Introduction

Cognitive behavior therapy (CBT) has been demonstrated to be efficacious in a range of formats (individual, group, or self-help) in treating social anxiety disorder (SAD). A recent meta-analysis indicates that individual CBT had greater effects on outcomes than psychological placebos and other psychological treatments (psychodynamic, interpersonal psychotherapy, mindfulness, and supportive therapy), suggesting that it should be offered as “the treatment choice” for SAD (Mayo-Wilson et al. 2014). However, there are some individuals who prefer and benefit more from the group format because group treatment is likely to offer more opportunities for normalization, positive peer modeling, social support, and exposure to social situations (Manassis et al. 2002). As about 40% of patients who received cognitive behavioral group therapy (CBGT) remained non-responders (Heimberg et al. 1998) and achieved high end-state functioning (Hofmann 2007; Mattick and Peters 1988), treatment innovation is greatly needed in CBGT for SAD.

Cognitive models of SAD have demonstrated that a negative self-image (NSI) is the key maintaining factor of the disorder (Clark and Wells 1995; Rapee and Heimberg 1997). NSIs are spontaneously occurring self-images of individuals with SAD as if seen from an observer’s perspective in social situations (Hackmann et al. 2000). Experimental studies manipulating NSIs in individuals with SAD or high social anxiety have demonstrated that when participants held a NSI in mind, they were more anxious, had more negative beliefs about their own social performances, and used more safety behaviors, and their social performance was worse (Hirsch et al. 2003a, b, 2004; Makkar and Grisham 2011; Vassilopoulos 2005). NSIs can also change self-esteem (Hulme et al. 2012), which may increase the chance of retrieving negative memories and thus undermine effective social performance. These findings converge to suggest that NSIs should be the direct target of treatments for SAD.

One of the widely used treatment technique of CBT to modify NSIs is video feedback. During a treatment session, patients engage in social situations and their performance is videotaped. Before patients and therapists watch the video together, therapists ask patients to predict in detail what they will see on the video and ensure they watch the video as objectively as possible. Video feedback usually helps patients find that they appear better than they think and thereby realize that their NSI is misleading (Clark 2005). Almost all CBT protocols use video feedback to deal with NSIs, but the dose of video feedback differs according to the models of SAD. For example, traditional CBGT does not include video feedback in its original protocol (Heimberg and Becker 2002). The enhanced CBGT includes one session of video feedback in an attempt to address distorted self-beliefs (Rapee et al. 2009). Imagery-enhanced CBGT (McEvoy and Saulsman 2014) also has only one session of video feedback.

By contrast, the CBGT protocol with social mishap exposure (Fang et al. 2013; Hofmann and Otto 2008) has 5 sessions of video feedback in which the patient’s exaggerated estimation of social cost is studiously addressed. Although the CBGT protocol with social mishap exposure showed a large effect size (Hofmann and Scepkowski 2006), it has not been compared with a waiting-list control group or other treatment groups. In the case of individual cognitive therapy (CT; Clark et al. 2006), which emphasizes the use of video feedback to modify NSIs, video feedback was used in the third session. Therapists could use more video feedback in CT if necessary, and this was possible because CT is an individual CBT. Given that video feedback is particularly useful to change NSIs, the CBGT might include more sessions of video feedback that could change NSIs with extra feedback from other group members.

Another powerful technique to address NSIs is behavioral experiments which use planned experiential activities, based on experimentation or observation, to correct judgment bias and test the validity of the NSIs. Based on the model of SAD which emphasized the role of NSIs, judgment bias, and other factors, CT utilized behavioral experiments extensively to modify NSIs and to de-catastrophize the impact of showing visible signs of anxiety. The excellent outcome of CT suggests that behavioral experiments can be an effective way to deal with NSIs and distorted beliefs. Similarly in CBGT, mishap exposure training can be carried out as behavioral experiments in which patients test the validity of overestimation of social cost and at the same time disconfirm their distorted images of how they come across in the mishap situations.

