Journal of Consulting and Clinical Psychology © 2017 American Psychological Association

2017, Vol. 85, No. 8, 803– 813 0022-006X/17/$12.00 http://dx.doi.org/10.1037/ccp0000224

Prediction and Moderation of Improvement in Cognitive-Behavioral and

Psychodynamic Psychotherapy for Panic Disorder
Dianne L. Chambless Barbara Milrod
University of Pennsylvania Weill Cornell Medical College

Eliora Porter Robert Gallop

University of Pennsylvania West Chester University

Kevin S. McCarthy Elizabeth Graf

Chestnut Hill College New York, New York
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Marie Rudden Brian A. Sharpless

Weill Cornell Medical College Argosy University, Northern Virginia

Jacques P. Barber
Adelphi University

Objective: To identify variables predicting psychotherapy outcome for panic disorder or indicating which of
2 very different forms of psychotherapy—panic-focused psychodynamic psychotherapy (PFPP) or cognitive-
behavioral therapy (CBT)—would be more effective for particular patients. Method: Data were from 161
adults participating in a randomized controlled trial (RCT) including these psychotherapies. Patients included
104 women; 118 patients were White, 33 were Black, and 10 were of other races; 24 were Latino(a).
Predictors/moderators measured at baseline or by Session 2 of treatment were used to predict change on the
Panic Disorder Severity Scale (PDSS). Results: Higher expectancy for treatment gains (Credibility/
Expectancy Questionnaire d ⫽ ⫺1.05, CI95% [⫺1.50, ⫺0.60]), and later age of onset (d ⫽ ⫺0.65, CI95%
[⫺0.98, ⫺0.32]) were predictive of greater change. Both variables were also significant moderators: patients
with low expectancy of improvement improved significantly less in PFPP than their counterparts in CBT,
whereas this was not the case for patients with average or high levels of expectancy. When patients had an
onset of panic disorder later in life (ⱖ27.5 years old), they fared as well in PFPP as CBT. In contrast, at low
and mean levels of onset age, CBT was the more effective treatment. Conclusions: Predictive variables
suggest possibly fruitful foci for improvement of treatment outcome. In terms of moderation, CBT was the
more consistently effective treatment, but moderators identified some patients who would do as well in PFPP
as in CBT, thereby widening empirically supported options for treatment of this disorder.

What is the public health significance of this article?

Panic disorder is associated with substantial societal and economic costs. With current psychother-
apies, as many as a third of patients do not benefit substantially from treatment. Identifying predictors
of better outcome may point to ways to improve treatment delivery, whereas identifying moderators
of response to particular psychotherapies may allow a more personalized approach to treatment. In
this study expectancy of improvement and later age of onset both predicted better outcome and
moderated response to cognitive-behavioral therapy (CBT) versus panic-focused psychodynamic
therapy (PFPP) such that CBT was the more beneficial treatment when patients were low in
expectation of improvement and when patients’ panic disorder began before age 27.5.

Keywords: panic disorder, agoraphobia, psychotherapy moderation, psychotherapy outcome prediction

Supplemental materials: http://dx.doi.org/10.1037/ccp0000224.supp

This article was published Online First June 26, 2017.
Dianne L. Chambless, Department of Psychology, University of Penn-
sylvania; Barbara Milrod, Department of Psychiatry, Weill Cornell Med-
ical College; Eliora Porter, Department of Psychology, University of
Pennsylvania; Robert Gallop, Department of Mathematics, West Chester
University; Kevin S. McCarthy, Department of Psychology, Chestnut Hill
College; Elizabeth Graf, Independent Practice, New York, New York;
Marie Rudden, Department of Psychiatry, Weill Cornell Medical College;
Brian A. Sharpless, American School of Professional Psychology, Argosy
University, Northern Virginia; Jacques P. Barber, Derner Institute of Ad-
vanced Psychological Studies, Adelphi University.
This research was supported by National Institute of Mental Health
grants to Barbara Milrod (Grant R01 MH70918) and to Jacques P. Barber
and Dianne L. Chambless (Grant R01 MH 070664).
Correspondence concerning this article should be addressed to Dianne L.
Chambless, Department of Psychology, University of Pennsylvania, 425 South
University Avenue, Philadelphia, PA 19104-6018. E-mail: chambless@
psych.upenn.edu

