In this RCT, participants were randomly allocated to 3 equally sized groups: a group that received app-based exposure therapy, a group that received app-based mindfulness meditation (active control), and a WL group (passive control). Panic and agoraphobia symptoms, depression and anxiety, QoL, anxiety sensitivity, and self-efficacy expectations were assessed before the intervention (T₀), after the 5-week intervention phase (T₁), and at follow-up 13 weeks after T₀ (T₂).

The study was conducted in accordance with the Helsinki Declaration, approved by the University of Bonn’s medical ethics committee (protocol number: 321/20), and preregistered in the German Clinical Trials Register (DRKS00022204). Before the start of data collection, we decided to raise the upper age limit for participants from 50 years to 60 years and removed the Beck Depression Inventory from the list of outcomes to reduce survey length, as depressive symptoms are also addressed by the Depression, Anxiety and Stress Scale (DASS; refer to “Enrollment and Assessment” below for details). The study protocol was updated accordingly. All participants gave their written informed consent and could opt out of the study at any time. Study data were deidentified using participant codes for collecting interview and questionnaire data, as well as for app access. Participants did not receive any financial compensation for taking part in the study.

Individuals between the ages of 18 and 60 years who met the diagnostic criteria for a PD with and without agoraphobia were eligible to participate in the study. Current psychotherapeutic treatment (face-to-face individual or group therapy) was an exclusion criterion. Other exclusion criteria were acute psychosis, pregnancy, unstable cardiovascular disease, inadequately controlled hypertension, other medical conditions that precluded the administration of exposure therapy, and insufficient knowledge of the German language, as we evaluated German versions of the studied apps. Mobile device literacy was not explicitly specified as an inclusion criterion but implicitly required for study participation. Participants were allowed to continue psychiatric medication during the study.

Recruitment and data collection for the study took place between November 2020 and January 2023 (first participant in: November 17, 2020 and last participant out: January 11, 2023). Participants were recruited via the outpatient clinic of the Department of Psychiatry and Psychotherapy of the University Hospital Bonn; by advertising among psychotherapists, psychiatrists, general practitioners, self-help groups, psychological study counseling; and via social media and newspapers. Since the study was conducted entirely online, the participants were recruited nationwide in Germany as well as in Austria and Switzerland.

A total of 210 individuals expressed interest in the study and were contacted by the study team. The study team outlined the study design and presented the inclusion and exclusion criteria. After this initial step, 117 individuals progressed further and underwent the full eligibility assessment. To establish the diagnosis of PD with and without agoraphobia and to determine comorbid disorders, participants underwent a structured clinical interview at the baseline assessment (T₀). The clinical interview was based on the Brief Diagnostic Interview for Mental Disorders Mini-DIPS [16] and carried out remotely, using a secure video communication platform [17]. If the diagnostic criteria were not met or any exclusion criterion was fulfilled, study participation was terminated at this point. If participants met the diagnostic criteria for PD with and without agoraphobia and did not show any exclusion criterion, we randomly assigned them to one of the 3 groups by drawing from an urn. The study team prepared the urn prior to the start of the study, and the allocation sequence was unpredictable. After screening, 111 individuals were included and assigned to 1 of the 3 groups; recruitment was stopped after reaching this prespecified sample size. Following the baseline assessment, the 5-week intervention period started for all included participants. During the intervention period, participants received weekly invitations to short surveys to assess usage patterns via SMS. At the end of the intervention period, all participants received a link to a post-intervention survey via email (T₁). At T₂, all participants received a link to a follow-up survey. Once the participants had completed this questionnaire, participation in the study was completed. Participants in the WL group equally underwent assessments at T₀, T₁, and T₂ and could request access to one of the 2 apps following T₂.

In total, 6 members of the study team rated the Mini-DIPS interview and performed randomization (1 physician, 1 postgraduate psychologist, 4 advanced-level psychology or medical students under supervision). Raters received training and conducted mock interviews before the first rating. All self-report data was obtained via an online survey tool [18].

