[1] 18w The mediational role of panic self-efficacy in cognitive behavioral therapy for panic disorder: A systematic review and meta-analysis

[1] 214w Cognitive models of panic disorder (PD) with or without agoraphobia have stressed the role of catastrophic beliefs of bodily symptoms as a central mediating variable of the efficacy of cognitive behavioral therapy (CBT). Perceived ability to cope with or control panic attacks, panic self-efficacy, has also been proposed to play a key role in therapeutic change; however, this cognitive factor has received much less attention in research. The aim of the present review is to evaluate panic self-efficacy as a mediator of therapeutic outcome in CBT for PD using descriptive and meta-analytic procedures. We performed systematic literature searches, and included and evaluated 33 studies according to four criteria for establishing mediation. Twenty-eight studies, including nine randomized waitlist-controlled studies, showed strong support for CBT improving panic self-efficacy (criterion 1); ten showed an association between change in panic self-efficacy and change in outcome during therapy (criterion 2); three tested, and one established formal statistical mediation of panic self-efficacy (criterion 3); while four tested and three found change in panic self-efficacy occurring before the reduction of panic severity (criterion 4). Although none of the studies fulfilled all of the four criteria, results provide some support for panic selfefficacy as a mediator of outcome in CBT for PD, generally on par with catastrophic beliefs in the reviewed studies.

[1] 219w Over the last decades, substantial research has demonstrated that cognitive behavioral therapy (CBT) is an effective treatment for panic disorder (PD) with or without agoraphobia (e.g. Hofmann & Smits, 2008;Norton & Price, 2007;Stewart & Chambless, 2009). Although CBT's efficacy for PD is well-documented, research on mechanisms of change has been scarce and we know little about how and why such treatment works (Kazdin, 2007). Studies on change processes in psychotherapy most often focus on so-called mediator variables or mediators. A mediator is an intervening variable that statistically accounts for the relationship between the independent variable (e.g. CBT) and the dependent variable (e.g. panic symptom severity; Kazdin, 2007). Establishing statistical mediation, however, does not determine whether the mediating variable causes therapeutic change. To test a causal relation between mediator and outcome requires establishing a timeline of change in the proposed mediator and therapeutic outcome over the course of therapy (i.e. that change in the proposed mediator precedes change in the outcome variable). Manipulating the assumed mediators in experimental designs may also prove a causal relation between mediator and outcome (Kazdin & Nock, 2003;Kraemer, Wilson, Fairburn, & Agras, 2002). To examine the specificity of a proposed mediator, Kazdin (2007) recommends including alternative theoretically derived mediators in mediational studies, thereby providing stronger support for the relation between the putative mediator and outcome.

[2] 54w Understanding change processes in CBT for PD is important since it can lead to prioritizing effective treatment components and methods, as well as augmenting treatment for the still large group of non-responders (Landon & Barlow, 2004). Knowing multiple change processes could clarify idiosyncratic pathways to change, and lead to more effective, individually tailored treatment.

[3] 144w Major theories of mechanisms of change in CBT for PD include the cognitive theories of Beck and Clark (Beck, Emery, & Greenberg, 1985;Clark, 1986), and anxiety control theories developed by Barlow (1988). In his widely recognized model of PD, Clark finds the most important factor to be the tendency to catastrophically misinterpret innocuous bodily symptoms or mental events (e.g. that a pounding heart may indicate a heart attack; or a feeling of unreality may mean one is about to lose control or going insane) (Clark, 1997;1999). These misinterpretations lead to a vicious cycle of escalating anxiety and bodily symptoms, ultimately resulting in a panic attack. CBT for PD disconfirms the catastrophic beliefs and substitutes them with more realistic interpretations of the threatening stimuli, reducing the intensity of anxiety, and breaking the vicious cycle of panic attacks (e.g. Austin & Richards, 2001;Clark et al., 1999).

[4] 422w Anxiety control theories stress the individual's perceived lack of control or inability to cope with anxiety-related bodily symptoms and anxiety-provoking situations as an important determinant for the development and maintenance of PD as well as anxiety disorders in general (Barlow, 1988(Barlow, , 2002)). Inspired by Lazarus' (1966) appraisal theory, Beck et al. (1985) considered anxiety a result of both an appraisal of situations as dangerous, and a perceived lack of ability to cope with the assumed dangers. Casey, Oei, and Newcombe (2004) explicitly applied Beck and Emery's conception into an integrated theoretical model of PD. They find that perceived lack of ability to cope with panic and catastrophic beliefs about danger play key roles in the development of PD as well as in the mediation of therapeutic change in CBT for PD. They refer to the individual's perceived ability to control or cope with perceived dangers in relation to panic attacks as "panic self-efficacy" (Casey, Oei, et al, 2004, p. 326). For individuals with PD, perceived dangers may comprise both external and internal events. Similarly, panic self-efficacy may include two subdimensions: a) perceived ability to perform a certain action within a feared (agoraphobic) situation, and b) perceived ability to control, discount, or cope with negative internal states such as feelings of panic, anxiety-related bodily symptoms, or negative thoughts (Mineka & Thomas, 1999;Zane & Williams, 1993). Barlow (1988) suggests that the concepts of perceived control and self-efficacy in relation to panic are overlapping constructs, and that "it may be possible to extend self-efficacy theory to include performance capabilities in dealing with intense anxiety and panic" (p. 298). Bandura's (1988) theory of self-efficacy ascribes a primary role to perceived self-efficacy in coping with panic in mediating panic symptom reduction, and sees changes in catastrophic beliefs as an effect of change in panic self-efficacy. According to Bandura (1997), perceived self-efficacy may determine whether individuals will initiate appropriate coping behavior in a feared situation rather than safety behavior or avoidance, how hard they will try, and how long they will sustain their effort in the face of obstacles and aversive experiences such as anxiety. Thus, panic selfefficacy seems important for an individual's motivation and willingness to engage in challenging new behavior. For instance, it could lead to a more adaptive interpretation of anxiety-provoking situations, thereby increasing one's willingness to expose oneself to them. In therapy, this could be critical in achieving new learning experiences, reducing the severity of anxiety, and improving the individual's ability to cope adequately in future similar situations (Bandura, 1997;Bandura & Locke, 2003).