803
 804 CHAMBLESS ET AL.
Panic disorder is associated with substantial societal and eco-
nomic costs, including high rates of health service utilization and
absenteeism from work (Batelaan et al., 2007; de Graaf, Tuithof,
van Dorsselaer, & ten Have, 2012), poor physical health, de-
creased quality of life, and impairments in social and occupational
functioning (e.g., Markowitz, Weissman, Ouellette, Lish, & Kler-
man, 1989). Fortunately, effective treatments for this condition
exist. A wealth of evidence supports the efficacy of CBT in the
treatment of panic disorder (see Mitte, 2005, for a meta-analysis
and review). More recently, Milrod, Busch, Cooper, and Shapiro
(1997) developed PFPP, a time-limited, manualized psychody-
namic treatment that specifically targets panic symptoms. In a
small RCT, PFPP outperformed applied relaxation training (ART)
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
in the reduction of panic symptoms and improvement in psycho-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
social functioning (Milrod, Leon, Busch, et al., 2007). Other
supportive evidence of PFPP’s efficacy comes from a study of its
effectiveness in clinical practice (Beutel et al., 2013).
No single psychotherapy is effective for all patients with panic
disorder. For example, in a recent large clinical trial of three
psychotherapies (Milrod et al., 2016), 54% of patients failed to
respond to ART, 41% to PFPP, and 37% to CBT. Thus, research-
ers have sought to identify which factors explain this variability in
treatment response. A key question is whether there are individual
differences that reliably predict better response to one treatment
over another. These variables are referred to as moderators and,
once identified, they can be used to guide treatment selection. A
related question is whether there are certain types of patients who
do not fare well in any existing treatment. Predictor variables
predict treatment outcome irrespective of the type of treatment,
and the identification of such variables is an important first step
toward developing more effective treatments for individuals who
do not benefit from existing interventions.
In the present study, we use data from the Milrod et al. (2016)
trial to examine predictors and moderators of response to CBT and
PFPP. Although both treatments are symptom-focused, they rep-
resent two very different approaches to treatment in theory and in
execution. CBT is a very directive treatment involving psychoe-
ducation, cognitive restructuring, and interoceptive and in vivo
exposure, with copious homework designed to help the patient
become less fearful by changing catastrophic thinking about panic
and relinquishing avoidance behaviors. In contrast, PFPP is a
short-term, transference-based psychodynamic treatment that es-
chews homework assignments and focuses on helping patients to
uncover the unconscious meanings of their panic symptoms and to
resolve core unconscious conflicts leading to panic. Frequent foci
are recognizing, managing, and expressing anger and guilt and
conflicts over separation and autonomy. The stark contrast be-
tween these treatments suggests it should be possible to identify
patients who would benefit more from one than the other.1
Predictors and Moderators of Response to CBT for
Panic Disorder
We began variable selection for this study by conducting a
systematic review and meta-analysis of the literature on predictors
and moderators of response to CBT for panic disorder and/or
agoraphobia (Porter & Chambless, 2015). Although the number of
studies for any single variable was small, we found that higher
levels of pretreatment agoraphobic avoidance, lower expectancy of
change, and greater pretreatment functional impairment consis-
tently predicted worse response to CBT in this population. There
was also some indication in this review that individuals with a later
age of onset and a shorter duration of disorder fared better in CBT,
but these variables were conceptualized somewhat differently
across studies, making the literature difficult to synthesize. Finally,
this meta-analysis indicated that the presence of a comorbid Clus-
ter C personality disorder was predictive of decreased improve-
ment in CBT for panic disorder and/or agoraphobia. However,
results from the individual studies included in the meta-analysis
were mixed, with one study finding that patients with Cluster C
disorders fared worse, and two studies finding no significant
effects. Despite exhaustive search procedures, we located no study
that examined moderators of response to a CBT treatment as
compared with another form of psychotherapy.
It seems highly likely that some of the identified predictors may
not be specific to CBT, but instead might predict response to
psychotherapy more broadly. For example, individuals who expect
more benefit from therapy at the outset of treatment are likely to
make greater gains in any form of psychotherapy (Constantino,
Glass, Arnkoff, Ametrano, & Smith, 2011). Similarly, we might
expect individuals with an earlier age of onset, longer duration of
disorder, greater agoraphobic avoidance, and greater functional
disability to do more poorly in any form of psychotherapy than
patients without these pretreatment characteristics in light of the
chronicity and severity of their condition. Although the literature
on Cluster C diagnosis as a predictor of outcome for anxiety
disorders is scant and mixed, in a 2003 review Reich concluded
that, overall, personality disorders (including but not limited to
Cluster C) are a predictor of poor outcome for anxiety treatments
across the board, with the possible exception of selective seroto-
nergic reuptake inhibitors (Reich, 2003).
Predictors and Moderators of Response to
Psychodynamic Therapy for Panic Disorder
There is little research on predictors of response to psychody-
namic treatments for panic disorder. Beutel et al. (2013) found that
higher emotional awareness and lower global illness ratings at
baseline predicted more change in treatment for patients with panic
disorder receiving either PFPP or CBT. The preponderance of
patients their study received PFPP, but the sample sizes were too
small to test prediction of PFPP alone. Existing papers on moder-
ation of response to psychodynamic treatments for panic disorder
have all investigated moderators of PFPP versus ART in a single
small, exploratory study (Milrod, Leon, Busch, et al., 2007), which
found that PFPP was superior to ART. No statistically significant
moderators were identified, which is unsurprising given the small
sample size in this trial. In several other papers Milrod and
colleagues sought to examine moderators in their study, but in light
of their small sample appropriately did not employ statistical tests
of their hypotheses (Kraemer, Wilson, Fairburn, & Agras, 2002).
However, there is some suggestion from the prepost effect sizes for
1
Milrod et al. (2016) included a third treatment group, ART. Because
this condition was, by design, randomized only half the number of patients
as the other two conditions, and because the drop-out rate was very high in
ART (41%), we judged the ART group too small to include in these
analyses.
 PSYCHOTHERAPY FOR PANIC DISORDER
different subgroups of patients that although PFPP was superior to
ART overall, the difference in favor of PFPP was even greater for
patients whose panic disorder onset was not precipitated by a
recent interpersonal loss (Klass et al., 2009) and for those with a
comorbid Cluster C diagnosis (Milrod, Leon, Barber, Markowitz,
& Graf, 2007). Patients with and without a comorbid Cluster B
diagnosis appeared to fare similarly well in each treatment, but the
number of Cluster B patients was so small as to limit any conclu-
sions (Milrod, Leon, Barber, et al., 2007).
Theorized Moderators of Response to CBT and PFPP
Because the empirical literature provides little guidance about
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
which variables might moderate response to CBT and PFPP, we
This document is copyrighted by the American Psychological Association or one of its allied publishers.
turned to theory and our clinical expertise in treating panic disor-
der to develop hypotheses about potential moderators. Suspecting
that some of our hypothesized predictors of response to treatment
might also serve as moderators, we predicted that patients with
greater functional disability and longer durations of disorder would
fare better in CBT as compared with PFPP. Because of the chro-
nicity of their disorder and the likelihood that these patients have
adapted their lives to panic disorder, they might benefit most from
a more structured psychotherapy like CBT, in which the therapist
is relatively directive and is prepared to break the change process
down into small steps. We also predicted patients higher in pre-
treatment agoraphobic avoidance would do better in CBT as com-
pared with PFPP, because CBT is highly directive in urging
patients to confront feared situations and to decrease their avoidant
behavior, whereas PFPP encourages such confrontation more in-
directly through transference interpretation.2
A goal of PFPP and of psychodynamic therapies more broadly
is to help patients to gain insight into the underlying meanings of
their symptoms as they occur in real time with the goal of devel-
oping greater control through improved insight and ability to
acknowledge, tolerate, and reflect on the feelings that lead to their
symptoms. Some writers have theorized that patients with higher
pretreatment insight may fare better in psychodynamic therapy
than their low insight counterparts (Barber, Muran, McCarthy, &
Keefe, 2013). Indeed, Beutel et al. (2013) found that patients with
panic disorder who were higher in emotional awareness fared
better in treatment. Closely related to the concepts of insight and
emotional awareness, reflective functioning (RF) is defined as the
ability to understand one’s own and others’ behavior in terms of
mental states (desires, intentions, feelings, and beliefs; Fonagy,
Target, Steele, & Steele, 1998). When designing this study, we
initially hypothesized that patients high in pretreatment RF would
fare equally well in CBT and PFPP, but that those with low
pretreatment RF would do better in CBT than in PFPP. However,
recent data raise doubt about this hypothesis. Ekeblad, Falken-
ström, and Holmqvist (2016) reported that having higher scores on
RF predicted better treatment outcome in a trial of CBT and
interpersonal therapy (IPT) for depression in a group of patients
with very impaired pretreatment RF. Contrary to our initial hy-
pothesis in this study, Ekeblad et al. (2016) found no moderation
by type of treatment. However, IPT is less insight oriented than