Primary outcomes were panic and agoraphobia symptoms based on the following self-reported questionnaires:

Secondary outcomes were general anxiety, depression, and stress symptoms, QoL, anxiety sensitivity, and self-efficacy expectations, and assessed by the following self-reported questionnaires:

Participants assigned to the exposure app group and to the meditation app group each were instructed to use the respective mobile app for a period of 5 weeks. During this period, they received no further instructions from the study team regarding the app content besides a short introduction at baseline (ie, self-guided therapy). To minimize the risk of performance bias, none of the 2 apps was presented as superior to the participants during the onboarding procedure. However, we could not guarantee that participants were kept completely blind, as information documents about the study stating that the exposure therapy intervention was the intervention of interest had to be made available online (eg, in the study registration) and had to be disclosed to obtain informed consent. Both apps were available for Android and iOS devices. After the 5-week intervention period, access to the app was not restricted, but no further instructions were given, so that participants could continue using their respective app as they wished.

Participants in the exposure app group received free access to Mindable: Panic & Agoraphobia [41], an e-health application otherwise available to individuals with PD with and without agoraphobia in Germany on prescription [42]. The app is based on the manual for cognitive-behavioral exposure therapy published by Lang et al [30] and focuses on preparing, accompanying, and debriefing various exposure exercises. After working through a base module featuring psychoeducative content (eg, psychological models of causes, maintaining factors, and consequences of the disorder), patients are guided through a second module focusing on interoceptive exposure (eg, accustoming to self-provoked body sensations such as dizziness or tachycardia) and a third module facilitating in-vivo exposure of agoraphobic situations. For the third module, patients define situations relevant for themselves (eg, public transportation, supermarket, and elevators) and are encouraged to gradually increase exposure difficulty (eg, first accompanied and later alone). Content is delivered through videos and accompanying texts. Patients interact with the application, for example, by answering self-reflection questions, logging anxiety during exposure exercises, and repetitively filling short symptom questionnaires. Patients are instructed to use the app daily and schedule at least one exposure exercise per week. Customizable mobile-operating-system-based notifications reminded participants to frequently use the app and complete scheduled tasks. We recommended participants to work through the 3 therapy modules chronologically, starting with the psychoeducation module in the first week. A study-specific “frozen” version of the app was used and not changed throughout the study. Figure 1 presents exemplary screenshots of the exposure therapy app. Panel A shows an app overview featuring the 3 main modules: psychoeducation, interoceptive exposure, and in-vivo exposure. Panel B features an exemplary video transcription from the psychoeducation module, and panel C an exemplary interoceptive exposure exercise focused on the elicitation of dizziness. An exemplary in-vivo exposure exercise guiding a visit to a supermarket together with a trusted person is displayed in panel D. The operating language was set to German for the study.

Participants in the meditation app group received free access to 7Mind [43], a meditation app which is otherwise available in Germany through selected health insurances, some employers, and on a self-payment basis. The application teaches mindfulness meditation through a range of courses (eg, “foundation” and “from stress to silence”) and themed exercises (eg, “falling asleep” and “bus or train”). The app dynamically recommends content and provides advice. Participants can schedule meditations within the program and receive reminders. The intervention is not specifically focused on PD with and without agoraphobia. Participants were encouraged to use the app daily and to choose courses and relaxation exercises according to their own needs and interests.

