Customize your JAMA Network experience by selecting one or more topics from the list below.
Barlow DH, Gorman JM, Shear MK, Woods SW. Cognitive-Behavioral Therapy, Imipramine, or Their Combination for Panic DisorderA Randomized Controlled Trial. JAMA. 2000;283(19):2529–2536. doi:10.1001/jama.283.19.2529
Context Panic disorder (PD) may be treated with drugs, psychosocial intervention,
or both, but the relative and combined efficacies have not been evaluated
in an unbiased fashion.
Objective To evaluate whether drug and psychosocial therapies for PD are each
more effective than placebo, whether one treatment is more effective than
the other, and whether combined therapy is more effective than either therapy
Design and Setting Randomized, double-blind, placebo-controlled clinical trial conducted
in 4 anxiety research clinics from May 1991 to April 1998.
Patients A total of 312 patients with PD were included in the analysis.
Interventions Patients were randomly assigned to receive imipramine, up to 300 mg/d,
only (n=83); cognitive-behavioral therapy (CBT) only (n=77); placebo only
(n=24); CBT plus imipramine (n=65); or CBT plus placebo (n=63). Patients were
treated weekly for 3 months (acute phase); responders were then seen monthly
for 6 months (maintenance phase) and then followed up for 6 months after treatment
Main Outcome Measures Treatment response based on the Panic Disorder Severity Scale (PDSS)
and the Clinical Global Impression Scale (CGI) by treatment group.
Results Both imipramine and CBT were significantly superior to placebo for the
acute treatment phase as assessed by the PDSS (response rates for the intent-to-treat
[ITT] analysis, 45.8%, 48.7%, and 21.7%; P=.05 and P=.03, respectively), but were not significantly different
for the CGI (48.2%, 53.9%, and 37.5%, respectively). After 6 months of maintenance,
imipramine and CBT were significantly more effective than placebo for both
the PDSS (response rates, 37.8%, 39.5%, and 13.0%, respectively; P=.02 for both) and the CGI (37.8%, 42.1%, and 13.0%, respectively).
Among responders, imipramine produced a response of higher quality. The acute
response rate for the combined treatment was 60.3% for the PDSS and 64.1%
for the CGI; neither was significantly different from the other groups. The
6-month maintenance response rate for combined therapy was 57.1% for the PDSS
(P=.04 vs CBT alone and P=.03
vs imipramine alone) and 56.3% for the CGI (P=.03
vs imipramine alone), but not significantly better than CBT plus placebo in
either analysis. Six months after treatment discontinuation, in the ITT analysis
CGI response rates were 41.0% for CBT plus placebo, 31.9% for CBT alone, 19.7%
for imipramine alone, 13% for placebo, and 26.3% for CBT combined with imipramine.
Conclusions Combining imipramine and CBT appeared to confer limited advantage acutely
but more substantial advantage by the end of maintenance. Each treatment worked
well immediately following treatment and during maintenance; CBT appeared
durable in follow-up.
Panic disorder (PD) is a chronic condition associated with substantial
reduction in quality of life,1 and lifetime
prevalence rates are approximately 3%.2,3
Role functioning is substantially lower in patients with PD than in patients
with diabetes, heart disease, or arthritis.4
Individuals with PD frequently use both emergency department and general medical
services,5 presenting with high rates of unexplained
cardiac symptoms,6,7 dizziness,
and bowel distress.6,8 These patterns
continue indefinitely.9,10 The
social and economic costs of PD are considerable.11-13
Successfully treating PD can produce medical cost offsets as high as 94%.12 Thus, efforts to ascertain effective treatments or
a combination of treatments for PD are important.
Earlier psychosocial treatments were directed specifically at unexpected
panic attacks and associated anxiety.14,15
A growing number of studies support the effectiveness of these psychosocial
approaches for PD compared with no treatment or credible psychosocial placebos.16 By the 1980s, the efficacy of pharmacological treatment
with imipramine in patients with PD had been well established.17-21
Imipramine was regarded as the pharmacological criterion standard for the
treatment of PD for more than 20 years,22-24
until the emergence of the selective serotonin reuptake inhibitors (SSRIs).
Despite the proven efficacy of both drug and psychosocial treatments
for PD and some emerging evidence on the possible synergistic effects of these
approaches, particularly on phobic behavior,25
studies of medication and psychosocial approaches have, until recently, run
on parallel and sometimes hostile tracks.15,26-29
We assembled a team of 4 investigators, 2 committed to each approach, to undertake
a comparative study that would determine optimal treatment for PD.
