Psychodynamic Therapy and Cognitive-Behavioral Therapy in Social Anxiety Disorder: A Multicenter Randomized Controlled Trial

Patients were recruited from April 11, 2007, to April 29, 2009, by outpatient clinics at the universities of Bochum, Dresden, Goettingen, Jena, and Mainz. Participants were recruited through advertisements or other information presented in the mass media or were referred by psychotherapists or physicians in private practice. At each of the five centers, one clinic performed CBT and another performed psychodynamic therapy. An investigator allegiance effect was controlled for by including experts in both CBT and psychodynamic therapy as local investigators at each center. The study protocol was approved by the Ethics Committee of the Medical Faculty of the University of Goettingen. The study was monitored by the Coordination Center for Clinical Trials (KKS Heidelberg), which is independent of the participating research centers. In addition, an independent data monitoring and safety committee was established.

To be included in the study, patients had to be 18–70 years of age, have a diagnosis of social anxiety disorder according to the German-language edition of the Structured Clinical Interview for DSM-IV (SCID) (15), have a score >30 on the Liebowitz Social Anxiety Scale (16), and have a primary diagnosis of social anxiety disorder according to the rating on the Anxiety Disorders Interview Schedule (17). In order to obtain a clinically representative sample, we allowed recruitment of patients with all comorbid mental disorders less severe than social anxiety disorder (according to the Anxiety Disorders Interview Schedule rating) except those listed among the exclusion criteria, which were as follows: psychotic and acute substance-related disorders; cluster A and B personality disorders; prominent risk of self-harm; organic mental disorders; severe medical conditions; and concurrent psychotherapeutic or psychopharmacological treatments. Providing informed consent was required for inclusion.

Randomization was carried out according to randomization lists that were computer-generated by and kept at KKS Heidelberg, which served as a central randomization unit disclosing the allocation after each patient was registered in the study. Lists were stratified according to center, age (<60 and ≥60), and Liebowitz Social Anxiety Scale total score at screening (<60 and ≥60). The original sequence of fixed-length blocks was locally exchanged to impede inferences on the next allocation. In order to avoid a lag at the end of the study, allocation to waiting list was terminated after the waiting list arm had achieved the necessary sample block lengths and distribution of arms within blocks were adapted to varying site-specific availability of therapists. Patients were allocated to psychodynamic therapy, CBT, or waiting list in a ratio of 3:3:1.

The method of CBT applied in this study was based on Clark and Wells’s model (18). This approach includes several components, such as establishing a personal version of the model, role-play-based behavioral experiments, practicing external focus of attention, and restructuring distorted self-image by video feedback or behavioral experiments (18). These procedures are assumed to reverse the maintaining factors identified in the Clark and Wells model of social anxiety disorder, especially safety behaviors and self-focused attention (18). Safety behaviors are a form of avoidance behavior by which an individual tries to reduce anxiety in a social situation (e.g., using make-up to reduce the fear of blushing). As an example of exploring safety behavior, the therapist may ask (19), “When you thought [specify the feared outcome] might/was happening, did you do anything to try to prevent it from happening? Did you do anything to try to prevent people from noticing?” We used the German version of the Clark and Wells manual published by Stangier, Clark, and Ehlers (20).

In order to make the results of psychodynamic therapy comparable with those of CBT, a manual-guided form of psychodynamic therapy was specifically developed for this trial. It was based on Luborsky’s model of psychodynamic therapy (21) and specifically adapted to treat social anxiety disorder (22). This model encompasses both supportive and expressive interventions that are assumed to lead to therapeutic change (21, 22). Establishing a secure helping alliance is one of the model’s most important supportive treatment elements. Expressive interventions relate the symptoms of social anxiety disorder to the patient’s underlying core conflictual relationship theme in order to reduce the symptoms of social anxiety disorder (21, 22). A core conflictual relationship theme comprises three components: a wish (e.g., “I wish to be affirmed by others”), an anticipated response from others (e.g., “Others will humiliate me”), and a response from the self (e.g., “I am afraid of exposing myself”) (21, 22). The response from the self represents the symptoms of social anxiety disorder. A therapist could link these components by an expressive intervention in the following way (22): “As we have seen, you are not only afraid of exposing yourself (response from the self), but you sometimes wish to be at the center of attention and to be affirmed by others (wish). However, you are afraid that they will humiliate you (response from others).” The intervention also includes a supportive component as it refers to the common work between patient and therapist (“As we have seen…”) (21). The core conflictual relationship theme is worked through in present and past relationships as well as in the relationship to the therapist. The treatment procedures are described in detail in a manual (22).

