Participants were aged between 18-72 years (mean age: 35-42). Most were female (range 56.1 [23]-79.0% [21]). Many had a University degree (range 31.3 [62]-57.0% [21]) and experience with psychological treatments (reported in four studies, 53 [23]-77.7% [24]). The proportion of PD/PDA cases varied between 25.7% [64] and 100% [57]. Comorbidity was high in some studies (e.g. [62,63]).

Out of web-based 10 interventions, four were disorder-specific [19,21,23,63] and six transdiagnostic treatments [for a summary of contents, Supplementary Table 1 (in the online-only Data Supplement)] [21,23,24,57,64,62]. Nine interventions were based on CBT [21,23,24,57,62,63] and one was ACT-based with CBT elements [64]. Most interventions had a length/availability of 8 weeks (range 4 [62]-12 [63]), with 4 to 8 lessons. Unguided self-help [21,24,57,62,64] was provided with automated messages and support on demand. Guidance [19,21,23,57,63,64], via email or telephone varied in the average time spent from a few minutes [19] up to one or two hours (per participant/program) [21,57,63,64]. Therapist qualifications differed (e.g., Master students, psychologists, CBT-therapists).

Seven RCTs compared eight active treatments against a control condition, three studies guided vs. unguided formats, and two studies transdiagnostic vs. disorder-specific approaches (i.e., 13 comparisons) at post-treatment.

(I) The Panic Disorder Severity Scale Self-Report version (PDSS-SR) was used in five studies [19,21,57,63,64], multiple measures [Agoraphobic Cognitions Questionnaire (ACQ), Body Sensations Questionnaire (BSQ), Mobility Inventory for Agoraphobia & Accompanied (MIB, MIA)] in two studies [23,24] and the Panic Agoraphobia Scale (PAS) in one study [62]. (II) As shown in Table 1, all studies reported attrition (9.8-42.1%) and adherence rates (7.8-75%). Participant satisfaction was assessed in three studies [23,24,62] using the Client Satisfaction Questionnaire (CQS-8) and in two studies [19,63] using other measures.

Taken together, five out of seven RCTs showed significant improvements in panic symptoms in guided disorder-specific iCBT [19,23] and in both unguided and guided transdiagnostic iCBT interventions [19,23,24,57,62,66], when compared to a control condition. These findings contribute to prior research targeting iCBT for anxiety disorders [14,30,38]. Future studies should thus focus on determining specific therapeutic effects of iCBT, for instance, by using study designs combining active and passive control conditions [19,24]. Another option could be to compare the efficacy of specific contents or modules such of internet treatments such as exposure or relaxation exercises in iCBT Moreover, the evidence base on predictors of outcome in iCBT for PD is limited [67]. Interestingly, the study on “ConfID” [62] showed that a more positive attitude at baseline moderated therapeutic outcomes. A further investigation of the impact of pre-treatment attitudes in the response to iCBT could be therefore useful to decide, which patient might most likely benefit [68]. Another interesting finding was that a good therapeutic working alliance could be established in different internet treatment approaches [23], which is consistent other research [66,69]. Overall, findings suggest the helpfulness of internet treatments for PD patients with a preference for online selfhelp as stand-alone treatments and for those wishing to bridge waiting times for face-to-face therapy.

In contrast, IIT-analyses indicated no significant improvements in panic symptoms in two included RCTs [63,64]. One of these studies assessed “Don’t Panic Online” [63] in a sample with mild or subclinical PD cases and proved to be effective in reducing panic symptoms only among program completers. In this context, van Ballegooijen et al. [63] noted the high proportion of comorbidity with Major Depressive Disorder (42.7%) might have contributed to poor -adherence rates. Furthermore, Allen et al. [19] argued that disorder-specific iCBT might be less helpful in subclinical PD. The other RCT [64] with no meaningful effects for PD evaluated “ACT-smart” with a supplemented app. Ivanova et al. [64] assumed the small proportion of PD cases in their study, the novelty of ACT protocols translated to internet-delivery and a probable superiority of iCBT over internet-delivered ACT for PD as possible reasons for non-significant findings [64]. Indeed, the few published studies on internet-delivered ACT mostly targeted depression [17]. Considering the scarce evidence base on the effectiveness of internet-delivered third-wave CBT programs on the one hand and the popularity of such trainings on the other hand, further research is recommended [15]. Moreover, given the lack of evidence-based apps for PD [70], a further investigation of “ACT-smart” appears worthwhile. In addition, other third-wave CBT components such as mindfulness exercises have been successfully applied in several internet treatments like “velibra” or “Free from Anxiety.”