[5] 111w Although, panic self-efficacy or anxiety control has a long tradition in theories of PD etiology and mechanisms of change, it has been less thoroughly researched than catastrophic beliefs (Oei, Llamas, & Devilly, 1999;Smits, Julian, Rosenfield, & Powers, 2012). No systematic review has yet been published on panic self-efficacy as a mediator in CBT for PD. The present paper defines panic self-efficacy as perceived ability to cope with or control panic attacks, anxiety-related bodily symptoms, negative thoughts about panic, or agoraphobic situations. We do not address general aspects of control or self-efficacy, for instance "locus of control" (Rotter, 1966), general self-efficacy (Schwarzer & Jerusalem, 1995), or self-efficacy in relation to other areas.

[6] 74w The study's primary aim was to conduct a systematic review of evidence for panic self-efficacy as a potential mediator of therapeutic outcome in CBT for PD, using both descriptive and quantitative (meta-analytic) procedures. We evaluated studies according to four criteria often required to establish mediation in an effective treatment (Baron & Kenny, 1986;Kraemer et al., 2002). The first three criteria were modeled after Baron and Kenny's test of mediation (1986) depicted in Fig. 1.

[7] 225w Criterion 1 examines whether the psychotherapeutic intervention (CBT) causes change in the proposed mediator (panic selfefficacy; the a path of Fig. 1). Criterion 2 examines the association between change in the proposed mediator and therapeutic outcome (the b path). Criterion 3 examines formal statistical mediation. According to Baron and Kenny (1986) this requires (a) a reduction of the association between treatment and outcome after controlling for the contribution of the proposed mediator (from c to c'), or (b) as suggested by Sobel (1982), an indirect mediational effect as indicated by the interaction of path a and path b. Theoretically, the indirect, mediated a  b pathway in Sobels's socalled "product of coefficients" approach should correspond to the c minus c' pathway in the Baron and Kenny "causal steps" approach (Preacher & Hayes, 2008). Criterion 4 examines whether a study establishes a causal relation between change in the proposed mediator and change in outcome either by establishing a timeline (mediator precedes outcome) or by experimental manipulation of the proposed mediator (Kazdin, 2007;Kraemer et al., 2002). It should be noted that mediation is sometimes conditional on moderator variables (e.g. treatment conditions or patient variables) in which case one may speak of "moderated mediation" (Preacher, Rucker, & Hayes, 2007). A variable may thus function as a mediator in one treatment condition, but not in another (i.e. moderated mediation).

[1] 255w The flow of information on study selection is summarized in Fig. 2. The electronic search strategy yielded 3371 publications. After duplicates were removed, we screened 2664 based on abstracts, and evaluated 470 full-texts. Thirty-eight of these fulfilled the inclusion criteria. We located one more relevant reference (Williams & Rappoport, 1983) from the reference lists of the included studies, resulting in 39 eligible studies. We excluded almost 99% of the identified studies in this broad literature search, 59% because the participants had a primary diagnosis other than PD or did not receive CBT. Twenty-two of the 39 eligible studies did not provide all the data relevant for examining criterion 1 or 2 in the meta-analyses. In these cases we contacted the primary or corresponding authors: the authors of 13 studies replied, but only five supplied the relevant data. In total, 19 studies provided data for evaluating criterion 1, 7 studies for criteria 1 and 2, one study for criteria 1, 2 and 3, one study for criteria 2 and 3, one study for criterion 3, three studies for criterion 4, and one study for criteria 1, 2 and 4 (see Table 1). Six of the 39 studies provided insufficient information on panic self-efficacy to evaluate any of the four criteria and they were therefore not included in the review (Bruce, Spiegel, Gregg, & Nuzzarello, 1995;Feske & Goldstein, 1997;Hoffart & Hedley, 1997;Mackay & Liddell, 1986;Meuret, Seidel, Hofmann, Rosenfield, & Rosenfield, 2012;Williams, Kinney, & Falbo, 1989). Thus, the final number of studies included in the review was 33.

