Processes of Therapeutic Change: Results from the Cornell-Penn Study of Psychotherapies for Panic Disorder.

Abstract

Objective:

To examine process of changes in two distinct psychotherapies - cognitive-behavioral therapy (CBT) and Panic-Focused Psychodynamic Psychotherapy (PFPP).

Method:

Two hypothesized processes of change – misinterpretation of bodily sensations and Panic Specific Reflective Function (PSRF) – were tested in the CBT and PFPP arms of the Cornell-Penn Study of Psychotherapies for Panic Disorder. The Brief Bodily Sensations Interpretation Questionnaire (BBSIQ) measures misinterpretation of bodily sensations - a focus of CBT interventions. PSRF, a target of PFPP, assesses the capacity to reflect on the underlying meaning of panic symptoms. A sample of 138 patients (37.7% men, 72.56% Caucasians, and 16.7% Latinx) with primary DSM-IV panic disorder were included in the present analyses. Mixed effects models tested the effects of early change in BBSIQ and PSRF (intake through Week 5) on subsequent change in the Panic Disorder Severity Scale (PDSS) (Week 5 through termination).

Results:

Early change on both PSRF and BBSIQ predicted subsequent change in panic severity across the two treatments. As predicted, PSRF changed more in PFPP than in CBT, but, contrary to expectation, BBSIQ showed comparable changes in both groups. Counterintuitively, CBT patients benefited more in terms of panic symptom improvement when their PSRF improved than did PFPP patients.

Conclusion:

This is the first demonstration of general processes of change (PSRF and BBSIQ) across psychotherapies for panic disorder, suggesting that to the extent patients change their beliefs about the meaning of panic, their panic symptoms improve in time-limited, panic-focused psychotherapies.

Testing specific processes of change leading to better treatment outcome of specific psychotherapies is an increasingly important goal of psychotherapy research. However, such studies are rarely conducted, in part because the task of collecting data on competing mechanisms in different treatments is extremely time-consuming, complex, and expensive. The Cornell-Penn Study of Psychotherapies for Panic Disorder (Milrod et al., 2016a) provided such an opportunity. In this study the authors compared the efficacy of two very different approaches to the treatment of panic disorder: Barlow and Craske’s cognitive-behavioral therapy (CBT;Craske, Barlow, & Meadows, 2000) and Milrod’s panic-focused psychodynamic psychotherapy (PFPP, Milrod, Busch, Cooper, & Shapiro, 1997).¹ The relatively large sample (N=138) of patients with panic disorder in these two treatments allowed to test two a priori-hypothesized theoretically relevant processes of change for CBT and PFPP respectively: misinterpretation of bodily sensations (e.g., Clark, 1986) and panic-specific reflective function (PSRF; Rudden, Milrod, Target, Ackerman, & Graf, 2006; Rudden, Milrod, Meehan, & Falkenström, 2009).

Misinterpretation of Bodily Sensations

According to the cognitive model of panic disorder, patients with panic disorder tend to misinterpret bodily sensations, which in turn increases the severity of their symptoms, thus further reinforcing the catastrophic cognitions and creating a vicious cycle of anxiety and panic (Clark, 1986). CBT focuses on breaking this cycle by helping patients learn how to interpret their bodily sensations more accurately.

In comparisons with waiting list control groups, several studies show that CBT leads to greater change in misinterpretations than the passage of time alone (Casey, Newcomb, & Oei, 2005; Clark et al., 1999; Smits, Powers, Cho, & Telch, 2004). Moreover, evidence suggests that CBT may lead to greater change in misinterpretations than other treatments: Clark and colleagues (1994) have shown that patients receiving CBT exhibited greater improvement in their tendency to misinterpret bodily sensations than patients receiving Applied Relaxation Therapy (ART) or medication post-treatment as well as 3 months post-termination (with up to three booster sessions). However, no treatment differences emerged in a 15-month follow up (Clark et al., 1994).

Several CBT studies have demonstrated an association between pre-post treatment reduction in misinterpretations and decrease in panic symptoms post-treatment (Arntz, 2002; Casey et al., 2005), as well as at 12- and 15-month follow ups (Clark et al., 1999; Clark, 1994; respectively). Additionally, in a session-by-session design of CBT for panic disorder, Teachman, Marker, and Clerkin (2010) found that change in misinterpretations predicted subsequent change in panic symptoms over the course of treatment.

Additional research has featured the comparison of the impact of changing misinterpretations in CBT vs. in other treatments. Arntz (2002) found that while misinterpretations were reduced during 12 weeks of both cognitive therapy and interoceptive exposure, only in the cognitive therapy condition were post-treatment levels of misinterpretations correlated with reduced panic symptoms post-treatment as well as at 1- and 6-months follow up. In two studies authors have gone further to explicitly conduct moderated mediation analyses, examining whether the misinterpretation-outcome association differs across treatments. Hofmann and colleagues (2007) investigated the role of catastrophic cognitions as a potential mediator of treatment and outcome. They found that a measure of catastrophic cognitions regarding physical symptoms was a mediator of outcome in conditions including CBT, but not in the medication alone condition. However, this model was based solely on data collected at three time points (pre- and post-treatment and 6-month follow up). Thus, as the authors indicate, it was not possible to establish temporal precedence of the mediator to the outcome. In a second study, Meuret, Rosenfield, Seidel, Bhaskara, and Hofmann (2010) conducted moderated mediation analyses and showed that change in symptom appraisal mediated the treatment-outcome association in the brief (4 weeks) CBT condition, but not in a capnometry-assisted respiratory training condition. While symptom appraisal was measured weekly, symptoms were only assessed pre and post treatment. Thus, as the authors acknowledge, temporal precedence could not be established.

