Introduction
Social anxiety is prevalent in youth (Wittchen et al.
1999), can disrupt academic performance and interpersonal interactions (Owens et al.
2008), persist into adulthood, and impact other disabling mental health conditions and quality of life (Woodward and Fergusson
2001). Cognitive behavioural therapy (CBT), the current gold-standard treatment, can reduce social anxiety in youth (Scaini et al.
2016) but many fail to show clinically significant responses (Kendall et al.
2012), respond but subsequently relapse (Ginsburg et al.
2014), or find it difficult to access. Identifying more effective, accessible methods so that young people can better manage their symptoms is a public health priority. Cognitive bias modification (CBM) training, which uses computerised tasks to target symptom-linked cognitive biases, has emerged as a potential adjunctive intervention (Butler et al.
2015; White et al.
2016) that may be amenable to delivery through computerised formats at home (Salemink et al.
2014) or in school (Fitzgerald et al.
2016). Yet, existing CBM packages remain weak at boosting more adaptive information-processing styles and at reducing symptoms (Cristea et al.
2015a,
b; Heeren et al.
2015; Mogoaşe et al.
2014). This study presents a newly developed, multi-session computerised training program that targets multiple cognitive biases using a variety of training techniques and stimuli, for adolescents with elevated social fears. We assess the viability of administering this training tool at school, it’s acceptability to young people and compare changes in biases and symptoms across a baseline and an intervention phase.
Drawing on cognitive models of social anxiety (Clark and Wells
1995; Rapee and Heimberg
1997), a large corpus of research has found a link between social anxiety and attention and appraisal biases in adults as well as adolescents (Bar-Haim et al.
2007; Haller et al.
2016; Klein et al.
2017; Miers et al.
2008; Rheingold et al.
2003). These manifest as: greater allocation of attention to threatening stimuli at involuntary and voluntary stages of processing (Roy et al.
2008; Stirling et al.
2006); a tendency to interpret ambiguous cues in threatening ways; and a tendency to disproportionately attribute negative events as caused by oneself (i.e., ‘internal’ reasons) and positive events as caused by others or circumstance (i.e., ‘external’ reasons). Computerised cognitive training methods, which encourage more adaptive styles of information-processing over repeated trials and practice, have been developed in adults to reduce general and social anxiety. Cognitive bias modification of attention (CBM-A) methods alter maladaptive attention-orienting patterns towards threat, and encourage selective attention towards neutral or positive stimuli. Most commonly, CBM-A methods use a modified dot-probe task in which probes only ever appear in place of non-threatening stimuli (MacLeod et al.
1986). In contrast, in ‘visual search’ CBM-A training the individual must locate a non-threatening stimulus from among threatening stimuli as quickly as possible (Waters et al.
2013). Cognitive bias modification for interpretations (CBM-I) targets biases in interpretation, mostly using the ‘ambiguous situations task’ (Mathews and Mackintosh
2000). Here, participants read a series of ambiguous sentences that end with a word fragment. Completion of the final word disambiguates the valence of the sentence in a positive direction. Participants receive a follow-up ‘yes/no’ comprehension question with ‘correct/incorrect’ feedback in order to reinforce the training. A few studies have developed programs to modify attributions in adults to reduce depressive mood (Peters et al.
2011) but not anxiety.
However, studies of adults with various anxiety conditions (including trait anxiety) have only found weak (but significant effects) in symptom change (Hakamata et al.
2010; Hallion and Ruscio
2011; Heeren et al.
2015, but also see; Cristea et al.
2015a; Mogoaşe et al.
2014). Reduction in symptoms typically occur when there is also successful bias modification (MacLeod and Clarke
2015), and possibly through multiple training sessions (Hallion and Ruscio
2011). Extensions of CBM-A and CBM-I for use in adolescents (Bar-Haim et al.
2011; Lau et al.
2013), using the same tasks but with modifications to the stimuli content and modality (audio/text/pictures) have found small to medium effects of CBM-I and CBM-A training on cognitive biases, but no effect on general indices of mental health (nor on anxiety specifically) (Cristea et al.
2015b). Looking at these packages separately, Lowther and Newman (
2014) identified that 8 out of 10 CBM-A studies reported positive changes in anxiety post-intervention (although only 4 of these 8 studies also found a change in attention bias). Through a meta-analysis, Krebs et al. (
2017) found that CBM-I had a statistically significant moderate effect on decreasing negative interpretations and boosting positive interpretations. A small but significant effect on self-reported anxiety immediately following training was also found. While adult studies have tried to alter cognitive processes relating to depression through attribution training (Peters et al.