[2] 202w Table 1 summarizes the studies' characteristics. Number of participants varied from 8 to 150 with a total of 1436. Treatment duration ranged from 3 to 24 sessions with an average of 10.5 sessions. The studies used 16 different scales of panic self-efficacy (see the Appendix A that includes an explanation of the abbreviations): the PAI-C (Telch, Brouillard, Telch, Agras, & Taylor, 1989) in nine, the SESA (or a shortened version of SESA; Kinney & Williams, 1988;Williams, 1990) in six, the ACQ (Rapee et al., 1996) in five, the SE-CPAQ (Gauthier, Bouchard, Cote, Laberge, and French, 1994) in five, the SES (Bandura & Adams, 1977) in three, and the PSEQ (Borden et al., 1991) in three. In nine studies, various measures of panic self-efficacy were applied only once. Three studies included more than one measure of panic self-efficacy. The scales assessed different aspects of panic self-efficacy: five of the sixteen concerned perceived ability to cope with or control anxiety symptoms or panic attacks, three agoraphobic situations, two both anxiety symptoms and agoraphobic situations, three both agoraphobic situations and scary thoughts, one both anxiety symptoms and scary thoughts, while two included all three aspects of panic self-efficacy: anxiety symptoms, agoraphobic situations, and scary thoughts.

[1] 80w In recent years, the need to investigate mediators of psychotherapeutic change has been stressed in the scientific field of clinical research to answer the central question of how specific processes or mechanisms underlying therapeutic change work (e.g. Arch, Wolitzky-Taylor, Eifert, & Craske, 2012;Prins & Ollendick, 2003). Although more sophisticated analytic strategies and meditational designs have been developed (e.g. Bouchard et al., 2007;Meuret et al., 2010), still relatively few studies rigorously examine putative mediators of therapeutic outcome in CBT for PD.

[2] 435w In the present study we sought to review the evidence for panic self-efficacy as a mediator of therapeutic outcome in CBT for PD by means of both descriptive and meta-analytic procedures. Our systematic literature search identified 39 studies that investigated change in panic self-efficacy during CBT for PD published over the last 30 years. In relation to Criterion 1, we found strong support for CBT effectively improving panic self-efficacy, as indicated by both a large pre-post-therapy overall ES (1.41) in twenty-eight studies, and a large between-group ES (d ¼ 1.46) compared to waitlist in nine RCTs. Controlled between-group ES for catastrophic beliefs (d ¼ 1.25) in the nine RCTs, although smaller, corresponded to the ES for panic self-efficacy. The ten studies related to Criterion 2 revealed a large correlation (r ¼ À0.47) between change in panic self-efficacy and change in outcome, similar to that for catastrophic beliefs (r ¼ 0.50), based on nine of the studies. The three studies that examined change in panic self-efficacy and catastrophic beliefs as predictors of therapeutic outcome gave varying results, not clearly favoring one of the two cognitive variables above the other. Only three studies attempted a formal statistical test for mediation (Criterion 3), and only one of these (Casey, Newcombe, et al., 2005) indicated mediation for panic self-efficacy, as well as for catastrophic beliefs. Finally, four studies were designed to investigate the causal relation between panic self-efficacy and therapeutic outcome (cf. Criterion 4). The small, complex study by Hoffart (1995b;1998) showed mixed results on the role of panic selfefficacy and catastrophic beliefs as mediators in CBT. The three studies testing temporally precedence of change in the proposed mediators to change in outcome variables during therapy (Borden et al., 1991;Bouchard et al., 2007;Fentz et al., 2013) all indicated that increases in panic self-efficacy preceded subsequent reductions of panic severity. Two of the four studies also included catastrophic beliefs. Fentz et al. (2013) found that within-person change in panic self-efficacy, but not in catastrophic beliefs, predicted subsequent within-person change in anxiety symptoms; while Bouchard et al.'s (2007) small, intensive study found individual trajectories of change over the course of therapy with panic self-efficacy as an independent mediator of outcome in 6 of 12 cases while both panic sel-efficacy and catastrophic beliefs mediated change in three of 12 cases. Only one study (Fentz et al., 2013) tested and found a bidirectional relationship between panic symptom severity and panic self-efficacy, not found for catastrophic beliefs. However, no study using repeated measurement included a waitlist-control condition necessary to establish that CBT drives the change in the proposed mediators during therapy (cf. Criterion 1).

[3] 141w A recent study by Gallagher et al. (Gallagher et al., 2013; published after the literature search and therefore not included in the review) also demonstrated temporal precedence of within-person change in both panic self-efficacy and catastrophic beliefs about bodily sensations (i.e. anxiety sensitivity) to within-person reduction in panic symptom severity during an 11-sessions CBT for 361 individuals with PD. Results further indicated that the greatest change in negative thoughts occurred in the first part of treatment, whereas the greatest enhancement of panic self-efficacy occurred in the latter. These results support the assumption that panic selfefficacy and catastrophic beliefs may be independent mediators of change in CBT for PD, and indicate that they may be differentially involved in different parts of the treatment. That different mediators may be differentially involved in different phases of treatment further complicates the design of mediator studies.

[4] 152w Overall, none of the studies in this review tested all of the suggested criteria for mediation for panic self-efficacy as a mediator of therapeutic outcome in CBT for PD. There was rather strong evidence that CBT causes change in panic self-efficacy (criterion 1), and that such change is associated with outcome (criterion 2); in both cases of a magnitude comparable to that found for change in catastrophic beliefs. Very few studies focused on formal mediation (criterion 3) or the sequence of change in mediator and outcome (criterion 4), but results here were at least on par with those for catastrophic beliefs. That none of the studies had designs suitable to test all of the four criteria for mediation is in itself an interesting finding, given the high priority such studies have been given in methodological discussions of psychotherapy research, at least since year 2000 (e.g. Kazdin & Nock, 2003;Kraemer et al., 2002).