The present study was designed to expand the existing evidence on change in catastrophic misinterpretations as a potential process of change, by investigating its role in two competing psychotherapy conditions and in doing so to account for temporal precedence: That is, to test whether change in misinterpretations precedes subsequent change in panic severity by measuring misinterpretations and severity across treatment. Moreover, we expand the literature by conducting the first test of the role of changes in misinterpretations of bodily sensations in an evidence-based psychodynamic therapy. Based on Clark’s (1986) model and previous findings showing that reduction in misinterpretation of bodily sensations was predictive of symptomatic outcome in CBT but not in alternative treatments (Arntz, 2002; Hofmann et al., 2007; Meuret et al., 2010), we predicted that misinterpretation of bodily sensations would precede and account for change in panic symptoms in CBT, but not in PFPP.

Reflective Function (RF) and Panic-focused Reflective Function (PSRF)

RF is an operationalization of mentalization – the capacity to identify mental states in one’s self and others, and to understand behaviors as reflecting underlying mental states and intentions (Fonagy, Gergely, Jurist, & Target, 2002). It was first studied in the context of attachment (Fonagy, Steele, Steele, Moran, & Higgitt, 1991), and measured by coding patients’ responses to specific questions designed to promote reflection on the Adult Attachment Interview (AAI; George, Kaplan, & Main, 1985). RF is considered a normative developmental achievement arising in part out of formative attachment relationships in childhood, but subject to change in response to life events such as trauma. It has been studied in the context of several psychiatric disorders in which dysregulated central attachment relationships are posited as relevant to symptom formation (e.g., Fonagy et al., 1996). Panic disorder, especially with agoraphobia, is an example of such a condition, as Bowlby (1973) posited that agoraphobia is developed in the context of anxious attachment and is the product of “pathogenic patterns of family interaction” (p. 300).

Based on RF theories, improvement in RF could potentially precede symptomatic change in psychodynamic psychotherapy (Fonagy & Allison, 2014). Psychotherapy researchers have begun studying the role of RF in treatment by testing whether RF changes during psychodynamic psychotherapy (e.g., Fischer-Kern et al., 2015; Levy et al., 2006; Rudden et al., 2006; Taubner et al., 2011; Vermote et al., 2010). One study has also investigated whether pretreatment RF levels are predictive of later symptomatic improvement (e.g., Ekeblad, Falkenström, & Holmqvist, 2016). RF studies have yielded mixed results, perhaps due to the large variance in methodologies, treatment protocols, lengths of treatment, patient populations, and sample sizes. Of note for the present study, Rudden et al. (2006) found no significant change in RF in a small study of the effects of 24 sessions of PFPP. Moreover, no prior studies included the statistical analyses required to empirically test whether RF is in fact a process of change.

An alternative approach to assessing RF, proposed by Rudden and colleagues (2006), is to focus on symptom-specific RF. Measurements of RF per se are broadly designed to capture reflective capacities in a mostly interpersonal framework and are thought to be unlikely to change significantly in a brief, symptom-focused psychotherapy. In contrast, symptom-specific RF was designed to capture the reflective functioning change about symptoms per se in time-limited, symptom-focused psychodynamic psychotherapy such as PFPP (Milrod et al., 1997; Rudden et al., 2006). The central organizing principle of PFPP is the necessity of decoding the psychological underpinnings (underlying meanings) of panic symptoms (Milrod et al., 1997). Indeed, promising results were demonstrated in a pilot study comparing three months of PFPP and ART for panic disorder (Rudden et al., 2006). As expected, RF did not change in this brief treatment, whereas panic specific reflective function (PSRF) improved significantly in the PFPP condition (and decreased in the ART condition). In light of their small sample (N=49), the authors did not examine whether RF and PSRF figured as processes of change in treatment (Rudden et al., 2006).

Given the lack of sufficient empirical work investigating the role of RF/PSRF as a process of change, the current study tested the following a priori hypothesis: PSRF will precede change in symptoms in PFPP, but not in CBT. Based on Rudden et al.’s work (2006), we expected little change in RF and therefore did not hypothesize that RF would predict outcome.