Regarding support formats, only one RCT [57] of three found significant improvements in panic symptoms and adherence rates in favor of guided treatments. This is in accordance with earlier research [22,71]. Oromendia et al. [57] assumed a positive impact of the scheduled support on participants’ commitment to “Free from Anxiety,” which might have increased adherence to treatment and therapeutic outcomes. Different reasons appear reasonable for the other two studies [21,64] that found no significant difference between support formats in outcomes. Fogliati et al. [21] argued that “new generation” iCBT work well as self-guided format, since they are grounded on well-established protocols, evaluated over several clinical trials and typically involve measures that aim at monitoring and engaging patients, such as automated messages [21]. Another possibility is the selection bias. For instance, in the study on the unguided “velibra,” [24] Berger et al. [24] argued that the need for consulting a primary care physician might have increased the threshold for less motivated patients. This assumption of selection bias is also supported by the fact that in some studies [21,57,62] only few participants in self-guided conditions accepted the offer of clinician-guidance on demand. Furthermore, although there were differences in the support quantity and modality in guided interventions, no patterns with outcomes or adherence appeared evident. Of the studies with a comparable average time spent per participant, some were effective [57], whereas others [63] were not. In fact, the current state of research is incoherent, especially regarding the dose-dependency of support in treatment outcomes [22]. A systematic review [22] found that guidance appears to be a helpful feature in internet treatments, but the impact they identified was smaller than stated earlier. Yet, studies on underlying processes and specific components making different support formats effective and acceptable are scarce [21]. For instance, in contrast to face-to-face treatments research indicates that the qualification of online-therapists appears to be irrelevant for outcomes [22,72]. For decisions on suitable treatment options for PD, patients’ preferences should be considered [20]. For instance, given a preference of many adults for self-help to manage milder psychological problems on one’s own as a barrier to seek help face-to-face [73], unguided programs can be recommended for subclinical symptoms and prevention [22]. It could be thus interesting to explore which subpopulations prefer guidance to provide stepped care models for PD [32]. Surveys further suggest that many persons prefer therapist-guided IPT and faceto-face services in case of more severe emotional problems [74]. Treatment preferences of patients with different symptomatology should be investigated to provide more or better suitable therapy options for specific subpopulations in routine care.

Furthermore, although of the most reviewed interventions were based on a transdiagnostic approach, only two studies [21,23] compared outcomes directly with the more established disorder-specific approach. Both studies showed no significant difference between the approaches. This is in contrast to findings of a review [21] that indicated a superiority of disorderspecific over transdiagnostic iCBT for anxiety disorders. It would be therefore interesting to explore the role of perceived benefits across treatment approaches from the perspective of patients [35]. Because tailoring allows for accurate assignment to treatment with respect to comorbid anxiety disorders, tailored transdiagnostic treatments have advantages for the dissemination of iCBT in routine care [24]. Transdiagnostic treatments can work well as self-guided format [24,62], which is a clear benefit for the dissemination regarding treatment gaps [21]. Thus, this combination appears efficient to provide additional treatment options.

Attrition rates varied largely across the RCTS. Interestingly, both the lowest (9.8%) [23] and the highest (42.1%) [63] dropout rate were found in different therapist-guided 8-week iCBT programs.

Relatively high attrition rates are commonly reported in iCBT trials for anxiety disorders [26]. For instance, in reviews on internet-based treatments for different diagnoses dropout rates were between 2% and 89%, with a weighted mean of 31% (19 studies) [75], between 1% and 50% (23 studies) [76], or between 0% and 55% (13 studies) [29]. In addition, adverse events as potential reasons for dropouts were not directly reported, which is a common issue in iCBT research [14,77].

The present review also found variance in adherence rates, ranging from 7.8% to 75%. For instance, in the study on the 5-lesson “Panic Program,” 19 adherence (63%) was lower than in a previous trial using a 6-lesson version (79%) [65]. This is interesting, because the briefer version aimed at improving adherence; the same study reported that the 5-lesson “Panic Program” was also effective (g=0.55), when delivered in a primary care setting (n=330 patients), albeit the adherence rate was lower (56.1%) than in the included RCT [19]. Clearing the relevance of iCBT length need further investigation. Consistent with other studies [60], two included studies [62,63] identified lack of time as a barrier to complete iCBT programs. Ivanova et al. [64] supposed that the relatively low degree of full adherence in the “ACT-smart” study could be attributed to multiple logins at different platforms and devices. Usability issues related to delivery modes (e.g., internet-based, computerized or smartphone-delivered) as reasons for non-adherence need detailed exploration. Moreover, minimal guidance may have improved adherence in two studies [64,57], and clinical effects in PD in one study [57]. According to Mewton et al. [12], guidance in iCBT is not needed for immediate improvements, but could be essential for the maintenance of clinical effects and adherence. In routine care, a combination of online and face-to-face format (i.e., blended treatments) can be useful as they can improve adherence or lower dropout rates [78].

In accordance with prior research [38], five RCTs that assessed participant satisfaction endorsed a relatively high acceptance of internet treatment with different modalities. Earlier research [14] indicated a high participant satisfaction in therapist-guided iCBT for anxiety disorders. However, in one included study [21], no difference in satisfaction was found between support formats. Furthermore, ratings stem from CCanalyses and might be thus biased towards higher satisfaction. Future studies should evaluate the (dis-) satisfaction of non-completers to improve the acceptability of Internet treatments. This could help identifying reasons for the discrepancy between high satisfaction and suboptimal adherence.