PFPP, and thus these results might not generalize to a more
intrapsychically oriented psychodynamic therapy. Accordingly,
additional research is needed: RF might prove to be a predictor of
805
outcome across both types of psychotherapy or a moderator of
treatment response.
We also suspected that patients’ personalities might moderate
response to CBT and PFPP, specifically that patients with comor-
bid Cluster C personality disorders would fare better in PFPP than
in CBT. Such a finding would be consistent with results from our
meta-analysis (Porter & Chambless, 2015), in which we found that
a comorbid Cluster C diagnosis was predictive of less improve-
ment in CBT, and Milrod, Leon, Barber et al.’s (2007) paper,
which suggested that patients with a comorbid Cluster C diagnosis
did especially well in PFPP as compared with ART. Some research
suggests that the impact of Cluster C comorbidity on response to
treatment may depend on the specific Cluster C disorder involved.
For example, Barber and Muenz (1996) found that depressed
patients with comorbid avoidant personality disorder fared better
in cognitive therapy (CT) than in a more affect-focused treatment
(IPT), whereas those with comorbid obsessive-compulsive person-
ality disorder (OCPD) fared better in IPT as opposed to CT. Given
the preponderance of OCPD among patients in our sample who
met criteria for a Cluster C diagnosis at pretreatment (40 of 54
Cluster C patients or 74%), these findings increase the likelihood
that patients with a Cluster C disorder will improve more in PFPP,
which, like IPT, is strongly affect focused, than in CBT.
Finally, we were interested in whether the experience of child-
hood loss might serve as a moderator. Such loss is a risk factor for
separation anxiety (Silove et al., 2015), which, in turn, is impli-
cated as a risk factor for a host of psychiatric disorders, including
panic disorder. Indeed, in his classic work, Bowlby (1973) posited
that separation and loss in childhood result in dysregulated attach-
ment and the lack of a sense of safety, which, in the extreme, may
then result in agoraphobia. We are unaware of any studies exam-
ining childhood loss as a predictor or moderator of response to
CBT or psychodynamic therapy for panic disorder. However,
Nemeroff et al. (2003) found that depressed patients who experi-
enced childhood loss fared significantly better in the cognitive-
behavioral analysis system of psychotherapy, a time-limited ther-
apy that combines techniques from CBT, IPT, and psychodynamic
therapy, than on antidepressant medication. We hypothesized that
patients in our trial who had experienced a childhood loss might do
better in PFPP than in CBT because in PFPP dysregulated attach-
ments, such as those that commonly occur in patients who have
sustained childhood loss, are a focus of treatment.3
The Present Study
We sought to examine predictors and moderators of response to
psychotherapy in the context of the Cornell-Penn Psychotherapies
for Panic Disorder Study, a multisite RCT in which the benefits of
CBT and PFPP for patients with a primary diagnosis of DSM–IV
2
In PFPP the therapist explores the underlying emotional meanings of
avoidance. If avoidance persists despite the patient’s improved understand-
ing, it is treated as an active communication to the therapist to “do
something” and hence can be approached through transference exploration.
3
At first blush, this might seem to be inconsistent findings from Klass
et al.’s (2009) exploratory study in which the pattern was for PFPP to do
especially well in comparison to ART when patients had not suffered a
recent interpersonal loss. However, the loss variable here and in Nemeroff
et al. (2003) concerns loss of parental figures in childhood rather than a
recent loss in adulthood.
 806 CHAMBLESS ET AL.
panic disorder with or without agoraphobia were compared (Mil-
rod et al., 2016).
For the reasons described above, we hypothesized that, in this
trial, greater expectancy for change, later age of onset, shorter
duration of disorder, lower functional impairment, lower agora-
phobic avoidance, and absence of a Cluster C diagnosis would be
predictive of greater pre- to posttreatment improvement across the
sample of patients who received CBT or PFPP. Thus, we seek to
replicate prior results in the CBT literature but extend these for the
first time to a psychodynamic treatment. With regard to modera-
tion, we predicted that patients with higher levels of agoraphobic
avoidance, longer duration of disorder, greater disability, and
lower RF would improve more in CBT as compared with PFPP,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
even though within the CBT condition such patients might do
This document is copyrighted by the American Psychological Association or one of its allied publishers.
worse than patients without these characteristics. Conversely, we
expected patients who had experienced a childhood loss and those
with Cluster C personality disorders to fare better in PFPP as
compared with CBT. Our design allows us (a) to test not only
whether prior findings for predictors of CBT outcome will repli-
cate in this trial but also whether they extend to a symptom-
focused psychodynamic therapy, and (b) to perform the first test of
moderation of these two very different forms of psychotherapy for
panic disorder. Moreover, given that we have two sites in our
study, our design allows us an internal test of replication of the
findings across sites. This permits us to rectify what Schneider,
Arch, and Wolitzky-Taylor (2015) identified as a major limitation
to the moderation literature in treatment of anxiety disorders—the
absence of two studies that explore the same treatments with the
same populations and use the same variables to define the mod-
erators.
Method
Participants
Participants were 161 patients with a primary diagnosis of panic
disorder with (n ⫽ 127, 79%) or without (n ⫽ 34, 21%) agora-
phobia drawn from the Cornell-Penn Psychotherapies for Panic
Disorder trial (Milrod et al., 2016). Patients in that trial were
randomly assigned to one of three forms of psychotherapy, and the
present sample comprises those in two of those conditions: CBT
(n ⫽ 81) and PFPP (n ⫽ 80). Of these, 104 (65%) were female;
118 (76%) patients were White, 31 (19%) were Black, 8 (5%) were
Asian or Asian American, 1 (⬍1%) were Native American, and 3
(2%) were other or missing. Twenty-four (15%) were Latino(a).
Participants ranged in age from 18 – 69 (M ⫽ 39.40, SD ⫽ 13.25);
the duration of their current panic episode ranged from 1 to 444
months (Mdn ⫽ 15).
Inclusion criteria included a primary diagnosis of panic disorder
and at least one spontaneous panic attack weekly for the month
before entry. Psychotropic medication was permitted if stable
for ⱖ2 months at intake and if the patient was willing for it to
remain constant across the trial. Exclusion criteria included active
substance dependence (⬍6 months’ remission), history of psycho-
sis or bipolar disorder, acute suicidality, organic mental syndrome,
evidence of medical conditions that might cause the patient’s
symptoms, involvement in legal or disability proceedings involv-
ing the patient’s mental state, and unwillingness to forego non-
study psychotherapy for the duration of the protocol. All patients
provided written informed consent; the Institutional Review
Boards of Weill Cornell Medical College and University of Penn-
sylvania approved the research.
Procedure
Patients at each site were stratified on diagnosis of agoraphobia
and of major depression before randomization. Subsequent to
baseline assessment and completion of 3 weeks of a panic diary,
they were assigned according to availability to one of eight CBT
therapists or 16 PFPP therapists for twice-weekly 45-min sessions
for 19 –24 sessions. Therapists were trained for this study or had
been trained in a prior trial to deliver one of the two treatments.
Therapists treated patients in a single protocol and were supervised
by an expert in their treatment modality throughout the trial.
Assessment of the primary outcome variable, the PDSS (Shear
et al., 1997), was conducted by trained diagnosticians kept unin-
formed as to treatment assignment and was carried out at roughly
monthly intervals: baseline, Weeks 1, 5, and 9 of treatment, and
termination. Predictor and moderator variables were assessed at
baseline, with the exception of treatment expectancy, which was
assessed prior to or immediately after Session 2. For additional
details of the procedure, see Milrod et al. (2016).
Interview measures.
Anxiety Disorders Interview Schedule for DSM–IV (ADIS).
Trained masters or doctoral level interviewers used the ADIS
(Brown, Di Nardo, & Barlow, 2004) for diagnosis of panic disor-
der, agoraphobia, and possible comorbid conditions. ADIS reli-
ability was excellent for panic disorder (␬ ⫽ 1.0) and agoraphobia
(␬ ⫽ .94). The diagnostician also collected information about
duration of the current episode of panic disorder and about the
patient’s age at the time of the first episode of panic disorder.
PDSS. The PDSS (Shear et al., 1997) was used to assess
change with treatment. Ratings on the seven PDSS questions range
from 0 (none) to 4 (extreme). Questions tap frequency and severity
of panic attacks; distress associated with attacks; severity of in-
teroceptive and situational avoidance and of anticipatory anxiety;
and social, familial, and vocational impairment, and form a uni-
factorial scale (Shear et al., 2001). Internal consistency of the
measure is acceptable (␣ ⫽ .695 in the present sample). Prior
research has demonstrated the interrater and test–retest reliability
and construct validity of this measure (Shear et al., 1997, 2001). In
the present study, the ADIS diagnostician, who was uninformed as
to treatment assignment, completed the PDSS as well. Interrater
PDSS reliability, calculated across sites with an intraclass corre-
lation coefficient, proved excellent, rI (2,1) ⫽ .95.
Childhood Trauma Scale: Loss. Using the Childhood Trauma
Scale (Lizardi et al., 1995; Nemeroff et al., 2003), diagnosticians
interviewed patients about their experiences prior to age 15 of loss
of or prolonged separation from a parent (coded as present/absent)
Loss was coded when the patient reported that before the age of 15
she or he experienced the death of a parent, separation from a
parent for more than 6 months, or parental divorce or separation of
more than 6 months. Seventy-two patients (45%) reported such a
loss.
RF. Originally designed to be used with the Adult Attachment
Interview and derived from its demand questions, RF ratings
(Fonagy et al., 1998) in the present study were based on material
collected via the Brief RF Interview-Revised (Rudden, 2009)
 PSYCHOTHERAPY FOR PANIC DISORDER