To evaluate the results under real-world conditions, we conducted an intention-to-treat (ITT) analysis for all primary and secondary parameters, in which we included all available data from all randomized participants. In addition, to examine the results under ideal conditions, we conducted an additional completer analysis on all primary outcome parameters. In both analyses, separate linear mixed effects models (LMM) were calculated for each analyzed outcome parameter. The model specification controlled for baseline values by including the factor time (T_0, T₁, and T₂) and the interaction of group (exposure, meditation, and WL) and time [44]. Further, participant ID was added as a random effects term, and age, gender, and education served as covariates. Post hoc tests were conducted by running Bonferroni-corrected t tests on the estimated marginal means from the model. All reported statistical tests are based on a significance level of ɑ=.05. Partial Eta-squared (η_p²) was used as a measure of LMM effect sizes. For significant effects, post hoc t test effect sizes were reported as Cohen d, using the estimated marginal means and the pooled SD from the observed means. We analyzed reliable changes based on the reliable change index suggested by Jacobson and Truax [45] using normative data for the PAS [8]. CIs for the number needed to treat (NNT) were estimated using the Wald method [46]. All statistical analyses were conducted in R 4.2.2 [47]. The packages lme4 [48] and emmeans [49] were used for model estimation and post hoc testing.

For the ITT analysis, data of all 37 participants per group were analyzed, including those with missing data at 1 or 2 timepoints. A detailed summary of the number of complete cases that were included in the ITT analysis for each questionnaire per group and timepoint can be found in the supplementary data S1 in Multimedia Appendix 1.

For the completer analysis, we considered only those participants who had no missing data in any questionnaires at all 3 time points. In total, 7 participants (3 in exposure app group, 2 in meditation app group, and 2 in WL group) indicated at the T₁ and T₂ assessment that, in addition to the apps evaluated in the study, they had used other meditation or therapy apps to reduce their PD with and without agoraphobia symptoms. We also excluded these individuals from the completer analysis. Thus, the completer analysis included 14 participants from the exposure app group, 25 participants from the meditation app group, and 27 participants from the WL group.

Demographic information, diagnostic status, and baseline symptom scores for all 3 groups are summarized in Table 1. A total of 111 participants (mean age: 36.78, SD 12.58; n=81, 73% female) were randomized. Of these 111 participants, 59% (n=65) fulfilled criteria for combined PD with and without agoraphobia, 22% (n=24) from mere panic disorder and 20% (n=22) from mere agoraphobia. Compared to the WL group, significantly more participants (P=.03) in the meditation app group had concomitant other anxiety disorders (eg, social anxiety disorder and generalized anxiety disorder). All other demographic and diagnostic status variables showed no significant group differences. The mean PAS score across all groups was 22.13 (SD 9.06) at baseline, corresponding to intermediate symptom severity. A significant baseline group difference was detected for the PAS score (P=.02), driven by significantly lower scores in the meditation app group compared to the WL group (exposure group vs WL group: P=.84; meditation group vs WL group: P=.02; mediation group vs exposure group: P=.07). Of note, however, we controlled for baseline differences in all statistical models (refer to the “Statistical Analyses” section above for details). There were no other significant baseline group differences in any of the other primary or secondary outcomes.

From a total of 111 participants enrolled (37 participants per group), 11 participants dropped out of the study at T₁. Furthermore, 3 participants did not participate at the T₁ assessment but returned for T₂. In total, T₁ data was missing for 5 (14%) participants in the exposure app group, 3 (8%) participants in the meditation app group, and 6 (16%) participants in the waiting list group. At T₂, an additional 18 participants had dropped out (total dropout rate in the exposure app group at T₂: n=16, 43%; meditation app group: n=8, 22%; waiting list group: n=5, 14%). In addition, at all 3 time points, individual participants only partially completed the questionnaires. Dropout rates at T₁ did not significantly differ between groups (Fisher exact test: P=.68). At T₂, we found significant differences in dropout rates between the groups (Fisher exact test: P=.01). Both at T₁ and at T₂, participants who dropped out did not significantly differ on any of the measured baseline characteristics including symptom severity (refer to Table 1 for a list of the tested parameters). For detailed information on the composition of dropouts, refer to the Consolidated Standards of Reporting Trials (CONSORT) flow diagram (Figure 2).

On a weekly basis, we evaluated usage patterns by asking participants how often and how long they used the app. The most selected category for frequency (average across all 5 wks) was “Several times a week” for both the exposure group (42.3%) and the meditation group (41.2%). Participants selected “daily” more often in the meditation app group than in the exposure app group (29.2% vs 12.0%). Notably, members of the exposure app group selected “Never” more frequently than those in the meditation app group (19.1% vs 8.7%).

