For this counterbalanced, randomized, cross-over designed study, the CONSORT guidelines (Eldridge et al., 2016) were followed to examine the pilot feasibility, acceptability and efficacy of the 2-AFC CBM transdiagnostically for adolescents. The study protocol was approved by the ethics committee of the medical faculty of the Ruhr-University in Bochum (Reference number: 207134). Informed consent was obtained from all patients and caregivers.

Patients were recruited from the inpatient setting of a university hospital for child and adolescent psychiatry in Germany. The inclusion criteria required patients to be female, aged 13–17 years and diagnosed with depression (n = 17) or (atypical) anorexia nervosa (n = 12) according to the International Classification of Diseases, 10th Revision (ICD-10, Weltgesundheitsorganisation, 2022) as is standard in the German healthcare system. The ICD-10-based clinical diagnoses were applied by experienced clinicians. Diagnostic procedures were routinely conducted whenever symptom-related abnormalities were observed, and comorbid conditions were systematically assessed to examine potential exclusion criteria. Depression was selected as one diagnostic group because it is a frequent disorder in adolescence, studies highlight the role of body image dissatisfaction, and therapeutic options addressing this aspect remain limited. Other diagnostic groups with transdiagnostic body image problems were either not available in sufficient numbers at the study site or were too acute for participation. (Atypical) Anorexia nervosa was chosen for the eating disorder group because body image symptomatology is particularly pronounced in this disorder and as the most frequent eating disorder requiring inpatient treatment, it represented the largest eating disorder cohort in the clinic. Atypical anorexia nervosa is defined as an eating disorder (ED) that meets almost all criteria for anorexia nervosa. In the present sample, n = 7 patients were diagnosed with anorexia nervosa and n = 5 with atypical anorexia nervosa; three of the ED patients also fulfilled criteria for comorbid depression. In these cases, anorexia nervosa was considered the primary diagnosis and served as the distinguishing criterion separating the ED group from the DEP group. All screened and included patients in the DEP group had moderate depression, the most diagnosed level in Germany (Stahmeyer et al., 2022). Patients with acute suicidality, severe depressive symptoms, BMI > 25 kg/m², autism spectrum disorder (ASD), IQ < 70, schizophrenia or other psychotic disorders, uncorrected vision impairments or insufficient German language skills of the patient or caregivers were excluded. These criteria ensured focus on the relevant diagnoses and avoided ethical concerns regarding experimental participation.

Power analysis (G*Power; Faul et al., 2009) based on data from previous studies (Gledhill et al., 2017) indicated a required sample size of n = 24 (d = 1.04, α = 5%, power = 80%). To account for potential dropouts, the target sample size was set at n = 30. Figure 1 shows the corresponding CONSORT flow chart.

This study used a 2-AFC/CBM paradigm by Gledhill et al. (2017). Pre-study interviews showed adolescents compared themselves to young adult bodies, so computer-generated images of a young adult in underwear were used. Fifteen photorealistic avatars with BMIs from 15 to 33.7 were shown, each presented three times to establish patients’ individual baseline by categorizing each avatar as “thin” or “fat” using their keyboard. Patients were randomized into two groups (A/B) in a counterbalanced design, receiving intervention and control conditions in different orders. Group A received the intervention first and then the control condition, while Group B received the control condition first and then the intervention. Training occurred over two four-day blocks (Days 1–4 and 15–18), with daily pre- and post-baseline assessments and six blocks of 31 trials. Avatars with extreme BMIs appeared once per block, while moderate BMIs appeared more frequently, based on prior research (Gledhill et al., 2017). In addition to baseline measures before and after each training session, questionnaires were administered at the beginning of the training and 10 days afterward (day 1, 15, 29). Day 15 thus served both as the 10-day follow-up after the first block and as the baseline of the second training block. In the intervention condition, feedback aimed to shift the categorical boundary (Catbound) two BMI points higher than baseline, correcting “old” boundaries with messages like “Wrong! This body was thin” or “Wrong! This body was fat.” In the control condition, participants completed the same assessments as in the intervention condition, but the confirmatory feedback was aligned with each patient’s baseline categorial boundary.

Patients were randomized 1:1 using the chit method (Singh, 2006) by study staff, with results securely stored for allocation concealment. Patients were blinded to their intervention group due to baseline-based feedback, but study staff were not blinded.

Statistical analyses were performed using SPSS 28. The sample was characterized using descriptive statistics and following the Intention-to-treat protocol, missing data were imputed by means of linear regression models. For each variable with missing data, a regression model was specified in which significant correlations between the available questionnaire scores, missing questionnaire scores, and Catbound values served as predictors to preserve the multivariate structure of the data.