There is a growing interest in imagery rescripting (IR) as a new method to change the NSIs and encapsulated beliefs of patients with SAD. IR is a procedure to help patients relive their negative past memories so that they can re-experience them through imagery and then rescript the adverse early experiences as “desirable results” (Arntz and Weertman 1999; Smucker et al. 1995; Wild et al. 2008). There are additional benefits of using IR as a technique to change NSIs. First, as NSIs are likely to have their roots in early socially aversive events, such as being bullied at school or abused by significant others, IR may provide a useful way to update the early memories linked to negative imagery and thus change the NSIs. Second, imagery may be a more powerful way than verbal techniques to update an aversive memory, process the emotion, and reconstruct the distorted meanings, as imagery is a vehicle to store meanings and emotions (Stopa 2009). Indeed, it has been demonstrated that IR reduced social anxiety and the distress and vividness of early memories as well as current NSIs, changed the core beliefs of self and others, and enhanced the state of self-esteem and positive affect (Çili et al. 2017; Lee and Kwon 2013; Nilsson et al. 2012; Norton and Abbott 2016; Wild and Clark 2011). IR has been used as a therapeutic intervention in the treatment of SAD, either as part of the CBT protocol (McEvoy et al. 2015; McEvoy and Saulsman 2014) or on its own (Frets et al. 2014; Lee and Kwon 2013; Reimer and Moscovitch 2015), but mostly in an individual treatment format. McEvoy and Saulsman (2014) attempted to use IR in a group format in their imagery-enhanced CBGT and their protocol produced an excellent outcome. Although IR has seldom been used in CGBT because of difficulties in implementing it in a group format, it is worth including in CGBT given its strong potential to change NSIs.

Most empirically supported treatment packages for SAD currently comprise a combination of exposure and cognitive restructuring as their principal therapeutic components (Hofmann 2007; Rapee et al. 2009; Rodebaugh et al. 2004). However, it is necessary to verify whether the form of cognitive restructuring should follow the traditional Beckian method. Cognitive changes can occur without specific cognitive interventions (Dobson and Dobson 2016; Newman et al. 1994), and may be even more likely to occur through experiential methods when the behavioral intervention can be modified effectively through specific cognitive interventions. For example, the individual CT protocol of SAD using behavioral experiments that dealt with safety behavior, attention shifting, video feedback, and in-vivo exposure exercises showed impressive effect sizes indicating pre-to post-treatment changes of around two standard deviations (Clark et al. 2003, 2006). In addition, Hofmann and Scepkowski (2006) showed a larger effect size in their CBT protocol, which predominantly used in-session public speaking and in-vivo social mishap exposures, than the traditional CBT protocol. If SAD patients could expose their exact “feared stimuli” repeatedly and find that their assumptions might be wrong, the efficacy of exposure might be greater (Moscovitch 2009). The dose of exposure training is also critical. Patients with SAD are habituated to avoiding socially anxious situations, so sufficient exposure in the CBT program itself could be important to alleviate social anxiety.

The main aim of this study was to compare an enhanced group CBT for SAD (ECBT) with a historical control group that used the conventional group CBT (CCBT) protocol. The ECBT protocol focused on changing NSIs through IR, repeated video feedback, and increased trials of exposure training. By contrast, the CCBT protocol was a modified Heimberg’s CBGT protocol (Heimberg and Becker 2002) to which three sessions of video feedback and one more session of dealing with core belief were added (Yoon and Kwon 2013). Compared with the wait-list control group, the CCBT was demonstrated to be more efficacious in reducing social anxiety (Yoon and Kwon 2013). The CCBT was demonstrated to have large effect sizes ranging from 1.30 to 1.71 in reducing social anxiety (Yoon and Kwon 2013). The first hypothesis was that ECBT would be acceptable to patients with SAD in the community. The second hypothesis was that ECBT would have a larger effect size than CCBT. The third hypothesis was that cognitive changes, such as fear of negative evaluation, self-focused attention, a dysfunctional belief of self, and judgment bias, would be greater in the ECBT group than in the CCBT group. Finally, ECBT would have a higher proportion of reliable and clinically significant change.

Method

Participants

Participants comprised 90 people who had social anxiety as their chief complaint and who wanted to take part in the treatment for SAD in a community social anxiety counseling center. The Structured Clinical Interview for DSM-IV Axis I disorders (SCID-I; First et al. 1997) was administered by master’s-level clinical psychologists to establish Axis-I disorders, and all the results of the SCID-I were supervised by doctoral-level licensed clinical psychologists. Participants who were primarily diagnosed with SAD could be enrolled in this study. Exclusion criteria were (a) schizophrenia or other psychotic disorders, (b) a history of organic brain injuries, (c) current bipolar disorder, (d) currently experiencing a severe major depressive episode requiring priority treatment over SAD, and (e) currently experiencing other anxiety disorders requiring priority treatment over SAD. People who were primarily diagnosed with SAD with secondary depressive or anxiety disorder could be included in the study. If participants had been taking medications related to social anxiety or other psychiatric disorders, those who were maintaining a stable dose of medication at least a month before participating in the treatment could be enrolled in this study. Written informed consent was obtained from all patients.

This study was conducted from March 2013 to May 2017. The CCBT group comprised 45 consecutive people who agreed to participate in this study from March 2013 to May 2015. The ECBT group included 45 consecutive patients who wanted to join in the study from August 2015 to May 2017. In the CCBT group, additional diagnoses included major depressive disorder or dysthymia (33.3%) and other anxiety disorders (20.0%). In the ECBT group, additional diagnoses included major depressive disorder or dysthymia (48.9%) and other anxiety disorders (11.1%). All participants of both groups were born in South Korea (100%).