[5] 304w Most of the reviewed studies included both panic self-efficacy and catastrophic beliefs in the analyses, however only three studies investigated the causal relationship between the two variables over the course of therapy. One study indicated that change in panic self-efficacy better predicted change in catastrophic beliefs than vice versa (Borden et al., 1991), Meuret et al. (2010) found no priority of change in the CBT condition, but temporal precedence of level of negative thoughts to later levels of panic self-efficacy in the CART condition. Fentz et al. (2013) found no temporal precedence of any of the two variables over the course of CBT. These few findings are inconclusive as to the relationship between panic self-efficacy and catastrophic beliefs during CBT. It is thus premature to conclude that one of the variables causes changes in the other -or even that they both change as a consequence of change in a third variable. Casey, Oei, Newcombe, and Kenardy (2004) have suggested that differing results on the association between panic selfefficacy and outcome in CBT research could be explained by the presence or absence of agoraphobia. Avoidance behavior characteristic of patients with agoraphobia may be associated with lower panic self-efficacy, and therefore panic self-efficacy may be especially important when agoraphobia is present (see e.g. Cox, Endler, Swinson, & Norton, 1992;Telch et al., 1989). It also seems natural to suppose that the two subdimensions of panic self-efficacy related to internal events (panic attacks, anxiety-related bodily symptoms, and negative thoughts) and external ones (feared agoraphobic situations) may be especially relevant to, respectively, PD and agoraphobia. Almost all studies in the review included both types of participants (i.e. patients with PD with or without agoraphobia), but none reported results separately for the two groups. Thus, the review cannot throw light on the possible influence of concurrent agoraphobia on panic self-efficacy.

[6] 77w The studies used various types of questionnaires that tapped into different aspects of panic self-efficacy with about an equal number of questionnaires related to each of the suggested subdimensions of panic self-efficacy (i.e. internal and external events). The divergence in the conceptualization and operationalization of the concept of panic self-efficacy complicates research on its mediational role. Future research could benefit from a clearer conceptual definition of panic self-efficacy and measurement by use of the same standardized scales.

[7] 351w The finding that change in panic self-efficacy, as well as change in catastrophic beliefs, are involved in therapeutic change in CBT for PD may have theoretical as well as clinical implications. We propose further integration of so-called positive cognitions concerning perceived ability to control or cope with panic in the cognitive models of PD, in line with the integrated model of PD developed by Casey et al (2004a). Our findings could also point to individual trajectories of change, with panic self-efficacy playing a major role in some cases, and catastrophic beliefs in others. As indicated by Gallagher et al. (2013) the two proposed mediators may be differently involved in different parts of treatment. Although the causal relationship between the variables is not established by the studies, it seems most likely that they are reciprocally related over the course of therapy. Individuals with low perceived ability to cope with panic could thus be more prone to catastrophic misinterpretations and increased anxiety; or high panic self-efficacy could help individuals disconfirm the catastrophic beliefs. Moreover, individuals with high panic self-efficacy may initiate appropriate coping behavior to enact more personally valued rather than anxiety-driven actions, and/or be more willing to tolerate and accept negative states related to anxiety (Borden et al., 1991;Casey, Oei, Newcombe, & Kenardy, 2004;Fentz et al., 2013). Thus, in therapy it may be important to increase the focus on building the individual's confidence in coping abilities, not only restricted to situation-specific actions, but also incorporate the individuals perceived ability to handle negative internal states such as bodily symptoms and scary thoughts. Increased acceptance of negative internal states is considered an important therapeutic goal in "third wave" therapies like Acceptance and Commitment Therapy (Hayes, Folette, & Linehan, 2004). Paradoxically, a patient with PD could improve panic self-efficacy (perceived control) through acceptance (letting go of control) or distancing ('defusion') from the internal negative states such as negative thoughts. Enhanced panic selfefficacy derived from emotion regulation strategies like avoidance or "over-control" (e.g. inhibition or suppression of negative thoughts and emotions; Gross, 2007) are thus important maladaptive coping strategies to address in the treatment of PD.

[8] 110w The present review has several limitations of which some have already been presented and discussed. Very few studies used advanced mediational methodology to examine panic self-efficacy as a mediator of CBT for PD. We could only perform meta-analyses on data for criteria 1 and 2, which are necessary but not sufficient conditions to establish mediation. The data precluded moderator or subgroup analyses that distinguished participants with or without agoraphobia or different measured sub-dimensions of panic self-efficacy. The large heterogeneity between studies in all conducted meta-analyses may hinder any firm conclusions. Finally, including various measures of a complex construct like panic selfefficacy could potentially limit the possibility of comparing the studies.

[9] 127w In conclusion, some support was found for panic self-efficacy as a mediator of therapeutic change in CBT for PD, although none of the studies fulfilled all of the four criteria for mediation. Mediational studies that included both panic self-efficacy and catastrophic beliefs generally found comparable support for the two putative mediators. Thus, results may indicate that helping patients develop a higher degree of panic self-efficacy, as well as to examine and change their catastrophic beliefs, could be important interventions in CBT for PD. More studies with rigorous mediation designs are, however, needed to draw firm conclusions on the mediational role of panic self-efficacy on therapeutic change in CBT for PD. scales: Self-efficacy scales, SE-thoughts: Self-efficacy in relation to negative thoughts, SE-Walk: Self-efficacy to walk a given distance.

