This pilot study (ethics approval no.: 2022-77) precedes a randomised controlled trial (RCT; ClinicalTrials.gov registration: NCT05554718) to test effectiveness, feasibility and acceptance of the app-based self-help program Mindable: Soziale Phobie. Results of this pilot study were used in the authorisation process of the app as digital health application within the German DiGA program (for reference, see Giebel et al., 2024). The study employed a quasi-experimental pre-post design in a naturalistic setting. Recruitment took place from September 2022 to February 2023 through two pathways: therapist referrals from outpatient centers at Goethe University Frankfurt and Technische Universität Dresden, and online recruitment, where interested respondents completed a pre-screening procedure assessing whether self-rated SAD symptom severity (SPIN) met the cut-off of 25 points. Eligible individuals from both pathways were contacted for an on-site intake screening conducted by independent, trained psychologists (at least master’s level), to assess inclusion and exclusion criteria. Inclusion required ages 18 to 65 years, a primary SAD diagnosis as determined by the Structured Clinical Interview for DSM-5 (SCID-5-CV), and absence of more severe or impairing psychiatric disorder. Exclusion criteria included acute suicidal ideation and current psychotherapeutic or psychopharmacological treatment. See Supplementary Figure 1 for the participant flowchart. Clinician-rated outcomes were assessed at baseline- and post-measurement (Week 12), self-reported questionnaires additionally at midpoint (Week 6). After enrollment, participants received a download link and access code, which was linked to the individual via a linking log stored separately from identifying data. The app was accessible on smartphones or tablets running iOS 12/Android 7 or higher, with a stable internet connection. For more details on procedure, see the RCT study protocol (Schittenhelm et al., 2023).

Eligible patients after pre-screening or based on referral (N = 39) were screened for intake. Of these, 33 were included in the pilot study. 76% of participants were female and mean age was M = 31.97, SD = 9.14 (Range 19 – 65). Sociodemographic data is shown in Supplementary Table 2.

Mindable: Soziale Phobie translates the German adaptation (Stangier et al., 2016) of the CT-SAD treatment (Clark & Wells, 1995) into a smartphone program. In this study, the program was implemented as a fully unguided self-help program; however, it can also be used to bridge waiting times for psychotherapy or in a blended treatment alongside face-to-face therapy.

The program consists of three structured and two unstructured modules. Structured modules, completed in a fixed sequence, include psychoeducation, development of an individualised case formulation, and a video-based processing experiment illustrating the effects of self-focused attention and safety behaviours. These could easily be completed within one week, though users proceed at their own pace. Afterwards, users access the unstructured modules, comprising attention training (AT) and behavioural experiments (BEs). BEs form a core therapeutic component, aimed at testing maladaptive beliefs through real-life exposure tasks without safety behaviours. The app supports users in planning and reflection, but provides no reminders. BEs can be undertaken flexibly, with participants encouraged to complete as many as possible during the 12-week program.

The sample size for this pilot study was determined through simulation-based power analysis using “simr” (Green & MacLeod, 2016). Based on a simplified linear mixed model (LMM) corresponding to a one-group pre-post design of the APP condition described in the study protocol by Schittenhelm et al. (2023), we iteratively increased the number of simulated cases until the estimated power exceeded 80%. Analyses were performed with the intention-to-treat (ITT) sample, consisting of all participants included. Missing data was imputed using a non-parametric Random Forest algorithm (Stekhoven & Bühlmann, 2012). Linear mixed models (LMMs) were used to determine change between measurements. We calculated partial η² as effect size, defining a small effect as η² > .01, medium effect as η² > .06 and η² > .14 as large effect (Cohen, 1988). Treatment response was defined as LSAS sum score reduction of at least 29% and remission as a LSAS sum score below 30 (von Glischinski et al., 2018). Reliable clinical change indices (RCI) were determined following Jacobson and Truax (1991). Moderation analyses were conducted using regression models including interaction terms between time and usage parameters, with baseline scores included as covariates. All analyses were performed using R Statistical Software (v4.2.2; R Core Team, 2021). The reporting of this pilot study was guided by the CONSORT-EHEALTH guidelines for digital health interventions (Eysenbach et al., 2011), with adaptations due to the scope and format of a short report.

Descriptively, all outcomes improved over the study period, with self-rated measures improving by midpoint and remaining stable thereafter. Means and standard deviations for all measures and time points are provided in Supplementary Table 3. LMMs were used to evaluate statistical significance of changes from baseline- to midpoint- and post-measurement (Table 1). LSAS significantly improved from baseline to post-measurement, while QIDS-C did not. SPIN and BDI-FS improved by study midpoint and remained stable afterwards. See Supplementary Table 4 and Supplementary Figure 2 for detailed results including additional measures.

Treatment response was achieved by 17 participants (52%). Only one patient reached remission. Reliable clinical change in SAD symptom severity was observed in 67% of participants based on clinician ratings (LSAS) and in 61% based on self-ratings (SPIN) at post-measurement. Notably, all self-rated improvements were achieved by midpoint. For an overview of RCI outcomes, see Supplementary Table 5.