Measures

Social Interaction Anxiety Scale (SIAS) and Social Phobia Scale (SPS) (Mattick and Clarke 1998)

The SIAS and SPS are widely used 20-item measures of social interaction and performance anxiety, respectively. The SIAS contains items reflecting cognitive, affective, and behavioral reactions to interaction situations, such as nervousness when speaking to authority, mixing with people, and talking to an attractive person of the opposite sex. The SPS describes situations in which the person is the focus of attention and observed by others, such as eating, drinking, and writing. The 5-point response scale for both scales is Not at all, Slightly, Moderately, Very, or Extremely characteristic of me. Internal reliabilities for the SIAS (a = 0.93) and SPS (a = 0.89) are high within clinical samples, and these scales have been shown to be sensitive to change (Cox et al. 1998). Twelve-week test–retest reliabilities are high for both the SIAS (r = .92) and the SPS (r = .93, Mattick and Clarke 1998). The internal consistency of the Korean version of SIAS and SPS in this study were high (Cronbach’s α of SIAS = 0.92, Cronbach’s α of SPS = 0.92).

Brief Fear of Negative Evaluation (BFNE; Leary 1983)

The BFNE assesses fear of negative evaluation from others, which is one of the core cognitive characteristics of social anxiety disorder. The original scale consisted of 30 true/false statements (Watson and Friend 1969), but the short-form version of the FNE or BFNE consists of 12 items rated on a scale from 1 (strongly disagree) to 5 (strongly agree). The possible scores on the BFNE range from 12 to 60 with a higher score indicating greater fear of negative evaluation. The internal consistency coefficient (Cronbach’s α) of the Korean version of BFNE in this study was 0.86.

Self-Focused Attention scale (SFA; Bӧgels et al. 1996 )

The SFA, which consists of 11 items, was developed by Bӧgels et al. (1996) to assess state self-focused attention during social situations. Six items measure a tendency to focus on one’s own arousal during social situations, and 5 items assess focus on one’s own social behavior. The SFA is a 5-point Likert scale ranging from 0 (never) to 4 (very much) with a total score ranging from 0 to 44. The higher the score, the greater the tendency to pay attention to self-awareness and performance rather than focusing on the task itself. The internal consistency of each subscale reported by Bӧgels et al. (1996) was 0.86 for self-focused attention—arousal and 0.77 for self-focused attention—performing. The internal consistency coefficient (Cronbach’s α) of the Korean version of SFA in this study was 0.81.

Dysfunctional Belief Test (DBT; Jang and Ahn 2010)

The DBT was originally developed to assess the dysfunctional belief of socially anxious people (Cho and Won 1999). The short version of the DBT was validated by Jang and Ahn (2010). We used the negative-social-self subscale of the short version of the DBT. A higher score indicates the tendency to perceive oneself as inferior and that the mental representation of one’s appearance, abilities, and behaviors is interpreted as negative. The internal consistency of DBT in this study was high (Cronbach’s α = 0.93).

Social Probability Questionnaire/Social Cost Questionnaire (SPQ/SCQ; McNally and Foa 1987)

This questionnaire was derived from Butler and Mathews (1983) and McNally and Foa (1987). It consists of 40 items, 20 on negative nonsocial events (e.g., “You will catch the flu”; “You will lose your house keys”) and 20 on negative social events. Half of the social items included social performance (e.g., “During a job interview, you will freeze”), and half did not (e.g., “Someone you know won’t say hello to you”). We used only social items. Participants first rated the probability that these hypothetical events would happen to them and then rated the cost associated with each event (i.e., how bad it would be if the event actually happened to them). All ratings used a 10-point Likert scale ranging from 0 (Not at all likely-bad) to 9 (Extremely likely-bad). The internal consistency of the Korean version of SPQ and SCQ in this study were high (Cronbach’s α of SPQ = 0.93, Cronbach’s α of SCQ = 0.93).

Procedure and Treatments

All participants answered questionnaires that assessed social anxiety symptoms (SIAS, SPS) and cognitive symptoms (BFNE, SFA, DBT, SPQ, SCQ) at pre-treatment, post-treatment, and 3-month follow-up. Social anxiety symptoms were assessed prior to each treatment session, and cognitive symptoms were additionally assessed every 4 weeks.