conducted by the diagnostician. Patients were asked to talk about
a parent, the relationship with that parent, and the changes in that
relationship across time, as well as their relationship with someone
important to them at the present time. Following transcription, one
of three coders rated overall RF on a ⫺1 (negative reflective
functioning/psychotic and idiosyncratic) to 9 (exceptional reflec-
tive functioning) scale. Rutimann and Meehan (2012) reported
good interrater reliability (␳I ⫽ .79) for an earlier version of the
Brief RF Interview as well as excellent concurrent validity with RF
as assessed from the Adult Attachment Interview (r ⫽ .71). One
coder (Marie Rudden) in the present study, who was trained in
coding RF by Mary Target of the Anna Freud Centre, trained the
other two coders in rating RF from the Brief RF Interview.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Reliability was assessed by having pairs of coders independently
This document is copyrighted by the American Psychological Association or one of its allied publishers.
code the same transcript (n ⫽ 62). Intraclass reliability coefficients
among the three coders ranged from acceptable to excellent, rI
(1,1) ⫽ .63–.91.
Structured Clinical Interview for DSM–IV Axis II Personality
Disorders (SCID-II, Version 2.0. The SCID-II (First, Spitzer,
Gibbon, Williams, & Benjamin, 1997) was administered by the
same diagnosticians as the ADIS. Cluster A (n ⫽ 24, 17%) and B
diagnoses (n ⫽ 18, 11%) and traits were too infrequent for analysis
of moderation. However, Cluster C diagnoses were sufficiently
common for analysis (n ⫽ 54, 33.5%). Interrater reliability for
presence/absence of any Cluster C personality disorder was very
good, ␬ ⫽ .83.
Questionnaire measures.
Avoidance Alone Scale of the Mobility Inventory for
Agoraphobia. The Avoidance Alone Scale of the Mobility In-
ventory for Agoraphobia (Chambless, Caputo, Jasin, Gracely, &
Williams, 1985) was used to assess agoraphobic avoidance. Pa-
tients are asked to rate 26 items reflecting situations patients with
agoraphobia may avoid on 1 (never avoid) to 5 (always avoid)
scales. Item scores are averaged to yield the total score. The
internal consistency, test–retest reliability, and convergent and
discriminant validity of this measure have been repeatedly sup-
ported (Chambless et al., 2011).
Credibility/Expectancy Questionnaire. The three expectancy
items from this measure (Devilly & Borkovec, 2000) yielded the
expectancy score. Patients are asked to indicate on 0 –100% scales
how much they think and feel they will improve by the end of
therapy and on a 1 (not at all) to 9 (very much) scale how helpful
they feel therapy will be in reducing their symptoms. Item scores
are standardized and averaged for analysis. The factor analytically
derived expectancy scale is internally consistent, has good test–
retest reliability, and has proved useful in predicting treatment
outcomes (Devilly & Borkovec, 2000).
The Sheehan Disability Scale. The Sheehan Disability Scale
(Sheehan, 1983) is a three-item measure of social, family, and
vocational impairment. Each item is rated on a 0 (not at all) to
10 (extremely) scale indicating the degree to which the patient’s
disorder disrupted functioning in that domain in the last month.
Expressed as a sum, total scores range from 0 –30. Leon, Shear,
Portera, and Klerman (1992) have demonstrated the reliability
and validity of this measure for panic disorder patients, which
evinced good internal consistency in the present sample,
␣ ⫽ .83.
807
Treatment
All treatments were delivered in 19 –24 twice weekly sessions of
45-min duration. Flexibility in treatment length was permitted in
CBT for cases of panic disorder without agoraphobia who did not
require the last sessions of the protocol, which focused on agora-
phobic avoidance, or who resolved their symptoms quickly. In
contrast, the goal for PFPP was 24 sessions with preplanned
termination dates to provide a clear endpoint and framework for
processing mixed feelings and distress about separation from the
therapist.4
CBT. CBT followed Barlow and Craske’s panic control ther-
apy protocol (Craske, Barlow, & Meadows, 2000), as modified to
fit the 24-session, 45- min per session format of the trial. This
equated CBT and PFPP in length and duration of sessions. The
protocol included psychoeducation about anxiety and panic; iden-
tification and correction of maladaptive thoughts about anxiety and
panic; training in slow, diaphragmatic breathing; and exposure to
bodily sensations designed to mimic those experienced during
panic. All sessions were followed by homework assignments and
readings. In vivo exposure via homework assignments was intro-
duced at Session 17 for those patients with significant agoraphobic
avoidance. Session 24 focused on review and relapse prevention.
Muscle relaxation was omitted to reduce overlap with ART.
PFPP. PFPP is a manual-based treatment (Milrod et al., 1997)
rooted in the assumption that panic symptoms have a psycholog-
ical meaning and that uncovering these unconscious meanings will
lead to relief. The therapist explores the circumstances and feelings
surrounding panic onset, the personal meaning of panic symptoms,
and the feelings and content of panic episodes. PFPP aims to lessen
vulnerability to panic by helping patients understand and alter core
unconscious conflicts, which are often identified and understood
through their emergence in the transference. Frequent themes
include conflicts over separation and autonomy; recognizing, man-
aging, and expressing anger; and guilt. Termination, addressed
prominently in the last third of treatment, permits patients to
reexperience conflicts directly with the therapist so that underlying
feelings can be articulated and rendered less frightening.
Adherence. Adherence to the therapy manuals was assessed
by teams of coders for each of the three treatments. As reported in
the main outcome paper (Milrod et al., 2016), we confirmed that
adherence was acceptable for both treatments, and that therapists
did not mix interventions from the other two protocols with their
assigned protocol.
Statistical Analyses
Change during treatment. Standard approaches to the lon-
gitudinal analysis of treatment data assume that missing data are
missing at random, but this was not the case with these data. For
this reason, we analyzed change in treatment over time with a
shared parameters approach (ten Have, Pulkstenis, Kunselman, &
Landis, 1998), rather than the typical multilevel modeling (MLM)
approach, which ignores the missing data mechanism. Each patient
was simultaneously modeled with (a) MLM for change over time
4
In rare cases where the patients had planned an extensive absence for
work or vacation toward the end of their treatment and did not have
extensive agoraphobia avoidance, 21-session treatments were planned.
 808 CHAMBLESS ET AL.
on the PDSS, and (b) a survival model for attrition. The two
processes share a common random effect, creating a quantifiable
correlation between change in the PDSS outcome and dropout
processes. In this sample this correlation was substantial,
r ⫽ ⫺.44, p ⫽ .03, demonstrating outcome-missing data depen-
dency: Patients who were improving at a slower rate were more
likely to drop out of treatment more quickly.
At Level 1 of the MLM, PDSS scores varied within patients
over time, expressed as weeks from baseline. Level 2 captured
between person variability as a function of treatment assignment.5
The attrition process was assessed with a discrete time survival
model: At each evaluation point, an individual was classified a
dropout, treatment completer, or continuing in treatment. By
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
jointly modeling outcome and attrition processes, we were able to
This document is copyrighted by the American Psychological Association or one of its allied publishers.
evaluate outcome while adjusting for the association between the
outcome and attrition models.
A priori covariates were use of anxiolytics and of antidepres-
sants, treatment condition, site, and Treatment ⫻ Site interaction
effects. In addition, because there were Treatment ⫻ Site differ-
ences in age and gender in the original study (Milrod et al., 2016),
these variables were included as covariates. To preserve power,
predictors or moderators of interest were introduced one per equa-
tion. Prediction and moderation of rate of change (the slope) are
the critical tests. Additionally, we also included the three-way
interaction of Moderator ⫻ Treatment Condition ⫻ Site to deter-
mine whether effects differed across sites. Continuous predictors/
moderators were centered for analysis. When interactions involv-
ing a continuous predictor/moderator were significant, we probed
these by estimating and contrasting slope estimates for the treat-
ments with the continuous predictor set at the mean and 1 SD
above and below the mean.
A number of predictors were substantially correlated with PDSS
baseline scores, thus risking a spurious relationship with slope
because of the steeper PDSS slope evinced by patients who started
with higher PDSS scores. To reduce the likelihood of artifactual
findings, we regressed these predictors on baseline PDSS scores
and used the residualized variables in the place of the original form
of the predictors in our analyses.
The main results of this trial were complicated by significant
Site ⫻ Treatment interaction effects: Among patients treated at
Cornell, there were no differences between the two treatments on
the primary outcome measure, change in symptom severity on the
PDSS (Shear et al., 1997), whereas among patients treated at Penn,
those who received CBT showed significantly greater improve-
ments in PDSS change than those who received PFPP. Thus,
before conducting shared parameters analyses, we ascertained
whether there were significant Site ⫻ Treatment differences on the
predictors/moderators. If this had been the case, the effects of the
predictor/moderator might have been confounded with Site ⫻
Treatment differences in treatment response. However, none of
these interactions were significant, all ps ⱖ .11.
SAS Version 9.4 (SAS Institute, 2015) was used for all analyses.
Degrees of freedom were estimated with the Kenward-Roger
(Kenward & Roger, 1997) approximation.
Power analyses. To preserve power, we set alpha at .05 for
each analysis. Due to participant or interviewer error in collecting
data or recording failure, sample sizes for predictors and moder-
ators ranged from 93–161. Power for tests of prediction was
estimated as .80 for a medium d of 0.48. Power for tests of
moderation depends on the form of any interaction obtained (or-
dinal vs. disordinal), the relative within-treatment condition effect
sizes, and the sample sizes of each treatment condition. Based on