[10] 68w Catastrophic belief measures: AgoCQ: Agoraphobia Cognition Questionnaire, AgoCQ-main: Agoraphobia Cognition Questionnaire e the individual's main dysfunctional belief, AgoCQ-physical: Agoraphobia Cognition Questionnaire e physical consequences subscale, ASI/BSQ combined: Anxiety Sensitivity Index and Body Sensation Questionnaire combined, BBSIQ: Brief Bodily Sensation Interpretations Questionnaire, CBS: Catastrophic Belief Scale, PACQ: Panic Attack Cognition Questionnaire, PAI-conseq: Panic Appraisal Inventory e physical consequences subscale, PAI-likelihood: Panic Appraisal Inventory e perceived likelihood of panic attacks.

[11] 252w Outcome measures: AgoCQ-physical: Agoraphobia Cognition Questionnaire e physical consequences subscale, AgoCQ-social: Agoraphobia Cognition Questionnaire e social consequences subscale, API: Acute Panic Inventory, ASI: Anxiety Sensitivity Index, BAI: Beck Anxiety Inventory, BDI: Beck Depression Inventory, BSQ: Body Sensation Questionnaire, FQ: Fear Questionnaire, FQ-Agoraphobia: Fear Questionnaire e agoraphobia subscale, MI: Mobility Inventory, MI-AAL: Mobility Inventory for Agoraphobiaalone, MI-ACC: Mobility Inventory for Agoraphobia-accompanied, Family life: the degree of impairment of family life, Fear of panic: Fear of panic self-monitoring record, GAS: Global Assessment of Functioning Scale, HARS: Hamilton Anxiety Rating Scale, HDRS: Hamilton Depression Rating Scale, IRGAD: Interview-rated general anxiety, IRpas: interview-rated panic symptoms, IRspon: interview-rated spontaneous panic, LOH: Likelihood of harm, Panic frequency: number of panic attacks during the last week, Panic record: Panic frequency self-monitoring record, PARS1: Phobic Avoidance Rating Scale 1 e avoidance of separation, PARS2: Phobic Avoidance Rating Scale 2 e avoidance of social situations, PARS3: Phobic Avoidance Rating Scale 3 e avoidance of simple phobic situations, PAS; Panic and Agoraphobia Scale, PASQ: Panic Attack Symptoms Questionnaire, Panic thoughts: time spent on negative thinking about panic, PDsit: Panic diary of situational panic attacks, PDspon: Panic diary of spontaneous panic attacks, PDSS: Panic Disorder Severity Scale, SCID-II: the Structured Clinical Interview for the DSMeIIIeR personality disorders, SDS: Sheehan Disability Scale, Social life: the degree of impairment of social life, SPRAS: Sheehan Patient-rated Anxiety Scale, SR-ago: Self-report agoraphobia, STAI: Sheehan Disability Scale, STAI-Y1: State-Trait Anxiety Inventory-state, STAI-Y2: State-Trait Anxiety Inventory-trait, TPAR: Texas Panic Attack Record Form, Work: the degree of impairment of work.

[1] 90w We searched electronic databases (PubMed and PsycINFO) through June 2013, with the following key words in combination with panic disorder or agoraphobia: CBT, cognitive behavior therapy, cognitive behaviour therapy, cognitive behavioural therapy, behavioral therapy, behavioural therapy, behavior therapy, behaviour therapy, exposure, cognitive therapy. Two searches were performed at different times (January and June 2013), one by the first author and one by a research assistant. We checked reference lists from all the included papers for additional relevant studies. H.N. Fentz et al. / Behaviour Research and Therapy 60 (2014) 23e33

[2] 104w To be included, studies had to investigate change in panic selfefficacy during CBT for adult patients (18 years and older) with a principal diagnosis of PD with or without agoraphobia according to a formal diagnostic system, such as the Diagnostic and Statistical Manual of Mental Disorders, Third Edition (American Psychiatric Association, 1980) and later editions, or the International Classification of Diseases, Tenth Revision (World Health Organization, 1992). Treatment was considered CBT if it included cognitive (e.g. problem solving skills or cognitive restructuring) or behavioral treatment methods (e.g. exposure, behavioral experiments, modeling, or social skills training). Only papers published in English-language peer-reviewed journals were evaluated.

[3] 68w The first author (HNF) completed data extraction and the second (MA) checked it with a data extraction sheet. It included verification of study eligibility, diagnostic criteria used, sample size, age, gender mix, type of treatment condition, number of sessions, information on measures of panic self-efficacy, catastrophic beliefs and outcome measures in the study, statistical procedures, and number of participants lost during therapy. Any discrepancies were solved by consensus.

[1] 148w We used meta-analytic syntheses of data to evaluate criteria 1 and 2, and qualitatively review on some studies related to criterion 2, and all studies related to criteria 3 and 4. None of the studies relevant for criterion 3 or 4 reported data necessary for a metaanalysis. In the meta-analyses, results for studies with more than one active CBT condition, more than one measure of panic selfefficacy, and/or more than one measure of therapeutic outcome were standardized and summarized so that each study only provided one mean effect size (ES) in the final meta-analysis (Borenstein, Hedges, Higgens, & Rothstein, 2009). When multiple papers reported findings from the same dataset, we considered data from the publication providing relevant data. In studies that also measured catastrophic beliefs, we compared results for panic self-efficacy with results for this alternative mediator; with regard to criterion 1 only in randomized controlled studies (RCTs).

