ORIGINAL ARTICLE

Fluoxetine, Comprehensive Cognitive Behavioral

Therapy, and Placebo in Generalized Social Phobia
Jonathan R. T. Davidson, MD; Edna B. Foa, PhD; Jonathan D. Huppert, PhD;
Francis J. Keefe, PhD; Martin E. Franklin, PhD; Jill S. Compton, PhD; Ning Zhao, PhD;
Kathryn M. Connor, MD; Thomas R. Lynch, PhD; Kishore M. Gadde, MD

Background: Generalized social phobia is common, per- Scale and Clinical Global Impressions scales as primary
sistent, and disabling and is often treated with selective outcomes. A videotaped behavioral assessment served as
serotonin reuptake inhibitor drugs or cognitive behav- a secondary outcome, using the Subjective Units of Dis-
ioral therapy. tress Scale. Adverse effects were measured by self-
rating. Each treatment was compared by means of ␹2 tests
Objective: We compared fluoxetine (FLU), compre- and piecewise linear mixed-effects models.
hensive cognitive behavioral group therapy (CCBT), pla-
cebo (PBO), and the combinations of CCBT/FLU and Results: Clinical Global Impressions scales response rates
CCBT/PBO. in the intention-to-treat sample were 29 (50.9%) (FLU),
31 (51.7%) (CCBT), 32 (54.2%) (CCBT/FLU), 30 (50.8%)
D e s i g n : Randomized, double-blind, placebo- (CCBT/PBO), and 19 (31.7%) (PBO), with all treat-
controlled trial. ments being significantly better than PBO. On the Brief
Social Phobia Scale, all active treatments were superior
Setting: Two academic outpatient psychiatric centers. to PBO. In the linear mixed-effects models analysis, FLU
was more effective than CCBT/FLU, CCBT/PBO, and PBO
Patients: Subjects meeting a primary diagnosis of gen- at week 4; CCBT was also more effective than CCBT/
eralized social phobia were recruited via advertisement. FLU and CCBT/PBO. By the final visit, all active treat-
Seven hundred twenty-two were screened, and 295 were ments were superior to PBO but did not differ from each
randomized and available for inclusion in an intention- other. Site effects were found for the Subjective Units of
to-treat efficacy analysis; 156 (52.9%) were male, 226 Distress Scale assessment, with FLU and CCBT/FLU su-
(76.3%) were white, and mean age was 37.1 years. perior to PBO at Duke University Medical Center,
Durham, NC. Treatments were well tolerated.
Interventions: Treatment lasted for 14 weeks. Fluox-
etine and PBO were administered at doses from 10 mg/d Conclusions: All active treatments were superior to PBO
to 60 mg/d (or equivalent). Group comprehensive cog- on primary outcomes. Combined treatment did not yield
nitive behavioral therapy was administered weekly for any further advantage. Notwithstanding the benefits of
14 sessions. treatment, many patients remained symptomatic after 14
weeks.
Main Outcome Measures: An independent blinded
evaluator assessed response with the Brief Social Phobia Arch Gen Psychiatry. 2004;61:1005-1013

S
OCIAL PHOBIA, NOW OFTEN RE- recognition of GSP was no more than 0.5%
ferred to as social anxiety dis- of all cases.4
order, is widespread with Controlled trials suggest benefit from
prevalence rates as high as medication and psychosocial ap-
Author Affiliations: 14% in the United States,1 be- proaches. Among medications, the selec-
Department of Psychiatry and gins early in life, and rarely remits.2 Of the tive serotonin reuptake inhibitor (SSRI)
Behavioral Sciences, Duke 2 types of social phobia, generalized so- group is the most extensively studied.5
University Medical Center, cial phobia (GSP) is believed to be more Cognitive behavioral treatments (CBTs)
Durham, NC (Drs Davidson,
severe than nongeneralized.3 A recent are also effective in social phobia, espe-
Keefe, Compton, Connor,
Lynch, and Gadde); and the study of GSP in a health maintenance or- cially a form of group CBT, which in-
Department of Psychiatry, ganization setting showed an 8% preva- cludes cognitive restructuring, exposure
University of Pennsylvania, lence rate, along with raised incidence of to simulated situations in sessions, and in
Philadelphia, Pa (Drs Foa, suicide attempts, greater health seeking, vivo exposure homework assignments.6
Huppert, Franklin, and Zhao). and reduced earning capacity; physician Group CBT has been found to be more ef-

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1005

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015
 Table 1. Demographic Characteristics*

FLU Group CCBT Group CCBT/FLU Group CCBT/PBO Group PBO Group Test P Value
No. of subjects 57 60 59 59 60
Women 42.9 53.3 54.2 36.2 45.8 ␹ 42 = 5.33 .26
White 71.4 71.2 84.7 75.9 82.8 ␹ 42 = 5.30 .26
Age, y, mean (SD) 36.3 (11.1) 36.7 (9.1) 38.2 (10.7) 37.8 (10.2) 36.9 (10.6) F4,283 = 0.31 .87

Abbreviations: CCBT, comprehensive cognitive behavioral therapy; FLU, fluoxetine; PBO, placebo.
*Values are expressed as percentages unless otherwise indicated.