hypotheses, we assumed any interactions would be ordinal, indi-
cating the effect was in the same direction across the intervention
arms but of different magnitudes, and calculated power under a
number of scenarios of larger and smaller effect sizes across the
two treatment conditions. Sample size for our study was deter-
mined on the basis of power for our randomized test of treatment
outcome differences and not for moderation. Thus, we only had
adequate power (⬎.80) when we assumed large differences (e.g.,
d of 0.9) in the efficacy of one of the treatments at low and high
levels of the moderator and very small effects for the second.
Accordingly, tests of moderation may well be underpowered and
should be taken as exploratory (Kraemer et al., 2002).
Results
Descriptive data for the sample and a correlation matrix for
predictor, moderator, and dependent variables may be found in
Tables 1 and 2 in the online supplemental materials. Additional
details of results from the shared parameters model are reported in
Table 3 in the online supplemental materials.
Expectancy
Unlike other predictors, expectancy was assessed once treatment
had begun (at Session 2 of the first week of treatment). Accord-
ingly, we calculated the slope effect beginning from Week 1 of
treatment rather than at baseline. Expectancy was a significant
predictor of change in treatment, t(84) ⫽ ⫺4.85, p ⬍ .0001,
d ⫽ ⫺1.05, CI95% [⫺1.50, ⫺0.60]. However, this effect was
moderated by an interaction with treatment condition, t(83) ⫽ ⫺2.19,
p ⫽ .03. Probes of the interaction indicated that patients with low
levels of expectancy improved more in CBT than in PFPP, t(83) ⫽
2.39, p ⫽ .02, d ⫽ 0.52, CI95% [0.09, 0.95], whereas at average and
high levels of expectancy the two treatment conditions did not differ
significantly, ds ⱕ 0.28, ps ⱖ .20. The expectancy effect was not
modified by interactions with site or Site ⫻ Treatment, ps ⬎ .97.
Age of Onset and Duration of Episode
Also consistent with prediction, age of onset (log transformed
to achieve normality) was significantly predictive of slope: pa-
tients who were older when they had their first episode of panic
disorder changed more in treatment, t(147) ⫽ ⫺3.95, p ⫽ .0001,
d ⫽ ⫺0.65, CI95% [⫺1.50, ⫺0.60]. However, this effect was
moderated by treatment, t(146) ⫽ ⫺2.61, p ⫽ .01. In the case of
patients who were relatively young when they had their first
episode of panic disorder (⬍14.32 years old), PFPP patients im-
proved significantly less than patients in CBT, t(146) ⫽ 3.77, p ⫽
.0002, d ⫽ 0.62, CI95% [0.29, 0.95]. The same was true at the
5
Because we previously found there were no significant therapist effects
(Milrod et al., 2016), we did not include therapist as a level in the analysis.
 PSYCHOTHERAPY FOR PANIC DISORDER
average age of onset (23.68 years of age6): CBT patients improved
more than PFPP patients, t(146) ⫽ 2.58, p ⫽ .01, d ⫽ 0.42, CI95%
[0.10, 0.75]. Among patients with later onset (⬎39.17 years of
age), there was no significant difference between treatments,
t(146) ⫽ ⫺0.06, p ⫽ .95, d ⫽ ⫺0.01, CI95% [⫺0.33, 0.31]. To
better pinpoint the age of onset at which treatment differences in
CBT and PFPP no longer approached significance, we tested
differences between these two treatments at 0.1 SD increments
above the mean. At 0.3 SD above the mean (27.53 years old), p for
the contrast exceeded .05. Using this cutpoint, we find that for the
top 37% of the age of onset distribution the two treatments were
comparable in outcome. These effects were not moderated by
interactions with site or Site ⫻ Treatment, ps ⱖ .35.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Contrary to hypothesis, duration of present episode (log trans-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
formed to achieve normality) was not associated with change in
treatment, t(150) ⫽ 0.23, p ⫽ .82, d ⫽ 0.04, CI95% [⫺0.28, 0.35].
The test of the hypothesis that patients who had a longer duration
of panic disorder would fare relatively better in CBT than in PFPP
indicated that this was not the case, t(149) ⫽ ⫺0.38, p ⫽ .70,
d ⫽ ⫺0.06, CI95% [⫺0.38, 0.26]. The effect of duration was not
modified by site or Site ⫻ Treatment interactions, ps ⱖ .62.
Sheehan Disability Scale
Because baseline Sheehan and PDSS scores were highly corre-
lated, r ⫽ .54, p ⬍ .001, residualized Sheehan scores were used for
analysis. The Sheehan was neither an overall predictor of change,
t(147) ⫽ 1.67, p ⫽ .10, d ⫽ 0.27, CI95% [⫺0.05, 0.60] nor a
moderator of treatment outcome, t(146) ⫽ ⫺1.91, p ⫽ .06,
d ⫽ ⫺0.31, CI95% [⫺0.64, 0.01]. We probed the interaction for
heuristic purposes (see Table 3 in the online supplemental mate-
rials). CBT proved to be more effective than PFPP at low and
moderate levels of disability but not at high levels of disability.
The effect of disability was not modified by site or Site ⫻ Treat-
ment interactions, ps ⱖ .41.
Agoraphobic Avoidance
Because avoidance was highly correlated with baseline PDSS
scores, r ⫽ .57, p ⬍ .001, residualized avoidance scores were used
for these analyses. As hypothesized, patients higher in agoraphobic
avoidance did worse in treatment, t(138) ⫽ 3.72, p ⫽ .001, d ⫽
0.55, CI95% [0.22, 0.89]. However, this effect was moderated by
site. Although the pattern was the same at both sites, the effect was
significant only at Cornell: Cornell t(137) ⫽ 3.66, p ⫽ .0004, d ⫽
0.62, CI95% ⫽ 0.28, 0.96; Penn t(137) ⫽ 1.38, p ⫽ .17, d ⫽ 0.23,
CI95% [⫺0.10, 0.57]. The interaction with treatment and the
Avoidance ⫻ Site ⫻ Treatment interaction were not significant,
ps ⱖ .15.
Cluster C Personality Disorder Diagnosis
Cluster C did not significantly predict outcome, t(152) ⫽ 1.84,
p ⫽ .07, d ⫽ 0.30, CI95% ⫽ ⫺0.02, 0.61. The effect size was in the
predicted direction but small. Contrary to prediction, the Cluster
C ⫻ Treatment interaction was not significant, t(151) ⫽ 1.21, p ⫽
.23, d ⫽ 0.20. There was no significant Cluster C ⫻ Site or Cluster
C ⫻ Site ⫻ Treatment interaction, ps ⬎ .74.7
809
RF
Contrary to hypothesis, patients higher in RF fared no better than
those with lower RF. Indeed the small effect size was in the opposite
direction, t(147) ⫽ 1.73, p ⫽ .09, d ⫽ 0.28, CI95% ⫽ ⫺0.04, 0.61.
Moreover, the Treatment ⫻ Reflective Functioning test of moderation
was not significant, t(146) ⫽ ⫺1.90, p ⫽ .06, d ⫽ ⫺0.31, CI95%
[⫺0.64, 0.01]. In light of our a priori hypothesis, we probed the
interaction for heuristic purposes (see Table 3 in the online supple-
mental materials). Consistent with prediction, patients who were low
or average on RF improved more in CBT than in PFPP, whereas there
was no significant difference between treatments for those high in RF.
Childhood Trauma Scale: Loss
Loss was not a predictor of treatment outcome, t(152) ⫽ ⫺1.68,
p ⫽ .095, d ⫽ ⫺0.27, CI95% [⫺0.59, 0.05]. Contrary to prediction,
the Loss ⫻ Treatment interaction was not significant, t(151) ⫽
0.90, p ⫽ .37, d ⫽ 0.15. There were no significant interactions of
loss with site or Site ⫻ Treatment, ps ⱖ .13.
Discussion
In a large sample of panic disorder patients treated with one of
two different forms of psychotherapy, we examined prediction of
treatment outcome across both PFPP and CBT by variables found
in our systematic review (Porter & Chambless, 2015) to predict
outcome of CBT: agoraphobic avoidance, functional impairment,
and expectancy of improvement. Of the three replicated predictors
Porter and Chambless (2015) identified, only one proved to predict
outcome across treatments and sites in this study— expectancy of
change. Another, agoraphobia avoidance, predicted poorer out-
come at one of the two sites. Of three other variables Porter and
Chambless (2015) identified as possible predictors, earlier age of
panic disorder onset proved to be significant in the present sample,
but duration of disorder and Cluster C personality disorders did
not. We further tested whether lower RF, higher functional im-
pairment, more severe agoraphobic avoidance, longer duration of
the disorder, childhood loss of a parent, and diagnosis of a Cluster
C personality disorder might serve as moderators of outcome,
indicating which treatment might be preferable for a given patient.
Our moderation hypotheses were not confirmed, but two unpre-
dicted significant moderators of treatment were identified and
proved to replicate across the two study sites— expectancy and age
of onset.
Consistent with two other studies of CBT for panic disorder
(Porter & Chambless, 2015) and with the results of a meta-analysis
of expectancy across disorders and types of treatment (Constantino
et al., 2011), treatment outcome expectancy as assessed at Session
2, on the whole, predicted better treatment response in the present
6
This figure differs slightly from that in Table 1 in the online supple-
mental materials because this mean is based on the log-transformed dis-
tribution whereas the other was computed on raw scores.
7
We repeated these analyses with OCPD alone, despite the loss of
power associated with the smaller sample size, for heuristic purposes. The
results were similar, in that the prediction and moderation effects were not
statistically significant. In light of the trend for a significant interaction
(p ⫽ .08), probes of the interaction are reported in Table 3 in the online
supplemental materials to guide future research.
 810 CHAMBLESS ET AL.
sample for both treatments with a medium-large effect size. We
assessed expectancy after the patient had one or two meetings with
the therapist to learn the rationale for the treatment. Thus, it is
conceivable that the expectancy measure could reflect not only the
treatment rationale but also the patient’s initial reaction to the
therapist, that is, not only will this treatment help me, but also will
this person help me? Research on why expectancy is related to
outcome is not extensive, but findings to date (Constantino et al.,
2011) suggest that expectancy may affect outcome through its
effects on completion of homework and on the therapeutic alli-
ance. Huppert et al. (2014) have demonstrated the positive rela-
tionship of working alliance at Session 3 with treatment outcome
on panic disorder symptoms. Thus, examining whether the work-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
ing alliance mediates the relationship of expectancy to outcome
This document is copyrighted by the American Psychological Association or one of its allied publishers.
would be a worthwhile pursuit. Moreover, in studies of CBT for
various disorders, both the quantity and quality of homework
completion is associated with better outcome (see the meta-
analysis by Kazantzis et al., 2016). It seems likely that expectancy
might also affect the level of engagement in challenging activities
in session such as interoceptive exposure in CBT and interpreta-