We conducted a randomized controlled trial comparing cognitive-behavioral
therapy (CBT), imipramine plus medical management, the combination of CBT
and imipramine (CBT+imipramine), pill placebo plus medical management, and
CBT+placebo for PD (Figure 1). Randomization
was stratified by site and presence of Diagnostic and Statistical
Manual of Mental Disorders, Revised Third Edition–defined current
major depression and was blocked within stratum. To improve trial efficiency,30 unequal numbers of patients were randomized to the
treatments (6 CBT, 6 imipramine, 5 CBT+imipramine, 25 CBT+placebo, and 2 placebo
per block of 24) based on expected pairwise comparison effect sizes. The acute
treatment phase consisted of 11 sessions during 12 weeks. Each CBT session
lasted approximately 50 minutes and each drug treatment session approximately
30 minutes. Patients in combined treatment saw 2 therapists for a total of
about 75 minutes per week. Responders to acute treatment entered a 6-month
maintenance phase without breaking the study blind. Maintenance-phase treatment
consisted of 6 monthly appointments in which treatment similar to the acute
treatment was continued. After 9 months of treatment (3 months acute and 6
months maintenance), patients were assessed again. Responders to maintenance
treatment were then assigned to discontinuation of treatment, except for 17
patients who were randomized to an extended maintenance pilot project. Remaining
patients were assessed again after 6 months of follow-up (15 months after
treatment was initiated). The trial was conducted between May 1991 and April
1998 and was approved by institutional review boards at each site.
All patients passing diagnostic screening for a principal diagnosis
of PD with or without mild agoraphobia (N=497) were entered in the pretreatment
phase after signing written, informed consent. Pretreatment included drug
washout for patients taking anxiolytic or antidepressant medication. Patients
were permitted up to 10 doses of benzodiazepine (0.5 mg of alprazolam-equivalent)
in the 2 weeks before the first treatment visit and up to 20 doses during
baseline and acute treatment combined. During the 2 weeks prior to the first
treatment visit, patients underwent physical and laboratory examinations,
and diagnosis was confirmed using the Anxiety Disorders Interview Schedule-Revised
(ADIS-R).31,32 Mild agoraphobia
was operationally defined as a score less than or equal to 18 on the ADIS-R
avoidance scale. In addition, inclusion required at least 1 full or limited
panic attack in the 2 weeks before the first treatment visit.
Exclusion criteria were psychotic, bipolar, or significant medical illnesses,
suicidality, significant substance abuse, contraindications to either treatment,
prior nonresponse to similar treatments, and concurrent competing treatment
or pending disability claims. More details are available on request from the
authors. Patients with comorbid unipolar depression were not excluded unless
suicidal. A detailed analysis of reasons for pretreatment attrition is provided
Therapists providing CBT were doctoral-level clinicians who underwent
extensive training and certification supervised by the Boston site prior to
treating study patients. Clinicians were not required to have prior training
in cognitive-behavioral procedures. Pharmacotherapists were experienced psychiatrists
who underwent additional training and certification in the study protocol
under the direction of the Columbia/Hillside site. All CBT therapists and
pharmacotherapists received ongoing supervision of their cases from the responsible
site during biweekly telephone calls, and adherence and competence ratings
were routinely collected after listening to a sample of audiotaped sessions.
Cognitive-behavioral therapy for PD, developed at the Boston site, combines
interoceptive exposure, cognitive restructuring, and breathing retraining.
This treatment was described in a manual containing detailed instructions
for the conduct of each session.34
The imipramine and placebo interventions were administered in a double-blind,
fixed flexible-dose design, according to a manual developed for this study.
Both the imipramine and placebo arms included a medical management component,
specified in the manual. The purpose of medical management was to monitor
adverse effects, clinical state, and the patient's physical/mental condition;
maximize compliance with the pharmacological treatment protocol; and proscribe
specific interventions included in CBT (eg, cognitive restructuring of anxiety
and panic symptoms). Patients who experienced clinical deterioration were
removed from the study and offered alternative treatment for 3 months free
Starting dosages of imipramine were 10 mg/d (or pill placebo equivalent),
increased every other day by 10 mg until 50 mg/d was reached. The dosage was
then increased more rapidly, with an effort made to reach 100 mg/d by the
end of week 3 and 200 mg/d by week 5, even if the patient became symptom-free
earlier, unless adverse effects became intolerable. If the patient was not
symptom-free, the dosage could be increased up to 300 mg/d by week 5. Blood
levels of imipramine were assessed at 6 and 12 weeks and benzodiazepine screening
of urine samples performed by local commercial laboratories.