fective than psychoeducation,6 atenolol,7 and buspirone pist. The study coordinator at each site enrolled and allocated
hydrochloride8 and equivalent to phenelzine sulfate.6 subjects to their treatment groups. This individual was blind
Comprehensive cognitive behavior therapy (CCBT) was to the sequence prior to assignment.
developed by Foa et al (unpublished data, 1994) for GSP. Medication was administered and monitored by a psychia-
trist. Medication sessions were audiotaped, and tapes were au-
In this group CBT, social skills training is added to ex-
dited at random by 1 of the investigators at each site. Pharmaco-
posure therapy and cognitive restructuring, because so- therapy was provided according to the study manual and adherence
cial skills deficits are a pertinent part of GSP and may rated according to a standardized checklist (available on re-
not respond well to programs if skill training is absent.9 quest) to assure that no CCBT was being conducted and that medi-
With the emergence of SSRIs as frontline and proven cation was being provided in a standard way. Medication visits
pharmacotherapy for GSP, we considered it important to occurred weekly for 4 weeks, then every 2 weeks. Study investi-
compare an SSRI with CCBT. To our knowledge, this is 1 gators (E.B.F., F.J.K., and M.E.F.) videotaped and evaluated CCBT
of only 2 completed GSP studies to include a combina- at each site and provided feedback to therapists in weekly super-
tion cell in which subjects received both CCBT and an vision sessions (adherence rating scale available by request). Early
SSRI.10 Our project had 5 overall goals: (1) to compare the in the trial, the sites provided feedback to therapists via weekly
supervisory conference calls to ensure consistency of treatment
effects of 14 weeks’ treatment with fluoxetine (FLU) alone,
across sites. Group CCBT was administered at weekly intervals.
group alone (CCBT), combined CCBT/FLU, CCBT/ Enrollment began in early 1995 and continued until Sep-
placebo (PBO) (to take into account nonspecific pill tak- tember 2001. Regular conference calls were held between the
ing), and PBO alone; (2) to evaluate maintenance of treat- 2 sites throughout. The protocol was approved by the institu-
ment effects following completion of treatment; (3) to tional review board at each site, and all subjects provided writ-
demonstrate transportability of treatments (ie, that CCBT ten informed consent.
can be successfully implemented in a center specializing
in pharmacotherapy [Duke University Medical Center, SAMPLE
Durham, NC] and vice versa for a center specializing in
Inclusion criteria were: (1) DSM-IV diagnosis of GSP; (2) age be-
medication [University of Pennsylvania, Philadelphia, Pa]);
tween 18 and 65 years; (3) fluency in English; and (4) provision
(4) to investigate mechanisms of therapeutic change by of written informed consent. Exclusion criteria were: (1) a pri-
examining the relationship between cognitive distortions mary comorbid anxiety disorder (defined by which disorder was
and social skills deficits; and (5) to explore predictors of the more debilitating and clinically salient); (2) lifetime history
treatment response. This report will focus on the first goal, of schizophrenia, bipolar disorder, or organic brain syndrome; (3)
short-term effects of the 5 treatments. major depression within the last 6 months; (4) substance abuse
or dependence within the past year; (5) mental retardation or per-
vasive developmental disability; (6) unstable medical condition;
METHODS
(7) prior failure of response to fluoxetine at 60 mg/d for at least
4 weeks or to 12 weekly sessions of CCBT for GSP; (8) concur-
OVERALL DESIGN rent psychiatric treatment or other psychoactive medications; (9)
positiveurinedrugscreenresults;(10)inabilitytomaintain2weeks’
The study was conducted at 2 academic medical centers with psychotropic drug-free washout; and (11) pregnancy or lactation.
outpatient programs specializing in anxiety disorder research. Table 1 presents demographic characteristics by treat-
At each research center, FLU, CCBT, CCBT/FLU, CCBT/PBO, ment group, showing no group differences. Between-site demo-
and PBO were all compared, with FLU and PBO being admin- graphic differences were observed for ethnicity, marital sta-
istered on a double-blind basis. An independent rater, blinded tus, and employment, with fewer African American subjects and
to treatment assignments, conducted the primary outcome as- more married and employed subjects in the Duke University
sessments. Eligible subjects met DSM-IV criteria for primary Medical Center site sample, as compared with the University
GSP, assessed by the Structured Clinical Interview for DSM- of Pennsylvania site sample. Baseline symptom scores for each
IV.11 Subjects underwent psychiatric and medical evaluation to treatment group are shown in Table 2.
establish inclusion and exclusion criteria. Subjects were as-
signed to treatment by block randomization, which was gen- TREATMENT ADMINISTRATION
erated by computer program at Duke University Medical Cen-
ter, in groups of 10, with 2 subjects assigned to each of the 5 Fluoxetine was started at 10 mg/d, increasing on day 8 to 20
conditions. There were some exceptions to the implementa- mg/d, on day 15 to 30 mg/d, and on day 29 to 40 mg/d. Unless
tion of this process because of a small number of prerandom- adverse effects became problematic, the goal was for subjects
ization dropouts. We balanced CCBT groups to include at least to reach 40 mg/d. At days 43 and 57, the dose could be raised
2 women and 2 men and typically had a male and a female thera- to 50 mg/d and 60 mg/d, respectively, if subjects failed to achieve

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1006

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015
 Table 2. Effects of Treatment and Response Rates: Raw Posttreatment and Linear Mixed-Effects Models (LMM) Analyses*

FLU Group CCBT Group CCBT/FLU Group CCBT/PBO Group PBO Group Test P Value
Effects of Treatment
No. of subjects
Week 0 57 60 59 59 60
Week 14 39 48 42 46 36
BSPS score
Week 0 38.4 (9.6) 39.2 (10.4) 39.1 (9.9) 37.6 (9.9) 37.3 (8.5) F4,274 = 0.49 .75
Week 0-4 F4,654 = 4.13 .003
Week 4-14 F4,647 = 2.89 .02
Week 14† 20.8 (13.2) 20.6 (9.9) 21.6 (12.6) 20.5 (12.1) 26.7 (13.5)
CGI-S score
Week 0 4.4 (0.1) 4.5 (0.1) 4.4 (0.1) 4.4 (0.1) 4.3 (0.1) F4,274 = 0.33 .86
Week 0-4 F4,654 = 4.08 .003
Week 4-14 F4,647 = 3.46 .01
Week 14‡ 2.7 (1.2) 2.9 (1.2) 2.7 (1.2) 2.8 (1.2) 3.3 (1.3)
SPAI score
Week 0§ 97.5 (24.8) 106.1 (21.3) 109.9 (23.8) 111.3 (21.0) 112.0 (21.0) F4,274 = 3.63 .007
Week 0-4 F4,654 = 1.03 .39
Week 4-14 F4,647 = 2.3 .05
Week 14† 69.3 (37.2) 77.1 (28.7) 76.1 (31.5) 75.4 (32.0) 94.8 (28.0)
Response Rate at Week 14
ITT sample, %㛳 50.9 51.7 54.2 50.8 31.7 ␹ 42 = 8.03 .09
Completer sample, %¶ 64.1 64.6 66.7 58.7 38.9 ␹ 42 = 8.09 .09

Abbreviations: BSPS, Brief Social Phobia Scale; CCBT, comprehensive cognitive behavioral therapy; CGI-S, Clinical Global Impressions Severity scale;
FLU, fluoxetine; ITT, intention to treat; PBO, placebo; SPAI, Social Phobia and Anxiety Inventory.
*Values are expressed as mean (SD) unless otherwise indicted. Statistics and P values were calculated by LMM analyses.
†All individual treatments were superior to PBO at week 14 (P⬍.05).
‡All individual treatments were superior to PBO, except CCBT/PBO, at week 14 (P⬍.05).
§At week 0, FLU treatment was superior to CCBT/FLU, CCBT/PBO, and PBO (P⬍.05).
㛳All individual treatment response rates were superior to PBO at week 14 (P⬍.05).
¶All individual treatment response rates were superior to PBO, except CCBT/PBO, at week 14 (P⬍.05).