2011), their extension to young people has focused on targeting aggressive behaviours and academic achievements (Sukariyah and Assaad
2015; Vassilopoulos et al.
2015). No studies to our knowledge have trained adaptive attributions in adolescents (or adults) to reduce anxiety. Thus, while CBM training packages have potential, efforts to boost bias change and symptom reduction are needed. Adult data advocate multi-session training but their extension to anxious adolescents yield mixed findings regarding symptom and bias change for CBM-A (Fitzgerald et al.
2016; de Voogd et al.
2016,
2017b; Pergamin-Hight et al.
2016) and CBM-I (de Voogd et al.
2017a; Reuland and Teachman
2014). Therefore, consideration of other methodological factors may be important in prompting significant symptom change.
The current study aimed to improve CBM training effects by incorporating several methodological features into the training package. Some of these features drew directly on findings around known contributions of cognitive factors to anxiety, while others aimed to increase user-engagement. Consistent with combined cognitive bias hypotheses of psychopathology (Hirsch et al.
2006; Everaert et al.
2012,
2014), we first included bias modification procedures to target both attention and interpretation biases in social anxiety, within the same package. Targeting biases together may produce a greater magnitude of change (because of their combined additive and interactive effects). Only one study we are aware of has utilised a combined-bias approach in socially anxious adolescents (Sportel et al.
2013; de Hullu et al.
2017), testing an internet-based CBM-A/CBM-I program and finding significant improvement across all groups but no significant difference between internet-based CBM, CBT and control group. Secondly, CBM-A tasks aim to modify maladaptive processes of selective attention towards, and difficulty disengaging from, threatening environmental stimuli, yet do less to target self-focused attention. Models of social phobia (Clark and Wells
1995) posit that the socially-anxious individual shifts their attention inwards to produce an (often negative) image of themselves, based on interoceptive sources, rather than actual monitoring of others’ responses to disconfirm these negative fears. This self-focused attention in turn reduces processing of environmental cues in adults as well as adolescents (Hodson et al.
2008; Judah et al.
2016), which suggests that targeting these maladaptive self-focused attentional processes during CBM-A training could be beneficial (Wells and Papageorgiou
1998). We therefore included a task within the CBM-A package that draws the individual’s attention toward their internal feelings and then encourages them to shift their attention externally to stimuli that challenge these beliefs of how others view them in a social situation. Thirdly, we also increased the scope of CBM-I by targeting attribution biases too, particularly the tendency to internally attribute responsibility for negative events compared to positive events (Haller et al.
2016). We included a second task within the CBM-I package, that asked young people to generate an internal attribution for a positively interpreted event.
Finally, trial repetition, boredom and disengagement are serious concerns for CBM training (Beard
2011). We increased user-engagement by varying the training techniques used and the modality of stimulus presentation. A combination of CBM-A techniques was used, from the visual dot-probe to the visual search tasks. In the dot-probe, we trained attention towards positive words and faces on some blocks, and attention towards neutral words and faces on other blocks—always away from negative stimuli. In the visual search, participants identified a smiling face from a grid of negative faces in one module, but also practiced shifting their attention from internal sensations and cues toward benign, external interpersonal cues in another module. For CBM-I, we used text-based scenarios to encourage benign/positive resolution of ambiguous situations, as well as visual presentations of ambiguous scenes that had to be resolved benignly/positively. The latter may allow for more effective visualisation, and therefore stronger emotional responses and bias modification, than material presented in word form (Holmes et al.
2006; Holmes and Mathews
2010).
To assess viability and acceptability of our enhanced, multi-session CBM intervention for social anxiety, we used an A–B case series design, in which adolescents selected for high social anxiety received 8 school-based CBM training sessions, in two 4-day blocks over a 2-week period. We also gathered quantitative data on changes on selected measures during the 2-week intervention phase but also during a 2-week baseline period. We expected a significant decrease in social anxiety symptoms, and a significant change in attention and interpretation biases. Due to these clear a priori hypotheses, we conducted significance testing on changes in social anxiety symptoms, real-life socially avoidant behaviour and measures of attention and interpretation biases during the baseline versus the intervention phases. We also calculated the correlation between changes in social anxiety and changes in cognitive measures. To explore specificity effects to social anxiety symptoms, we measured changes on general anxiety and depression symptoms.