The Present Study

Overall, CBT theorists assert the underlying cause of panic symptoms is distorted cognitions and perceptions. Psychodynamic theorists contend that panic symptoms are biopsychological symbolic events that, like conversion symptoms, contain unconscious, affectively charged significance, accompanied by deficits in one’s awareness and understanding of one’s own emotional world at the core. Thus, patients have difficulty mentalizing panic symptoms and becoming aware of their distressing psychological underpinnings. Whereas both processes of change are based on the assumption that panic patients struggle to make sense of their symptoms, the cause for this difficulty and the interventions formulated to address it are theory-based and distinct.

We tested our hypotheses in the Cornell-Penn Study of Psychotherapies for Panic Disorder. Specifically, we tested whether early change in catastrophic thinking about bodily sensations, RF, and PSRF would predict subsequent change in panic disorder severity in CBT and PFPP.

Method

Participants

Patients with primary DSM-IV PD with or without agoraphobia (N = 201) were recruited at Weill Cornell Medical College (Cornell) and University of Pennsylvania Perelman School of Medicine (Penn). All patients provided written informed consent; both universities’ Institutional Review Boards approved the study. Exclusion criteria included active substance dependence (<6 months’ remission), history of psychosis or bipolar disorder, acute suicidality, organic mental syndromes, medical conditions potentially explaining panic symptoms, involvement in legal or disability proceedings involving the patient’s mental state, and unwillingness to forgo non-study psychotherapies or new medication trials during the protocol. Stable medications (≥2 months’ duration) were permitted and held constant. The sample for the present study comprised that subset (N=138; 70 Penn, 68 Cornell) who provided the necessary data for testing effects of early change of process variables on later change on the outcome variable. Of these, 37.7% were male (n=52), 72.56% were Caucasian (n=100) and 16.7% (n=23) Latinx, 46.4% were cohabiting (n=64), and 76.1% employed (n=102). There was no evidence of imbalance across the treatments (all ps>0.05; for details, see Milrod et al. (2016).

Procedure

Patients at each site were stratified on diagnosis of agoraphobia and major depression before randomization to psychotherapies in a 2:2:1 ratio of CBT, PFPP and ART, respectively. In this study, we omitted the ART condition due to its small sample size and its high dropout rate. Outcome was assessed five times at monthly intervals: at baseline, during treatment (Weeks 1, 5, 10), and at termination (Week 12); process variables were evaluated at intake, Week 5, and termination.

Measures

Panic Disorder Severity Scale

(PDSS; Shear et al., 1997), a 7-item clinician-rated scale providing a diagnosis-based, composite, global rating of panic severity, was our primary dependent variable. Scores are based on the interviewer’s ratings of panic attack frequency, distress due to panic attacks, severity of anticipatory anxiety about panic attacks, avoidance behavior, and impairment in work and social/family functioning. The PDSS has been found to be valid, reliable, and sensitive to change (e.g., Shear et al., 2001). MA or Ph.D. level trained diagnosticians who were uninformed as to treatment assignment administered the PDSS. The PDSS was internally consistent (α = .88) in this sample, and interrater reliability across sites proved excellent (ρI=.95).

Brief Bodily Sensations Interpretation Questionnaire

(BBSIQ; Clark et al., 1997) is a 7-item measure of catastrophic misinterpretation of panic-related bodily sensations. This measure has been found to be internally consistent (α=.86) and reliable over three months (r=.73), with good convergent and discriminant validity relative to other measures of anxious cognition, and criterion-related validity relative to diagnosis (Clark et al., 1997). The BBSIQ was internally consistent in our study (α=.87).

Reflective Function Interview.

This brief, semi-structured interview assesses patients’ general reflective capacity (RF) in relationships. The RF scales derive from the demand questions of the Adult Attachment Interview (AAI; George, Kaplan, & Main, 1985). PSRF assesses patients’ RF in relationship to their panic, that is, the connection between their panic symptoms and awareness of their emotional life (Rudden et al., 2006).

The RF interviewer asks specific questions about the patient’s relationship with early attachment figures (e.g., “Describe your parents”); then the interviewer follows with additional questions designed to promote reflection on mentalization and attachment (e.g., “How do you think your parent came to be that way?”). An example of impaired RF (in response to questions about mother) is the following: “I don’t know. I have no idea. I couldn’t tell you even if I tried.” In contrast, an example of a higher level of RF is “I think she had a pretty tough life. Which is why in the past I felt very sorry for her. Her father isolated her and her sister from their biological mother. She didn’t meet her real mother until she was 35 years old.”

PSRF questions include: Why do you think you have panic attacks? Do you get panic when you are more upset about something? Impaired PSRF protocols in response to “reasons for your panic attacks” may include: “No, it’s the heat. The heat brings them on.” A less impaired PSRF response would be: “I notice that I get them when I am feeling a lack of control in my personal relationships. I fear that others will leave me, or that I may want to leave them.”