a Clinical Global Impressions (CGI) Improvement12 score of 1 INDEPENDENT EVALUATOR RATINGS

or 2 and were tolerating medication. Compliance was moni- AND SELF-REPORT MEASURE
tored by reviewing daily medication logs and pill counts at each
visit. Normally, the dose was given in the morning. Primary outcome assessments by the blinded independent evalu-
Comprehensive cognitive behavioral therapy is a 14-week ator (IE) were as follows: (1) CGI Improvement scale, a 7-point
group treatment that combines in vivo exposure, cognitive re- rating measured improvement wherein response is defined as hav-
structuring, and social skills training. Derived in part from a ing a CGI Improvement score of 1 (very much improvement) or
treatment developed by Heimberg et al,13 CCBT differs from 2 (much improvement), and the 7-point CGI Severity scale12 and
Heimberg group CBT in that CCBT includes specific social skills (2) the Brief Social Phobia Scale (BSPS), an 18-item scale com-
training (eg, how to begin a conversation with a stranger) and prised of fear, avoidance, and physiological symptoms.14 Inde-
improves specific behaviors (eg, eye contact), the role-plays are pendent evaluator ratings were conducted at baseline and at weeks
much shorter, and CCBT is 2 sessions longer than group CBT. 4, 8, and 14. Success of the blinding procedure was not evalu-
Two therapists (1 male, 1 female) who received extensive train- ated. Altogether, 8 IEs were associated with this trial from start
ing prior to this study conducted the treatment; each group con- to finish (3 at Duke University Medical Center and 5 at Univer-
sisted of 5 to 6 subjects. The first 2 sessions were educational, sity of Pennsylvania). Intraclass correlation pairwise agreement
with therapists presenting a cognitive behavioral model of so- coefficients for between raters ranged from 0.77 to 0.98.
cial anxiety and explaining treatment techniques. Sessions 3 The Social Phobia and Anxiety Inventory,15 which is reli-
and 4 were devoted to social skills training; patients received able, valid, and sensitive to treatment effects,16 served as a sec-
instruction and role-played short (30-60 second), repeated (5-7 ondary measure.
times) scenarios devoted to initiating, maintaining, and end-
ing conversations, as well as compromise/negotiation. In ses-
sions 5 through 13, patients participated in longer (3-4 minute) ADVERSE EFFECTS
role-plays tailored to their specific social concerns. Prior to each
role-play, subjects identified a core dysfunctional thought as- Symptoms possibly attributable to treatment were evaluated by
sociated with that situation and a relevant rational response to means of the Severity of Symptoms Scale,17 which was given to
replace it. Next, social skills training instructions were pro- subjects at each visit, regardless of treatment assignment.
vided before the role-play, and specific aspects of each role-
play were repeated to facilitate skills acquisition. Between ses- BEHAVIORAL ASSESSMENT
sions, subjects completed homework assignments designed to
help them confront fearful social situations using the tech- A role-play test evaluated social skills before and after treat-
niques learned in therapy. Session 14 included a discussion of ment. The subject was videotaped in each of 3 social sce-
treatment gains and recommendations for future practices. narios: initiating a conversation with a stranger; expression of

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1007

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015
 negative assertiveness; and an impromptu speech. Subjects were most common reasons for exclusion at this stage were
asked to indicate their level of anxiety using the Subjective Units the presence of major depression, social phobia being non-
of Distress Scale (SUDS), an analogue scale ranging from 0 (re- generalized, presence of another primary anxiety disor-
laxed and calm) to 100 (extremely fearful).The SUDS ratings der, and substance abuse. Of these, 295 were random-
were assessed 7 times: immediately after receiving role-play in-
ized to 1 of the 5 treatments, with the following
structions (ie, at baseline); before each of the 3 role-play sce-
narios (to measure anticipatory anxiety); and after each sce- dispositions: 16 (5%) dropped out prior to receiving treat-
nario (to measure “aftermath”). An experimenter and a ment; 68 (23%) dropped out between week 1 and week
confederate, both blinded to the subject’s treatment status, were 14; and 211 (72%) completed 14 weeks of treatment. Ran-
present during the test. Each scenario lasted 3 minutes. domized subjects who were analyzed composed the ITT
(n=265) and completer (n=211) samples, respectively,
STATISTICAL ANALYSIS as follows: FLU (n=57, 39), CCBT (n=60, 48), CCBT/
FLU (n=59, 42), CCBT/PBO (n=59, 46), and PBO (n=60,
The following hypotheses were tested: (1) all active treatments 36). Figure 1 provides details of subject disposition. The
would be superior to PBO and (2) combined CCBT/FLU treat- overall significance for rate of dropout by treatment type
ment would be superior to other active treatments. Post hoc, ex- was not statistically significant (␹24 =8.22; P=.08). How-
ploratory, 2-tailed pairwise contrasts were performed to deter- ever, in pairwise contrasts, the PBO group had signifi-
mine whether the active treatments differed from one another.
cantly more dropouts than the CCBT group (␹21 =6.53;
Sample size was calculated based on pilot data obtained earlier.
Three power estimates were made for the primary contrasts, with P =.01) and the CCBT/PBO group (␹21 =4.48; P=.03), re-
a single pooled estimate of the error term and 1-tailed hypoth- spectively.
esis testing. Power to detect a difference for 60 subjects per group, The proportion of subjects who discontinued prema-
with an effect size of 0.30 at ␣=.05, ranged from 0.7 to 0.9 in turely did not differ statistically across site (␹28 = 9.67;
the combined site sample according to comparison. P=.29). Reasons for early discontinuation are provided
First, to detect possible pretreatment differences among in Figure 1.
conditions, 1-way analyses of variance were performed on pre- There were no significant differences among groups
treatment data for the BSPS, CGI, Social Phobia and Anxiety In- in respect of mean medication doses. For the combined
ventory, and SUDS ratings. Second, the differential efficacy of the PBO groups (CCBT/PBO and PBO), mean (SD) maxi-
conditions at 14 weeks or an earlier end point was analyzed us-
mum and final attained visit doses were 49.3 (15.2) mg
ing (1) ␹2 tests to compare the proportion of subjects who achieved
treatment response status within each treatment and (2) linear equivalents per day and 46.4 (17.9) mg equivalents per
mixed-effects models (LMM), from SAS procedure MIXED,18 to day, respectively. For the combined FLU groups (CCBT/
assess differences in course of treatment response. For these analy- FLU and FLU), they were 47.3 (13.9) mg/d and 43.6
ses, piecewise linear growth curve models, with a change point (16.4) mg/d, respectively.
at week 4, provided the best fit for the data. An unstructured vari-
ance model best fit the data for all analyses. We applied this piece-
wise LMM to each of the 3 continuous outcome measures. These INDEPENDENT EVALUATIONS
analyses were followed by pairwise analyses to determine whether
treatments differed in how fast they induced change. In addition, The ITT response rates determined by a CGI Improve-
piecewise LMMs were applied to each outcome variable to deter-
ment score of 1 or 2 were as follows: FLU, 50.9%; CCBT,
mine whether each treatment induced change across the 14 treat-
ment weeks. Site effects were examined by including a binary 51.7%; CCBT/FLU, 54.2%; CCBT/PBO, 50.8%; and PBO,
indicator and its interactions with treatment and time in the piece- 31.7%. The overall ␹2 test showed a trend toward sig-
wise LMMs mentioned earlier. When the interaction of site by nificance (␹42 =8.03; P=.09), and pairwise significance was
treatment or site by time was significant, we conducted follow- found for all treatment comparisons against PBO: FLU
up analyses to explain these interactions. Sex, age, marital status, vs PBO (␹12=4.46; P=.03); CCBT vs PBO (␹12=4.94; P=.03);
ethnicity, and employment status were all examined to determine CCBT/FLU vs PBO (␹21 =6.19; P=.01); and CCBT/PBO vs
whether they differentially affected treatment outcome. None of PBO (␹21 = 4.52; P = .03). For the completer sample, re-
these variables had significant interactions with treatment con- sponse rates were as follows: FLU, 64.1%; CCBT, 64.6%;
dition by time and were therefore not included in final analyses. CCBT/PBO, 59.6%; CCBT/FLU, 66.7%; and PBO, 40.5%.
Linear mixed-effect model analyses included all random-
The overall ␹2 test showed a trend toward significance
ized subjects and were conducted using pretreatment and post-
treatment behavioral measures, with the behavioral measure (␹24 =8.09; P=.09). Pairwise significance was found for the
as the dependent variable. following treatment comparisons against PBO: FLU vs
Responder analyses were conducted separately for all ran- PBO (␹21 =4.77; P=.03); CCBT vs PBO (␹21 =5.46; P=.02);
domized subjects for whom any data were available (ITT) and and CCBT/FLU vs PBO (␹21 =6.02; P=.01); CCBT/PBO vs
subjects who completed 14 weeks (completer sample). PBO showed a trend toward significance (␹12=3.17; P=.08).
Effect size, along with 95% confidence intervals, was cal- There were more subjects in the responder sample in the
culated according to the Cohen formula19 for a difference be- CCBT/PBO group at Duke University Medical Center than
tween groups of m1 (postactive treatment) – m2 (post-PBO)/ at the University of Pennsylvania. No other site differ-
pooled standard deviation (m1m2), in order to determine effect ences were detected.
sizes of active treatments at posttreatment.
Effect sizes (95% confidence interval) for each com-
parison vs PBO, based on the ITT analysis of variance,
RESULTS provided the following results on the BSPS: 0.40 (0.02
to 0.77) for FLU; 0.30 (−0.07 to 0.66) for CCBT; 0.24
A total of 4306 subjects were screened by telephone, of (−0.13 to 0.60) for CCBT/FLU, and 0.52 (0.14 to 0.89)
whom 722 were invited for a full study assessment. The for CCBT/PBO vs PBO.

