Intensive group cognitive treatment and individual cognitive therapy vs. treatment as usual in social phobia: a randomized controlled trial


Ewa Mörtberg, The Unit for Psychotherapy in City (Psykoterapienheten City), Karlavägen 53, SE-114 49 Stockholm, Sweden.


Objective:  To compare the effects of an intensive group cognitive treatment (IGCT) to individual cognitive therapy (ICT) and treatment as usual (TAU) in social phobia (DSM-IV).

Method:  Hundred patients were randomized to: IGCT involving 16 group sessions spread over three weeks; ICT involving 16 shorter weekly sessions in 4 months and; TAU involving an indicated selective serotonin reuptake inhibitor (SSRI) with therapy sessions as required for 1 year. The main outcome measure was a Social Phobia Composite that combined several standardized self-report measures. Diagnostic assessment was repeated at 1-year follow-up.

Results:  Significant improvements were observed with all treatments. ICT was superior to IGCT and TAU, which did not differ in overall effectiveness.

Conclusion:  The study confirms and extends previously reported findings that ICT is more effective than group cognitive treatment and treatment with SSRIs. IGCT lasts only 3 weeks, and is as effective as more protracted TAU.

Significant outcomes

  • Individual cognitive therapy is more effective than medication-focused treatment as usual or intensive group cognitive treatment.
  • A 3-week intensive group cognitive treatment programme without medication is as effective as a 12-month medication-focused treatment as usual programme but requires more therapist contact.


  • There was a high dropout rate in treatment as usual.
  • Most assessments were patient self-report. Independent psychiatric ratings were only completed at 1 year.


Social phobia (1) is a common and disabling anxiety disorder (2, 3) associated with considerable social and occupational handicap that is unlikely to remit without treatment (4).

Within psychological approaches, the efficacy of behavioural and cognitive–behavioural therapies (CBT) is relatively well established. Meta-analytic reviews comparing these treatments conclude that exposure therapy alone and the combination of exposure with cognitive restructuring are more effective than waiting list control conditions (5–8). Within pharmacological approaches, the best validated treatments are: benzodiazepines, monoamine oxidase inhibitors and selective serotonin reuptake inhibitors (SSRI) (9) of which the well-tolerated and safer drugs are the SSRIs, which are recommended as first-line treatments (10, 11). Among the SSRIs, controlled trials have established the superiority over placebo medication for fluvoxamine, fluoxetine, sertraline and paroxetine (9, 12).

Meta-analyses comparing pharmacological treatments and CBT have shown contrasting results, one (6) found no evidence for differences between the approaches (both were superior to control conditions) while the other (7), found pharmacological treatments to be more effective in the short term. In the longer term, some studies have reported higher relapse rates for pharmacological treatments (13, 14).

Although effective psychological and pharmacological treatments are available, there is a need for further improvement, as a substantial portion of patients do not fully recover with existing treatments (15–17), and because treatments tend to be protracted. For example, CBT usually takes around 4 months and adequate drug treatment takes 6–12 months in routine practice.

In an attempt to speed up therapy, Mörtberg et al. (18, 19) tested a intensive Cognitive Behavioural Group Treatment. The intensive format provided more concentrated and prolonged exposure that it was thought might be better at overcoming phobic avoidance than weekly treatment. The intensive treatment lasts 3 weeks and demonstrated effectiveness in a pilot study (18) and in a randomized controlled trial in which it was compared with no treatment (waiting list control) (19). The improvements observed with intensive Cognitive Behavioural Group Treatment were similar to those usually reported with commonly used, and more protracted, group CBT programmes.

The present study is a further evaluation of the intensive group treatment. The content of the treatment has been modified to take into account some of the latest developments in psychological treatment. In particular, Clark et al. (20) have recently reported that a new individual cognitive therapy (ICT) programme is associated with particularly large effect sizes. ICT contains a range of therapeutic procedures, which focus on key maintaining processes in social phobia. Recently, ICT has been found to be superior to treatment with an SSRI (fluoxetine) (20); to a weekly group-version of the cognitive therapy (21) and to exposure therapy combined with training in applied relaxation (22). Many of the key procedures used in ICT have been incorporated into the intensive group treatment programme which has, a consequence, been renamed Intensive Group Cognitive Therapy (IGCT).

To test the clinical utility of IGCT, it is compared with ICT and with ‘treatment as usual’ (TAU). TAU comprises medication plus standard psychiatric care. In contrast to the protocol used in most drug trials (designed per protocol including placebo control, predetermined dose regimens and fixed numbers of visits) the TAU condition reflected treatment in a regular clinical practice. There have been no previous studies comparing TAU with cognitive treatments for social phobia.