In both CBT and psychodynamic therapy, up to 25 individual 50-minute treatment sessions were applied. (In addition, up to five preparatory sessions were conducted, which are compulsory in the German health care system and cover diagnostic and administrative issues.) In CBT, up to six sessions lasted for 100 minutes, counted as twelve 50-minute sessions. Thus, the dose of treatment (sessions minutes) was consistent with Clark et al. (23), who applied up to fourteen 90-minute sessions plus up to three booster sessions. In CBT, sessions were conducted weekly; in psychodynamic therapy, sessions were also weekly, except during the middle part of the treatment (sessions 7 to 16), which allowed for two sessions a week in order to intensify the treatment. By these procedures, an identical weekly dose and length of treatment was ensured for CBT and psychodynamic therapy. The calendar minimum duration of both treatments was 6 months. After the five preparatory sessions, however, the start of the treatment was regularly delayed by administrative procedures inherent in the German health care system (up to 6 weeks). Further delays resulted from vacations or illness of patients or therapists. Thus, the actual calendar treatment duration (time between the first preparatory session and the end of treatment) was longer than 6 months.

All therapists held degrees as clinical psychologists or physicians and had completed their psychotherapeutic training or were in advanced psychotherapeutic training. They regularly conducted either CBT or psychodynamic therapy. Fifty-five cognitive-behavioral therapists conducted CBT (37 of them women) and 53 psychodynamic therapists conducted psychodynamic therapy (30 of them women). No therapist conducted both CBT and psychodynamic therapy.

The mean age of the cognitive-behavioral therapists was 31.1 years (SD=4.98), and that of the psychodynamic therapists was 39.4 years (SD=8.06). The average general clinical experience was 1.7 years (SD=0.9) for the cognitive-behavioral therapists and 8.0 years (SD=9.0) for the psychodynamic therapists (p=0.0002). On the other hand, significantly more cognitive-behavioral therapists reported having already used both manual-guided psychotherapy in general (67% compared with 24%; χ²=14.40, p<0.0001) and specifically the manual-guided approach applied in this study (35% compared with 0%, p<0.0001) before the trial.

Before inclusion in the trial, therapists were specifically trained in their respective treatment approach by authors of the treatment manuals (U.S., M.B., F.L.). To be included in the trial, therapists were required to treat two pilot cases in accordance with the respective manual. In order to maintain treatment fidelity during the trial, therapists received regular site-level and cross-site supervision, which was comparable in structure and amount between treatments. At each center, supervision of therapists was performed as a 90-minute group supervision conducted by supervisors specifically trained by the authors of the treatment manuals. Supervision was performed every 2 weeks during the first 6 months of the trial, and monthly thereafter. All treatment sessions were videotaped. Videotapes were used for both supervision and the assessment of treatment integrity. For assessing treatment integrity, established instruments were used: the Penn Adherence and Competence Scale for Supportive-Expressive Therapy (24) adapted for social phobia; and the Cognitive Therapy Competence Scale for Social Phobia (25). To ensure treatment fidelity during the trial, two tapes from every second treatment carried out by a therapist were randomly selected and rated for adherence and competence—two tapes each of the early, middle, and late phases of therapy. As a minimum target for competence, a rating of ≥3 on a 7-point Likert scale (0=poor, 6=excellent) was used for both psychodynamic therapy and CBT (25). In the case of therapists falling below this level, feedback was given to both the therapist and the supervisor. For cases of therapists who continued to fall below that level, the study procedures allowed for the treatment to be excluded from the trial, but this proved unnecessary, and no specific treatments were excluded. The procedures and results of treatment integrity assessments are described in more detail below.