Both ECBT and CBGT were delivered by three CBT-trained doctoral-level licensed clinical psychologists using its treatment manual. All the therapists had received CBT training for at least 5 years and had experience treating SAD for at least 3 years. ECBT and CCBT were administered by one doctoral-level licensed clinical psychologist and one or two masters-level interns. All the treatments were conducted under supervision with a corresponding author who is highly experienced with CBT (over 20 years). Treatment integrity was not assessed formally but encouraged by using a detailed treatment manual with therapist instructions, patient handouts, and worksheets. To ensure treatment adherence, a co-therapist acted as a rater who was present in each ECBT and CCBT session to conduct real-time rating of adherence using an adherence checklist based on each manual and who gave real-time feedback to therapists to maintain the essential treatment component. As CCBT and ECBT were administered at different points of time (CCBT was conducted earlier than ECBT) and the therapists for both CCBT and ECBT were supervised to bring about the best outcome, possible allegiance effects could have been minimized. All groups comprised of between four and seven participants (M = 5.88, SD = 0.64).

ECBT and CCBT comprised 12 weekly 2.5–3-h sessions plus a 3-month follow-up. The CCBT protocol was a modified Heimberg’s CBGT protocol (Heimberg and Becker 2002), to which three sessions of video feedback and one more session of dealing with core belief were added (Yoon and Kwon 2013). ECBT emphasized changing NSIs through IR, the increased number of speech tasks (in-session exposure) and video feedback, and social mishap exposures. All the ECBT treatment components were performed to change participants’ NSIs (see Table 1).

Table 1 Some key differences between ECBT and CCBT
Full size table

There were three main differences between the two groups. First, the emphasis of psychoeducation was different between the two protocols. In the first session, ECBT involved psychoeducation about anxiety, social anxiety, and key mechanisms of SAD following Clark and Wells’ model (1995). The key mechanisms of SAD (NSIs, self-focused attention, judgment bias, and safety behaviors) were explained to group members and each member filled out his/her own mechanism model of social anxiety. On the other hand, the CCBT also involved psychoeducation about anxiety and SAD in the first session, but unlike ECBT, the CCBT emphasized a traditional “Beckian” cognitive model and group members were encouraged to understand the interactions among thoughts, physical symptoms, and behaviors.

Second, methods of cognitive restructuring were different. Traditional cognitive restructuring techniques were used in CCBT from sessions 2–4. By contrast, ECBT relied on IR, repeated video feedback and mishap exposure training for challenging and modifying distorted cognition. In CCBT, participants learned how to identify automatic thoughts in feared social situations and tried to detect cognitive errors. Participants were encouraged to practice these methods with therapists in the session, and they had to do homework identifying automatic thoughts and cognitive errors in social situations. In session 4, the downward arrow technique and guided discovery were used to identify and modify core beliefs. In ECBT, experiential methods such as IR and mishap exposure training were used instead of traditional cognitive restructuring techniques. In session 2 and 8, IR was administered individually in the group with the same early memory, based on the previous studies suggesting that IR may be the most effective when delivered across multiple sessions (Lee and Kwon 2013; Norton and Abbott 2016). We followed the instruction used in Lee and Kwon (2013), but did not include cognitive restructuring because of a time limit and positive findings on IR without cognitive restructuring (Nilsson et al. 2012). It took about 30 min for each member to complete IR and share their experience in the group. Participants were given the assignment of listening the tape-recorded session of their IR and encouraged to rescript their memories on their own if they wanted to. The reason why we gave this assignment was to facilitate the emotional processing of the early memory which could have been insufficiently achieved in the session because of limited time (Foa and Kozak 1986; Lee and Kwon 2013).

Third, exposure trainings were different. The dose of exposure training was much higher in the ECBT group than the CCBT group. While the CCBT group had six sessions of exposure training, the ECBT group had nine exposure sessions. Additionally, ECBT group had two exposure trainings per participant in sessions 5–7 and at least three instances of in-vivo social mishap exposure per person in sessions 9–12. The dose of video feedback was also higher in the ECBT group than the CCBT group. The CCBT group did video feedback in sessions 6–8, and the ECBT group in Sessions 3–7. In addition, participants in the ECBT group had to watch their own exposure video at least three times a day at home and fill in a video-feedback checklist for homework. The target of exposure was different as well. Participants in the CCBT group were simply exposed to feared social situations, such as public speaking or conversations with strangers, with gradually increasing anxiety levels. In ECBT, however, all the exposure training, including social mishap exposure, was conducted by identifying clients’ most feared symptoms (e.g., blushing in front of others or speaking to the audience in a trembling voice) and confirming whether these symptoms were as negative as the participants expected. In this way, participants were able to identify what they really wanted to hide and tested whether displaying their feared symptoms in front of others would result in catastrophic outcomes.