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1008

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015
 Telephone
Contact
Preliminary
Assessment

4368 Called
62 Abandoned
4306 Screened
3602 Ineligible

Excluded:
173 Not Meeting Inclusion Criteria
722 Assessed for Eligibility
65 Refused to Participate
189 Other Reasons

295 Randomized

57 Allocated to FLU Group 60 Allocated to CCBT Group 59 Allocated to CCBT/FLU Group 59 Allocated to CCBT/PBO Group 60 Allocated to PBO Group

5 Dropped Out Pretreatment 3 Dropped Out Pretreatment 3 Dropped Out Pretreatment 1 Dropped Out Pretreatment 4 Dropped Out Pretreatment

39 Completed Treatment 48 Completed Treatment 42 Completed Treatment 46 Completed Treatment 36 Completed Treatment
13 Dropped Out 9 Dropped Out 14 Dropped Out 12 Dropped Out 20 Dropped Out

Reason for Dropping Out: Reason for Dropping Out: Reason for Dropping Out: Reason for Dropping Out: Reason for Dropping Out:
5 Adverse Effects 4 Time Commitment 4 Unclear 5 Unclear 5 Not Improving
3 Unclear 2 Treatment Too Difficult 2 Depressed 2 Lost to Follow-up 4 Unclear
2 Depression 2 Lack of Efficacy 2 Not Improving 2 Treatment Too Difficult 3 Treatment Too Difficult
1 Not Improving 1 Worsening 2 Treatment Too Difficult 2 Depression or Lack of 2 Adverse Effects
1 Treatment Too Difficult 1 Adverse Effects Efficacy 2 Depression
1 Scheduling 1 Pregnant 1 Scheduling 2 Job
1 Moving 1 Tasted Pill
1 Scheduling 1 Unknown

Figure 1. Flowchart of subject progress through the phases of the study. There were 295 subjects in the intention-to-treat analysis and 211 subjects in the
completer analysis. CCBT indicates comprehensive cognitive behavioral therapy; FLU, fluoxetine; PBO, placebo.

For the CGI Severity scale, effect size (95% confi-

dence interval) results for each treatment relative to PBO PBO CCBT CCBT and PBO CCBT and FLU FLU
were as follows: 0.42 (0.04 to 0.80) for FLU; 0.27 (−0.10
to 0.64) for CCBT; 0.30 (−0.07 to 0.67) for CCBT/FLU; 45

and 0.42 (0.04 to 0.79) for CCBT/PBO.