The average duration per use was similar in both groups. Here, the most selected category was “less than 15 minutes” in both the exposure group (50.0%) and the meditation group (50.2%), followed by “15 to 30 minutes” (34.8% and 36.6%, respectively). On average, only 1.82% of participants in the exposure group and 0% in the meditation group used the respective app for “more than 1 hour” per use. For detailed information on usage patterns per week, refer to Table S2 in Multimedia Appendix 2.

Changes in PD with and without agoraphobia symptom severity during the study are depicted in Figure 3. In the exposure app group, scores decreased from T₀ to T₁ as well as from T₁ to T₂ on all 6 symptom severity scales. Also, the meditation app group showed a decrease in all 6 symptom scores from T₀ to T₁; however, here the decreases were descriptively smaller, and they increased again at T₂ in 4 of the measures (overall PD with and without agoraphobia symptoms, agoraphobic cognitions, fear of bodily sensations, and safety maneuvers). The WL group showed varying courses of outcome with no clear overall tendencies.

Table 2 summarizes the LMM results of our ITT analysis (for more detailed results, refer to supplementary data in Multimedia Appendix 3). For overall PD with and without agoraphobia symptoms (PAS), we found a significant time×group interaction (F_6,148.48=3.00, P=.009, η_p²=0.11). Post hoc t tests revealed significant differences between the exposure app group and the WL group at both T₁ (t₁₇₇=2.48, P=.04, d=0.55) and T₂ (t₂₁₁=2.50, P=.04, d=0.60), but no other significant differences in group comparisons. In addition, for the exposure app group, we observed significant reductions in overall PD with and without agoraphobia symptoms between T₀ and T₁ (t₁₇₄=4.24, P<.001, d=0.72) and between T₀ and T₂ (t₁₈₁=4.18, P<.001, d=0.90), while for all other within-group comparisons, we did not find any significant differences.

For safety maneuvers (TSMS), we likewise found a significant time×group interaction (F_6,142.69=2.88, P=.01, η_p²=0.11). Here, our post hoc t tests did not reveal any significant between-group effects, but significant within-group decreases for the exposure app group between T₀ and T₁ (t₁₇₂=2.74, P=.02, d=0.32) and T₀ and T₂ (t₁₇₅=4.69, P<.001, d=0.62). All other within-group comparisons revealed no significant differences.

Regarding agoraphobic cognitions, fear of bodily sensations, and avoidance, we found no significant time×group interactions in the ITT analysis. However, across groups, the LMM for fear of body sensations (BSQ) revealed a significant main effect of time (F_2,173.61=5.06, P=.007, η_p²=0.06).

In the additional completer analysis, we again found a significant time×group interaction (F_6,91.90=2.69, P=.02, η_p²=0.15) for overall PD with and without agoraphobia symptom severity (PAS). When following up the interaction with post hoc t tests, we, however, did not find significant between-group differences at any of the 3 time points. Regarding within-group comparisons, we found a significant reduction in overall PD with and without agoraphobia symptom severity in the exposure app group between T₀ and T₁ (t₁₂₆=3.50, P=.002, d=0.79) and between T₀ and T₂ (t₁₂₆=3.65, P=.001, d=0.84), but no other within-group comparison reached level of significance. Regarding safety maneuvers (TSMS), we again observed a significant time×group interaction (F_6,91.90=2.94, P=.01, η_p²=0.16) that we followed up using post hoc t tests. These indicated a significant within-group decrease for the exposure app group between T₀ and T₂ (t₁₂₆=3.95, P<.001, d=0.56), while all other post hoc tests did not show significant effects. Detailed results of the completer analysis can be found in supplementary data in Multimedia Appendix 4.

Changes in secondary outcomes are depicted in Figure 4. Descriptively, the exposure app group improved in all secondary outcomes over the course of the study, while most measures stayed stable or even deteriorated in the other 2 groups.