The combined treatment conditions were administered according to a manual
essentially consisting of the CBT and the imipramine manuals.
Prior to beginning the study, independent evaluators participated in
training and certification on the rating instruments, supervised by the Pittsburgh
site. Bimonthly conference calls were held for evaluators to discuss any questions
and/or discrepancies among sites, and regular, random interviews were monitored
for continued reliability of diagnosis and ratings. Reliability on main measures
remained above 90%. Evaluator assessments occurred at baseline and after acute,
maintenance, and follow-up phases, and evaluators were blind to treatment
The primary continuous outcome measure was the average item score for
the Panic Disorder Severity Scale (PDSS), a clinician-rated scale of PD severity.35 We also present results for a 40% reduction from
baseline on PDSS. The primary categorical outcome measure involved determination
of a responder based on the Clinical Global Impression Scale (CGI),36 consisting of 7-point scales rating overall improvement
and severity. We added specific PD-based anchor points for rating the CGI
in this study. To be a responder, a patient needed to achieve a score of 2
(much improved) or better while being rated as 3 (mild) or less on CGI severity.
Receipt of nonstudy treatment was evaluated at each assessment point. Patients
who received nonstudy treatment for anxiety or panic were determined to be
nonresponders, both for PDSS and CGI responder definitions.
The study was designed to address if both CBT alone and imipramine alone
performed better than placebo; if either CBT or imipramine performed better
relative to each other; and if there was an advantage to combining CBT and
imipramine as evidenced by superiority of CBT+imipramine to CBT+placebo, imipramine
alone, and CBT alone. These questions were addressed at each of 3 assessment
points: postacute, postmaintenance, and follow-up. At each point, we performed
an intent-to-treat (ITT) analysis, using an early-termination assessment.
The baseline was carried forward as an assumption of nonresponse or return
to baseline if the early-termination assessment was missing or contaminated
by nonstudy treatment. Similarly, the baseline was also carried forward in
follow-up analyses for the same reasons. Intent-to-treat analyses included
all randomized patients except those discovered to be ineligible after randomization
(Figure 1). At postmaintenance and
follow-up assessments, we also conducted intent-to-continue (ITC) analyses,
restricted to patients completing the preceding phase as responders. At postacute
assessment, we present results for completers only.
Data were subjected first to omnibus testing of all 5 treatments and
then to pairwise post hoc testing when the omnibus test indicated statistical
significance. The Freeman-Halton and Fisher exact tests were used for categorical
measure analyses, analysis of variance for PDSS baseline analysis, and analysis
of covariance for subsequent analyses using baseline as the covariate. When
analyses indicated a significant baseline-by-treatment interaction on the
dependent measure, analysis of covariance was replaced by repeated-measures
analysis of variance, with treatment assignment as the independent (between
groups) variable and time as the dependent (within groups) variable. Significance
was defined as P≤.05 with a 2-tailed test.
Of the 326 patients randomized, 312 are included in the analysis. Thirteen
were excluded following uniform screening for loss of eligibility, and 1 was
removed because of inadvertent unblinding. Proportions of excluded patients
were not significantly different among treatment assignments.
There were no significant differences on demographic measures or on
baseline PDSS score among the 5 randomized groups (Table 1). The mean PDSS scores indicate a moderate-to-average severity.
No significant site effect, depression stratum effect, site-by-treatment interaction,
or stratum-by-treatment interaction was observed in the acute ITT analyses
for any of the 3 measures; thus, all data were aggregated across sites and
across strata. Similarly, there were no significant interactions between treatment
and other baseline variables (Table 1)
in multivariate acute ITT analyses for any of the 3 measures. No important
differences at baseline were observed between treatment groups entering vs
not entering maintenance or follow-up. It has been suggested that when P<.10 level heterogeneity of the slope is present across
treatments, treatment effects that apply to an important portion of the baseline
severity spectrum may be obscured in analyses of covariance.37
For slope heterogeneity, P>.25 for 8 of 9 omnibus
PDSS analyses of covariance, suggesting little heterogeneity.
Acute completion rates did not differ significantly across treatment
groups. Among the 8 noncompliant patients was 1 patient receiving imipramine
alone who violated protocol by using benzodiazepines at rates exceeding those
allowed, and 1 patient receiving CBT who began nonprotocol antidepressant
treatment during acute treatment. Figure 1 also shows that 82% to 90% of maintenance-eligible patients completed
maintenance across active treatments, compared with 33% of placebo patients.