Statistical Analysis

Multilevel modeling (MLM; Mirman 2016) was used to analyze significant changes in the weekly social anxiety scores and monthly cognitive symptoms from pre- to post-treatment and whether the rates of change differed by treatment condition. MLM allowed all participants to be included in the analyses despite missing data through therapy. MLMs consider and adjust for any bias in standard errors and statistical tests resulting from the interdependence that is typical in such data. We modeled time at Level 1 and individual participants at Level 2. The Level 1 predictor was a session, and the Level 2 predictor was a group. We modeled session as a continuous variable. Growth curve modeling and visual inspection of the shape of change in weekly social anxiety items revealed that the most parsimonious and fitted models differed across groups and measurements (e.g., linear/modest quadratic change in one group vs. linear/cubic change in the other group). We included only a linear effect of time to address the primary question of interest—that is, whether the treatment groups differed in rates of change over treatment. The data were fitted using a linear growth model with fixed effects of group (CCBT vs. ECBT) on the intercept (baseline assessment) and linear (rate of recovery) terms and random effects of participants on the intercept and slope to model individual differences in initial severity and rate of recovery. The fixed effects of group were added individually, and their effects on model fit were evaluated using model comparisons. Improvements in model fit were evaluated using—2 times the change in log-likelihood, which is distributed as χ2 with degrees of freedom equal to the number of parameters added. We added fixed effects of group on time. Group was set by coding CCBT as 0 and ECBT as 1. The CCBT group was treated as the baseline, and parameters were estimated for the ECBT group. Random effects of the intercept and slope terms were included. To test whether the change in each weekly item differed by group, we included the session by group interaction. We used an unstructured covariance structure. MLM analyses were carried out in R version 3.4.2 using the lme4 package (version 1.1–14).

To analyze the reliable change (RC) and clinically significant change (CSC) of social anxiety symptoms during the treatment, the scores on the SIAS and SPS were selected. We used Jacobson and Truax’s (1991) method. Specifically, the magnitudes of change required to achieve RC on the SIAS and SPS were 12.00 and 11.41, respectively. The CSC cutoff scores were 39.97 (SIAS) and 29.82 (SPS). To be classified as CSC, participants scored above the CSC cutoff score at pre-treatment, showed RC during the treatment, and had below the CSC cutoff score at post-treatment (McEvoy et al. 2015). We analyze pre- to post-treatment RC and CSC first and then calculated pre- to follow-up RC and CSC. For the CCBT group, in the pre-post analysis, 37 (out of 45) participants were included in the SIAS–CSC comparison, and 34 (out of 45) patients were included in the SPS–CSC comparison. For the ECBT group, from pre- to post- comparison, 36 (out of 45) patients were included in the SIAS–CSC calculation, and 25 (out of 45) participants were included in the SPS–CSC comparison. For the CCBT group, from pre- to follow-up comparisons, 25 (out of 29) participants were included in the SIAS–CSC, and 24 (out of 29) patients were included in the SPS–CSC calculation. For the ECBT group, 25 (out of 31) patients were included from pre- to follow-up comparison of the SIAS–CSC calculation, and 18 (out of 31) participants were included in the SPS–CSC calculation. Independent-samples t tests and Chi square analyses were used to compare ECBT to CCBT on demographic data, clinical variables, RC, and CSC using SPSS 20.0.

Results

Attrition

To be classified as treatment completers, participants had to attend a minimum of 8 (out of 12) sessions. For CCBT, 41 (out of 45) participants completed treatment (91.2%). For ECBT, 43 (of 45) completed treatment (95.6%). The dropout rates were not significantly different from one another, χ2(1) = 0.71, p = 0.40.

Preliminary Analysis

The ECBT and CCBT groups did not differ significantly on age, gender, jobs, marital status, education, current medication, and comorbidity, all ps > 0.05 (Table 2). In addition, all pre-treatment scores, such as SIAS, SPS, SFA, DBT, SPQ, and SCQ, did not differ significantly in both ECBT and CCBT groups, all ps > 0.05 (Table 3). There were no differences in demographic data and pre-treatment scores between the two groups before conducting treatment.

Table 2 Comparison of pre-treatment demographic and clinical characteristics for the ECBT and CCBT
Full size table
Table 3 Means and standard deviation at pre-treatment, post-treatment, and follow-up and within-group effect sizes
Full size table

Trajectories of Social Anxiety Symptoms

As shown in Table 4, MLM revealed significant linear decreases during treatment and follow-up for both social interaction anxiety and performance anxiety. In addition, ECBT and CCBT showed large within-group effect sizes (Cohen 1988) on social anxiety symptoms (Table 3). For social interaction anxiety, adding a fixed effect of group on the intercept did not improve model fit, χ2(1) = 0.79, p = 0.38, but adding a fixed effect of group on the slope did improve model fit, χ2(1) = 6.69, p = 0.01, indicating faster improvement in the ECBT group compared to the CCBT group in social interaction anxiety. This difference was approximately 0.58 (SE = 0.22) points per week faster for the ECBT group compared to the CCBT group (Fig. 1). On the other hand, for performance anxiety, adding a fixed effect of group on the intercept improved model fit, χ2(1) = 5.36, p = 0.03, but there was no significant effect of group on the linear slope term, χ2(1) = 0.84, p = 0.36, indicating that CCBT and ECBT group did not differ in the performance anxiety improvement rate during treatment.