In the ITT analysis, the LMM on psychological QoL (as measured by the psychological subscale of the WHO-QOL) revealed a significant main effect of time (F_2,175.68=4.45, P=.01, η_p²=0.05) as well as a significant time×group interaction (F_6,144.29=3.12, P=.007, η_p²=0.11). The post hoc t tests attributed this interaction to a significantly higher psychological QoL in the exposure app group than in the WL group at T₂ (t₁₆₀=–2.88, P=.01, d=0.63), since all other between-groups comparisons showed no significant differences. Moreover, for the exposure app group, we found significant within-group increases in the psychological QoL between T₀ and T₂ (t₁₇₈=−3.14, P=.006, d=0.36) and between T₁ and T₂ (t₁₇₅=−2.98, P=.01, d=0.32). For the WL group, in turn, we found significant decreases in psychological QoL between T₀ and T₁ (t₁₇₅=2.54, P=.04, d=−0.26) and between T₀ and T₂ (t₁₇₅=2.59, P=.03, d=−0.28), while all other within-group comparisons showed no significant differences.

Further, there was a significant time×group interaction for depressive symptoms as measured by the DASS questionnaire (F_6,148.84=3.00, P=.009, η_p²=0.11). Our post hoc between-group comparison thereby revealed significantly lower depressive symptoms in the exposure app group compared to the WL group at T₂ (t₂₀₆=3.09, P=.007, d=0.75). Moreover, within the exposure app group, we found a significant decrease in the severity of depressive symptoms from T₀ to T₂ (t₁₈₄=2.87, P=.01, d=0.48), while all of our other comparisons did not show any significant effects.

Regarding our ITT analyses on all remaining secondary outcomes (physical, social, and environmental QoL, stress and overall anxiety, anxiety sensitivity, and perceived general self-efficiency), we found no significant time effects or time×group interactions. Detailed results of our analyses for secondary outcomes can be found in Multimedia Appendix 5.

Because we found a significant between-group effect for overall PD with and without agoraphobia symptoms, we analyzed individual change for each participant based on this outcome (refer to Tables 3 and 4). At T₁, we observed reliable improvement for 35% (n=11) of participants in the exposure app group (meditation app: n=2, 6%; WL: n=2, 7%; Fisher exact test: P=.002). In the exposure app group, 23% (n=7) of participants showed reliable improvement to a score below the PAS cut-off value of 8 suggested for remission [8] compared to 3% (n=1) in the WL group and no participant in the meditation app group (Fisher exact test: P=.12). Compared to WL, these differences translate into a NNT of 3.47 for reliable improvement and 5.20 for reliable improvement to remission (compared to meditation app: NNTs of 3.39 and 4.43, respectively). Follow-up data showed a similar pattern with an NNT of 3.85 for reliable improvement compared to WL and an NNT of 3.16 compared to the meditation app. However, 3 remitted participants in the exposure app group dropped out between T₁ and T₂, and we observed lower rates of remission at T₂ compared to T₁ for the exposure group. NNT CIs for this measure included ± ∞, implying that no significant differences in remission rates were observed between the groups at T₂.

To assess the safety of the meditation app, we also analyzed data for reliable individual deterioration. At T₁, 1 participant (3%) in the exposure app group showed reliable deterioration (meditation app: n=1, 3%; WL: n=0, 0%). The observed score for this participant increased from 12 (mild symptoms) to 24 (intermediate symptom severity). At T₂, reliable deterioration was not observed in the exposure app group (meditation app: n=3, 10%; WL: n=2, 6%). Between-group NNT comparisons for reliable deterioration showed no significant group differences (all CIs included ± ∞). One participant from the WL group and 1 participant from the meditation app group were admitted to the hospital and subsequently dropped out of the study. No adverse events were reported to the study team in the exposure app group.