LMMS ANALYSIS
35
BSPS Score, Mean

At week 14, significant differences existed in favor of all

active treatments compared with PBO on the BSPS and
Social Phobia and Anxiety Inventory (P⬍.05) (Table 2).
25
The CGI Severity scale evidenced the same pattern, with
FLU and CCBT/FLU both being superior to PBO (P=.01)
but not reaching significance for CCBT or CCBT/PBO
(P⬍.10).
15
For the BSPS, treatment by time models (ie, slope or 0 4 8 14
rate of change) was significant for weeks 0 to 4 (P=.002) Time, wk
and for weeks 4 to 14 (P=.02) (Figure 2). Pairwise analy-
sis suggested that FLU alone produced more change than Figure 2. Mean Brief Social Phobia Scale (BSPS) score by treatment group
PBO, CCBT/FLU, and CCBT/PBO from weeks 0 to 4 (n = 295). Piecewise linear mixed-effects models analysis. CCBT indicates
comprehensive cognitive behavioral therapy; FLU, fluoxetine; PBO, placebo.
(P⬍.01), while CCBT evidenced more change than CCBT/
FLU and CCBT/PBO (P⬍.05) and a trend toward supe-
riority over PBO (P = .10). From weeks 4 to 14, CCBT/ more change than FLU in the last 10 weeks (P=.03), and
FLU and CCBT/PBO conditions produced more change CCBT/PBO showed a trend in the same direction (P=.07).
than PBO (P⬍.05), and CCBT evidenced a trend in the Similar results were found for the CGI Severity scale. In
same direction (P=.09). Furthermore, CCBT/FLU showed summary, FLU demonstrated the fastest onset of action

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1009

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015
 followed by CCBT. By the final visit, combined treat- response rate (31% and 33%, respectively) and broadly
ments were not different from monotherapies, and all ac- comparable rates of response to group therapy (52% and
tive treatments were superior to PBO. 58%, respectively). Response rates to FLU and phenel-
zine were 51% and 65%, respectively. Given a very broad
BEHAVIORAL RATINGS spectrum of activity for monoamine oxidase inhibitors,
it is conceivable that phenelzine carries greater benefit
Posttreatment SUDS ratings are presented in Table 3. than an SSRI, a hypothesis consistent with a recent meta-
The multiple tests were combined to produce average rat- analysis by Hidalgo et al.5
ings for all 3 scenarios. For each treatment, we provide One SSRI, sertraline hydrochloride, has been studied
the average postinstruction, anticipatory, and aftermath in conjunction with CBT in a Swedish primary care set-
scores for all 3 scenarios. Significant time ⫻ treatment ting.10,20 Sertraline/CBT and sertraline alone, but not CBT/
⫻ site interactions emerged for all 3 analyses (P⬍.05), PBO, exceeded the response from PBO alone. Com-
leading us to break down the treatment comparisons by bined sertraline/CBT tended to produce greater benefits.
site. At the University of Pennsylvania, no comparisons However, on 28-week posttreatment follow-up, initial CBT
were significant. At Duke University Medical Center, sub- alone was associated with further gain, whereas sertra-
jects in the FLU group reported significantly less anxi- line alone or combined with CBT was associated with de-
ety than the PBO group on all behavioral measures. Com- terioration. This study differed from ours in sample char-
prehensive cognitive behavioral therapy/fluoxetine was acteristics, type of CBT, and lack of a CBT-alone group,
superior to PBO in anticipatory anxiety before scenarios which may account for some of the different findings.
but not in baseline anxiety. Fluoxetine was associated with The IE and self-report analyses consistently demon-
significantly lower anticipatory distress than CCBT, and strated an advantage of all active treatments over PBO, other
CCBT/PBO was associated with less distress than CCBT than the CGI Severity scale, on which only FLU and CCBT/
at anticipatory and posttest tasks. FLU were superior to PBO at end point. Effect sizes (which
Further examination of Table 3 indicates a difference were not derived from the LMM analyses) indicate a mod-
in outcome between those who received CCBT in any form erate effect of all treatments, with CCBT/PBO and FLU tend-
and those who did not. While there were no differences ing to be the highest. The fact that CCBT/PBO had a large
in outcome per se, the pattern of SUDS scores from postin- effect size on the BSPS but only a trend toward superiority
struction to anticipatory to aftermath is consistent at week over PBO on the CGI measures suggests that the CGI scale
0 across conditions; at each step, there is an increase in may be somewhat less sensitive to changes in social anxi-
anxiety. However, at week 14, subjects who partici- ety symptoms. Alternatively, the BSPS may detect subtle
pated in CCBT evidenced, on average, no change or even changes in social anxiety that do not reflect the global im-
a decrease in SUDS score from anticipatory to after- pairment as well as the CGI scale.
math, while the PBO and FLU subjects continued to show Measures of social interaction were evaluated in vivo,
an increase in anxiety from anticipatory to aftermath. This in the form of a series of videotaped social interactions
difference in SUDS ratings was significant (P⬍.05). These and speech delivery. These assessments proved gener-
data provide some evidence that exposure during group ally robust in picking up treatment effects relative to PBO
therapy did have an effect on behavioral ratings. at Duke University Medical Center but not at the Uni-
versity of Pennsylvania. This is because at the Univer-
ADVERSE EFFECTS sity of Pennsylvania, but not at Duke University Medi-
cal Center, the patients treated with PBO responded well
Significantly higher rates of treatment emergent events to the behavioral task, even though their symptoms did
were noted on 4 symptoms (Table 4). Insomnia and not change substantially. We do not understand why
headache both occurred more often in the PBO and CCBT/ CCBT alone should have failed to distinguish from PBO,
PBO groups relative to CCBT alone and in the CCBT/ although a numerically superior effect was observed in
FLU and FLU groups relative to CCBT alone. Nausea oc- the posttest evaluation.
curred more frequently in the PBO and CCBT/PBO groups Mixed-effects models have become well accepted for
relative to the group undergoing CCBT alone and in the interpretation of clinical trial data and include all data
FLU group relative to the CCBT and CCBT/PBO groups. without imputing missing values.21 We used a 2-level ap-
Relative to the PBO group, anorgasmia was more com- proach (piecewise linear growth model) and found FLU
monly seen in the CCBT/PBO, CCBT/FLU, and FLU generated a faster response than the other treatments; at
groups, as well as in the FLU vs CCBT/PBO groups. week 4, FLU showed superiority to CCBT/FLU, CCBT/
PBO, and PBO. However, the degree of change at post-
treatment did not differ between FLU and CCBT. Thus,
COMMENT our findings are similar to Heimberg et al,6 who re-
ported an earlier advantage for phenelzine over CBT, with
In adults with GSP, this study demonstrated efficacy for the 2 treatments being comparable by week 12. Such a
FLU and CCBT relative to PBO but no evidence for greater finding suggests that greater advantage would accrue from
benefit of combined treatment over monotherapies. Re- a strategy of initial treatment with an SSRI, followed by
sponse rates indicated a treatment vs placebo difference augmentation with psychosocial treatment after 4 to 8
ranging from 15% to 24%. A comparison of our findings weeks. This indeed was found in a study of posttrau-
with the study of phenelzine and group CBT by Heim- matic stress disorder,22 and we are now planning such a
berg et al6 indicates that the 2 studies have a similar PBO study in social anxiety disorder. Besides taking advan-