We predicted that all three treatments would bring about substantial changes and that IGCT would be as effective as ICT. As ICT had been shown to be superior to a single SSRI (fluoxetine), it might be expected that ICT and IGCT would be superior to TAU. However, the greater flexibility of drug choice in TAU could mean that it is more effective than the single medication that was studied in the previous trial (20). As a consequence, we did not have a clear prediction about how TAU would compare to ICT and IGCT.

Aims of the study

In summary, the aim of the study was to examine the effects of a brief intensive groups cognitive therapy programme compared with ICT and TAU, in patients suffering from social phobia.

Material and methods


Patients were randomly assigned to ICT, IGCT or TAU. In ICT, patients had up to 16 sessions in 4 months followed by a booster session at 8 and 12 months. In IGCT, patients had up to 16 prolonged treatment sessions during a 3-week period, followed by a booster session at 4, 8 and 12 months. In TAU, patients were given an indicated medication on which they were maintained for 12 months. Each psychiatrist decided the number of visits. Assessments were at pre-treatment, 3 weeks (IGCT only), 4, 8 and 12 months. The Ethics Committee at the Karolinska Institute approved the study.


Figure 1 shows the flow of patients through the study. Patients were recruited through advertisements in a local newspaper and were offered inclusion in the trial if they fulfilled the following criteria: i) had a primary diagnosis of social phobia according to DSM-IV (1); ii) were 18–65 years, and were without iii) current depressive episode, bipolar disorder, stress disorders, addiction or psychoses; iv) present psychotropic medication; or v) current psychotherapy. Fifteen patients did not meet the inclusion criteria and an additional two patients declined further participation after the psychiatric interview. The remaining patients (n =100) fulfilled criteria i–v, gave written consent and were randomly assigned to ICT, IGCT or TAU.

Figure 1.

 Flowchart of patients’ progress through phases of the trial. ICT, individual cognitive therapy; IGCT, intensive group cognitive therapy; TAU, treatment as usual.


Patients were screened on telephone for possible social phobia and were mailed self-report questionnaires and written information of the trial. Subjects who expressed interest in participating in the trial and returned the screening questionnaires were invited to a diagnostic interview using the Structured Clinical Interview for DSM-IV (SCID) (23) for establishing the diagnosis of social phobia and assessing additional psychiatric and personality disorders. Three senior SCID-trained psychiatrists, who would not be trial therapists, conducted the interviews. Inter-rater reliability for the social phobia diagnosis was satisfactory (kappa = 0.76).

Randomization was based on an extension of Efron's biased coin design procedure for assigning subjects to more than two conditions and when subjects are arriving sequentially into a trial (24). An independent administrator who was not informed about the patient's characteristics made the allocation procedure using a random number table.


Individual cognitive therapy was based on the model of Clark and Wells (25) of the maintenance of social phobia and was guided by a therapist manual in English (D.M. Clark, unpublished data) and a related book for therapists in Swedish (26). The main steps in the treatment were: i) Deriving an individualized version of the model using patient's own thoughts, images, anxiety symptoms, safety-behaviours and strategies of attention. ii) Safety behaviours experiment. Key safety behaviours were identified and their adverse effects demonstrated with an experiential exercise in which patients’ role played a difficult social situation while using their safety behaviours and then while attempting to drop safety behaviours. iii) Video feedback to modify distorted self-imagery. Patients viewed the video of themselves in the safety behavioural experiment under conditions that were intended to make the discrepancy between their negative distorted self-image and their objective social performance particularly evident. iv) Shifting focus of attention away from oneself and onto to the social situation. Patients had repeated role-plays in which they were instructed to shift between focusing attention on themselves and monitoring their performance and focusing externally while reducing self-monitoring and getting lost in the interaction. The aim of the role-plays was to demonstrate the adverse effects of self-focus and to enhance patients’ ability to become externally focused. External focus of attention was encouraged during subsequent social interactions. v) Behavioural experiments in which patients specified in advance their feared outcome for various social situations and tested whether the outcomes occurred during planned exposure to the situation in within-session role-plays and in vivo assignments in sessions and as homework. In order to maximize belief disconfirmation, patients were encouraged to drop safety behaviours and focus externally during behaviour experiments. vi) Problematic anticipatory and post-event negative processing was identified and modified. vii) Dysfunctional assumptions were identified and modified by behavioural experiments and cognitive restructuring techniques. The first two treatment sessions were approximately 90 min. Subsequent sessions were approximately 60 min.