One of the coders in the present study (Marie Rudden) was trained in coding RF by Mary Target of the Anna Freud Centre. She then trained the other two coders in rating RF from the Brief RF Interview. Reliability in RF and PSRF ratings among the three independent coders, calculated with intraclass correlations, ranged from acceptable to excellent (Shrout & Fleiss, 1979), between .63 and .91 for RF and from .72 to .80 for PSRF

Treatment

Treatments were delivered individually in 19–24 twice-weekly, 45- to 50-minute sessions within a 16-week interval (Milrod et al., 2016).²

CBT

(Craske et al., 2000) followed the Panic Control Therapy protocol, modified to fit the 24-session format of the trial. CBT emphasized psychoeducation about anxiety and panic, identification and correction of maladaptive thoughts about anxiety and panic, training in diaphragmatic breathing, and interoceptive exposure to bodily sensations designed to mimic those experienced during panic. Patients with persisting agoraphobic avoidance received homework involving in vivo exposure in the final third of treatment.

PFPP

(Milrod et al., 1997) aims to uncover psychological meanings underlying panic and to help patients acknowledge and tolerate underlying feelings connected with panic and anxiety. To this end, treatment explores circumstances and feelings surrounding panic onset, personal meanings of panic symptoms, and feelings and content of panic episodes. Common psychodynamic conflicts in panic disorder addressed include (a) separation and autonomy; (b) recognizing, managing, and coping with expression of anger; and (c) panic as an expression of guilt. This treatment has no exposure component, and no structured psychoeducation.

Therapists

This analysis included 24 therapists (11 women), with a mean of 13.66 years of post-degree experience (SD = 7.36); 6.88 years of experience in time-limited treatment models (SD = 6.14); and ≥1 year of experience in panic disorder treatment. Twelve were MDs, 11 were PhDs, and 1 a PsyD. Each therapist delivered one treatment modality (PFPP or CBT). All therapists received training from experts in the treatment modality as well as ongoing supervision during the trial. We found high adherence in both modalities at both sites (Milrod et al., 2016).

Statistical Analyses

Primary analyses.

For the primary analyses we used mixed effects modeling to test the effects of early change in BBSIQ and PSRF (intake through Week 5) on subsequent change in PDSS scores (Week 5 through termination), controlling for site and the a priori specified covariates of anti-anxiety medication, anti-depressant medication, age, and gender (Milrod et al., 2016). All analyses were conducted in SAS (SAS Institute Inc., 2015). Time was measured as actual calendar time from screening; therefore, post Time 0, each individual could have unique time measurements. The mixed effects structure allowed us to model the repeated BBSIQ and PSRF assessments, while additionally treating time as calendar time (continuous weeks from screening), represented by the two-level hierarchical linear model equations as such:

$$\text{Level 1:}{Y}_{ij}={\beta }_{0i}+{\beta }_{1i}{t}_{ij}+e_{ij}$$

Level 2:

$$\begin{gathered} {\beta }_{0i}\text{= }{\gamma }_{00}\text{+ }{\gamma }_{01}Antideprres{s}_i\text{+ }{\gamma }_{02}Antianxiet{y}_i\text{+ }{\gamma }_{03}Sit{e}_i\text{+ }{\gamma }_{04}T{X}_i\text{+ }{r}_{0i} \\ {\beta }_{1i}={\gamma }_{10}+{\gamma }_{11}Antidepres{s}_i+{\gamma }_{12}Antianxiet{y}_i+{\gamma }_{13}Sit{e}_i+{\gamma }_{14}T{X}_i+r_{1i} \end{gathered}$$

where e_ij is the within-subject error and r_0i and r_1i are the between-subject errors for the intercept and slope component, respectively. For each subject we can get a level 1 estimated intercept (β_0i)and slope terms (β_1i) corresponding to the estimated subject-specific initial and early change estimates. With the limited number of repeated measures, more simplistic covariance structures may need to be specified in the presence of variance/covariance terms going to 0. The most simplistic covariance structure will be a compound symmetry model fit through specifying only a random intercept term.

With the Week 5 through termination portion of the analysis consisting of repeated measures, the mixed effects modeling structure allows us to account for the within-subject correlation of the repeated measures, as well as allowing us to include the early change in BBSIQ or PSRF (i.e., difference score of Week 5 and intake), as well as include the initial level of the respective predictor (BBSIQ or PSRF). Moreover, to establish the temporal precedence of any prediction of outcome by change in the BBSIQ or PSRF, we controlled for early PDSS change in these analyses. Finally, we also examined the potential interactive effects with site, treatment, and treatment by site.

The mixed effects structure for assessing the predictability of early change in the BBSIQ predicting subsequent change in the PDSS is represented by the two-level hierarchical linear model equations as such:

$$\text{Level}1:PDS{S}_{ij}={\beta }_{0i}+{\beta }_{1i}{t}_{ij}+e_{ij}$$

$$\begin{gathered} \text{Level}2:{\beta }_{0i}={\gamma }_{00}+{\gamma }_{01}Antidepres{s}_i+{\gamma }_{02}Antianxiet{y}_i+{\gamma }_{03}{\mathit{\text{Site}}}_i+ \\ {\gamma }_{04}T{X}_i+{\gamma }_{{05}_i}{\beta }_{0i}(PDSS)+{\gamma }_{{06}_i}{\beta }_{1i}(PDSS)+{\gamma }_{{07}_i}{\beta }_{0i}(BBSIQ)+{\gamma }_{{08}_i}{\beta }_{1i}(PDSS)+r_{0i} \\ {\beta }_{1i}={\gamma }_{10}+{\gamma }_{11}Antidepres{s}_i+{\gamma }_{12}Antianxiet{y}_i+{\gamma }_{13}Sit{e}_i+{\gamma }_{14}T{X}_i \\ +{\gamma }_{{15}_i}{\beta }_{0i}(PDSS)+{\gamma }_{{16}_i}{\beta }_{1i}(PDSS)+{\gamma }_{{17}_i}{\beta }_{0i}(BBSIQ)+{\gamma }_{{18}_i}{\beta }_{1i}(PDSS) \\ +r_{0i} \end{gathered}$$