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1010

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015
 Table 3. Effects of Treatment on Behavioral Measure (Subjective Units of Distress Scale) at Weeks 0 and 14 by Site:
Raw Posttreatment and Linear Mixed-Effects Models (LMM) Analyses*

FLU Group CCBT Group CCBT/FLU Group CCBT/PBO Group PBO Group Test P Value
No. of subjects
University of Pennsylvania
Week 0 27 29 28 27 29
Week 14 20 22 20 21 17
Duke University Medical Center
Week 0 27 30 30 29 27
Week 14 21 24 25 25 21
University of Pennsylvania
Postinstruction score
Week 0 31.0 (18.7) 29.5 (21.4) 28.7 (19.4) 27.4 (20.4) 35.6 (22.5) F4,123 = 0.62 .65
Week 14 24.3 (18.1) 33.4 (17.9) 30.5 (17.6) 26.1 (14.0) 25.0 (21.0) F4,81 = 0.64 .63
Anticipatory score
Week 0 45.1 (16.8) 44.8 (18.3) 47.9 (23.7) 39.2 (19.8) 47.8 (16.2) F4,123 = 0.87 .48
Week 14 30.3 (16.4) 37.0 (15.0) 34.9 (19.2) 34.2 (17.7) 35.4 (21.3) F4,81 = 0.92 .45
Aftermath score
Week 0 49.1 (21.0) 50.4 (20.1) 54.6 (24.2) 44.2 (19.8) 52.5 (20.6) F4,123 = 0.91 .46
Week 14 36.8 (18.2) 33.3 (17.8) 35.6 (18.9) 36.3 (20.6) 44.0 (22.1) F4,81 = 0.95 .44
Duke University Medical Center
Postinstruction score
Week 0† 33.9 (21.4) 29.4 (20.5) 42.8 (24.1) 28.2 (14.5) 29.6 (15.4) F4,135 = 2.68 .03
Week 14‡ 21.5 (12.6) 32.3 (20.2) 33.7 (17.8) 28.5 (18.1) 36.6 (22.7) F4,106 = 2.05 .09
Anticipatory score
Week 0 43.1 (18.9) 43.2 (19.2) 54.4 (23.7) 46.6 (18.8) 45.6 (16.6) F4,135 = 1.60 .18
Week 14§ 25.0 (16.9) 36.4 (20.7) 30.8 (13.1) 31.8 (14.2) 44.6 (21.1) F4,106 = 3.60 .009
Aftermath score
Week 0 48.1 (22.7) 47.4 (21.5) 60.1 (22.8) 51.7 (15.9) 52.7 (15.2) F4,135 = 1.84 .13
Week 14㛳 31.1 (21.5) 38.1 (22.1) 28.0 (13.7) 31.3 (16.7) 47.9 (20.2) F4,106 = 3.52 .01

Abbreviations: CCBT, comprehensive cognitive behavioral therapy; FLU, fluoxetine; PBO, placebo.
*Values are expressed as mean (SD) unless otherwise indicated. Statistics and P values were calculated by LMM analyses. Given the treatment time by site
interaction, data are presented separately for each site. Final Subjective Units of Distress Scale scores given (0 = relaxed and calm; 100 = extremely fearful).
Pairwise differences were significant at P⬍.05, 2-tailed.
†At week 0, CCBT/FLU treatment was superior to PBO, CCBT, and CCBT/PBO treatments (P⬍.05).
‡At week 14, FLU treatment was superior to PBO, CCBT/FLU, and CCBT treatments (P⬍.05).
§At week 14, CCBT/FLU, FLU, and CCBT/PBO treatments were superior to PBO; FLU treatment was superior to CCBT treatment (P⬍.05).
㛳At week 14, CCBT/FLU, FLU, CCBT/PBO treatments were superior to PBO treatment (P⬍.05).

Table 4. Significant Differences in Adverse Effects by Treatment Group*

FLU Group CCBT Group CCBT/FLU Group CCBT/PBO Group PBO Group P
Parameter (n = 48) (n = 44) (n = 53) (n = 55) (n = 52) ␹ 42 Value
Insomnia†‡§㛳 47.9 (33.3-62.8) 13.6 (5.0-27.3) 45.3 (31.6-59.6) 41.8 (28.7-55.9) 42.3 (28.7-56.8) 14.82 .005
Headaches†‡§㛳 31.2 (18.7-46.2) 6.8 (1.4-18.7) 34.3 (21.5-48.3) 27.3 (16.1-41.0) 38.5 (25.3-53.0) 13.76 .008
Nausea†‡§㛳¶ 18.8 (9.0-32.6) 0.0 (0.0-8.0) 17.0 (8.1-29.8) 9.1 (3.0-22.0) 15.4 (6.9-28.1) 10.08 ⬍.04
Anorgasmia‡§㛳#**††‡‡ 32.4 (22.2-50.5) 4.6 (0.86-15.5) 28.3 (16.8-42.4) 7.3 (2.0-17.6) 9.6 (3.2-21.0) 26.79 ⬍.001
Erectile dysfunction‡㛳** 10.4 (3.5-22.7) 0.0 (1.0-8.0) 5.7 (1.2-15.7) 14.6 (6.5-26.7) 1.9 (0.04-10.3) 11.55 ⬍.02

Abbreviations: CCBT, comprehensive cognitive behavioral therapy; FLU, fluoxetine; PBO, placebo.
*Values are expressed as rate percentage (95% confidence interval). Adverse effect identified by an increase of at least 2 points (on a 0-3 scale) relative to
baseline at any assessment point and counted only once per patient.
†Significant pairwise control PBO vs CCBT; PBO was worse.
‡Significant pairwise control CCBT/PBO vs CCBT; CCBT/PBO was worse.
§Significant pairwise control CCBT/FLU vs CCBT; CCBT/FLU was worse.
㛳Significant pairwise control FLU vs CCBT; FLU was worse.
¶Significant pairwise control CCBT/FLU vs FLU; CCBT/FLU was worse.
#Significant pairwise control CCBT/FLU vs PBO; CCBT/FLU was worse.
**Significant pairwise control FLU vs PBO; FLU was worse.
††Significant pairwise control CCBT/FLU vs CCBT/PBO; CCBT/FLU was worse.
‡‡Significant pairwise control FLU vs CCBT/PBO; FLU was worse.

tage of the therapeutic impact of an SSRI, when psycho- and adverse effects or pseudo adverse effects that may
social treatment is added to established SSRI therapy, its appear early in the course of treatment. Our findings about
effects would not be obfuscated by issues of pill taking adverse effects are interesting in this regard, because they