[2] 151w Related to criterion 1, we calculated a pre-post ES of change in panic self-efficacy in the form of Cohen's d corrected for the prepost correlation between measures (Borenstein et al., 2009). In case no correlation was given or could be calculated (for instance, from t-values), we assumed a correlation of 0.7 as recommended by Rosenthal (1993). We also calculated between-group ESs in the form of Cohen's d from RCTs including a waitlist-condition; both for panic self-efficacy and for the alternative mediator, catastrophic beliefs. For criterion 2, we calculated the correlation between residualized pre-post change in panic self-efficacy and catastrophic beliefs, respectively, and residualized change in outcome based on the Pearson's correlation coefficient (r). Thus, the analyses took into account the pre-therapy level of both the proposed mediators and outcome variable (Cronbach & Furby, 1979). All ESs were calculated as weighted means, thus taking number of participants in each study into account.

[3] 249w We conducted meta-analyses using the Comprehensive Meta-Analysis Program, version 2.2.057 (Comprehensive Meta-Analysis, 2006; see Borenstein et al., 2009). Analyses were based on random-effects models, which yield superior results in terms of clinical interpretability and external generalizability (Borenstein et al., 2009). Q-statistics evaluated significance of heterogeneity in the sample of studies (Cochran, 1954), and I 2 statistics assessed the variance accounted for by heterogeneity. I 2 values of 25%, 50% and 75% indicate low, moderate, and high degrees of heterogeneity, respectively (Higgins & Thompson, 2002). We checked the risk of publication bias, that studies reporting insignificant results or lower ESs are more likely not to be published than studies with higher ESs, by visually inspecting funnel plots (Light, Singer, Willet, 1994) and by means of Egger's test (Egger, Davey, Schneider, & Minder, 1997). A funnel plot is a scatterplot of treatment effect against a measure of study size. A skewed funnel plot with a longer tail in direction of larger ESs for smaller studies may indicate publication bias. Egger's test is a formal, statistical analysis of funnel plot asymmetry. In case of suggested publication bias, we used Duval and Tweedie's (2000) trim and fill method to adjust for possible bias in the overall ES by imputing the ESs of the estimated number of missing studies and recalculating the overall ES. Furthermore, Rosenthal's (1991) fail-safe N addressed potential filedrawer problems. If N exceeded 5K þ 10 (K being the number of studies included in the meta-analysis), no file-drawer problem was suspected.

[4] 199w Twenty-eight studies provided data for a meta-analysis of change in panic self-efficacy from pre-to post-treatment. The overall within-group ES for panic self-efficacy was 1.41 (95% CI 1.21e1.62; ranging from 0.64 to 2.96) indicating a large ES. There was a large between-studies heterogeneity (Q(27) ¼ 154.42, p < .001, I 2 ¼ 82.5%). Furthermore, nine out of the ten randomized controlled studies comparing CBT with a waitlist-condition (not Klein & Richards, 2001) provided data for calculating a controlled between-group ES for panic self-efficacy. The overall controlled ES was 1.46 (95% CI 1.08e1.85; ranging from 0.62 to 2.52) in favor of the treatment condition. There was a moderate to large betweenstudies heterogeneity (Q(8) ¼ 24.6, p ¼ .002, I 2 ¼ 67.5%). The same nine RCTs also provided data on change in catastrophic beliefs with a resulting controlled ES of 1.25 (95% CI 1.66 to 0.83; ranging from 0.19 to 2.22). There was a significant moderate to large between-studies heterogeneity (Q(8) ¼ 25.9, p ¼ .001, I 2 ¼ 69%). No publication bias was suspected in any of the analyses. Thus, the meta-analytic evaluations supported that CBT causes a large increase in panic self-efficacy, comparable to its decrease in catastrophic beliefs.

[5] 215w Criterion 2: demonstrating an association between change in panic self-efficacy and outcome Ten studies provided information on the correlation between change in panic self-efficacy and change in outcome from pre-to post-treatment (Casey, Newcombe, et al., 2005;Casey, Oei, et al., 2005;Cho et al., 2007;Fentz et al., 2013;Feske & de Beuers, 1997;Hoffart, 1995b;Williams & Falbo, 1996;Reilly et al., 2005;Schmidt et al., 1997;Van Apeldoorn et al., 2010). The metaanalysis showed a large correlation between improvement of panic self-efficacy and reductions of panic severity (r ¼ À0.53; CI À0.64 to À0.39; ranging from À0.26 to À0.82). There was a large between-studies heterogeneity (Q(9) ¼ 36.5, p < .001, I 2 ¼ 75.4%). No file-drawer problem was suspected. However, Egger's test indicated significant asymmetry in the funnel plot (p ¼ .04). Duval and Tweedie's trim and fill imputation method, adjusting the overall ES, only reduced the ES marginally (r ¼ À0.47; 95% CI À0.53 to À0.40). Nine of the ten studies (not Reilly et al., 2005) also provided information on the correlation between change in panicrelated catastrophic beliefs and change in outcome from pre-to post-treatment showing an overall r ¼ 0.50 (CI 0.39 to 0.59; ranging from r ¼ 0.30e0.71). There was moderate between-studies heterogeneity (Q(8) ¼ 18.1, p ¼ .021, I 2 ¼ 55.7%). No publication bias was suspected.