Assessments were conducted at baseline, at weeks 8 and 15 of treatment, and at end of treatment. Analogous to the procedure used by Davidson et al. (26), the assessments at weeks 8 and 15 were carried out to compensate for possible missing data. In these additional assessments, only the Liebowitz Social Anxiety Scale was administered. Diagnoses were made using the SCID (15). The primary (i.e., most severe) mental disorder was assessed using the Anxiety Disorders Interview Schedule severity rating scale (17). Twenty-three specifically trained and independent assessors (clinical psychologists) blind to treatment condition conducted the interviews. All interviews were videotaped in order to assess interrater reliability. Reliability for the Liebowitz Social Anxiety Scale was assessed by comparing the individual results of 23 diagnosticians with an expert’s rating of three videotaped interviews. High interrater reliability was observed for the Liebowitz Social Anxiety Scale score (Kendall’s W=0.98, df=2, p<0.001). Blinding was ensured by instructing the patients not to discuss their treatment during assessment. Furthermore, personnel involved in handling study documentation and scheduling interviews were required to keep all treatment-specific documents separate from other patient information. Waiting list status was kept open-label.

Rates of remission and response were used as the primary outcome measures. Following recommendations by Liebowitz et al., remission was defined as a score ≤30 on the Liebowitz Social Anxiety Scale (16, 27). We assessed response rates so that results could be compared with those from other studies of social anxiety disorder. Response was defined as a reduction of at least 31% in Liebowitz Social Anxiety Scale score, which is demonstrably comparable to a Clinical Global Impression improvement subscore ≤2, the measure usually used to define response (28). As secondary outcome measures, well-established self-report instruments were used, such as the Social Phobia and Anxiety Inventory, the Beck Depression Inventory, and the Inventory of Interpersonal Problems (29–31).

Adverse events were defined as any significant unfavorable change in the patient’s pretreatment mental condition, regardless of its relationship to treatment, in particular the occurrence of any additional mental disorder. Serious adverse events included mortality, hospitalization, suicide, or attempted suicide. Adverse events and serious adverse events were reported to an independent data and safety monitoring board.

Based on previous studies of the Clark and Wells approach (23, 32, 33), we hypothesized that patients in the CBT and waiting list groups would have response rates of 70% and 5%, respectively. We expected the new psychodynamic therapy method to achieve response a rate slightly lower than that of CBT, at 55%. For differences in proportions, h is the appropriate measure of effect size (34). A difference of 15% (70% for CBT and 55% for psychodynamic therapy) corresponds to an effect size (h) of 0.30 (34). We defined a difference 0.30 in h values as meaningful; h=0.30 can be considered a small effect size (34). In order to detect a difference in h of 0.30 with a power of 0.80, we determined that 174 patients were required in each active treatment group when using two-tailed tests at an alpha level of 0.05 (8). Following the study protocol, we planned to compensate for dropouts by including additional patients. Previous studies suggest a dropout rate of 15% (6), so we planned to include 31 additional patients per treatment arm (174+31=205).

The data were analyzed using SAS 9.2 (SAS Institute, Cary, N.C.). Categorical data were analyzed using the chi-square test, Fisher’s exact test, or mixed-effects logistic regression models, as appropriate, and dimensional measures were analyzed by mixed-effects linear regression models. We also performed sensitivity analyses. All regression-based analyses included study center and baseline scores as covariates. In addition, therapists were included as a random factor in the mixed-effects model (35). Model-based pairwise comparisons were planned a priori. To analyze remission and response, alpha was set at 0.05. For the comparison of CBT and psychodynamic therapy with regard to the secondary outcome measures, alpha was adjusted for multiple comparisons (alpha=0.01; 0.05/4). For categorical data, between-group differences were assessed using Cohen’s h and odds ratios, and for dimensional data, Cohen’s d was used, with d based on mixed-model-analysis-adjusted means (34). For the intent-to-treat analysis, we applied multiple imputation by chained equations to account for the uncertainty resulting from missing outcome data (36, 37). Multiple imputation is superior to the last-observation-carried-forward procedure for several reasons. First, multiple imputation adequately incorporates the uncertainty arising from missing information and introduces additional variance leading to larger standard errors. Furthermore, under the assumption of missing at random, multiple imputation yields unbiased results that are more accurate than those provided by the last-observation-carried-forward procedure (38). To generate conservative estimates, 50 imputations were created and all available variables were included in the imputation process. We also report data from an analysis of participants who completed treatment, for which no imputation method was applied.