The omnibus test for this analysis was significant. Pairwise comparisons were
significant for all active treatments vs placebo, including CBT alone (P=.006), imipramine alone (P=.007),
CBT+imipramine (P=.001), and CBT+placebo (P=.004).
Patients receiving medication reported taking doses of 175 to 180 mg/d
across treatments by week 6 and 214 to 239 mg/d by week 12. Mean (SD) imipramine+desipramine
plasma levels at week 6 were 219 (186) ng/dL and 223 (127) ng/dL for CBT+imipramine
and imipramine alone, respectively, and 225 (148) ng/dL and 263 (156) ng/dL
at week 12.
Rates of urine samples that tested positive for benzodiazepines were
low. At week 6, 3 (1.5%) of 197 samples tested positive (1 CBT+imipramine,
1 CBT, 1 imipramine), and at week 12, 3 (1.8%) of 164 (1 CBT+imipramine, 1
CBT, 1 imipramine) tested positive. Rates of missing urine samples were not
significantly different across treatments.
CBT Alone and Imipramine Alone vs Placebo. In the acute ITT analysis, both CBT alone and imipramine alone were
superior to placebo for the PDSS continuous measure (Table 2). Both treatments had significantly fewer dropouts for lack
of efficacy than did placebo (4/18 for CBT and 10/30 for imipramine vs 8/10
for placebo [Figure 1]), and the
imipramine group had significantly more dropouts for adverse effects than
did the placebo group. Maintenance ITT analysis (Table 2) confirms the superiority of both CBT alone and imipramine
alone to placebo.
CBT vs Imipramine. We found no significant difference between CBT alone and imipramine
alone in the acute ITT or completer analyses or in the maintenance ITT or
ITC analyses (Table 2). As expected,
significantly more patients in the imipramine group than in the CBT group
dropped out because of adverse effects (Figure
1). The follow-up analyses (Table
2) show trends favoring CBT over imipramine.
Combined CBT+Imipramine vs Single Treatment. We hypothesized that CBT+imipramine would be better than all 3 of the
relevant comparison groups. Table 2
shows that CBT+imipramine was superior to CBT alone on 1 of 3 ITT analyses
and 1 of 3 completer analyses but was not superior to CBT+placebo. In the
maintenance ITT analysis (Table 2),
combined treatment was better in the PDSS average analysis than in all 3 comparison
treatments (thereby meeting our criteria for superiority) and better than
CBT and CBT+placebo on ITC analyses.
Intent-to-continue analyses showed that responders to imipramine, with
or without CBT, fared significantly worse in the no-treatment follow-up period
than those who received either CBT alone or CBT+placebo (Table 2). After treatment discontinuation in the follow-up ITT analyses,
the treatments that continued to show evidence of superiority to placebo were
CBT alone (Table 2) and CBT+placebo
(statistically significant for all 3 measures).
In a secondary analysis restricted to responders based on the CGI definition
(Table 3), acute imipramine responders
had significantly lower PDSS average scores than acute CBT responders, indicating
a higher quality of response. Maintenance responders showed the same pattern
at the trend level. Responders to combined CBT+imipramine had higher-quality
responses than CBT responders at acute and maintenance points, as well as
a higher quality of response than patients taking CBT+placebo at maintenance.
Timing of loss of response could be determined in 4 of 5 imipramine
follow-up completers, losing response during months 3, 4, 5, and 6 (1 case
each) and in 1 of 2 CBT follow-up completers, losing response during month
3. Among 9 CBT+imipramine follow-up completers losing response, relapse occurred
during months 2, 4, and 5 (2 cases each), months 1, 3, and 6 (1 case each),
and in the CBT+placebo relapser during month 1.
Our results demonstrate that both imipramine and CBT are better than
pill placebo for treatment of PD. Imipramine produced a superior quality of
response, but CBT had more durability and was somewhat better tolerated.
In our study, ITT placebo response did not differ from active treatment
on acute-phase assessment when CGI was used to determine responder status.
By contrast, the 7-item PDSS successfully discriminated between conditions.