[6] 313w In addition, three of the ten studies examined panic self-efficacy and catastrophic beliefs as predictors of panic severity during therapy and were also qualitatively evaluated as to criterion 2 (Casey, Oei, et al., 2005;Cho et al., 2007;Reilly et al., 2005). These studies examined associations between predictor and criterion variables by means of either multiple regression analyses or ANCOVAs. Cho et al. (2007) simultaneously examined change in panic self-efficacy, catastrophic beliefs, and anticipated anxiety as predictors of improvement on four measures of outcome in 120 PD participants in the short and long term, immediately after treatment and at six-month follow-up (FU). Enhancement in panic selfefficacy during therapy only predicted one measure of outcome and only at short-term, whereas change in catastrophic beliefs and anticipated anxiety both predicted two outcome measures at shortterm, and three at six-month FU. Reilly et al. (2005) investigated change in catastrophic beliefs and three measures of panic selfefficacy as predictors of six outcome measures by means of separate ANCOVAs. Results indicated that one measure of panic selfefficacy significantly predicted improvement in three out of six outcome measures at post-treatment, while the remaining two measures of panic self-efficacy predicted improvement in all six outcome measures. Change in catastrophic beliefs only predicted change in two out of six outcome measures. Casey, Oei, et al. (2005) investigated working alliance, panic self-efficacy and catastrophic beliefs as predictors of outcome in three different treatment conditions. In the first six weeks (Phase 1 of a two-phased treatment), all treatment conditions received weekly therapistadministered CBT, after which only two groups continued treatment either as usual or computer-administered the last 6 weeks (Phase 2). Hierarchical regression analyses, with working alliance at step 1, and panic self-efficacy and catastrophic beliefs simultaneously entered at step 2, demonstrated that changes in panic selfefficacy and catastrophic beliefs significantly contributed to the prediction of panic severity reduction in the first phase of therapy.

[7] 103w In the second phase, only change in panic self-efficacy was a significant predictor of improvement. Working alliance did not significantly add to the prediction of panic severity in any of the treatment phases. The meta-analysis of the ten studies showed a large negative association between pre-post improvement in panic self-efficacy and reduction in panic severity with comparable ESs as those found for catastrophic beliefs. Results of the three studies comparing panic self-efficacy and catastrophic beliefs as predictors of therapeutic outcome in CBT were more mixed, although results also lended some support for an association between change in panic self-efficacy and anxiety symptom improvement.

[8] 329w Only three studies reported formal statistical analyses of mediation. Hoffart (1995b) examined change in panic-self-efficacy, catastrophic beliefs, and perceived control of scary thoughts as mediators of reductions in situational fear in 46 patients randomly assigned to CBT or Guided Mastery Therapy (GMT; Williams, 1990) in a six-week intensive inpatient program. Hoffart (1995b) assumed cognitive therapy to work through a decrease in catastrophic beliefs, and GMT through the enhancement of panic self-efficacy. His analysis was thus based on a theoretical hypothesis that mediation was moderated by therapeutic condition. A multiple regression analysis with treatment condition, change in panic-self-efficacy, catastrophic beliefs, and perceived control of scary thoughts entered simultaneously, revealed that only change in panic selfefficacy significantly predicted situational fear in both treatment conditions. To test for statistical mediation of change in the two treatment conditions, interaction terms (mediators  treatment condition) were added. None proved significant. Thus, the analysis did not show a causal relationship between different treatments and specific mediators as hypothesized, nor formal mediation by means of an interaction effect. The design of the study precludes any conclusion as to panic self-efficacy as a non-specific mediator in both treatment conditions, since no waitlist control condition was included. Casey, Newcombe, et al. (2005) examined panic self-efficacy and catastrophic beliefs as mediators of panic severity in a waitlist control RCT of CBT, including 60 patients with PD. Regression analyses indicated that CBT was significantly more effective in reducing panic severity than the waitlist condition, and that both panic self-efficacy and catastrophic beliefs changed significantly more in the treatment condition. Furthermore, both change in panic self-efficacy and catastrophic beliefs predicted outcome in separate regression analyses. Finally, when entered simultaneously, together with treatment condition, both proposed mediators significantly contributed to the prediction of outcome, while treatment condition did not. Thus, treatment effect was no longer significant after controlling for the influence of panic self-efficacy and catastrophic beliefs, suggesting that changes in the two cognitive variables fully explained (i.e. mediated) the treatment effect.