In this longitudinal RCT, we evaluated the efficacy of an exposure therapy app for PD with and without agoraphobia in comparison to an active (mindfulness meditation app) and passive (WL) control group. We found that the primary outcome parameter overall PD with and without agoraphobia symptom severity was significantly lower in the exposure app group compared to WL at both post-treatment and follow-up. Within the exposure app therapy group, the use of safety maneuvers declined between baseline and posttreatment and between baseline and follow-up. Regarding secondary outcomes, we found significant between-group effects for the exposure app versus waiting list comparisons in depressive symptoms and psychological QoL at follow-up.

The findings demonstrate the efficacy of the exposure app as a treatment for core symptoms of PD with and without agoraphobia. Between-group effect sizes compared to WL were medium according to the classification suggested by Cohen [50]. Treatment effects were sustained throughout the entire study period of 13 weeks. PD with and without agoraphobia symptom scores declined even further between post-assessment and follow-up, suggesting potential for further therapeutic gains in long-time users. Compared to WL, participants who used the exposure therapy app showed stronger improvement in overall severity of PD with and without agoraphobia symptoms. This finding is in line with a recently published RCT that investigated the same exposure therapy app and also found significantly greater improvement in overall PD with and without agoraphobia severity in the exposure app group compared to a waiting list control group [51].

Further, in the exposure app group, we also found a reduction in maladaptive safety maneuvers. These strategies have been found to contribute to the persistence of symptoms by reaffirming the assessment that the associated situations are dangerous and by preventing full exposure and subsequent anxiety reduction [28,52]. Safety maneuvers are explicitly addressed within the exposure app, and our findings suggest that participants engaged in less of these maneuvers following exposure app use. Descriptively, scores for the remaining symptom clusters associated with PD with and without agoraphobia (agoraphobic cognitions, fear of bodily sensations, and avoidance behavior) also declined within the exposure app group, but the effects failed to reach statistical significance. As the exposure therapy app possibly targets all these symptom subsets through framed interoceptive and in-vivo exposure, the overall improvement in PD with and without agoraphobia symptoms observed might have been the result of the sum of smaller improvements in all these symptom subsets. In consequence, the study might not have been sufficiently powered to detect significant group differences for the possibly smaller improvements in symptom subset scores.

In addition to comparing the app-based exposure therapy to WL, we evaluated its efficacy against an alternative active control intervention, the mindfulness meditation app. The mindfulness meditation app was not specifically designed to treat PD with and without agoraphobia but provides a similar experience of using a high-quality digital intervention. In theory, comparing the 2 apps should permit separating general effects of using digital interventions (eg, stemming from the sense of being able to self-treat symptoms with a novel type of intervention) from specific effects of app-guided exposure therapy. Descriptively, symptom scores declined more in the exposure app group compared to the mindfulness meditation app group. Additionally, significant within-group symptom reductions as well as significantly lower post-treatment symptoms compared to WL were only observed for the exposure therapy app. These observations seem to support the assumption that app-guided exposure therapy is specifically suitable as a digital intervention in PD with and without agoraphobia. However, direct comparisons between the exposure app and the meditation app did not demonstrate a statistically significant difference in symptom improvement favoring the exposure app group. One possible explanation might be that the meditation app itself had additional specific therapeutic effects exceeding those generally associated with app usage. These may have mitigated exposure therapy effects in between-group comparisons, which is supported by previous research that found positive effects of mindfulness meditation in anxiety disorders including PD with and without agoraphobia [53,54]. App usage pattern might have also played a role here. Our weekly survey indicated that more people in the meditation group than the exposure group used the app every day. This increased engagement could have had a reinforcing effect on the therapeutic benefit of the meditation app. Another possible explanation is that the degree to which exposure therapy effects exceed general effects associated with mobile intervention usage was too small to be detected with our study design and sample size. As previous literature on similar comparisons is very limited [12], it is difficult to judge what effect size is to be expected for this type of comparison, and our study might have been underpowered. Another aspect to consider is that recent meta-analyses showed that digital interventions in PD with and without agoraphobia are often provided as guided interventions [12]. Also, the exposure therapy app studied here was built based on material from a therapeutic manual intended to be used with therapist guidance [30]. The effects of using the intervention unguided might have been smaller than those of a guided intervention, and this might have resulted in insignificant comparisons with the meditation app. Furthermore, despite randomization, baseline PD with and without agoraphobia symptom severity scores were significantly lower in the meditation app group. Meditation might be particularly difficult to practice when symptoms are more severe. Therefore, meditation could have lower effects among participants with higher symptom scores, a subgroup that was underrepresented in the meditation app group of our study. In addition, our between-app comparisons suggested that advantages of the exposure app become more pronounced at follow-up compared to post-assessment. This suggests that our overall study period might have been too short to detect significant between-group effects of app-guided exposure therapy compared to the meditation app.