Intensive group cognitive treatment was a development of the intensive group CBT used in our previous trial (19) and was manual based. The treatment procedures in ICT (D.M. Clark, unpublished data) were adapted for group administration and incorporated into a programme that also included psycho-education and training in applied relaxation. Delivery of the ICT interventions was less individualized in IGCT than would be normal in ICT. Each treatment group contained six to seven patients and was lead by two therapists. Patients usually attended a 3-h morning session and a 2-h afternoon session and had lunch with other members of the group. There were nine sessions in the first week (23 h over 4.5 days) and seven sessions (18 h over 3.5 days) in the third week, each followed by homework. During the second week patients completed homework assignments in their normal work and home environment but did not attend the clinic.

The morning sessions consisted of psychoeducation, exercises for deriving an idiosyncratic version of the cognitive model, identifying key safety behaviours and alternative behaviours, exercises for shifting self-focused attention and cognitive restructuring. Exercises were individualized within the group so that each patient could discuss, e.g. her/his version of the cognitive model, including identification of safety behaviours with one of the therapists. An additional treatment component was applied relaxation training (27), which was delivered in a shortened form (7 h) by a physiotherapist.

The afternoon sessions comprised video-recorded exposure with graded tasks of verbal presentations in front of the group. Safety behaviour experiments were conducted and patients were instructed to explore alternative behaviours and to shift from self-focused to externally focused attention. All exercises were followed by video feedback. Patients’ expected self-image (negative) and actual self-image were compared and the discrepancies between felt and observable anxiety were discussed. There were homework assignments in all three weeks to facilitate for processing each step in treatment.

‘Treatment as usual’ followed routine psychiatric practice. All patients were prescribed medication. For 89% this consisted of an antidepressant with indicated efficacy for social phobia in either randomized controlled trials (fluoxetine, paroxetine, sertraline, moclobemide) (9–12) or open trials (citalopram) (28). The remaining 11% received a benzodiazepine (oxazepam). SSRIs were most common (83%), and were maintained for 12 months, in line with current clinical recommendations (29). Daily stabilized doses (mean, range) were: fluoxetine (35 mg, 20–60), paroxetine (25 mg, 20–40), sertraline (67 mg, 50–100), moclobemide (600 mg), citalopram (50 mg, 20–60) and oxazepam (15 mg). Prior to starting medication, patients were told that medication rectifies the neurochemical imbalances that maintain social phobia. They were told that as the medication dose build up it should help them to become more confident in social situations. The psychiatrist and the patient discussed and decided numbers of visits, telephone consultations between visits, and choice of medication and other steps in treatment. Sessions usually lasted 45 min and included routine psychiatric procedures such as support, a review of progress, side-effects and any other adverse effects. TAU did not involve exposure instructions or cognitive interventions.

Therapists and supervision

Seven therapists (five clinical psychologists, one nurse and one psychiatrist) with 5–25 years of practice and experience of treatment of anxiety disorders delivered ICT and IGCT. Five delivered both treatments. The remaining two, whose outcomes did not differ from the others, only delivered ICT. In Sweden there are two levels of cognitive therapy and CBT training. Five of the therapists had the highest (level 2) cognitive therapy qualification and two had the lower (level 1) CBT training. Four of the therapists had attended 1–2 days of workshops on ICT by the second author (D.M.C.). During the trial, therapists had supervision once a month to check adherence to the protocol and for planning of future sessions. The supervisor (Anna Kåver) was a highly experienced cognitive therapist who had also supervised ICT in an earlier Stockholm trial. All sessions were video-recorded and the tapes were viewed during supervision to check treatment integrity. D.M.C. also reviewed a random selection of session videotapes as a further check. No examples of deviation from the treatment protocol as described in the study manuals were observed. However, the quality of treatment delivery was not rated. Five senior psychiatrists with 10–30 years of clinical practice with anxiety disorders and extensive experience in the use of SSRIs delivered TAU. The project leader had regular meetings with the psychiatrists in order to monitor that part of the trial.

Assessment and outcome measures

All groups were assessed at pre-treatment, 4 (end of ICT), 8 and 12 months. In addition, the IGCT condition had an extra assessment at 3 weeks (the end of that treatment).

Screening instruments  The questionnaires used in the telephone interview were: the Screener for the SCID Axis-I Disorders (SCID Screener) (23); the Liebowitz Social Anxiety Scale (LSAS) (30, 31); and the Beck Depression Inventory (BDI) (32, 33).