A total of 1,450 potential patients were screened. Of these, 495 patients met the inclusion criteria and did not meet any exclusion criteria (a diagram of participant flow through the study is available in the data supplement that accompanies the online edition of this article). Initially, 209, 207, and 79 patients were randomly assigned to CBT, psychodynamic therapy, and waiting list, respectively. Eight percent of the patients assigned to CBT and 5% of those assigned to psychodynamic therapy stopped treatment but were included in the assessment (treatment withdrawals). In the CBT, psychodynamic therapy, and waiting list groups, 24%, 28%, and 27%, respectively, of patients stopped treatment or assessment (treatment or study withdrawals). Fisher’s exact test revealed no significant differences between groups for study or treatment withdrawals. The outcome data of these dropouts were included in intent-to-treat analyses using multiple imputation. Thus, data analysis was based on all patients who were assigned to one of the three study arms.

Table 1 summarizes age and sex of the participants, along with the distribution of treatment assignments by study site. The sample included a relatively high rate of patients suffering from comorbid mental disorders; 63% of the sample had at least one comorbid disorder, 25% at least two, 7% at least three, and 1% at least four. The most frequent comorbid disorders were avoidant personality disorder (25%), major depressive disorder (21%), dysthymia (11%), specific phobia (7%), agoraphobia (5%), and generalized anxiety disorder (2%). Personality disorders other than avoidant personality disorder were seen in 2% of the patients.

The mean number of sessions completed was 25.8 (SD=9.13) for CBT and 25.7 (SD=9.61) for psychodynamic therapy. The mean duration of treatment was 38.7 weeks (SD=16.03) for CBT and 37.4 weeks (SD=18.03) for psychodynamic therapy.

To examine treatment integrity, videotapes of 50 randomly selected treatments (25 each from the CBT and psychodynamic therapy groups) were rated by trained evaluators (clinical psychologists) who were blind to treatment condition. One session was randomly selected from each treatment. Three evaluators trained in CBT rated the Cognitive Therapy Competence Scale for Social Phobia, and another three evaluators trained in psychodynamic therapy rated the Penn Adherence and Competence Scale for Supportive-Expressive Therapy. These scales allow the assessment of both adherence to the manual and competence in delivery of interventions. Evaluators received 10 hours of training in rating the scales, by experts in the respective approach (U.S., M.B., J.W.). Each evaluator rated 34 sessions, half of them psychodynamic therapy sessions and half CBT sessions. Each session was rated four times, twice by psychodynamic therapy raters and twice by CBT raters. For statistical analyses, the ratings of the respective pair of raters were averaged. Each pair of raters included one experienced rater who had rated between 183 and 194 training sessions. Mean interrater reliability values (intraclass correlation coefficient) for adherence and competence were 0.59 and 0.58, respectively, for the Penn scale and 0.98 and 0.94, respectively, for the Cognitive Therapy Competence scale. For the Penn scale, interrater reliability was comparable to that reported by Barber and Crits-Christoph (24). Consistent with the treatment models, cognitive therapists used significantly more interventions of the CBT approach (Cognitive Therapy Competence scale, adherence subscale mean score, 2.00 compared with 0.12; t=13.42, p<0.001) and applied them more competently (Cognitive Therapy Competence scale, competence subscale mean score, 3.19 compared with 1.26; t=10.68, p<0.001). In contrast, psychodynamic therapists used significantly more interventions of the psychodynamic therapy model (Penn scale, adherence subscale mean score, 3.02 compared with 2.23; t=5.48, p<0.001) and applied them more competently (Penn scale, competence subscale mean score, 3.27 compared with 2.20; t=6.23, p<0.001). Thus, the treatments differed significantly in accordance with the treatment model.