Table 4 Multilevel modeling results for social anxiety
Full size table
Fig. 1
figure 1

Weekly social anxiety scores during ECBT and CCBT (vertical lines indicate 95% C.I.)

Full size image

Trajectories of Cognitive Symptoms

ECBT and CCBT showed large within-group effect sizes (Cohen 1988) on all the cognitive symptoms (Table 3). MLM analysis showed that fear of negative evaluation decreased significantly during both CCBT and ECBT (Table 5). There was a significant effect of group on the linear slope term, χ2(1) = 6.92, p = 0.01, indicating faster improvement in the ECBT group compared to the CCBT group in fear of negative evaluation. This difference was a decrease of approximately 1.36 (SE = 0.50) points per week faster for the ECBT group compared to the CCBT group.

Table 5 Multilevel modeling results for cognitive symptoms
Full size table

MLM analysis showed that self-focused attention decreased significantly during both CCBT and ECBT (Table 5). There was no significant effect of group on the linear slope term, χ2(1) = 3.29, p = 0.07, however, indicating that CCBT and ECBT group did not differ in self-focused attention improvement rate during treatment.

MLM revealed that dysfunctional belief of the self decreased significantly during both CCBT and ECBT (Table 5). There was a significant effect of group on the linear slope term, χ2(1) = 5.78, p = 0.02, indicating faster improvement in the ECBT group compared to the CCBT group in dysfunctional belief. This difference was a decrease of approximately 2.24 (SE = 0.91) points per week faster for the ECBT group compared to the CCBT group.

MLM revealed that judgment bias improved significantly during both CCBT and ECBT (Table 5; Fig. 2). For the probability bias, there was a significant effect of group on the linear slope term, χ2(1) = 7.32, p = 0.01, indicating faster improvement in the ECBT group compared to the CCBT group in overestimation of negative consequences. This difference was a decrease of approximately 5.62 (SE = 2.01) points per week faster for the ECBT group compared to the CCBT group. For the social cost, there was a significant effect of group on the linear slope term, χ2(1) = 10.50, p = 0.001, indicating faster improvement in the ECBT group compared to the CCBT group in judgment bias of social cost. This difference was a decrease of approximately 6.78 (SE = 2.01) points per week faster for the ECBT group compared to the CCBT group.

Fig. 2
figure 2

Judgment bias during ECBT and CCBT (vertical lines indicate 95% C.I.)

Full size image

Clinical Significance

Chi square analyses failed to find a significant difference between the two groups on the proportion achieving RC (yes vs. no) on the SIAS, χ2(1) = 0.53, p = 0.47, and the SPS, χ2(1) = 1.32, p = 0.25, at post-treatment. In addition, there were no significant differences between the two groups on the proportion achieving CSC on the SIAS, χ2(1) = 0.12, p = 0.73, and on the SPS, χ2(1) = 2.71, p = 0.10, at post-treatment. At follow-up, CCBT and ECBT groups did not differ significantly on the proportion achieving RC on the SIAS, χ2(1) = 0.57, p = 0.45, and on the SPS, χ2(1) = 0.57, p = 0.45. On the contrary, a significantly higher proportion achieved CSC in the ECBT group on the SIAS, χ2(1) = 5.33, p = 0.02, and on the SPS, χ2(1) = 4.71, p = 0.03 (Table 6).

Table 6 Number (proportion) of each sample achieving reliable change (RC) and clinically significant change (CSC)
Full size table

Discussion

This study aimed to improve the outcomes of CBGT for SAD in a community setting. The ECBT program is composed mainly of treatment components that change the NSI of SAD patients. To achieve this goal, ECBT used IR and video feedback much more than the previous treatments for SAD. In addition, ECBT increased the trials of exposure training, including social mishap exposures, to ensure cognitive and behavior changes. We compared the acceptability and effectiveness of an ECBT protocol to a historical control that completed a modified Heimberg’s well-established program. As mentioned before, both interventions were delivered in the same community social anxiety counseling center and had the same length, patients were recruited through the same referral routes, and the groups did not differ significantly on a range of clinical and demographic variables, suggesting that both samples were drawn from a similar population.