Social phobia  The main measure of social phobia symptoms was the Social Phobia Composite which was created by averaging scores from five standardized self-report social phobia scales: the LSAS (30, 31); the Social Phobia Scale (SPS) and the Social Interaction Anxiety Scale (SIAS) (34); the Fear Questionnaire Social Phobia sub-scale (FQ-SOC) (35); the Fear of Negative Evaluation scale (FNE) (36). In addition, the Social Phobia Weekly Summary Scale (SPWSS) (20) was included. The 5-item SPWSS has good internal consistency (Cronbach's alpha = 0.81) and consists of 0–8 ratings of social anxiety, social avoidance, self-focused vs. external attention, anticipatory processing and post-event rumination. The social phobia composite was generated by the procedure recommended by Rosenthal and Rosnow (37). Patients’ scores on each of the six social phobia measures were standardized (mean = 0, SD = 1) across pre- and post-treatment assessment by converting to Z-scores. The composite at each assessment occasion was the mean of the Z-scores on that occasion.

A process composite assessed several of the key process variables in the model of Clark and Wells (25). Standardized scores from three questionnaires were combined to create the composite. The questionnaires were: the Social Cognitions Questionnaire (SCQ) (A. Wells, L. Stopa and D.M. Clark, unpublished data) which assesses the frequency and believability of social phobia-related negative automatic thoughts; the Social Behaviours Questionnaire (SBQ) (D.M. Clark, G. Butler, M. Fennell, A. Hackmann, F. McManus and A. Wells, unpublished data), which assesses safety behaviours; and the Social Attitudes Questionnaire (SAQ) (D.M. Clark et al., unpublished data) which assesses social phobia-related negative beliefs.

Social phobia-related disability was assessed by the Sheehan Disability Scale (SDS) (38) which comprises 0–10 patient ratings of impairment in work, social life and family life.

Depression  The BDI (32, 33) was used to assess depressive mood.

Statistical analyses

The main analyses were intention-to-treat (ITT) with the last available observation carried forward (LOCF) if necessary. To identify any differences between conditions on the symptom and process measures before treatment, initial scores for the three conditions were compared with the liberal procedure of separate one-way analyses of variance (anovas) for each measure. To identify any differences between conditions after the start of treatment, condition by time (4, 8 and 12 months), repeated measures analyses of covariance (ancovas) with initial scores as covariates were used. To deal with multiple measurement of social phobia-related symptoms, a two-stage analytic procedure was agreed in advance. Analysis of possible differences in outcome on individual measures could only be conducted if a significant effect was first observed in analysis of the social phobia composite. The same procedure was adopted for the cognitive model measures. To check the validity of using LOCF in the ITT analysis, the data from all randomized patients was also analysed using the Linear Mixed Model (LMM) approach (39, 40), in which all available data are utilized. The LMM analysis produced essentially the same pattern of results to that reported here.


Characteristics of patients

Table 1 shows the demographic and clinical characteristics of patients in each treatment condition. There were no significant differences between the conditions on any characteristic.

Table 1.   Demographic and clinical characteristics
Demographic and clinical characteristicsICT (n = 32)IGCT (n = 35)TAU (n = 33)All patients (n = 100)Group effect F (2, 97), χ2 (2)
  1. *Higher education = University/University College.

  2. †Previous psychiatric treatment = patients who previously had received a pharmacological (from general practitioners and psychiatrists) or a psychological treatment.

  3. ‡The Axis-I life-time comorbid conditions were: depressive disorder (27%), obsessive–compulsive disorder (8%), panic disorder (7%), anxiety disorder not otherwise specified (NOS) (6%), eating disorder (5%), alcohol or substance abuse (4%), dysthymia (3%), specific phobia (3%) and generalized anxiety disorder (GAD) (3 %).

  4. §Specification of social phobia subtypes was based on a procedure described by Baker et al. (31), categorizing patients ratings of the Liebowitz Social Anxiety Scale into four domains of social anxiety; formal speaking/interaction, informal speaking/interaction, assertive interaction and observation by others. Patients were considered to have generalized social phobia if they rated one or more social situations from each as at least inducing moderate levels of fear (rating of 2 or higher on a 0–3 point scale). All other patients were classified as non-generalized (31).

  5. ¶The SCID-II Screener (DSM-IV) was used for diagnoses of personality disorders (PD), with the cut-off level adjusted upwards so that one additional criterion was required for each of the personality disorders. The procedure that has shown a high overall kappa agreement (0.78) between the SCID interviews and the SCID-II Screener (45).