where e_ij is the within-subject error and r_0i and r_1i are the between-subject errors for the intercept and slope component, respectively. The parameters β_0i(PDSS), β_1i(PDSS), β_0i(BBSIQ), and β_1i(BBSIQ), represent the subject-specific intercept and slope for PDSS and BBSIQ respectively from the early change models. The above model can be extended to include interaction terms involving the subject-specific intercept and slope terms.

We conducted preliminary tests for therapists’ effects to determine whether these needed to be incorporated into the models. Clarke (2008) indicates that group level clustering can be ignored when <5% of the variance in change is accounted for by, in this case, therapist effects. In our data, therapist effects accounted for <5% of the variance in early change in the process variables and in the effects of early change in the process variables on later change in PDSS. Indeed, of the six models we tested, therapists effects accounted for almost 0% of the variance in four. Accordingly, we incurred no appreciable loss of model fit in eliminating therapist effects from the models. Power. With a sample size of 138 subjects, we had 95% power to detect a medium effect for our primary prediction analyses (corresponding to a partial eta squared of 0.30).

Missing data.

The present analyses relied upon the sample of patients for whom we had any data on the PDSS during the later portion of the study - Week 10, Week 12 (termination) or both. Of these patients, 94% had a termination PDSS (outcome) score. While this is a small amount of missingness, sensitivity analyses were conducted using multiple imputation techniques to complete the data (Yuan, 2011).

Results

Descriptive data are presented in Table 1. These results indicate that on average on the PDSS patients were rated as moderately to markedly ill before treatment (Furukawa et al., 2009) and scored comparably to other panic disorder patients on the BBSIQ (Clark et al., 1997). Scores on the RF and PSRF interviews indicate that these patients, on average, were lower than normal (a score of 5) on RF and were particularly impaired on PSRF (a score of 3 is considered low/impaired). Correlational analyses examining the overlap among the three process variables indicated that BBSIQ was quite distinct from PSRF and RF (rs ≤.12). As would be expected from their common method variance and focus on the ability to be reflective, PSRF and RF were substantially correlated, r = .50.

Table 1.

Means and Standard Deviations of Potential Predictors and Outcome

	CBT			PFPFP
	CORNELL M (n, SD)	PENN M (n, SD)	TOTAL M (n, SD)	CORNELL M (n, SD)	PENN M (n, SD)	TOTAL M (n, SD)
Intake
BBSIQ	1.87 (36, 0.34)	1.93 (31, 0.45)	1.90 (67, 0.39)	1.96 (30, 0.43)	1.92 (33, 0.38)	1.94 (63, 0.40)
RF	4.38 (36, 1.38)	4.41 (32, 1.27)	4.39 (68, 1.32)	4.02 (32, 1.33)	4.07 (37, 1.16)	4.04 (69, 1.23)
PSRF	3.67 (36, 1.25)	3.69 (31, 1.17)	3.68 (67, 1.21)	3.52 (32, 1.24)	3.49 (37, 1.17)	3.50 (69, 1.19)
PDSS	12.28 (36, 3.38)	14.67 (33, 3.97)	13.42 (69, 3.84)	14.19 (32, 3.25)	14.14 (37, 3.43)	14.16 (69, 3.32)
Week 5
BBSIQ	1.71 (25, 0.38)	1.90 (21, 0.40)	1.80 (46, 0.40)	1.88 (17, 0.40)	1.89 (23, 0.41)	1.89 (40, 0.40)
RF	4.25 (34, 1.47)	3.90 (26, 1.30)	4.10 (60, 1.40)	3.84 (29, 1.43)	4.02 (31, 1.08)	3.93 (60, 1.26)
PSRF	3.67 (33, 1.31)	3.64 (25, 1.11)	3.66 (58, 1.22)	4.09 (29, 1.24)	4.26 (31, 1.06)	4.18 (60, 1.14)
PDSS	8.88 (34, 3.78)	10.12 (26, 5.04)	9.42 (60, 4.37)	10.07 (28, 4.23)	10.06 (31, 3.41)	10.07 (59, 3.79)
Week 12
BBSIQ	1.62 (32, 0.31)	1.51 (26, 0.26)	1.57 (58, 0.29)	1.80 (25, 0.39)	1.85 (28, 0.42)	1.83 (53, 0.41)
RF	4.32 (31, 1.17)	4.43 (23, 1.09)	4.37 (54, 1.13)	4.33 (27, 1.36)	4.50 (31, 1.20)	4.42 (58, 1.27)
PSRF	3.69 (31, 1.07)	3.65 (23, 1.17)	3.68 (54, 1.10)	4.12 (25, 1.63)	4.67 (31, 1.11)	4.42 (56, 1.38)
PDSS	5.83 (29, 3.65)	5.22 (27, 3.84)	5.54 (56, 3.72)	6.71 (28, 4.57)	7.47 (32, 4.15)	7.12 (60, 4.33)