The ECBT protocol was highly acceptable to the community patients. The attrition rate of ECBT was low (4.4%), and CCBT also had a low dropout rate (8.8%). These completion rates were comparable to those of previous efficacy and effectiveness studies (see McEvoy 2007), which means that the ECBT protocol was highly acceptable to the SAD patients in the community.

Both treatments were very effective and had large effect sizes for social anxiety and cognitive symptoms of SAD, but ECBT were significantly better and faster in reducing social interaction anxiety, fear of negative evaluation, judgment bias, and dysfunctional self-belief than CCBT. Given that CCBT was demonstrated to be an efficacious protocol for SAD (Yoon and Kwon 2013) with large within-group effect sizes, it is noteworthy that ECBT produced a better outcome and faster improvement than CCBT, and these differences were even greater at follow-up, indicating that therapeutic gains of ECBT remained longer than in the CCBT group.

The effect sizes of several CBT protocols may be compared as follows. The effect size of CBGT following the Heimberg model was 0.80, and that of ECBGT, which enhanced Heimberg’s CBGT (Rapee et al. 2009), was 1.10 (Mayo-Wilson et al. 2014). The effect size of individual CT following Clark and Wells’ model was 1.56 (Mayo-Wilson et al. 2014). In a more recent study, the pre-to-post effect sizes of social anxiety symptoms in imagery-enhanced CBGT were 1.33 to 2.18, and pre-to-follow-up effect sizes were 1.33 to 2.18 (McEvoy et al. 2015). As shown earlier, the pre-to-post effect sizes of ECBT were 1.39 to 1.66, and the pre-to-follow-up effect sizes were 1.62 to 2.05, which are comparable to those of the most efficacious treatments, such as the individual CBT of Clark and Wells’ model and the more recent protocol of imagery-enhanced CBGT.

Participants who received ECBT fell more within the normative range of social anxiety than patients in the CCBT group. ECBT participants showed a higher rate of CSC than CCBT participants at 3 months after the treatment. Participants showed a high proportion of RC and CSC in both groups. Patients who received ECBT demonstrated a high rate of RC (76–78%), and 78–84% of participants achieved CSC after the treatment. For the CCBT group, 64–71% of patients also achieved RC after treatment, and 65–81% of participants showed CSC after treatment, indicating that both ECBT and CCBT were very effective treatments for SAD. In terms of the long-term effects of the treatment, however, there were significant differences in the proportion of CSC between ECBT and CCBT. The rates of fulfilling RC were not significantly different (81% for ECBT vs. 72.4% for CCBT), but 76–89% of the participants who received ECBT achieved CSC 3 months after treatment. For the CCBT group, only 44–58% of the patients attained CSC at the 3-month follow-up. These results can be interpreted as indicating that the clinical improvement of ECBT participants lasts longer than in CCBT participants.

Although there were many components of ECBT that enhanced the treatment efficacy in ECBT, the main focus on altering NSIs contributed greatly to the better outcome of ECBT. Although the NSIs of patients with SAD play a key role in developing and maintaining the disability, contemporary CBT protocols have not focused on NSIs “sufficiently” in the course of the treatment. Some might argue that NSIs can be changed through a variety of techniques used in CBT and therefore need not to be “directly targeted” in an effective treatment for SAD, but we believe that NSIs should be directly targeted throughout the treatment. As mentioned before, recent experimental research on the self-image of SAD has demonstrated that maintaining a neutral or negative self-imagery could affect social anxiety, self-focused attention, safety behaviors, and self-esteem (Hirsch et al. 2003a, 2004, 2006; Makkar and Grisham 2011; Vassilopoulos 2005). In other words, as changes in NSIs can lead to the reduction of social anxiety symptoms, it is important to apply these results of laboratory studies to actual clinical settings.

Use of imagery techniques such as IR might be more powerful than verbal techniques for changing NSIs in the long term. In the CCBT group, mainly verbal methods were used to change cognition, although exposure training could have also contributed to the changes in cognition. The fact that the differences were larger at the 3-month follow-up means that changing NSIs via imagery could bring about long-term effects. As McEvoy and Saulsman (2014) have reported, using positive images to create and utilize new core beliefs or to build up an action plan in a social context are also good examples of using imagery. Their imagery-enhanced protocol was also very effective in improving a variety of social anxiety symptoms, and the authors speculated that this result was probably due to using imagery in their protocol. We agree with their point of view and consider IR to play a similar role in our ECBT protocol. It should be noted that IR was conducted twice on the same early memory in the ECBT protocol to ensure that the early adverse memory would be fully updated and lead to changes in NSIs. In a previous study (Lee and Kwon 2013), two consecutive sessions of IR working on the same early memory enhanced the effect of IR. In the current protocol, IR was conducted in Sessions 2 and 8 to bring about changes in NSIs and core beliefs before the exposure training. To date, IR has largely been delivered individually, but even when it was conducted in a group format, most participants were deeply involved in the IR process and were able to relive their deepest hotspots. All participants said that they were able to get engaged fully with IR without worrying about the other group members. Future research is needed to address the question of the best dosage and the best way of conducting IR in individual or group CBT.