To check the assumption of negligible carryover effects pre-tests were conducted using Welsh t-tests and Mann–Whitney U-tests comparing within-subject sums of the results from both periods (Wellek and Blettner, 2012). Treatment effects were tested with Welch t-tests or Mann–Whitney U-tests based on parametric conditions, comparing within-subject outcome differences between periods. A mixed ANOVA analyzed the effect of the 2-AFC task with corrective feedback on the BSQ, using baseline as covariate, time as within-subject factor, and diagnosis and cross-over group as between-subject factors. The direct effect on general psychopathology was examined with multiple mixed ANOVAs using the same corresponding covariate and factors. Differences between each training day’s post-baseline and the pre-baseline were plotted against zero by diagnosis and cross-over group. Linear regressions examined the relationship between post-training Catbound values and follow-up questionnaire scores by diagnosis. Treatment effects, including 95% confidence intervals, p-values, and Cohen’s d effect sizes, were calculated separately for each diagnosis using Welch t-tests or Mann–Whitney U-tests. Non-parametric effect sizes were converted following Dunlap (1994).

Table 1 summarizes the sample characteristics of the DEP and AN group before the trial and the Welsh t-test referring to pre-measurement comparisons.

The groups differed in relation to age and BMI, with the AN group presenting younger and with lower BMI’s. As consistent with their diagnoses, patients with AN showed significantly higher scores than patients with DEP on BSQ and ANIS. Furthermore, the level of DEP measured with the PHQ was comparable between the groups. The later indicating severe symptoms of DEP in both groups. Baseline comparisons between the randomization groups revealed no significant differences. When separating by diagnosis, no differences emerged within the depression group, whereas in the eating disorder group a small baseline difference was found for the BSQ (p = 0.049).

No statistically significant effects of the 2-AFC task on body image disturbance were detected, with the BSQ showing non-significant interactions between time and baseline F(1,24) = 0.523; p = 0.477, time and crossover F(1,24) = 0.3; p = 0.865 and time and diagnosis F(1,24) = 0.134; p = 0.717. The psychometric task performance means are presented in Table 2.

Table 3 shows the direct effect of post-training Catbound values on the respective follow-up questionnaire scores, highlighting a significant training impact on ANIS and BSQ at Follow-Up 1 in both groups and at Follow-Up 2 in the DEP group.

Change of Catbound: In addition to the significant within-subject interaction of time*crossover described above, a significant between-subject effect of diagnosis was found during training, F (1,24) = 6.957, p = 0.014, ηp² = 0.225, and at follow-up, F(1,24) = 6.596, p = 0.017, ηp² = 0.216. Figure 3 displays the difference between post- and pre-baseline for each training day, plotted against zero and separated by diagnosis and crossover group, with mean values in Appendix Table A2. The above analyzed training effect was observed in both diagnosis groups, pronounced in Group A (intervention first), where post-baseline increased. In Group A (DEP and AN) and particularly in Group B (AN), post-baseline decreased during the control condition, with this decrease accumulating over the 4 days of training. Patients with AN in group B showed the lowest post-baseline increase during the following intervention. The confirmed personal boundary (control condition) corresponded to a BMI of 24.65 kg/m² (Group A, DEP), 23.42 kg/m² (Group B, DEP), 22.17 kg/m² (Group A, AN), and 22.79 kg/m² (Group B, AN). Thus, patients with DEP classified most normal-weight avatars as thin, while patients with AN corrected avatars ≥22 kg/m² as “fat.”

Welch t-tests/Mann–Whitney U-tests showed significant treatment effects for both diagnoses with high effect sizes during training (Post 4/8: DEP t(14.96) = 2.706, p = 0.016, d = 1.31; AN t(8.05) = 2.852, p = 0.021, d = 1.57) and at follow-up (Pre 5/9: DEP U = 14.5, Z = –2.092, p = 0.036, d = 1.19; AN U = 3.5, Z = –2.185, p = 0.028, d = 1.62).

Change of disorder specific psychopathology: A treatment effect was found in the DEP group for the PHQ, t(13.58) = −2.434, p = 0.029, d = −1.198. No significant changes were observed in the AN group. Table 4 presents the treatment effect analysis results for questionnaire and Catbound changes by diagnosis.

The 2-AFC task significantly shifted the categorical boundary for both diagnostic groups, with a stronger effect during training than at follow-up, indicating a decrease over time. Preliminary studies in adults showed that body perception can change through feedback and repetition (Gledhill et al., 2017; Irvine et al., 2020). This study replicated these findings in adolescents, demonstrating the task’s feasibility in this age group. However, the control condition had a non-neutral effect, not observed in adult studies (e.g., Gledhill et al., 2017). Adolescents are considered more suggestible than adults (Zimmermann et al., 2022) and emotionally vulnerable during puberty due to hormonal and brain changes (Konrad et al., 2013; Herpertz-Dahlmann et al., 2013), which may explain this.

Particularly in the AN group, the control condition with confirmatory feedback negatively shifted the boundary toward a lower BMI, accumulating over time and weakening the subsequent intervention effect in Group B. In the DEP group, the control condition had a positive effect with no weakened intervention effect. Patients with DEP had a higher boundary, labeling most normal-weight avatars as thin, while patients with AN labeled avatars ≥22 kg/m² as “fat,” confirming and reinforcing their negative body image during the control condition. This may explain the different feedback effects between diagnoses.

The influence of “fat talk,” often underestimated, contributes to body dissatisfaction (Curtis and Loomans, 2014). The results emphasize the relevance of critical feedback. Feedback should serve as orientation and confirm a healthy normal weight limit, especially for suggestible patients who suffer from body image distortion.