Age, mean (SD)36.1 (9.8)32.1 (7.8)35.8 (9.1)34.6 (9.1)2.3
Women, n (%)22 (69)21 (60)20 (61)63 (63)0.67
Married/cohabiting, n (%)13 (40.6)17 (48.6)19 (57.6)49 (49)1.9
Higher education*, n (%)16 (50)12 (34.3)15 (45.5)43 (43)2.9
Occupation, n (%)
 Employed19 (59.3)25 (71.4)22 (66.7)66 (66)1.1
 Unemployed3 (9.4)2 (5.7)3 (9.1)8 (8)0.38
 Students6 (18.8)5 (14.3)8 (24.2)19 (19)1.1
 Sick list4 (12.5)3 (8.6)0 (0)7 (7)4.1
Age at onset, mean (SD)14.3 (6.9)15.2 (6.5)16.6 (8.3)15.4 (7.3)0.82
Duration of social phobia years, mean (SD)21.8 (10.3)16.9 (8.9)19.6 (13.3)19.5 (11)1.7
Previous psychiatric treatment, n (%)†16 (50)19 (54.3.)15 (45.5)50 (50)0.77
Axis I comorbidity, n (%)‡13 (40.6)16 (45.7)16 (48.5)45 (45)0.42
Generalized social phobia, n (%)§25 (78)21 (60)17 (52)63 (63)5.1
Avoidant PD, n (%)¶25 (78)21 (60)19 (58)65 (65)0.86
Any PD, n (%)28 (87.5)28 (80)27 (81.8)83 (83)0.72

Dropouts and number of sessions attended

Eighteen patients dropped out before starting their assigned treatment and 10 patients dropped out during treatment. There was a significant difference [χ2(2) = 8.9 P = 0.012] in dropout rates between treatments: TAU (45%); IGCT (26%); ICT (14%). The reasons for dropping out are presented in Fig. 1. There were no significant differences between completers (n = 72) and dropouts (n = 28) on the baseline social phobia measures, the BDI, duration and onset of social phobia, proportion of patients with APD, social phobia subtype (generalized vs. non-generalized) or any of the other characteristics described in Table 1.

For completers (n = 72), the mean numbers of sessions attended in each treatment were: ICT, 14.9 (SD = 1.9) treatment sessions and 1.8 (SD = 0.49) booster sessions; IGCT, 15.3 (SD = 1.4) treatment sessions and 1.9 (SD = 1.4) booster sessions; TAU, 5.2 (SD = 1.8) treatment sessions and two (SD = 1.6) telephone consultations for checking adaptation to the medication.

Effects of treatment on social phobia symptoms

Table 2 shows the social phobia composite and the individual measures at each time point. At pre-treatment, anovas indicated that the groups did not differ on the social phobia composite or any individual measure except the LSAS, where patients in ICT had significantly higher scores than patients in IGCT. Within treatment repeated measures anovas on the social phobia composite revealed a significant time effect for each treatment [ICT: Wilk's Lambda = 0.24, F(3, 29) = 30.1, P < 0.001; IGCT: Wilk's Lambda = 0.46, F(3, 32) = 12.4, P < 0.001; TAU: Wilk's Lambda = 0.70, F(3, 30) = 4.4, P < 0.05]. Paired comparisons revealed that for each treatment, patients’ scores at 4, 8, and 12 months were significantly lower than at pre-treatment (P < 0.05). Each treatment was therefore associated with significant improvement.

Table 2.   Effects of treatment on questionnaire measures at each assessment occasion (intention to treat analysis)
AssessmentICT (n = 32)IGCT (n = 35)TAU (n = 33)Group effectGroup comparisons
Mean (SD)Mean (SD)Mean (SD)
  1. At pre-treatment, group effect was based on one-way analysis of variance. On all other assessment points the group effect is based on group by time (4, 8 and 12 months) ancovas with pre-treatment scores as covariate. Last observation carried forward was used in the case of missing observations. To check the validity of this procedure, Linear Mixed Models (LMM) analyses of covariance that did not carry forward last observations but instead used all available data from each patient were also conducted. The LMM analyses produced an identical pattern of results for the composites and very similar results for the individual measures that make up the composite.