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1011

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015
 indicate that pill taking itself (whether drug or PBO) is acquisition of funding, recruitment into the study, and
associated with a high rate of headaches and insomnia, subsequent analysis of data. Such trials can be success-
which occurred in 27% to 42% of subjects. Thus, a per- fully accomplished, and this one has demonstrated that
son who is in therapy and taking a pill may be initially widely used treatments (ie, CBT and FLU) are effective
distracted by somatic concerns such as impaired sleep for individuals with generalized social phobia, many of
and headaches. The same was true to a lesser extent for whom are significantly impaired. Although FLU and
erectile dysfunction and nausea. Anorgasmia occurred CCBT were more successful than PBO, substantial symp-
more frequently in patients receiving FLU than PBO; we toms still remained after 14 weeks’ treatment. We there-
do not know if this in any way compromised the integ- fore wonder whether longer-term pharmacological treat-
rity of the double-blind design. Although the IEs were ment is necessary and if changes in the delivery of CCBT
instructed not to discuss adverse effects and to remind would improve results. Furthermore, recent evidence sug-
subjects only to address symptoms of social phobia, ad- gests that individual CBT produces better results than
verse effects were discussed in the medication manage- group CBT in GSP.32 Finally, combining FLU with CCBT
ment sessions. did not provide any greater therapeutic benefit. One pos-
Similar to the findings of Heimberg et al,6 we did not sible contributing reason may be the distracting physi-
detect a site ⫻ treatment effect, indicating that medica- cal complaints that arise early with medication and may
tion and psychosocial treatments, as well as PBO, can be be a reason for sequential treatment rather than simul-
given in a consistent manner at sites that differ in their taneous initiation.
therapeutic allegiance. It is possible that the between-
site differences on the behavioral test were related to the Submitted for Publication: December 29, 2003; final re-
fact that less attention was devoted to intersite standard- vision received April 12, 2004; accepted April 20, 2004.
ization of test delivery, relative to medication and CBT. Correspondence: Jonathan Davidson, MD, Box 3812,
Some limitations of our study need to be acknowl- Duke University Medical Center, Durham, NC 27710
edged. First, group treatment restricted flexibility of sched- (jonathan.davidson@duke.edu).
ule for potential subjects, some of whom may have been Funding/Support: This study was supported by grant R10-
deterred through fear of the group itself. Second, some MH49339-05A1 from the National Institute of Mental
subjects declined to take part in treatment after being ran- Health, Bethesda, Md (Drs Davidson and Foa).
domized, since they did not want to receive medication Additional Information: Medication and matching pla-
alone. This issue has been addressed in a separate report cebo were provided by Eli Lilly, Indianapolis, Ind, who
by our group23 and is partly taken care of in the ITT analy- have reviewed the manuscript. They were uninvolved in
sis. It could be questioned whether FLU was the most study design, data analysis, or manuscript preparation.
appropriate drug, particularly since there are 2 reports Acknowledgment: We thank Alan Bellack, PhD, and
demonstrating no difference between FLU and PBO.24,25 Richard Heimberg, PhD, for significant assistance in study
Given that SSRIs as a class have very broad-spectrum anx- design and/or training. Michael J. Kozak, PhD, and Don-
iolytic properties but are not invariably positive in all trials, ald J. Bux, PhD, served as coordinators at the University
we are as yet unpersuaded that this argument has great of Pennsylvania, Philadelphia, Pa. Nader Amir, PhD; No-
merit. However, we do not know if other SSRIs might rah Feeny, PhD; Lisa Jaycox, PhD; Simon Rego, PhD;
have been more effective (eg, paroxetine hydrochloride David Tolin, PhD; and Elna Yadin, PhD, served as thera-
or sertraline).10,26-30 No published head-to-head compari- pists at the University of Pennsylvania/Eastern Pennsyl-
sons of SSRIs exist to our knowledge. As in most social vania Psychiatric Institute. Pat Segee, PhD, and Rita Davi-
phobia/social anxiety disorder trials to date, we ex- son, BA, served as study coordinators at Duke University
cluded subjects with a diagnosis of major depression. Data Medical Center. Nicholas L. S. Potts, MD; Suzanne M.
from a subsection of the subject screens administered at Sutherland, MD; Stewart D. Barnett, MD; John Travers,
the University of Pennsylvania suggest that depression MD; Cherri Miner, MD; Brian A. Stabler, PhD; Ayesha
was the most frequent reason for study exclusions, com- Chaudhary, MD; Susmita Kashikar-Zuck, PhD; and Mar-
prising approximately one third of the patients ex- lene Sandstrom, PhD, served as therapists at Duke Uni-
cluded. It is essential that the next generation of studies versity Medical Center. L. Erik Churchill, MA, provided
determine effective treatment strategies for subjects with data management support.
comorbid social phobia and depression, especially given
their impairing effects.23 We did not assess the success REFERENCES
of IE binding, as recommended by the CONSORT guide-
lines.31 As such, this could be seen as a limitation. On 1. Kessler RC, Stein MB, Berglund P. Social phobia subtypes in the National Co-
the other hand, our experience from a similar previous morbidity Survey. Am J Psychiatry. 1998;155:613-619.
trial suggests that “guessing” increases attention and/or 2. Davidson JRT, Hughes DL, George LK, Blazer DG. The epidemiology of social
phobia: findings from the Duke Epidemiological Catchment Area Study. Psychol
sensitivity to the matter of what treatment is being ad- Med. 1993;23:709-718.
ministered. As to the increased anorgasmia in subjects 3. Brown EJ, Heimberg RG, Juster H. Social phobia subtype and avoidant person-
receiving FLU, the IE was instructed not to discuss or ality disorder: effect on severity of social phobia impairment and outcome of cog-
encourage reports of adverse events, which formed part nitive-behavioral treatment. Behav Ther. 1995;26:467-481.
of the pharmacotherapy assessment. 4. Katzelnick DJ, Kobak KA, DeLeire T, Henk HJ, Greist JH, Davidson JRT, Schneier
FR, Stein MB, Helstead CP. Impact of generalized social anxiety disorder in man-
To conduct and complete a trial of this magnitude is aged care. Am J Psychiatry. 2001;158:1999-2007.
a considerable undertaking, representing approxi- 5. Hidalgo RB, Barnett SD, Davidson JRT. Social anxiety disorder in review: two
mately 10 years of work from inception of the idea, to decades of progress. Int J Neuropsychopharmacol. 2001;4:279-298.