Compared to WL, psychological QoL and depressive symptoms improved significantly among exposure therapy app participants at follow-up but not at post-treatment. The observed delayed significant improvement in these secondary outcomes might suggest that app-guided exposure therapy effects on these measures might be downstream effects of initial improvements in core PD with and without agoraphobia symptoms. The observed effects are partially consistent with the results of the 2 previous meta-analyses [12,13] which also found that IMIs for PD improved QoL and comorbid depressive symptoms. Of note, however, in the current study, the QoL effect was specifically found for psychological QoL, while for the remaining QoL domains, no significant effect was observed. Psychological QoL might be the primary domain of improvement as it is the domain most closely related to mental health. This view is supported by our finding that, across all QoL domains, the lowest baseline values were consistently reported in the psychological domain for all groups. Also, a recent validity analysis of the WHO-QOL [55] found that, out of all QoL domains, the psychological domain is the best predictor for overall QoL.

Of note, no significant treatment effects were found for comorbid general anxiety, although in the previous meta-analyses [12,13], similar positive effects of IMIs were found for anxiety as for comorbid depression and QoL. Also, we found no effects for anxiety sensitivity and perceived general self-efficacy. These findings are unexpected, given that CBT interventions, including interventions with interoceptive exposure or symptom provocation, seem to effectively reduce anxiety sensitivity [56] and increases in self-efficacy as well as decreases in anxiety sensitivity appear to predict subsequent symptom severity reductions in CBT for PD with and without agoraphobia [57]. Potential explanations for these findings might, however, be insufficient statistical power to detect such changes in secondary outcomes and divergent mechanisms of change in IMIs compared to classical CBT. Furthermore, the focused nature of the therapeutic material featured in the exposure therapy app might also have reduced its ability to improve anxiety symptoms in a broader sense beyond improvements in core PD with and without agoraphobia symptoms.

Our analyses estimated that approximately 1 in 4 participants achieved substantial improvement in the exposure app group compared to WL and the meditation app, and that rates for reliable improvement and remission were highest in this group. As no adverse events were reported for the exposure app and rates of deterioration were lower or comparable to the other groups, our results indicate that using the exposure app is safe for individuals with PD with and without agoraphobia. In sum, these findings suggest that the exposure app can be a valuable treatment option in addition to established psychological and pharmacological strategies. However, to avoid patient demotivation, clinicians should clearly indicate that, while some patients might greatly benefit from using the exposure app, many patients might also see no relevant change in their symptom severity.

One major limitation of the current study is the high dropout rate, especially the high rate of participants who were lost at follow-up in the exposure app group (43%). This attrition clearly reduced statistical power. The particularly high number of dropouts in the exposure app group might be caused by the study design. After completing the core part of the intervention, some participants in the exposure app group presumably no longer saw any benefit in continuing their study participation. This might be particularly applicable for those participants who responded best, as 3 out of 7 participants who had reached remission according to the PAS at post-assessment subsequently dropped out of the study. Conversely, the WL group exhibited the highest completer rate, as participants in this group might have been highly motivated to complete the study to gain access to the treatment. Future studies should further investigate long-term effects and implement strategies that increase participants’ incentives to complete the study, regardless of their group allocation.