tion of intrapsychic conflict and the transference in PFPP. Why
low expectancy predicted worse outcome for PFPP than CBT is
not clear. However, this might be the result of CB therapists’ use
of cognitive restructuring for low expectations of change—an
approach readily available in their armamentarium. That is, had we
had measures of expectancy at later sessions, we might have found
that in CBT initially low expectancy patients had a more positive
view of their prospects for improvement later in therapy. Unfor-
tunately, such measures are lacking in our sample.
Constantino et al. (2011) noted the poor methodological quality
of many of the 46 studies they included in their meta-analysis.
Thus, the present study makes a meaningful contribution to the
literature by assessing expectancy in a reliable and valid fashion
and by measuring change with treatment subsequent to expectancy
assessment, thus avoiding confounding the effects of early panic
disorder improvement and expectancy on outcome. Although cor-
relational, these findings argue for the importance of therapists’
efforts to increase early treatment expectancy. Constantino et al.
(2011) suggested a number of strategies including explicit discus-
sions of patients’ expectancies during which the therapist is re-
spectful of any doubts patients hold, providing information on the
known efficacy of an intervention plus reasons the therapist be-
lieves the intervention will be beneficial for this particular person,
and the like.
Age of onset of the first episode of panic disorder was both a
predictor of poorer treatment outcome and a moderator of out-
come, in that regardless of treatment condition patients did worse
when they had an earlier age of onset, but this was especially
marked for PFPP. For roughly 63% of the sample, CBT was likely
to be more effective than PFPP, but for patients in the upper 37%
of the distribution (ⱖ27.5 years old at first onset), there was no
advantage to CBT over PFPP. These data solidify the uncertain
findings reported in Porter and Chambless (2015) as to age of
onset’s predictive utility and are consistent with reports that age of
onset is related to worse outcome in psychotherapy for social
anxiety disorder as well (e.g., Borge, Hoffart, & Sexton, 2010).
Interpreting why socially anxious patients with earlier onset did
worse in both CBT and IPT, Borge et al. (2010 hypothesized that
their findings might reflect greater personality problems in the
younger onset patients, such as higher neuroticism. We do not
have measures of neuroticism but note that Cluster C diagnosis
was not related to worse outcome in the present study. In our
clinical experience the younger the onset the more normal devel-
opment is disrupted by panic disorder, and the less patients have a
compelling image of the life they could have if they confronted
their fears in treatment. Progress may seem punishing in that it
may confront the patient with the challenge of delayed develop-
mental experiences, for example, seeking further education or
employment or establishing an independent residence. Longer
term treatment with support to develop life skills may be required.
PFPP may fare worse than CBT with patients of younger onset
because CB therapists are more likely to provide such concrete
coaching.
Cluster C was not a predictor of poor treatment outcome overall,
thus adding another negative study to the column of those reported
by Porter and Chambless (2015) and making it less likely that this
is a negative prognostic sign for CBT. In light of the high prepon-
derance of patients with OCPD in our Cluster C patients (74% of
those with a Cluster C diagnosis), we predicted that Cluster C
patients would improve more if in PFPP than in CBT (cf., Barber
& Muenz, 1996). We found no statistically significant moderation
by Cluster C diagnosis. However, in analyses conducted for heu-
ristic purposes, we found that, contrary to prediction, CBT was
superior to PFPP for patients with OCPD. Thus, these results are
inconsistent with Milrod, Leon, Barber, et al.’s (2007) small study
in which the pattern of results suggested PFPP was particularly
more effective than ART for Cluster C patients (albeit that ART is
a different form of cognitive-behavioral treatment than panic con-
trol therapy). Under these circumstances, no clear statement can be
made about moderation other than to say that the findings do not
suggest that a psychodynamic therapy should be selected over
CBT when a panic disorder patient has a Cluster C diagnosis.
Given the small number of patients with Cluster A and B disorders
in our study, our results do not speak to treatment selection where
other personality disorder types are concerned.
We tested whether RF would predict better treatment outcome
across therapies, as was the case in treatment of depressed patients
with very low RF (Ekeblad et al., 2015), or would moderate
outcome such that those low in RF would do better in CBT than in
PFPP due to their low capacity for reflection on mental states. The
effects for prediction and moderation both failed to reach signifi-
cance. For heuristic purposes we probed the moderation effect and
found that, consistent with the hypothesis, CBT was superior to
PFPP at low and average levels of RF but not at high levels. We
note that average RF scores were substantially higher in our study
than in that of Ekeblad et al. (2015), and this may have made it less
likely that we would find this variable to be a predictor of outcome
in our sample.
Contrary to hypothesis, we did not find that patients who had
suffered a significant loss in childhood benefited more from PFPP
than from CBT. This is not to say that a panic disorder patient who
is suffering from such a loss and needs to talk about it might not
do better in a treatment that has space for such a discussion (which
would have been difficult in our highly structured CBT unless
therapists went off protocol). However, the mere fact of such a loss
does not indicate that the patient needs to focus on this in treat-
ment: childhood loss neither predicted poorer outcome nor mod-
erated results of treatment.
 PSYCHOTHERAPY FOR PANIC DISORDER
Limitations
There are several limitations to our tests of moderation: The
obtained moderation effects, although replicated across sites, were
not those we predicted, our study had adequate power only for
large moderation effects thus risking Type II error, and to maintain
power we conducted a number of tests without correcting for Type
I error. Moreover, we acknowledge that the substantial majority of
our patients were White and non-Latino. Whether our results
would generalize to a more racially and ethnically diverse sample
remains to be determined. In addition, we conducted CBT on a
slower schedule than is often the case, spreading a treatment often
delivered in 11–12 sessions (e.g., Barlow, Gorman, Shear, &
Woods, 2000) over 19 –24 sessions. It is impossible to assess how
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
this might have affected the results, although we note the consis-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
tency of our three predictors of outcome with the prior literature.
CBT does not work for all patients (e.g., Milrod et al., 2016),
and we had hoped to detect moderators identifying patients who
might benefit particularly from a relatively new treatment with
empirical support of efficacy—PFPP. These hopes were not ful-
filled. At best, we identified patients for whom PFPP was a
reasonable alternative to CBT, the treatment with the most empir-
ical support: These were patients whose panic disorder began later
in life (age 27.5 or later) and patients who were average or above
average on expectancy for improvement with treatment. Given that
psychologists are enjoined to consider patients’ preferences as well
as empirical evidence in selecting a treatment approach (American
Psychological Association Presidential Task Force on Evidence-
Based Practice, 2006), these results are important for practitioners
treating patients with panic disorder who would prefer a psychody-
namic approach.
We failed to identify patients who would benefit preferentially
from a psychotherapy other than CBT using the set of moderators
we had available. However, there were important potential mod-
erators that we did not assess or had inadequate sample size to test,
such as the effect of Cluster B diagnoses and of separation anxiety.
In light of a recent report by Milrod et al. (2014) of preliminary
data indicating separation anxiety moderated the outcome of PFPP
versus ART in treatment of panic disorder (with PFPP being
superior for patients with separation anxiety) and Aaronson et al.
(2008)’s study finding that adult separation anxiety predicted
worse treatment outcome for CBT for panic disorder, it is unfor-
tunate that we did not have a measure of separation anxiety in our
data set. Finally, we note that our measure of childhood loss was
heterogeneous, and it is highly likely that some kinds of loss (e.g.,
death of a parent) have greater implications than others (e.g.,
parental separation for 6 months or more). It is possible that had
we had the sample size to test the effects of the more severe forms
of loss, our results would have been different, and we would have
found that PFPP would be advantageous for such patients.
Conclusions and Clinical Implications
We identified two treatment moderators— expectancy and age
of onset—that indicate when patients are significantly more likely
to benefit from CBT than from PFPP. These effects were repli-
cated across the two study sites, enhancing confidence in their
reliability. In addition, for heuristic purposes we have reported two
cases where moderation effects were not statistically significant in
the overall tests of moderation but where the tests of simple effects
811
were significant. These were RF and OCPD. Noting the difficulty
of amassing sufficient statistical power to test medium but none-
theless potentially meaningful moderators in a single study, Sch-
neider et al. (2015) advocated reporting such results to make them
available for meta-analysis, and in that spirit we highlight them as
potential targets for future research.
Finally, we have identified two predictors of poor treatment
outcome (again, expectancy and age of onset) and have suggested
how these might guide therapists’ efforts to improve outcomes for
patients with these characteristics. An important next step is to
elucidate the mechanisms behind these prediction effects, and we
plan both qualitative and quantitative studies to test such mecha-
nisms. Such research, along with the present study, will provide
direction to continued efforts to improve treatment outcome for
panic disorder.
References