[9] 360w Meuret et al. (Meuret et al., 2010) used an advanced experimental design to directly manipulate proposed mediators; catastrophic beliefs and respiratory regulation in a comparative study of five sessions of cognitive training (CT) or respiratory skill training (capnometry-assisted respiratory training, CART) for 47 patients with PD. Like Hoffart (1995b), their study also primarily investigated moderated mediation, since the mediators were hypothesized to be differentially involved in the two treatments. Panic selfefficacy was, however, also included as a modality-nonspecific mediator, not directly targeted in the two treatment conditions. In each treatment condition, all proposed mediators were assessed five times, once at pre-treatment and once at sessions 2, 3, 4, and 5. Longitudinal multilevel modeling with asymmetric distribution of products test for mediation (MacKinnon, 2008), which calculates the product of the two segments of the mediated pathway, examined indirect mediated pathways. Results showed no difference in post-treatment outcome (panic severity) between the two treatments. The hypothesized specificity of the two treatment mediators was only partially confirmed; respiratory regulation improved specifically during CART while both catastrophic beliefs and panic self-efficacy equally improved in both treatment conditions. The researchers tested mediation by including all proposed mediators simultaneously and an interaction term between (the product of path a and b) in each of the two treatment conditions. Results showed that change in panic self-efficacy mediated change in panic severity regardless of treatment condition. A reversed relation was also seen in both conditions. Change in respiratory regulation mediated change in panic severity only in the CART condition, while catastrophic beliefs mediated reductions in panic severity only for patients receiving CT, although a bidirectional relationship was also found. However, because CT was not proved to specifically cause change in catastrophic beliefs as hypothesized: full statistical mediation was only found for respiratory regulation in the CART condition. As to the role of panic self-efficacy this proposed modality-nonspecific mediator did mediate panic severity in both treatment conditions. However, to fully establish panic self-efficacy as a (nonspecific) mediator of outcome in both CT and CART, a waitlist control condition would have been necessary to establish if treatments did in fact cause the changes in panic self-efficacy (cf. the a path).

[10] 38w Thus, as for criterion 3, three studies tested, but only one study (Casey, Newcombe, et al., 2005) established formal statistical mediation of panic self-efficacy and catastrophic beliefs in accordance with the Baron and Kenny (1986) criteria for mediation.

[11] 57w We qualitatively reviewed four studies examining the causal relation between change in panic self-efficacy and change in outcome; one study (Hoffart, 1998) by including change in outcome during the FU, and three studies (Borden et al., 1991;Bouchard et al., 2007;Fentz et al., 2013) by repeated measurements of the proposed mediators and outcome over the course of therapy.

[12] 556w Based on his aforementioned earlier study (Hoffart, 1995b), Hoffart (1998) examined whether change in panic self-efficacy (measured by S-SESA, a different measure than used in the first paper) and catastrophic beliefs during treatment predicted various outcome measures at one-year FU. A multiple regression analysis revealed that change in catastrophic beliefs during therapy predicted change in outcome from post-treatment to FU (residual gain scores) on five of the 17 measures regardless of treatment condition. Panic self-efficacy was not a significant predictor of any of the outcome measures. Treatment condition (CBT or GMT) had no influence on change in the proposed mediators, rendering a formal mediation test irrelevant according to Baron and Kenny's causal steps approach to mediation (1986). Borden et al. (1991) measured panic self-efficacy and level of panic symptoms nine times (before treatment, at five weekly assessments during treatment, after treatment and at one-and two-month FU) in ten sessions of CBT (guided imaginal coping) against non-specific panic education for 19 PD patients. A time  group ANOVA for the nine assessments of panic self-efficacy revealed a time effect, but no group or interaction effect, indicating similar change of the variable in both groups. Cross-lagged panel analyses for the combined treatment group correlated panic selfefficacy and symptoms at subsequent data points. Five of the eight cross-lagged panels indicated changes in panic self-efficacy preceding changes in symptoms; however, most were nonsignificant and not larger than correlations in the opposite direction. Bouchard et al. (2007) investigated panic self-efficacy and catastrophic beliefs as mediators of change in a study on 15 sessions of exposure or cognitive restructuring for 12 patients with PD. Participants were recruited from an RCT (N ¼ 31) and only included if they were panic-free after treatment. They filled out daily panic diaries from the beginning of a six-week pre-treatment selfmonitoring period to the end of the six-week post-treatment period, a total of 30 weeks (i.e. > 200 daily observations). The diary included scales from 0 to 100 on the most important catastrophic belief, self-efficacy in the presence of panic sensations and panic cognitions, respectively, and degree of panic apprehension. Examining individual change trajectories over the course of therapy by means of time series analyses (Tiao & Box, 1981) and causality testing (Boudjellaba, Dufour, & Roy, 1992) revealed no significant differences between the two treatment conditions. Altogether, change in panic self-efficacy predicted subsequent reductions in panic apprehension in six cases, catastrophic beliefs in three; while both of the cognitive variables were involved in three cases. Fentz et al. (2013) examined change in panic self-efficacy and catastrophic beliefs as mediators of change in anxiety symptoms measured thirteen times during a 13-session group CBT for 46 PD patients referred to a specialized clinic for anxiety disorders. Data were analyzed using linear mixed modeling with separation of the between-and within-person effects of the proposed mediators. Such a strategy makes a more exact evaluation of the within-person change process of the proposed mediators and outcome during therapy possible (Curran & Bauer, 2011). Results revealed a significant treatment effect of both potential mediators and outcome during CBT. Multilevel analyses showed that within-person enhancement in panic self-efficacy in a specific session, but not in catastrophic beliefs, preceded within-persons reductions in anxiety symptoms the subsequent week. A significant reversed relation between anxiety symptoms and panic self-efficacy indicated a reciprocal causal relationship between the variables.

[13] 64w In relation to criterion 4 we found four studies that tested the temporal precedence of change in panic self-efficacy to outcome. Three of the four studies found change in panic self-efficacy occurring before the reduction of panic severity. The one study that tested for a reversed relationship (Fentz et al., 2013) also found that prior change in anxiety caused subsequent change in panic self-efficacy.