  2. ***P < 0.001, **P < 0.01, *P < 0.05 and P < 0.1.

inline imageinline image

To determine whether the treatments were associated with differential improvement, the 4-, 8- and 12-month data were subjected to repeated measures ancovas with pre-treatment scores as covariates. For the social phobia symptom composite there was a significant effect of treatment condition, F(2, 96) = 11.7, P < 0.001, a significant effect of time, Wilk's Lambda = 0.91, F(2, 95) =4.8, P < 0.01, and no interaction between time and condition, Wilk's Lambda = 0.97, F(4, 190) = 0.8, P = 0.54. The time effect indicates that patients continued to improve between the 4- and 12-month assessments. The condition effect indicates the presence of a difference between the treatments throughout the period from 4 to 12 months. Paired comparisons using Duncan's multiple range test, indicated that ICT was superior to both TAU (P < 0.001) and IGCT (P < 0.001). Analysis of the individual social phobia symptom measures showed that ICT differed from TAU on all six measures and differed from IGCT on four measures (SIAS, SPS, FNE and SPWSS) and showed non-significant trends on LSAS (P = 0.07). Although IGCT and TAU did not differ on the social phobia symptom composite, IGCT was superior to TAU on the LSAS (P < 0.05) and the FQ-SOC (P < 0.05).

As IGCT was concentrated in a 3-week period, a separate analysis was conducted to determine whether the improvements that were observed with this treatment were already evident after 3 weeks or whether more time was required for the improvements to emerge. Social phobia symptom composite scores for IGCT were compared at pre-treatment, 3 weeks and 4 months. There was a significant overall effect of time [Wilks’ Lambda = 0.55, F(2, 33) = 13.7, P < 0.001]. Pair-wise comparisons show that scores on the social phobia composite significantly reduced between pre-treatment and 3 weeks (P < 0.001) and between pre-treatment and 4 months but there was no significant difference between the 3-week and 4-month assessments (see Fig. 2). It therefore appears that IGCT achieved almost all of its effects after only 3 weeks.

Figure 2.

 Mean social phobia composite scores for intensive group cognitive therapy at pre-treatment, 3 weeks and 4 months. Intention to treat data (n = 35).

Figure 3 shows the social phobia symptom composite scores for completers only. As one might expect, symptom changes are somewhat larger than in the ITT analysis. However, ancova revealed an identical pattern of results. There was a significant effect of treatment condition, F(2, 53) = 9.8, P < 0.001, a significant effect of time, Wilk's Lambda = 0.80, F(2, 52) = 6.4, P < 0.01, and no interaction between time and condition. Paired comparisons indicated that ICT was superior to IGCT and TAU, which did not differ from each other.

Figure 3.

 Mean social phobia composite scores at each assessment for completers only. ICT, individual cognitive therapy; IGCT, intensive group cognitive therapy; TAU, treatment as usual.

Effects of treatment on disability and depressed mood

At pre-treatment, the groups did not differ on the SDS or the BDI. For both measures, ancovas performed on the 4-, 8- and 12-month data revealed a significant effect of treatment condition [for SDS, F(2, 96) = 4.8, P < 0.01; for BDI, F(2, 96) = 10.9, P < 0.001], no effect of time and no condition by time interaction. Paired comparisons indicated that for both measures ICT was superior to IGCT and TAU, which did not differ from each other.

Effects of treatment on process measures

Table 2 also shows the cognitive model process composite and the individual process measures. At pre-treatment, anovas indicated that the groups did not differ on the process composite or any individual process measure. In line with the social phobia symptom composite findings, repeated measures ancova of the process composite revealed a significant effect of treatment condition, F(2, 96) = 13.7, P < 0.001, a significant effect of time, Wilk's lambda = 0.93, F(2, 95) = 3.8, P < 0.05, and no interaction between time and condition (F < 0.1). The time effect indicates that patients continued to improve on the process composite between the 4 and 12 months. Pair-wise treatment comparisons indicated that ICT was superior to both TAU (P < 0.001) and IGCT (P < 0.001), which did not differ from each other. Analysis of the four individual process measures (SCQ frequency and belief scales, SBQ and SAQ) showed similar pattern of results with ICT associated with significantly greater change on all the measures than either TAU or IGCT.

The model of Clark and Wells (25) of the maintenance of social phobia would predict that patients who changed most on the model process variables would also change most in social phobia symptoms. To test this prediction, correlations were computed between pre-treatment to 4 months change in the model process composite and the social phobia symptom composite among patients who completed treatment. Process change was significantly related to symptom change in all three treatments (for ICT, r27 = 0.64, P < 0.001; for IGCT, r25 = 0.74, P < 0.001; for TAU, r17 =0.87, P < 0.001).

Effect sizes, clinically significant change and loss of diagnosis

To gain a clear impression of the magnitude of change associated with each treatment, pre-treatment to 4 (post-ICT), 8 and 12 months effect sizes were calculated using the formula: ES = (pre-treatment mean minus 4, 8, or 12 months mean)/pre-treatment standard deviation. Table 3 shows the ITT data. Cohen (41) proposed a three-fold classification of effect sizes: small (0.20–0.49), medium (0.50–0.79) and large (0.80 and above). According to this classification, TAU was associated with a small to medium effect size (0.44–0.58), IGCT was associated with a medium effect sizes at 3 weeks post-treatment and at 4 months (0.70–0.79) and large effect sizes at 12 months (1.00). ICT was associated with effect sizes that were approximately twice those of IGCT.