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1012

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015
 6. Heimberg RG, Liebowitz MR, Hope DA, Schneier FR, Holt CS, Welkowitz CA, Juster 20. Haug TT, Hellstrom K, Blomhoff S, Humble M, Madsbutt P, Wold JE. The treat-
HR, Campeas R, Bruch MA, Cloitre M, Fallon B, Klein DF. Cognitive behavioral ment of social phobia in general practice: is exposure therapy feasible? Fam Pract.
group therapy vs phenelzine therapy for social phobia: 12-week outcome. Arch 2000;17:114-118.
Gen Psychiatry. 1998;55:1133-1141. 21. Gueorguieva R, Krystal JH. Move over ANOVA: progress in analyzing repeated-
7. Turner SM, Beidel D, Jacobs RG. Social phobia: a comparison of behavior therapy measures data and its reflection in articles published in the Archives of General
and atenolol. J Consult Clin Psychol. 1994;62:350-358. Psychiatry. Arch Gen Psychiatry. 2004;61:310-317.
8. Clark DB, Agras WS. The assessment and treatment of performance anxiety in 22. Connor KM, Rothbaum BO, Foa EB, Davidson JRT, Cahill SP, Cary CM. A con-
musicians. Am J Psychiatry. 1991;148:598-605. trolled trial of combined sertraline and prolonged exposure therapy in posttrau-
9. Franklin ME, Jaycox LH, Foa EB. Social skills training. In: Hersen M, Bellack A, matic stress disorder. Eur Neuropsychopharmacol. 2002;12(suppl 3):S35.
eds. Handbook of Comparative Treatments for Adult Disorders. 2nd ed. New York, 23. Huppert JD, Franklin ME, Foa EB, Davidson JRT. Study refusal and exclusion from
NY: John Wiley and Sons; 1999:317-339. a randomized treatment study of generalized social phobia. J Anxiety Disord.
10. Blomhoff S, Haug TT, Hellstrom K, Holme I, Humble M, Madsbutt P, Wold JE. 2003;17:683-693.
Randomized controlled general practice trial of sertraline, exposure therapy and
24. Kobak KA, Greist JH, Jefferson JW, Katzelnick DJ. Fluoxetine in social phobia: a
combined treatment in generalized social phobia. Br J Psychiatry. 2001;179:
double-blind, placebo-controlled pilot study. J Clin Psychopharmacol. 2002;
23-30.
22:257-262.
11. Spitzer RL, First MB, Gibbon M, Williams JBW. Structured Clinical Interview for
25. Clark DM, Ehlers A, McManus F, Hackmann A, Fennell M, Campbell H, Flower T,
Axis I DSM-IVR Disorders: Patient Edition. New York: Biometrics Research De-
Davenport C, Louis B. Cognitive therapy vs fluoxetine in generalized social pho-
partment, New York State Psychiatric Institute; 1996.
bia: a randomized placebo-controlled trial. J Consult Clin Psychol. 2003;71:
12. Guy W. ECDEU Assessment Manual for Psychopharmacology. Washington, DC:
1058-1067.
Department of Health, Education, and Welfare; 1976:217-222. Publication (ADM)
26. Stein MB, Liebowitz MR, Lydiard RB, Pitts CD, Bushnell W, Gergel I. Paroxetine
76-338.
treatment of generalized social phobia (social anxiety disorder): a randomized
13. Heimberg RG, Dodge CS, Hope DA, Kennedy CR, Zollo L, Becker RE. Cognitive-
controlled trial. JAMA. 1998;280:708-713.
behavioral group treatment of social phobia: comparison to a credible placebo
control. Cognit Ther Res. 1990;14:1-23. 27. Baldwin D, Bobes J, Stein DJ, Scharwachter I, Faure M. Paroxetine in social phobia/
14. Davidson JRT, Miner CM, de Veaugh Geiss J, Tupler LA, Colket JT, Potts NLS. social anxiety disorder: randomized, double-blind, placebo-controlled study. Br
The Brief Social Phobia Scale: a psychometric evaluation. Psychol Med. 1997; J Psychiatry. 1999;175:120-126.
27:161-166. 28. Allgulander C. Paroxetine in social anxiety disorder: a randomized placebo-
15. Turner SM, Beidel DC, Dancu CV, Stanley MA. An empirically derived inventory controlled study. Acta Psychiatr Scand. 1999;100:193-198.
to measure social fears and anxiety: the Social Phobia and Anxiety Inventory. 29. Katzelnick DJ, Kobak KA, Greist JH, Jefferson JW, Mantle JM, Serlin RC. Ser-
Psychol Assess. 1989;1:35-40. traline for social phobia: a double-blind, placebo-controlled crossover study. Am
16. Beidel DC, Turner SM, Cooley MR. Assessing reliable and clinically significant J Psychiatry. 1995;152:1368-1371.
change in social phobia: validity of the Social Phobia and Anxiety Inventory. Be- 30. van Ameringen MA, Lane RM, Walker JR, Chokka PR, Goldner EM, Johnston
hav Res Ther. 1993;31:331-337. DG, Lavalle YJ, Nandy S, Pecknold JC, Hadrava V, Swinson RP. Sertraline treat-
17. Barnett SD, Tharwani HM, Hertzberg MA, Sutherland SM, Connor KM, Davidson ment of generalized social phobia: a 20-week, double-blind, placebo-controlled
JRT. Tolerability of fluoxetine in posttraumatic stress disorder. Prog Neuropsy- study. Am J Psychiatry. 2001;158:275-281.
chopharmacol Biol Psychiatry. 2002;26:363-367. 31. Moher D, Schulz KF, Attman D; CONSORT group. The CONSORT statement: re-
18. Gibbons RD, Hedeker D, Elkin I, Waternaux C, Kraemer HC, Greenhouse JB, Shea vised recommendations for improving the quality of reports of parallel-group ran-
MT, Imber SD, Sotsky SM, Watkins JT. Some conceptual and statistical issues in domized trials. JAMA. 2001;285:1987-1991.
analysis of longitudinal psychiatric data. Arch Gen Psychiatry. 1993;50:739-750. 32. Stangier U, Heidenreich T, Peitz M, Lauterbach W, Clark DM. Cognitive therapy
19. Hedges L, Olkin I. Statistical Methods for Meta Analysis. New York, NY: Aca- for social phobia: individual versus group treatment. Behav Res Ther. 2003;
demic Press; 1985:88. 41:991-1007.

(REPRINTED) ARCH GEN PSYCHIATRY/ VOL 61, OCT 2004 WWW.ARCHGENPSYCHIATRY.COM

1013

©2004 American Medical Association. All rights reserved.

Downloaded From: http://archpsyc.jamanetwork.com/ by a New York University User on 05/16/2015