Moreover, the placebo response in the ITT analysis of 37.5% based on CGI criteria
after 3 months drops to 13% after 9 months of treatment. Several studies have
made similar observations in the treatment of depression and PD.38-41
While attrition in the placebo group may have compromised comparisons at the
end of maintenance, placebo response is weak in magnitude and transient in
There are several differences between active treatments. Although no
differences emerged on a priori planned completer or ITT analyses, patients
treated with imipramine and designated responders based on CGI criteria following
the acute phase showed significantly more improvement on the PDSS than patients
who responded to CBT. A trend level of significance remained at the end of
maintenance. Thus, among those who did well with either treatment, patients
receiving imipramine responded more completely. However, at follow-up, patients
who had received CBT alone maintained their improvement significantly better
(4% relapse) than those treated with imipramine (25% relapse) based on PDSS
responder criteria. We discontinued imipramine by tapering during a 1- to
2-week period, following standard practice at the time. Relapses in medication-treated
patients appeared to be evenly distributed across follow-up, suggesting that
withdrawal played little role in the results. We did not aim to determine
optimal tapering strategy or to ascertain the duration of medication maintenance
that produces the best long-term outcome. Our results indicate the need for
Findings of high acceptability and durability of CBT are consistent
with previous reports,16 although attrition
in the CBT-alone group was higher than reported previously.6,14,15,42
Adverse effects with imipramine and relapse following discontinuation are
also consistent with previous reports.38-40
However, our results that show a superior quality of response with imipramine
among responders to both treatments in the acute phase underscore the need
to reduce attrition and develop optimal maintenance and discontinuation procedures
for those receiving medication.
Acute coadministration of imipramine and CBT resulted in limited benefit
over monotherapy. Adding medication to CBT achieved significantly better results
than CBT alone at postacute assessment on some measures, but this combination
was never better than CBT+placebo. By the end of maintenance, CBT+imipramine
was superior to both CBT alone and CBT+placebo (as well as imipramine alone)
on the PDSS average measure. However, this robust combination treatment produced
the highest relapse rate at follow-up assessment. Surprisingly, the addition
of CBT to imipramine did not mitigate relapse following medication discontinuation;
addition of imipramine appeared to reduce the long-term durability of CBT.
More work is needed to elucidate this result. A selection effect may account
for the relatively poor outcome of combined treatment after discontinuation.
Combined treatment could conceivably select patients who had been in particular
need of long-term treatment from the beginning. We have not been able to detect
important differences at baseline on variables in Table 1 between patients who did and did not enter the follow-up
phase, but these analyses do not exclude the possibility of selection on an
unmeasured prognostic factor.
There are several limitations of this study. First, to avoid the complications
and possible confounding factors of adding exposure-based interventions to
each condition, we enrolled patients with only limited degrees of phobic avoidance,
and results are generalizable only to this group. Second, it could be argued
that we underestimate the benefits of medication by using a tricyclic antidepressant
instead of an SSRI. Current recommendations43
(not yet in place when this study began) consider SSRIs to be the first-line
medication. While there is little question that SSRIs are more convenient
and have a more limited adverse-effect profile, studies examining differences
from tricyclic antidepressants do not consistently find higher efficacy for
SSRIs. In fact, of 4 randomized studies, none found significant differences
in the end-of-study acute ITT analyses.44-47
One study found evidence for a more rapid response for the SSRI,44
and another found that the tricyclic antidepressant but not the SSRI was more
effective than placebo.46 Moreover, there have
also been refinements in CBT since we began our study. Thus, we believe it
likely that results of a similar comparative study using an SSRI would not
differ substantially from ours. Third, at the time we designed this study,
a consensus existed that an adequate dose of imipramine should be at least
200 mg/d. A recent study, however, suggests that imipramine/desipramine plasma
levels of about 150 mg/mL may be optimal in treating PD.48
Hence, it is conceivable that we underestimate the therapeutic potential of
imipramine in this study. Finally, the use of nonstudy medication is a possible
confounding factor to any anxiety study. We made the decision to permit limited
use of benzodiazepines, because we sought to simulate clinical practice. Rates
of urine samples that tested positive for benzodiazepine use among the 5 treatments
were equivalent and low. Only 1 subject's data were censored because of excessive
benzodiazepine use. Thus, we believe benzodiazepine use did not play a significant
role in our results.
This study represents, to our knowledge, the first multicenter trial
comparing medication and psychosocial therapies and their combination for
PD. Prior studies contrasting the 2 approaches have been criticized because
of possible investigator-allegiance bias in study design, implementation,
and/or analysis. Our study sites included 2 with psychotherapy and 2 with
pharmacotherapy expertise. In this context, the absence of site differences
in ITT outcome supports the important implication that both types of treatment
should be transportable to most clinical settings and confirms the generalizability
of the results.