Aaronson, C. J., Shear, M. K., Goetz, R. R., Allen, L. B., Barlow, D. H.,
White, K. S., . . . Gorman, J. M. (2008). Predictors and time course of
response among panic disorder patients treated with cognitive-
behavioral therapy. The Journal of Clinical Psychiatry, 69, 418 – 424.
http://dx.doi.org/10.4088/JCP.v69n0312
American Psychological Association Presidential Task Force on Evidence-
Based Practice. (2006). Evidence-based practice in psychology. Ameri-
can Psychologist, 61, 271–285. http://dx.doi.org/10.1037/0003-066X.61
.4.271
Barber, J. P., & Muenz, L. R. (1996). The role of avoidance and obses-
siveness in matching patients to cognitive and interpersonal psychother-
apy: Empirical findings from the treatment for depression collaborative
research program. Journal of Consulting and Clinical Psychology, 64,
951–958. http://dx.doi.org/10.1037/0022-006X.64.5.951
Barber, J. P., Muran, J. C., McCarthy, K. S., & Keefe, J. (2013). Research
on psychodynamic therapies. In M. J. Lambert (Ed.), Bergin and Gar-
field’s handbook of psychotherapy and behavior change (6th ed., pp.
443– 494). New York, NY: Wiley.
Barlow, D. H., Gorman, J. M., Shear, M. K., & Woods, S. W. (2000).
Cognitive-behavioral therapy, imipramine, or their combination for
panic disorder: A randomized controlled trial. Journal of the American
Medical Association, 283, 2529 –2536. http://dx.doi.org/10.1001/jama
.283.19.2529
Batelaan, N., Smit, F., de Graaf, R., van Balkom, A., Vollebergh, W., &
Beekman, A. (2007). Economic costs of full-blown and subthreshold
panic disorder. Journal of Affective Disorders, 104, 127–136. http://dx
.doi.org/10.1016/j.jad.2007.03.013
Beutel, M. E., Scheurich, V., Knebel, A., Michal, M., Wiltink, J., Graf-
Morgenstern, M., . . . Subic-Wrana, C. (2013). Implementing panic-
focused psychodynamic psychotherapy into clinical practice. The Cana-
dian Journal of Psychiatry/La Revue canadienne de psychiatrie, 58,
326 –334. http://dx.doi.org/10.1177/070674371305800604
Borge, F.-M., Hoffart, A., & Sexton, H. (2010). Predictors of outcome in
residential cognitive and interpersonal treatment for social phobia: Do
cognitive and social dysfunction moderate treatment outcome? Journal
of Behavior Therapy and Experimental Psychiatry, 41, 212–219. http://
dx.doi.org/10.1016/j.jbtep.2010.01.005
Bowlby, J. (1973). Attachment and loss. New York, NY: Basic Books.
Brown, T. A., Di Nardo, P. A., & Barlow, D. H. (2004). Anxiety Disorders
Interview Schedule Adult Version (ADIS-IV): Client interview schedule.
New York, NY: Oxford University Press.
Chambless, D. L., Caputo, G. C., Jasin, S. E., Gracely, E. J., & Williams,
C. (1985). The Mobility Inventory for Agoraphobia. Behaviour Re-
search and Therapy, 23, 35– 44. http://dx.doi.org/10.1016/0005-
7967(85)90140-8
 812 CHAMBLESS ET AL.
Chambless, D. L., Sharpless, B. A., Rodriguez, D., McCarthy, K. S.,
Milrod, B. L., Khalsa, S.-R., & Barber, J. P. (2011). Psychometric
properties of the Mobility Inventory for agoraphobia: Convergent, dis-
criminant, and criterion-related validity. Behavior Therapy, 42, 689 –
699. http://dx.doi.org/10.1016/j.beth.2011.03.001
Constantino, M. J., Glass, C. R., Arnkoff, D. B., Ametrano, R. M., &
Smith, J. Z. (2011). Expectations. In J. C. Norcross (Ed.), Psychotherapy
relationships that work: Eveidence-based responsiveness (2nd ed., pp.
354 –376). New York, NY: Oxford University Press. http://dx.doi.org/
10.1093/acprof:oso/9780199737208.003.0018
Craske, M. G., Barlow, D. H., & Meadows, E. A. (2000). Mastery of your
anxiety and panic: Therapist guide for anxiety, panic, and agoraphobia.
New York, NY: Oxford University Press.
de Graaf, R., Tuithof, M., van Dorsselaer, S., & ten Have, M. (2012).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Comparing the effects on work performance of mental and physical
This document is copyrighted by the American Psychological Association or one of its allied publishers.
disorders. Social Psychiatry and Psychiatric Epidemiology, 47, 1873–
1883. http://dx.doi.org/10.1007/s00127-012-0496-7
Devilly, G. J., & Borkovec, T. D. (2000). Psychometric properties of the
Credibility/Expectancy Questionnaire. Journal of Behavior Therapy and
Experimental Psychiatry, 31, 73– 86. http://dx.doi.org/10.1016/S0005-
7916(00)00012-4
Ekeblad, A., Falkenström, F., & Holmqvist, R. (2016). Reflective func-
tioning as predictor of working alliance and outcome in the treatment of
depression. Journal of Consulting and Clinical Psychology, 84, 67–78.
First, M. B., Spitzer, R. L., Gibbon, M., Williams, J. B. W., & Benjamin,
L. (1997). Structured Clinical Interview for DSM–IV Axis II Personality
Disorders (SCID-II). Washington, DC: American Psychiatric Press.
Fonagy, P., Target, M., Steele, H., & Steele, M. (1998). Reflective-
functioning manual (Version 5). Unpublished manuscript, Psychoanal-
ysis Unit, Sub-Department of Clinical Health Psychology, University
College London, London, UK.
Huppert, J. D., Kivity, Y., Barlow, D. H., Gorman, J. M., Shear, M. K., &
Woods, S. W. (2014). Therapist effects and the outcome-alliance cor-
relation in cognitive behavioral therapy for panic disorder with agora-
phobia. Behaviour Research and Therapy, 52, 26 –34. http://dx.doi.org/
10.1016/j.brat.2013.11.001
Kazantzis, N., Whittington, C., Zelencich, L., Kyrios, M., Norton, P. J., &
Hofmann, S. G. (2016). Quantity and quality of homework compliance:
A meta-analysis of relations with outcome in cognitive behavior therapy.
Behavior Therapy, 47, 755–772. http://dx.doi.org/10.1016/j.beth.2016
.05.002
Kenward, M. G., & Roger, J. H. (1997). Small sample inference for fixed
effects from restricted maximum likelihood. Biometrics, 53, 983–997.
http://dx.doi.org/10.2307/2533558
Klass, E. T., Milrod, B. L., Leon, A. C., Kay, S. J., Schwalberg, M., Li, C.,
& Markowitz, J. C. (2009). Does interpersonal loss preceding panic
disorder onset moderate response to psychotherapy? An exploratory
study. The Journal of Clinical Psychiatry, 70, 406 – 411. http://dx.doi
.org/10.4088/JCP.08m04248
Kraemer, H. C., Wilson, G. T., Fairburn, C. G., & Agras, W. S. (2002).
Mediators and moderators of treatment effects in randomized clinical
trials. Archives of General Psychiatry, 59, 877– 883. http://dx.doi.org/
10.1001/archpsyc.59.10.877
Leon, A. C., Shear, M. K., Portera, L., & Klerman, G. L. (1992). Assessing
impairment in patients with panic disorder: The Sheehan Disability
Scale. Social Psychiatry and Psychiatric Epidemiology, 27, 78 – 82.
http://dx.doi.org/10.1007/BF00788510
Lizardi, H., Klein, D. N., Ouimette, P. C., Riso, L. P., Anderson, R. L., &
Donaldson, S. K. (1995). Reports of the childhood home environment in
early-onset dysthymia and episodic major depression. Journal of Abnor-
mal Psychology, 104, 132–139. http://dx.doi.org/10.1037/0021-843X
.104.1.132
Markowitz, J. S., Weissman, M. M., Ouellette, R., Lish, J. D., & Klerman,
G. L. (1989). Quality of life in panic disorder. Archives of General
Psychiatry, 46, 984 –992. http://dx.doi.org/10.1001/archpsyc.1989
.01810110026004
Milrod, B., Busch, F., Cooper, A., & Shapiro, T. (1997). Manual of panic-
focused psychodynamic psychotherapy. Washington, DC: American
Psychiatric Association.
Milrod, B., Chambless, D. L., Gallop, R., Busch, F. N., Schwalberg, M.,
McCarthy, K. S., . . . Barber, J. P. (2016). Psychotherapies for panic
disorder: A tale of two sites. The Journal of Clinical Psychiatry, 77,
927–935. http://dx.doi.org/10.4088/JCP.14m09507
Milrod, B. L., Leon, A. C., Barber, J. P., Markowitz, J. C., & Graf, E.
(2007). Do comorbid personality disorders moderate panic-focused psy-
chotherapy? An exploratory examination of the American Psychiatric
Association practice guideline. The Journal of Clinical Psychiatry, 68,
885– 891. http://dx.doi.org/10.4088/JCP.v68n0610
Milrod, B., Leon, A. C., Busch, F., Rudden, M., Schwalberg, M., Clarkin,
J., . . . Shear, M. K. (2007). A randomized controlled clinical trial of
psychoanalytic psychotherapy for panic disorder. The American Journal
of Psychiatry, 164, 265–272. http://dx.doi.org/10.1176/ajp.2007.164.2
.265
Milrod, B., Markowitz, J. C., Gerber, A. J., Cyranowski, J., Altemus,
M., Shapiro, T., . . . Glatt, C. (2014). Childhood separation anxiety
and the pathogenesis and treatment of adult anxiety. The American
Journal of Psychiatry, 171, 34 – 43. http://dx.doi.org/10.1176/appi
.ajp.2013.13060781
Mitte, K. (2005). A meta-analysis of the efficacy of psycho- and pharma-
cotherapy in panic disorder with and without agoraphobia. Journal of
Affective Disorders, 88, 27– 45. http://dx.doi.org/10.1016/j.jad.2005.05
.003
Nemeroff, C. B., Heim, C. M., Thase, M. E., Klein, D. N., Rush, A. J.,
Schatzberg, A. F., . . . Keller, M. B. (2003). Differential responses to
psychotherapy versus pharmacotherapy in patients with chronic forms of
major depression and childhood trauma. Proceedings of the National
Academy of Sciences of the United States of America, 100, 14293–
14296. http://dx.doi.org/10.1073/pnas.2336126100
Porter, E., & Chambless, D. L. (2015). A systematic review of predictors
and moderators of improvement in cognitive-behavioral therapy for
panic disorder and agoraphobia. Clinical Psychology Review, 42, 179 –
192. http://dx.doi.org/10.1016/j.cpr.2015.09.004
Reich, J. (2003). The effect of Axis II disorders on the outcome of
treatment of anxiety and unipolar depressive disorders: A review. Jour-
nal of Personality Disorders, 17, 387– 405. http://dx.doi.org/10.1521/
pedi.17.5.387.22972
Rudden, M. (2009). Brief RF Interview. Unpublished manuscript, West
Stockbridge, MA.
Rutimann, D. D., & Meehan, K. B. (2012). Validity of a brief interview for
assessing reflective function. Journal of the American Psychoanalytic
Association, 60, 577–589. http://dx.doi.org/10.1177/0003065112445616
SAS Institute. (2015). SAS® 9.4 language reference: Concepts (5th ed.).
Cary, NC: Author.
Schneider, R. L., Arch, J. J., & Wolitzky-Taylor, K. B. (2015). The state of
personalized treatment for anxiety disorders: A systematic review of
treatment moderators. Clinical Psychology Review, 38, 39 –54. http://dx
.doi.org/10.1016/j.cpr.2015.02.004
Shear, M. K., Brown, T. A., Barlow, D. H., Money, R., Sholomskas, D. E.,
Woods, S. W., . . . Papp, L. A. (1997). Multicenter collaborative Panic
Disorder Severity Scale. The American Journal of Psychiatry, 154,
1571–1575. http://dx.doi.org/10.1176/ajp.154.11.1571
Shear, M. K., Rucci, P., Williams, J., Frank, E., Grochocinski, V., Vander
Bilt, J., . . . Wang, T. (2001). Reliability and validity of the Panic
Disorder Severity Scale: Replication and extension. Journal of Psychi-
atric Research, 35, 293–296. http://dx.doi.org/10.1016/S0022-
3956(01)00028-0
Sheehan, D. V. (1983). Sheehan disability scales. In D. V. Sheehan (Ed.),
The anxiety disease (p. 151). New York, NY: Charles Scribner.
 PSYCHOTHERAPY FOR PANIC DISORDER 813