Note. CBT = cognitive behavioral therapy; PDSS= Panic Disorder Severity Scale; BBSIQ=Brief Bodily Sensations Interpretations Questionnaire; PFPP = panic focused psychodynamic psychotherapy; RF = Reflective Function; PSRF=Panic-Focused Reflective Function.

BBSIQ

Change on the BBSIQ from intake to Week 5.

Unexpectedly, we found no differences between treatments in early change on the BBSIQ, F(1,80) = 0.03, p = .87, d = .04. These results were consistent across sites in that the Site X Treatment interaction was not significant, F(1,78) = 0.06, p = .82, d = .05. Model-based change estimates from intake to Week 5 were −0.08 (SE = 0.04) for CBT, and −0.06 (SE = 0.05) for PFPP, reflecting minimal change for both groups on average for the BBSIQ by this point in treatment (negative slopes on the BBSIQ indicate improvement). While the rates of change did not differ, we investigated whether there were intervention differences in the variance-covariance matrix of the repeated assessment between CBT and PFPP. Heterogeneity analysis of the variance-covariance matrix yielded no significant difference between the two treatments (χ²(3)=2.7, p=0.44).

Early change on the BBSIQ as a predictor of subsequent symptom change.

We tested whether early BBSIQ change predicted subsequent PDSS change. We focused first on the higher order terms (3-way and 2-way interactions) to examine whether any BBSIQ association with subsequent change was dependent on site and/or treatment. Thus, we fitted the multilevel model sequentially starting from higher order models to the most parsimonious model. We found a non-significant BBSIQ X Site X Treatment interaction (F(1,69) = 0.81, p = 0.37, d = 0.20), a non-significant BBSIQ X Treatment interaction (F(1,70) = 0.01, p = 0.95, d = 0.01), and a non-significant BBSIQ X Site interaction, F(1,70) = 2.59, p = 0.11, d = 0.36. In contrast, the prediction effect for BBSIQ on subsequent PDSS change was significant and medium in size, F(1,70) = 4.41, p=0.04, d = 0.66). Thus, contrary to prediction, early change in BBISQ was not only significantly associated with subsequent change in the PDSS in CBT but also in PFPP. We again investigated whether there were intervention differences in the variance-covariance matrix of the repeated assessment between CBT and PFPP. Heterogeneity analysis of the variance-covariance matrix yielded no significant difference between treatments (χ²(6)=6.9, p=0.33).

Reflective Functioning

We found no significant differences between treatments on early change in RF (F(1,117) = 0.15, p = .69, d = .07). These results were consistent across sites in that the Site X Treatment interaction was not significant (F(1,115) = 0.35, p= .55, d = .11). RF did not improve in either group. Model-based change estimates from intake to Week 5 were −0.09 (SE = 0.16) for CBT, and −0.17 (SE = 0.15) for PFPP, reflecting on average minimal change for both groups (positive slopes in RF indicate improvement). Heterogeneity analysis of the variance-covariance matrix of the repeated measures between CBT and PFPP yielded no significant difference (χ²(2)=1.8, p=0.41). Applying again the approach we used with the BBSIQ by focusing first on the higher order terms and reducing terms subsequently to the most parsimonious model, we found a non-significant RF X Site X Treatment interaction (F(1,96) = 0.05, p = 0.82, d = 0.05), a non-significant RF X Treatment interaction (F(1,97) = 0.69, p = 0.41, d = 0.17), and a non-significant RF X Site interaction, F(1,97) = 0.34, p = 0.56, d = 0.12). Critically, the prediction effect for RF was not significant, and the effect size was very small (F(1,98) = 0.02, p = 0.88, d = 0.03). Thus, consistent with our expectations, early change in RF is not significantly associated with subsequent change in the PDSS for PFPP or CBT. Heterogeneity analysis of the variance-covariance matrix of the repeated measures between PFPP and CBT yielded no significant difference (χ²(4)=6.6, p=0.16).

Panic Specific Reflective Functioning

Change on the PSRF from intake to Week 5.

In the case of early PSRF change, we found for the first time a significant difference between the treatments (F(1,114) = 5.55, p = .02, d = 0.42). Despite the on-average difference in magnitude, we did not see significant heterogeneity in the variability between interventions (χ²(1)=1.53, p=0.16). As illustrated in Figure 1, consistent with prediction, from intake to Week 5 there was a significant improvement in PSRF in PFPP (t(114) = 4.02, p < .0001, d = 0.68) (positive change indicates greater PSRF improvement), whereas in CBT there was not (t(114) = 0.53, p = .60, d = 0.10). These results were consistent across sites in that the Site X Treatment interaction was not significant (F(1,112) = 0.05, p = .82, d = .04).