Previous studies indicated that modification of the categorical boundary is associated with body dissatisfaction (Cornelissen et al., 2022) and can be shifted through training programs (Gledhill et al., 2017; Irvine et al., 2020). In the present study, hypothesis-specific analyses examined this effect and showed multiple direct training effects on disorder-specific characteristics. In the AN group, this effect was limited to the first follow-up, likely due to the increased influence of the control condition for this diagnosis. A mixed ANOVA of all measurement points found no significant improvement in body image disturbance. This may be due to the control condition’s influence preventing significant improvement. Studies with adults showed reduced body shape and weight concerns (Gledhill et al., 2017; Irvine et al., 2020), including preclinical samples or patients with AN without a control group after leaving the intention-to-treat protocol (Gledhill et al., 2017). Thus, previous studies using other study designs had no negative effects of the control condition on the training effect. Another difference to previous studies is the timing of the intervention during hospitalization when patients had severe symptoms and cognitive impairments typical of underweight patients with AN (AWMF S3-Leitlinie Diagnostik und Therapie der Essstörungen, 2018). Improvement in these symptoms typically occurs only with weight normalization (AWMF S3-Leitlinie Diagnostik und Therapie der Essstörungen, 2018). Thus, the lack of improvement may be due to the intervention starting too early, with an average baseline BMI of 18.12 kg/m². Later training at a higher weight or more sessions may yield more consistent, clinically relevant results.

This is the first study to examine CBM for body image in adolescent inpatients with DEP, comparing results to patients with AN. Both groups benefited, though the control condition had a more negative effect in patients with AN, who did not benefit more from the intervention, as expected. As in prior studies (Delinsky and Wilson, 2006), patients with DEP showed reduced depressive symptoms (PHQ), even while participating in the hospital’s DEP-specific inpatient treatment. Comparing study periods (Group A/B) demonstrated that the intervention caused a significant symptom reduction. The results support previous research and demonstrate how body image training can be integrated transdiagnostically, as suggested by Rawana (2013). While body image disturbance is central to AN diagnosis, the intervention significantly impacted body image in patients with AN despite their lower response to feedback, difficulty changing behavior (Foerde and Steinglass, 2017), need for control, and rigid thinking (AWMF S3-Leitlinie Diagnostik und Therapie der Essstörungen, 2018). These disorder-specific differences may explain the varying training effects between diagnostic groups highlighting the intervention’s potential to modify body perception even in challenging clinical populations.

In addition to disorder-specific characteristics that influence body image and, consequently, the outcomes of the 2-AFC task, contextual factors such as social, cultural, economic, and environmental determinants of health behavior may differ between diagnostic groups (e.g., Lorant et al., 2003; Swanson et al., 2011) and shape body image, as outlined in the Introduction. Such factors may help explain both interindividual and diagnosis-specific differences in responses to the 2-AFC task and should be systematically addressed in future research.

This pilot study aimed to test the paradigm’s feasibility in adolescents in a clinical context. As a pilot, crossover-designed study, only small group sizes were included, and although we informally asked adolescents of a comparable age group about avatar preference, this does not constitute a systematic validation; thus, the avatar originally developed for adults in the UK was neither culturally adapted nor validated for adolescents. In the 2-AFC task, the avatar was used to measure shifts in perceptual boundaries, while questionnaires assessed disorder-specific symptoms such as body dissatisfaction. Previous studies have shown that variation in the categorical boundary is associated with body dissatisfaction (Cornelissen et al., 2022). Future studies should therefore incorporate direct perceptual body image measures to complement questionnaire-based assessments of body dissatisfaction. Many patients were excluded due to the long duration of the study with a shorter planned hospital stay, acute suicidality or an obese BMI in the DEP group. Participants with an obese BMI were excluded to avoid reinforcing a potentially negative self-perception of being “fat.” Besides IQ, no demographic covariates or emotional and motivational factors were assessed or controlled for in the structural model. Future studies should therefore consider including additional variables such as parental income, education, personal health experiences, and emotional or motivational alignment with the task. Furthermore, the confrontation with body ideals and corrective feedback may have induced stress in some participants, which could be reflected in the dropout of two patients with depression. The compatibility of participation during the clinical setting was sometimes difficult. Despite many appointments and repetitions, the tasks were completed reliably. The control condition’s non-neutral effect was another limitation. The analysis slightly deviated from the preregistered plan on AsPpredicted (# 68371) to align with CONSORT guidelines for crossover studies. Importantly, no changes were made to the hypotheses or outcome variables. Deviations were limited to analytic procedures, specifically variables differed to appropriately reflect the crossover structure and instead of the preregistered mediation analysis, a linear regression was conducted, as the temporal order required for mediation could not be assumed in the crossover context. A repeated-measures ANOVA was added so the second hypothesis could still be answered clearly. To ensure data usability despite deviations, the carryover effect was analyzed. While the study was transparently preregistered, it was subsequently registered with the German Clinical Trials Register (DRKS), a WHO-recommended trial registry for Germany, in response to additional journal requirements (Trial Registration Number: DRKS00037169).