The other factor that deserves attention in producing a better outcome in ECBT was clarifying what to expose during exposure training and increasing the sufficient dose of exposure training in the treatment. As Moscovitch (2009) argued, it is important to repeatedly expose each client’s own feared stimuli, called “attributes of self,” rather than simply exposing them to feared social situations. Thus, what should be clearly identified in the actual exposure sessions are not what situations the patients avoid but what symptoms they want to conceal. Exposure training should be targeted to verify whether symptoms that patients want to conceal would really cause the expected negative consequences. In ECBT, all exposure training, including social mishap exposure, was conducted by identifying clients’ most feared symptoms and confirming whether these symptoms were as negative as the clients expected. In this way, participants were able to identify what they really wanted to hide and verified the results of repeatedly revealing their feared symptoms to others. The ECBT protocol was intended to change patients sufficiently through the treatment process alone. One of the greatest differences between the ECBT and CCBT programs was the repetition of this emphasis, which seems to be another cause of the superior effect of ECBT over CCBT.

It was surprising that the cognitive symptoms of SAD were changed more when using mainly behavioral interventions than when using traditional cognitive interventions. Cognitive symptoms such as fear of negative evaluation, overestimation of probabilities and social cost, and dysfunctional belief of self were changed faster in the ECBT group than in the CCBT group. These results indicate that therapists do not always have to use traditional cognitive restructuring to cognitive errors, as by finding and modifying automatic thoughts when patients feel anxious, finding evidence, and so on. Rather, aggressively using empirical methods, such as behavioral interventions, may cause a faster and more accurate modification of the dysfunctional belief of patients with SAD. Further studies, such as dismantling studies, will provide some clues to effective methods for changing cognitive symptoms of SAD.

On the other hand, no significant differences in SPS were found between the two groups. This is consistent with the results of McEvoy and Saulsman (2014) and McEvoy et al. (2015), conducted in a similar way. Their studies also found no significant differences in SPS. The authors concluded that image-enhanced CBGT was a significantly more effective program for social interaction anxiety because the patients had prioritized social interactions for homework tasks in both treatments. In addition, they suggested that imagery-enhanced CBGT was highly effective but not superior for performance anxiety. Nevertheless, our result is surprising in that the exposure exercises, which were mainly carried out in ECBT, focused on performance tasks such as speaking in front of others and doing something while others were watching them. There might be a floor effect in the SPS scores, which were very low at post-treatment (M = 15.07) and follow-up (M = 12.52) in the ECBT group. Additionally, CCBT programs might also have the advantage of reducing anxiety in performance situations. In other words, a cognitive intervention, which was mainly used in CCBT, might be as effective as a behavioral intervention in lowering performance anxiety. Therefore, further study is needed to investigate the differences between cognitive and behavioral interventions in reducing the level of performance anxiety.

Certain limitations in this study should be kept in mind. First, participants in this research were not randomized to different treatments, which might have led to systematic errors across the samples. This issue was to some extent alleviated by the similarities of the samples in terms of therapists, treatment settings, recruitment procedures, and clinical and demographic variables, but there might be systematic errors that we did not take into account. Nonetheless, it is noteworthy that excellent therapeutic effects have been demonstrated in patients who came to actual treatment settings. Therefore, future studies might be needed to validate the efficacy of ECBT protocols through randomized controlled studies. Second, it is difficult to know exactly which treatment component of ECBT caused it to have a better therapeutic effect than CCBT. We believe that attempts to modify an NSI repeatedly through video feedback and IR led to these changes. In addition, we think that high doses of behavioral interventions that focused on “the feared stimuli” of SAD patients also resulted in better effects. However, it is unclear whether the better effect is due to a change in NSIs, a high dose of behavioral interventions, or their interaction. Dismantling studies will be needed to clarify the effective treatment component of ECBT, which should be elucidated in future research. Third, although we focused on modifying NSIs with various techniques in order to show the better therapeutic effects of ECBT, there was insufficient evidence that differences in NSIs actually caused the different therapeutic effects between the two groups.

Despite these limitations, this study indicates that a newly developed ECBT can be a promising group CBT for SAD. Further research will be needed to clarify the role of NSIs as a treatment mechanism of ECBT, and to identify the optimal combination of techniques for changing NSIs.