Figure 1.

Change in Panic Specific Reflective Function (PSRF) from intake to Week 5 as a function of treatment. Positive slopes on the PSRF indicate improvement. CBT = cognitive behavioral therapy, PFPP = panic focused psychodynamic psychotherapy, Pre = Intake, Wk = Week. Improvement is significant in PFPP (p < .001) but not in CBT, p = .60.

Early change on the PSRF as a predictor of subsequent symptom change.

Repeating our approach with the BBSIQ and RF by focusing first on the higher order terms, we found a non-significant PSRF X Site X Treatment interaction (F(1,96) = 0.00, p = 0.95, d = 0.01) and a non-significant PSRF X Site interaction (F(1,97) = 0.98, p = 0.22, d = 0.25). The prediction effect for PSRF was significant with a medium effect size (F(1,98) = 7.74, p = 0.006, d = 0.56), but in this case it was modified by a significant PSRF X Treatment interaction (F(1,97) = 5.68, p = 0.018, d = 0.48). In contrast to BBSIQ findings, the magnitude of the association between early change in PSRF and subsequent change in the PDSS is dependent on the intervention arm (see Figure 2).

Figure 2.

Early Change in Panic Specific Reflective Function (PSRF) from intake to Week 5 as a function of treatment predicting subsequent reduction on the Panic Disorder Severity Scale (PDSS). Positive slopes on the PSRF indicate improvement. Mean represents the average level of early change in PSRF; −1SD, −2SD, +1SD, and +2SD are the respective standard deviation units below/above the mean early change in PSRF. Average early rate of change per week on the PSRF is 0.027 (SD = 0.1501). Observed range on the standardized scale for our sample was −2.7 to 2.1.

Dissecting the PSRF X Treatment interaction (which held across both sites), we found that for both treatment groups more early change on the PSRF was associated with significantly greater later change on the PDSS. However, opposite to our prediction, that association was stronger for CBT than for PFPP: On average CBT patients experienced a 1.66 point drop in the PDSS per each standard deviation of change in PSRF (SE = 0.31, t(125) = 5.41, p < .0001, d = 1.41), whereas the comparable decrease for PFPP patients was only 0.62 (SE = 0.31, t(125) = 2.03, p = .04, d =.52). Heterogeneity analysis of the variance-covariance matrix of the repeated measures between CBT and PFPP yielded no significant difference (χ²(2)=3.0, p=0.22)

Sensitivity analyses.

Sensitivity analyses using multiple imputation approaches (Yuan, 2011) yielded results consistent with those reported from the observed data. All significant findings were maintained.

Discussion

We examined processes of change in two different forms of brief psychotherapy for panic disorder. In all our analyses, we carefully took time sequence into consideration (Kraemer, Wilson, Fairburn & Agras, 2002; DeRubeis & Feeley, 1990).

We have three main findings: First, early change on BBSIQ predicted subsequent change in panic disorder severity in both CBT and PFPP, which we had not predicted. This result held at both of our sites, thus providing an internal replication. Correcting patients’ catastrophic misinterpretations of panic-related bodily sensations early in treatment led to symptomatic improvement, regardless of psychotherapy type or treatment strategy. Thus, the impact of cognitive change is not limited to treatments in which maladaptive cognitions are directly targeted per se. These findings are consistent with Clark et al.’s (1994) follow-up study in which termination scores on the BBSIQ were predictive of patients’ symptom change during follow-up regardless of whether they had received CBT, ART, or pharmacotherapy. Others have also emphasized the lack of strong evidence supporting the specificity of this mechanism of change to CBT (Smits, Julian, Rosenfield, & Powers, 2012). Our results provide strong evidence for Clark’s (1986) cognitive model of panic disorder.

Our second main finding was that, consistent with our a priori prediction, RF did not significantly improve in these short term treatments, and change in RF did not significantly predict subsequent change in panic symptoms, regardless of treatment condition. Although our results cannot prove the null hypothesis, they make sense: Improvements in overall RF, which subsumes broad improvements in ego strength and capacities (Fonagy et al. 1996), are unlikely to occur after brief psychotherapies designed primarily to unravel meanings of symptoms.

Our third main finding was that, consistent with our a priori prediction, patients showed significant improvement in PSRF early in treatment in the PFPP condition but not in CBT.Moreover, like BBSIQ, PSRF predicted subsequent change in PDSS across both treatments. Counterintuitively, however, we found that early PSRF improvement was more strongly related to better outcome for CBT patients than for those in PFPP. Given that these results were unexpected, we believe these data must be taken with caution pending replication.