Secondly, overall PD with and without agoraphobia symptom severity at baseline was intermediate in this study according to the classification of Bandelow [8]. In addition, due to the lack of stratified randomization, group differences in baseline symptom severity were observed in that the meditation group had lower overall PD with and without agoraphobia symptom scores than the other groups. While we controlled for symptom severity at baseline in our statistical models, it is still possible that treatment effects of both the exposure therapy app as well as the meditation app to some degree depend on the symptom severity level. For example, some types of exercises, such as longer meditations or longer in-vivo exposure sessions, might be difficult for participants with very severe symptoms. Therefore, further studies of relatively small or moderate sample sizes should use a stratified randomization procedure and explore the effects of app-guided exposure therapy systematically in samples with high and low symptom severity.

A further caveat of the current study is the relatively short intervention period of 5 weeks. Previous RCTs on IMIs for PD with and without agoraphobia featured in the meta-analysis of Polak et al [13] typically used treatment periods ranging from 6 to 12 weeks. This may also be reflected in the delayed treatment effects for our secondary outcomes, which only became apparent at follow-up, but not immediately at post-treatment. Therefore, the intervention period of this study might not have been sufficient to reveal all potential effects, and a longer intervention period could possibly lead to stronger post-treatment effects in future studies.

An additional limitation of our study is that we could not guarantee that participants were kept completely blind, as information documents about the study stating that the exposure therapy intervention was the intervention of interest had to be made available online (eg, in the study registration) and had to be disclosed to obtain informed consent.

Another important note about the current study is that a large part of data collection took place during Covid-19 restrictions, including temporary lockdowns. This might have seriously restricted the realization and effectiveness of in-vivo exposure exercises, especially exercises that required using public transport and being in crowded places. Hence, it is conceivable that the effects found in the exposure app group would have been stronger if there had been no Covid-19 restrictions and exposure exercises could have been carried out more easily.

More generally, future research should attempt to systematically investigate where, when, and how participants used the exposure therapy app, in order to investigate what factors do influence the efficacy of this type of IMI. In addition, the relationship between participant characteristics, such as attitudes to digital technology or previous treatment experiences, and the efficacy of app-guided exposure therapy should be analyzed. A further point to systematically investigate is the optimal level of guidance required for the use of IMIs. In our study, participants received only a brief introduction before beginning the unguided self-help interventions. Of note, however, a recent meta-analysis suggests that guided self-help interventions in PD with and without agoraphobia, ie, self-help interventions with regular therapist contact, are more beneficial than unguided interventions [58]. Hence, while an unguided intervention like the exposure therapy app might offer benefits such as greater cost efficiency compared to guided interventions, user engagement and efficacy might be lower for this type of IMIs. Therefore, future research should also investigate how systematic variation of therapist guidance might influence the efficacy of the here tested app.

This RCT found evidence for the efficacy of app-guided exposure therapy for PD with and without agoraphobia. The evaluated app improved overall PD with and without agoraphobia symptom severity as well as depressive symptoms and psychological QoL and reduced safety maneuvers. Greater improvement for overall PD with and without agoraphobia symptoms in exposure app versus WL comparisons was found. However, we were unable to demonstrate any superiority of the exposure app compared to our active control condition, an app focused on mindfulness meditation. While descriptively, participants assigned to the exposure app showed greater and more sustained improvements than participants in the meditation app group, these differences were not statistically significant. The extent to which this was due to a lack of statistical power needs to be determined in future large-scale confirmatory studies, since the general feasibility, safety, and efficacy of mobile exposure therapy for PD with and without agoraphobia has been indicated here. Additional directions for future research include investigations aimed at the role of therapist guidance, usage intensity, situational factors of app use, and patient characteristics. The observed response and remission rates indicate that, while some patients might greatly benefit from using the exposure app, many patients might also see no relevant change in their symptom severity. Overall, our findings suggest that app-guided exposure therapy can be a useful further treatment option in addition to established psychological and pharmacological strategies available for PD with and without agoraphobia.