Tables 2 and 3 present the key outcome data for the CBT, psychodynamic therapy, and waiting list groups. Remission rates for the three groups were 36%, 26%, and 9%, respectively, and the corresponding response rates were 60%, 52%, and 15%. Pairwise comparisons from a logistic regression model including study center, baseline Liebowitz Social Anxiety Scale score, and therapist as covariates showed that both CBT and psychodynamic therapy were superior to waiting list with regard to remission and response rates (CBT: remission, h=0.71, p<0.0004; odds ratio=7.42, 95% CI=2.45–22.51; response, h=0.98, p<0.0001; odds ratio=8.63, 95% CI=3.53–21.08; psychodynamic therapy: remission, h=0.50, p=0.011; odds ratio=4.27, 95% CI=1.39–13.14; response, h=0.82, p<0.0001; odds ratio=6.28, 95% CI=2.54–15.54). For the comparison of CBT and psychodynamic therapy, logistic regression models revealed significant differences in favor of CBT for remission rates (h=0.22, p=0.034; odds ratio=1.75, 95% CI=1.04–2.92) but not for response rates (h=0.16, p=0.198; odds ratio=1.36, 95% CI=0.85–2.18). The differences of 10% and 8% in remission and response rates correspond to between-group effect sizes (h) of 0.22 and 0.16, which are below the value of 0.30 defined a priori as clinically meaningful. For the completer analysis, we included those patients who completed treatment per protocol, that is, those who either received 25 sessions or who remitted before completion. The completer analysis yielded the same pattern of results for CBT and psychodynamic therapy, with remission rates of 42% and 30%, respectively, and response rates of 66% and 56%, respectively. Thus, no differences in the pattern of results were observed between the intent-to-treat analysis using multiple imputation and the completer analysis.

For continuous measures, planned pairwise comparisons for posttherapy data via linear-effects models with study center, baseline score, and therapist as covariates revealed significant differences in favor of CBT for the scores on Liebowitz Social Anxiety Scale (p=0.01, d=0.25, 95% CI=0.06–0.44), the Social Phobia and Anxiety Inventory (p=0.0009, d=0.33, 95% CI=0.13–0.52), and the Inventory of Interpersonal Problems (p=0.003, d=0.29, 95% CI=0.10–0.49). No significant differences were observed for the Beck Depression Inventory. Again, all between-group effect sizes were small (d values ≤0.33) (34). The proportion of variance explained by treatment condition was between 1% and 3% (34). The proportions of variance (intraclass correlation coefficients) explained by therapist for the Liebowitz Social Anxiety Scale, the Social Phobia and Anxiety Inventory, the Beck Depression Inventory, and the Inventory of Interpersonal Problems were 0.05, 0.07, 0.05, and 0.07, respectively. There was neither a center-by-treatment interaction nor a significant center effect in any of the analyses. Within-group effect sizes for the Liebowitz Social Anxiety Scale score were large for both CBT and psychodynamic therapy (d values, 1.32 compared with 1.02) and small for the waiting list (d=0.24) (34). CBT was associated with a reduction in Liebowitz Social Anxiety Scale score by 29.12 points (95% CI=25.50–32.73), psychodynamic therapy by 22.55 points (95% CI=18.96–26.13), and waiting list by 5.19 points (95% CI=–0.14 to 10.52).

In a sensitivity analysis using logistic regression with baseline Liebowitz Social Anxiety Scale score as a covariate, we found that general clinical experience (in years) was not associated with either remission or response. This was also true for having used treatment manuals in general, and it applied to all other outcome measures as well. Having already used the specific manuals applied in this study had a significant impact on outcome on ratings on the Social Phobia and Anxiety Inventory (p=0.003) and the Inventory of Interpersonal Problems (p=0.013), but not on other measures. Controlling for this variable, however, showed that results remained stable, with CBT superior to psychodynamic therapy (p values, 0.005 and 0.006).

Serious adverse events occurred in seven patients (1%) during the trial, including psychiatric hospitalization for the following conditions: major depression (one each in the CBT and psychodynamic therapy groups), acute stress disorder (one in the psychodynamic therapy group), alcohol abuse (one in the CBT group), severe increase in severity of social anxiety disorder symptoms (one in the CBT group), deliberate self-poisoning (one in the psychodynamic therapy group), and diverticular disease (one in the CBT group).

Adverse events occurred in seven patients (1%) during the trial, including major depression (three in the CBT group, two in the psychodynamic therapy group, and one in the waiting list group) and adjustment disorder (one in the psychodynamic therapy group). Given the small number of participants with adverse events and serious adverse events, no significance tests were performed.