Based on previous studies that did not find change in RF in CBT (Karlsson & Kermott, 2006) and short-term psychodynamic therapy (Rudden et al.,2006; Taubner et al., 2011; Vermote et al., 2010;), and because PFPP is a brief therapy that specifically uses panic symptoms as a lens through which to understand patients’ symptoms and life experiences, we predicted that patients in our study would show significant change in PSRF but not in RF. Accordingly, we expected PSRF but not RF to predict outcome. That PSRF and not RF proved to be a significant process in change of panic disorder severity buttresses arguments for the importance of a focus on symptoms in short-term dynamic therapy for panic disorder.

Improvement in PSRF was linked to better outcome in both treatments at both sites. Nonetheless, PSRF is likely more relevant to the practice of PFPP than CBT as practiced in this protocol, given the emphasis on facilitating PSRF in panic-focused psychodynamic therapy. In another study of the current sample, we found that greater focus on patients’ interpersonal difficulties and their association with panic facilitated change in PSRF in both treatments (Solomonov et al., 2019). Thus, although therapists in CBT and PFPP may differ in how they approach interpersonal problems the patient brings into the session, when they do address such problems, both treatments may help patients improve their capacity to reflect on their panic symptoms. Theoretically, PSRF taps into the self domain of attachment: the ability to emotionally understand one’s own affective experiences and to soothe oneself (Bowlby, 1980, 1988; Fonagy, Gergely, Jurist, & Target, 2002), which specifically relates to development and maintenance of anxiety. Early childhood attachments are hypothesized to affect development of psychological capacities for anxiety tolerance, affect modulation, and reflective function, and to have far-reaching consequences on interpersonal functioning and symptomatic distress in adulthood (Bateman & Fonagy, 2012).

From our data, improvement in the ability to reflect on self-states connected with panic symptoms reduces panic symptoms per se, and perhaps suggests one way that panic might arise, namely through dysregulated attachments established as childhood premorbid patterns, and underdeveloped capacities for anxiety tolerance/tolerance of uncertainty, affect modulation, and autonomy (Milrod et al., 2016b). These process findings help to articulate specific attachment elements that are dysregulated in panic disorder, advancing our understanding of the syndrome.

How do we make sense of our finding that CBT patients got a greater boost from changes in PSRF than did PFPP patients? Keeping in mind that these findings require replication and that on average CBT patients did not improve on PSRF, we tentatively suggest the following: Although in CBT, therapists do not assume that unacknowledged emotion underlies a patient’s panic attacks, some patients do bring such material into the session. If so, this then becomes a topic of discussion. CBT patients are equipped by their treatment explicitly to use what they have learned about the connection between unacknowledged emotions, interpersonal difficulties, and panic attacks to challenge their catastrophic thinking about panic. This training may allow them to make quicker use of the enhanced insight reflected in PSRF to ameliorate their panic severity. Moreover, these findings may suggest that CBT clinicians would do well to be more alert to emotional conflict that might trigger panic in their patients. Fear of anxiety symptoms (highlighted in CBT) has been found to be strongly related to fear of the experience of other emotions (Williams, Chambless, & Ahrens, 1997). While an early cognitive-behavioral model of panic disorder with agoraphobia (Goldstein & Chambless, 1978) emphasized the importance of unacknowledged or disavowed emotions as stimuli for panic attacks, over time the predominant treatment models (e.g., Clark, 1986) focused more exclusively on misinterpretation of bodily sensations as the driving force in panic disorder. Something useful may have been lost.

Our study has two significant limitations. First, as in most clinical trials, we did not have complete data for the intention-to-treat sample. For all analyses, we had to omit patients who did not provide any outcome data from Week 5 to termination or who did not complete the process measures. Second, as is typically the case for studies of processes of change, our approach is partly correlational, and we cannot ignore the possibility that unmeasured variables might have accounted for the apparent relationship between the processes of change and treatment outcome. However, strengths of our study are that we carefully took time sequence into consideration in our analysis, which allows us to come closer to a causal demonstration than was possible in past research, and we conducted sensitivity analyses using multiple imputation that suggested our results were not distorted by missing data.

In summary, we identified two significant processes of change in psychotherapy for panic disorder: Early change on both the BBSIQ and PSRF predicted subsequent change in panic disorder severity across both sites and both treatments. In future analyses we will investigate which specific therapeutic process elements in these treatments are associated with changes in BBSIQ and PSRF. Looking at the common theme in our two constructs suggests the following conclusion: Therapists can help panic patients change by altering their understanding of their panic attacks (Summers & Barber, 2009). This can be done by helping them make fewer catastrophic interpretations (Clark, 1986), or by increasing their capacity to reflect and mentalize (Fonagy, Luyten, & Bateman, 2015) specifically about their panic attacks. Either way, patients improve because they have reduced their belief that something awful, out of control, and damaging is happening to them.

Public Health Significance Statements.

Panic disorder is associated with disability and personal suffering, posing a significant public health problem not fully remediated by available treatments. Identifying processes of change associated with symptomatic improvement can guide development of targeted interventions and potentially improve response rates. We found that both the ability to reflect on panic symptoms and to accurately interpret one’s bodily sensations predict subsequent symptomatic improvement in both CBT and brief psychodynamic therapy. This suggests that a sharper focus on these elements has the potential to improve outcomes in short-term treatments of panic disorder.