Silove, D., Alonso, J., Bromet, E., Gruber, M., Sampson, N., Scott, K., . . . response data with informative drop-out. Biometrics, 54, 367–383.
Kessler, R. C. (2015). Pediatric-onset and adult-onset separation anxiety http://dx.doi.org/10.2307/2534023
disorder across countries in the World Mental Health Survey. The
American Journal of Psychiatry, 172, 647– 656. http://dx.doi.org/10
.1176/appi.ajp.2015.14091185 Received August 3, 2016
ten Have, T. R., Kunselman, A. R., Pulkstenis, E. P., & Landis, J. R. Revision received April 27, 2017
(1998). Mixed effects logistic regression models for longitudinal binary Accepted April 28, 2017 䡲
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Members of Underrepresented Groups:

Reviewers for Journal Manuscripts Wanted
If you are interested in reviewing manuscripts for APA journals, the APA Publications and
Communications Board would like to invite your participation. Manuscript reviewers are vital to the
publications process. As a reviewer, you would gain valuable experience in publishing. The P&C
Board is particularly interested in encouraging members of underrepresented groups to participate
more in this process.

If you are interested in reviewing manuscripts, please write APA Journals at Reviewers@apa.org.
Please note the following important points:

• To be selected as a reviewer, you must have published articles in peer-reviewed journals. The
experience of publishing provides a reviewer with the basis for preparing a thorough, objective
review.

• To be selected, it is critical to be a regular reader of the five to six empirical journals that are most
central to the area or journal for which you would like to review. Current knowledge of recently
published research provides a reviewer with the knowledge base to evaluate a new submission
within the context of existing research.

• To select the appropriate reviewers for each manuscript, the editor needs detailed information.
Please include with your letter your vita. In the letter, please identify which APA journal(s) you
are interested in, and describe your area of expertise. Be as specific as possible. For example,
“social psychology” is not sufficient—you would need to specify “social cognition” or “attitude
change” as well.

• Reviewing a manuscript takes time (1– 4 hours per manuscript reviewed). If you are selected to
review a manuscript, be prepared to invest the necessary time to evaluate the manuscript
thoroughly.

APA now has an online video course that provides guidance in reviewing manuscripts. To learn
more about the course and to access the video, visit http://www.apa.org/pubs/authors/review-
manuscript-ce-video.aspx.