Table 3.   Effect sizes for the Social Phobia Composite at 4, 8 and 12 months
Assessment occasionICTIGCT*TAU
  1. ICT, Individual Cognitive Therapy; IGCT, Intensive Group Cognitive Therapy; TAU, Treatment as Usual.

  2. Effect size = (pre-treatment scores minus mean scores at 4, 8 and 12 months)/pre-treatment standard deviation.

  3. *Effect size for IGCT at 3 weeks was 0.70 for all randomized patients and 0.84 for completers.

  4. †For patients who dropped out, last available assessment was carried forward for comparing effect sizes. This procedure probably gives a conservative estimate of effect sizes. Linear Mixed Model analysis using all available data for each patient tended to generate larger effect sizes.

 All randomized†
4 months1.620.790.44
8 months1.780.960.45
12 months1.890.990.58
 Completers only
4 months1.810.960.89
8 months1.951.180.92
12 months2.111.231.17

Clinically significant change was calculated using Jacobsen and Truax's (1991) stringent criterion that a patient's post-treatment symptom score should indicate a reliable change and be at least two standard deviations below the pre-treatment mean. Using this criterion, clinically significant improvement rates on the social phobia composite were 56% in ICT, 26% in IGCT and 24% in TAU. ICT was associated with a significantly higher clinically significant improvement rate than IGCT (χ2 = 6.48, d.f. = 1, P < 0.025) and TAU (χ2 = 10.13, d.f. = 1, P < 0.01), which did not differ from each other. Psychiatric assessors who were blind to treatment condition repeated the SCID social phobia module at 12 months, when 24 patients in ICT (75%), 23 patients in IGCT (66%) and 16 patients in TAU (48 %) were judged to have lost the diagnosis of social phobia. On this measure, ICT was superior to TAU (χ2 = 4.83, d.f. = 1, P < 0.05) but did not differ from IGCT.

Post-hoc analyses

Two further post-hoc analyses were conducted to explore the nature of the treatment response in IGCT and TAU. With the exception of two social phobia symptom measures that favoured IGCT, these two treatments did not differ in outcome but TAU tended to have a higher dropout rate. It could be suggested that although TAU was less acceptable overall, individuals who persisted with medication may have a particularly good response. To explore this possibility completers-only effect sizes on the social phobia composite were calculated (see Table 3). Contrary to expectation, the completers-only effect sizes for TAU were not larger than those for IGCT. Clinically, it is often suggested that medication is likely to be particularly helpful in generalized, as opposed to non-generalized, social phobia. To explore this suggestion, separate social phobia composite ancova contrasts between IGCT and TAU were calculated for the different social phobia subtypes. In non-generalized social phobia, the two treatments were similarly effective [condition main effect: F(1, 27) = 0.1, P = 0.73]. Contrary to expectation, in generalized social phobia IGCT showed a tendency to be more effective than TAU [condition main effect: in LOCF analysis F(1, 35) = 3.2, P = 0.08; in LMM analysis F(1, 75) = 5.8, P < 0.05].


Consistent with our first prediction, all three treatments were associated with significant and enduring improvements in social phobia. Contrary to our second prediction, the modified IGCT programme was less effective than ICT. No prediction was made about the relative efficacy of ICT and IGCT in comparison with TAU. In the event, ICT was consistently superior to TAU. IGCT did not differ from TAU on most measures but was superior to TAU on the LSAS and the Fear Questionnaire. In addition, there was a trend for IGCT to be superior to TAU on the overall measure of social phobia (the social phobia composite) when the analysis was restricted to patients with generalized social phobia.

The results for IGCT confirm our previous trial's (19) finding that brief, intensive treatment is effective. IGCT works fast, showing a credible effect size on the social phobia composite (0.70 for the ITT sample and 0.84 for completers) after only 3 weeks. For comparison, meta-analyses of CBT programmes that take 12–16 weeks report effect sizes ranging from 0.51 to 1.08 at post-treatment (5–8). The addition of CT procedures to the intensive treatment in this trial is associated with an improvement in effect size compared with our previous trial (19), which suggests that the procedures may have enhanced the effectiveness of the intensive programme.

