism and externalization did not improve fidelity. We speculate that, while the didactic content in the ET-FBT group may have improved intellectual understanding of these interventions, participating in CCC provided additional learned application of these skills in a clinical case, which leveled up fidelity in both groups. Alternatively, it may be that the preliminary studies overestimated the benefits of these additional training modules and that standard training is already sufficient for clinicians’ understanding and use of agnosticism and externalization.

While the EOT results were comparable between groups, the predicted attrition of 50% posttraining participation in CCC was evident in this trial. However, a large proportion of those who did not complete CCC were unable to start (38%) because they could not find an eligible patient in time. As suggested above, this may be a result of the fact that much of the study took place during COVID-19, when there was a significant uptick in eating disorders, and therapists treating eating disorders were overwhelmed with referrals and did not have the time or energy for training. However, a more important factor might have been the study eligibility requirement that patients treated in the study be underweight at the start of FBT, which precluded the enrollment of adolescents who were partially weight-restored in other settings or who had atypical AN without low weight. Many clinicians in private practice reported seeing mostly patients who were weight-restored in other settings (hospitals, day programs, and residential programs). Of those who did complete CCC, there was an early weight response by session 4 of FBT after only 4 hours of CCC. At the same time, those who completed all 12 hours of CCC saw continued improvement in fidelity and self-efficacy (TVAN) in both training arms, as well as WAI in the ET-FBT arm (Table 2).

Because early weight response is a strong predictor of end-of-treatment weight restoration, these data provide support that training plus CCC led to clinical implementation effects on patient outcomes similar to those expected in RCTs. The early weight response did not differ by treatment group (odds ratio 0.99, 95% CI 0.67-1.47). However, the study design did not include the collection of patient outcome data beyond session 4 of FBT. This was because weight gain at this early time point is a reliable predictor of end-of-treatment remission, with a predictive value of 70% to 80% in RCTs [32], and also because the study’s funding duration was limited to a 3-year window. However, by not including non-low weight adolescents with AN in this sample, we cannot know for certain whether session 4 weight gain will be a predictor of end-of-treatment response for this clinical group, and future studies that include all adolescents with AN, regardless of starting weight, are needed to clarify this.

This study was designed using a practical approach to real-world challenges related to future training and implementation procedures, as well as clinical outcome measurement. As a result, we focused on clinician-reported data as a feasible approach to diagnostic assessment, fidelity, and patient weight outcomes. Therefore, the clinician-participants identified eligible patients in their practice (ie, diagnosed with AN both prior to and during the training period). The clinician participants also reported weight at baseline and session 4 for the identified patients they treated or would treat during the training period. These design features allowed for real-world evaluation of training and patient outcomes but also resulted in limitations regarding related findings resulting from self-report. It is possible that some clinicians inaccurately diagnosed or reported inaccurate weight change data. However, there was no evidence or any clear motivation for false reporting, and there were no differences in the changes between the 2 online training platforms, suggesting that if these inaccuracies were present, they were equally distributed between the randomized groups.

This study used a standard dose of CCC post training because studies suggest that CCC post training is a necessary aspect of mastering the skills described in a didactic training [31,47,48]. However, CCC is a challenge to scale because it is costly, and there are limited experts to provide it. A next step would be to systematically examine cost-effective strategies for CCC, including determining more specifically how much CCC is needed to ensure fidelity and clinical outcomes, as well as strategies for more efficient procedures of CCC, including potentially group approaches. This is an important area for future investigations. Furthermore, it is unknown if the additional 8 hours of CCC beyond the first 4 had any effect on patient outcomes. CCC post online training is likely necessary to implement EBT effectively [31], so future studies should identify how many hours of CCC are needed in addition to the online training to develop an evidence-based and scalable FBT training model combining these two essential elements.

Another limitation of this study design is the exclusive focus on private practice and majority White clinicians, thereby limiting generalization to clinicians of other ethnicities or races and those who practice in other settings. However, there is little reason to expect that clinicians in other clinical settings or even those outside the United States would not respond similarly, especially given our previous pilot study [28]. We used training incentives, specifically CME, to encourage clinicians in private practice to attend training. We did this because clinicians in private practice need CMEs for continued licensure and to help mitigate the loss of income related to training. Clinicians in public practice settings or those with organizational support may not need this type of incentive. Thus, for clinicians in these other settings, costs to provide training would likely be lower without the costs of CME included in the training.

By design, most of our outcome measures are self-reported, and additionally, some do not have psychometric studies documenting their reliability. However, this study used measures that were short and scalable so that they had viability in a real-world clinical context where the training of therapists and evaluation of fidelity must be efficient and practical. Thus, while the measures have only face validity, they appear to capture self-reported change on key variables related to fidelity and self-efficacy in the delivery of FBT. Importantly, the key measure of therapist self-reported fidelity was correlated with expert consultant fidelity ratings, suggesting that self-report may be a reasonable and practical strategy to assess fidelity in a clinical setting, but this requires additional study. In addition, future studies should determine the psychometric properties of these fidelity measures with greater precision.

Finally, although patients treated during CCC more than doubled the rates of early response (15% vs 34%) from pretraining to posttraining, supporting an important patient outcome change effect, there was no comparison group that did not receive CCC posttraining; therefore, it is not possible to determine if and how much additional CCC is needed to improve early weight gain.

This study demonstrates that it is feasible, acceptable, and practical to train therapists in private practice in FBT-AN using 2 different online strategies, with little evidence that one has advantages over the other. It is important to note that the main reason participants were excluded from training was that they were not in private practice or could not find an eligible patient with baseline weight data, rather than due to a lack of interest, as 59% of eligible participants enrolled and most of those (95%) completed online training. In addition, the outcomes of the study suggest that there were improvements in all measured training and fidelity metrics, as well as patient early response rates. These data, taken together with our previous studies, suggest that asynchronous online FBT training could be offered to clinicians in private practice at low cost using either of the approaches tested in this study. These training materials can be distributed easily through learning platforms and translated into non-English versions using relatively inexpensive artificial intelligence translation technologies. However, several key questions remain: What are the benefits of adding CCC to online training, and how much CCC is needed to achieve fidelity to the key components of FBT and to achieve expected patient outcomes? Challenges in the uptake of CCC were evident in this trial, and additional strategies to increase uptake and completion of CCC are needed. Because CCC is potentially costly and the availability of expert consultants is limited, it is important to clarify for individual providers and health care systems an expected effective strategy. Making CCC as short as feasible and associated with a predictable outcome would likely encourage more therapists to take up and complete CCC.

This study portends a new era for FBT training availability, which should, in turn, lead to an increase in trained and competent providers, as well as improved access for young persons with AN and their families to an important evidence-based intervention for this potentially lethal disorder.