22 participants (67%) completed all study procedures, eleven participants (33%) dropped out either before (N = 4) or after (N = 7) midpoint-measurement. The mean duration of study participation among drop-outs was M = 41.82 days, SD = 15.65. Reasons for drop-out included initiating psychotherapeutic or psychopharmacologic treatment (N = 3) and dissatisfaction with the intervention (N = 2). For six participants, the drop-out reasons were unknown. On average, patients used the app on M = 8.18 days, SD = 6.78, spread out over a mean duration of M = 49.45 days, SD = 31.47. Most participants completed all three psychoeducation modules (97%) and all three model creation modules (70%), which were both part of the structured modules. 39% of participants conducted at least one BE, with a maximum of 13 experiments. The number of usage days was correlated more strongly with the number of unstructured modules (BE: r_s = .85, p < .001; AT: r_s = .83, p < .001) than structured modules (r_s = .49, p = .004). A priori treatment expectancy was moderate with M = 68.52, SD = 19.41, and was not correlated with any usage parameter. See Supplementary Table 6 for descriptive data on usage parameters. An overview of open feedback is presented in Supplementary Table 7.

The number of conducted BEs moderated SAD symptom severity (SPIN) and both intensity and frequency of anxiety-related cognitions (SPQ), but not clinician-rated symptom severity (LSAS). Specifically, a higher number of BEs was associated with greater reductions in SAD symptoms and cognitions. A similar effect was observed for AT: both the number of ATs and the number of usage days moderated self-rated SAD symptom severity (SPIN) and the frequency of anxiety-related cognitions (SPQ), with greater engagement linked to stronger symptom improvement. The number of completed structured modules did not moderate any outcome. See Table 2 for detailed results.

The app-based self-help program Mindable: Soziale Phobie demonstrated a very large pre-post treatment effect on clinician-rated (LSAS: η² = .54) and self-reported SAD symptoms (SPIN: η² = .47. These effects are in line with, and in part exceed, those reported in previous studies on unguided web-based iCBT programs for SAD (e.g. Kählke et al., 2019; McCall et al., 2018; Titov et al., 2008). Large effects were also observed in anxiety-related cognitions and depressive symptoms. Over half of the participants experienced reliable clinical improvement in SAD symptom severity and beliefs. Consistent with prior evidence, our study’s pre-post treatment effects are in the same range as face-to-face CBT (d = 1.22, Bandelow et al., 2015). Notably, these effects were achieved without clinician guidance, highlighting the potential of app-based self-help for SAD.

Despite large treatment effects, attrition and adherence remain challenges in unguided app interventions. Our program’s drop-out rate of 33% was higher than in face-to-face CBT for SAD (e.g. Stangier et al., 2011; Clark et al., 2006), but lower than most unguided internet-based CBT trials, which report rates up to 66% (Chen et al., 2020; McCall et al., 2018; Titov et al., 2008). Our comparatively low drop-out may be attributed to the app-based format, which offers greater flexibility and easier integration into daily life.

Adherence declined over time, with most patients completing the initial structured modules, but 61% of participants not conducting any BEs. This aligns with Dryman et al. (2017), who report that one third of users dropped out after cognitive restructuring but before completing exposures, potentially due to avoidance or task difficulty. Alternatively, users might have disengaged after an initial symptom reduction, as reflected in our finding that improvements occurred mostly in the first half of the study. Notably, patients who engaged more in unstructured modules, specifically BEs, showed greater reductions in SAD symptoms and beliefs, consistent with findings identifying BEs as a key mechanism in CT-SAD (Yilmaz et al., 2025).

Results on attrition and adherence raise questions about the impact of lacking therapist guidance. While some participants independently engaged fully, others who dropped out early might have benefited from additional support, as suggested in participant feedback indicating a desire for external encouragement (see Supplementary Table 2). Other participants, however, valued the flexibility and anonymity of the self-guided format, consistent with previous observations (Kählke et al., 2019). These findings highlight individual differences in treatment needs, emphasising the importance of future research to identify differentiating patient characteristics in order to optimise adherence, effectiveness, and resource allocation in digital interventions (Ibeh et al., 2024).

First, as this was a pilot study assessing the feasibility of an app-based self-help intervention, the small sample size limits external validity. Second, the uncontrolled design prevents firm conclusions about efficacy, so results should be interpreted with caution. Third, selective sampling may restrict the generalizability to the population of SAD patients. For example, digital affinity may be higher among younger individuals, while older patients may prefer face-to-face interaction. This could lead to an overestimation of efficacy in the general population. Fourth, the 12-week study duration limits assessment of long-term engagement or sustained effects and short-term improvements observed may not persist over time. Fifth, participants were not blinded due to the nature of the intervention, which may further contribute to expectancy effects or reporting bias. Finally, the app was tested as a standalone tool, whereas in real-world clinical settings such tools are more commonly used as part of therapist-guided treatments.

Future research should investigate the effectiveness of app-based self-help interventions in an RCT with a larger and more diverse patient sample. A key area for further study is the specific contribution of app usage to the efficacy of therapist-guided treatments (Schittenhelm et al., 2023; Stolz et al., 2018), and their benefit for subgroups of patients less inclined to blended care, such as older adults. Several of these questions are currently being explored in an ongoing RCT (Schittenhelm et al., 2023), which compares unguided app use, guided use with additional therapy sessions, and a waitlist control condition. This study also examines predictors of treatment success, including patient characteristics and mechanisms of change, monitored through weekly symptom assessments.

iCBTs offer a quickly available, effective, and potentially resource-efficient treatment option for highly self-motivated patients or patients who feel too ashamed to pursue face-to-face psychotherapy. This pilot study provides initial evidence for the effectiveness of an app-based self-help program implementing CT-SAD based on the Clark and Wells model. Despite moderate overall treatment engagement, the results demonstrate significant improvements in SAD symptomatology. Further research on its implementation and the potential benefit of therapist guidance is warranted.