Intensive group cognitive treatment was less effective than ICT, even though IGCT involved more within session exposure exercises, which are often seen as particularly important for successful treatment of social phobia. One possible explanation could be that ICT is better suited to maximizing the effectiveness of exposure. According to the model (25) on which cognitive therapy is based, exposure exercises should be designed in a way which maximizes patients’ opportunities to test their predictions about how dangerous a situation is, rather than as exercises that rely on repetition to promote habituation. In order to maximize belief disconfirmation, it is necessary to: i) identify patients’ idiosyncratic predictions; ii) to help patients identify and drop their overt and covert safety behaviours, and iii) to assist patients to reconfigure attention so that they attend to the social interaction rather than to themselves. In IGCT, it is difficult to provide each patient with the individual attention that is needed for these detailed procedures. The same applies for other cognitive therapy procedures such as video feedback and imagery modification, each of which focuses on information that is highly idiosyncratic. If this analysis is correct, the effectiveness of IGCT might be improved by paying more attention to individualization. An alternative explanation for the superiority of ICT could be the length of treatment. The duration of IGCT may be too brief for some patients, who might need therapist support and feedback during an extended of period of time to be able to keep up the motivation to explore and practise new behaviours in a range of situations, some of which occur infrequently. In addition, some patients may find a group format too interpersonally demanding, at least initially.

‘Treatment as usual’ and IGCT were associated with similar improvement overall, although IGCT was superior on two measures and showed a trend towards better outcome in patients with generalized social phobia. TAU tended to have a larger dropout rate, suggesting that it was less well accepted. A more structured treatment involving a larger number of visits may have increased patients’ motivation to remain in therapy. However, the attendance schedule in TAU was modelled on existing, routine clinical practice. A notable feature of TAU was that it achieved broadly similar outcomes to IGCT with less therapist contact (on average 5 h/patient). Nevertheless, 3 weeks of IGCT (41 h for 6–7 patients) was as effective as 12 months of medication. Relapse rates in TAU once medication was withdrawn could not be determined as follow-up was limited to 12 months.

Individual cognitive therapy was consistently superior to IGCT and TAU on measures of social phobia and was associated with substantially larger effect sizes. This result confirms and extends the findings of previous trials in which ICT was superior to an SSRI (20) and to two alternative psychosocial interventions (21, 22). An important issue with psychological treatments is the extent to which a treatment can be successfully disseminated to other groups and settings. The initial evaluations of ICT (20, 22) were conducted by the group in the UK that developed the treatment. The present study was a largely independent investigation in a different country and a different setting (a psychiatric clinic vs. a specialist psychological treatment centre). The therapists in the current trial attended a 2-day workshop on ICT given by one of the authors (D.M.C.), but there was no within-trial supervision from, or contact with, the originating group and the trial therapists expectations were that ICT would not be the most effective treatment. Rather it was anticipated that the locally developed IGCT programme would be at least as effective as ICT. The observed superiority of ICT is therefore encouraging evidence that the treatment can be successfully disseminated and that positive results can be obtained by groups who do not have a strong allegiance to that particular treatment. The procedures that made successful dissemination possible were not studied and will need to be delineated in future research. However, it was our impression that regular supervision from a local expert in cognitive therapy and the availability of therapist manuals in both English and Swedish were particularly helpful.

The superiority of ICT to TAU in the present trial compliments the findings of a previous trial in which ICT was superior to fluoxetine (20). In that trial, a single SSRI was given in a highly structured per protocol treatment. In contrast, the present trial aimed to model routine practice by allowing psychiatrists to choose from a range of indicated medications and to individually tailor treatment. While treatment per protocol studies are essential for establishing drug efficacy, TAU studies may be seen as example of effectiveness in routine practice.

Consistent with recent psychological treatment studies in other psychiatric disorders (43, 44), it is an encouraging finding that both ICT and IGCT showed enduring effects well after the end of treatment. Although positive effects were demonstrated with all three treatments, there were a significant number of patients who remained impaired at the end of therapy. A further study will examine the characteristics of these patients compared with those who benefited from treatment.

A limitation of the study is the fact that outcome was largely assessed by patient self-report, albeit with standardized measures. Psychiatric assessment that was blind to treatment allocation was restricted to a repeat diagnostic interview at 12 months. While the findings of this interview were in line with the self-report measures, it would have been better if psychiatric ratings had been obtained at each assessment point. Inclusion of a behaviour test would have further strengthened the assessment protocol.


This study was supported by grants from the Boethius Foundation, the Söderström-Königska Foundation, the Organon Foundation and the FOU (The Research and Development Centre) of Stockholm County Council and the St. Göran Psychiatric Clinic Foundation.