Discussion
This case series explored the value of a combined-bias, multi-session CBM program, for adolescents with elevated social anxiety. While targeting both attention and interpretation biases for threat, new training modules targeting self-focused attention and internal attributions were included as well as a variety of training techniques that used both verbal and pictorial stimuli to enhance user-engagement. The data obtained suggest that it is viable to deliver this CBM program in a school in individual sessions with a trained researcher. Under experimental conditions, the program showed itself to be feasible in terms of its applicability and accessibility in a school setting: Only 4 of the 23 participants withdrew from the study prior to completion, thus it appears to have a good acceptability from participants. Although not directly assessed, school teachers were largely supportive of this research and we had good recruitment rates amongst schools. It should be noted that participants were always accompanied by a researcher and some participants received several reminders of their appointment and needed supervision by a researcher in order to remain engaged in the training task. Some participants were fully engaged throughout the entire study without additional support from the researcher. This has implications for determining whether a CBM program such as the one used in this study, delivered in a school, is engaging enough for individuals to complete without supervision.
The significant reduction in symptoms on the SAS-A following eight sessions of CBM over 2-weeks, compared to no significant reduction in SAS-A scores following a 2-week baseline phase—and similar findings using a diary measure of daily social interactions—suggests that there is some potential in reducing social anxiety levels in adolescents reporting elevated symptoms. However, there are two caveats to this conclusion. First, although 13 participants showed changes on social anxiety symptoms, these varied between a decrease of 1–19, across the training phase, possibly suggesting that a few individuals with large changes drove the significant decreases. Also, only around 9 showed reductions across items on the social diary assessment. This suggests variability in how useful this training was for targeting social anxiety across individuals, reflected somewhat in the qualitative feedback too. Second, data from other outcome measures showed that these effects were not specific to social anxiety, and instead reductions in depressive and general anxiety symptoms were also observed. It is possible the observed decrease in socially-avoidant behaviour led to increased exposure to potentially rewarding social situations, thus having an impact on these general affective indices. However, as all of these measures were self-report, we cannot rule out the possibility that these broader symptoms changes indicate the presence of demand effects.
Also challenging for our findings of symptom improvement was mixed findings around changes in interpretation and attention bias. Although post-hoc analysis showed that interpretation bias scores did show a significant change pre- to post-training with no significant change pre to post baseline, the absence of a significant interaction effect between phase and time suggests that the degree of change was not significantly greater. However, individual scores show that for most participants the interpretation bias went in the intended direction, and several participants showed a greater jump from Assessment 3 to Assessment 4 than from Assessment 1 to Assessment 2. The feedback, on the whole, also points to several participants feeling the CBM-I tasks were beneficial. Finally, there was a significant association between this change in interpretative style and change in social anxiety symptoms. It should be noted that, whilst there was a lack of significant correlation between initial baseline interpretation bias and SAS, the correlation reported was in the expected direction. With a larger sample size, we would expect this to reach significance. Furthermore, a weak correlation between initial interpretation bias and SAS may not be a prerequisite of a correlation between changes in these two variables, if the common factor explaining this correlation is the administration of a training tool designed to effect changes on both. Therefore, we tentatively suggest biased interpretations could provide a promising target for symptom improvement for some young people.
In contrast, we found no significant effects for attention bias change, or any correlation between change in attention bias to threat and change in symptoms. There was also a lack of significant correlation between cognitive biases and symptom measures at baseline. It may be that our current method for assessing attention bias is problematic. Previous research has shown the dot-probe task has poor reliability, comprising internal consistency and test–retest reliability in children and adolescents (Brown et al.
2014; White et al.
2016) and in adults (Van Bockstaele et al.
2017). Indeed, the current results display an extremely low test–retest reliability for attention bias r = − 0.01, compared to the interpretation bias measure (r = 0.88). Some studies using eye-tracking have demonstrated that certain measures, such as dwell time across trials on socially threatening stimuli, are more reliable across time, but also more consistent in their associations with social anxiety (Lazarov et al.
2017). More generally, others have argued that a visual search grid could be more effective than the dot probe as a tool for more reliably measuring and more effectively modifying attention processes that are linked to anxiety (Mogg and Bradley
2016; Van Bockstaele et al.
2017). The development and application of potentially more stable and reliable measures like these are essential to better understand the nature and modification of attentional biases. Furthermore, as participant feedback suggests that the rigidity of the dot-probe task may result in a lack of motivation and task engagement, incorporating extrinsic motivators, such as real-time performance feedback (Bernstein and Zvielli
2014) and using this real-time performance data to tailor the task to the individual’s optimal rate of learning (Schnyer et al.
2015), may increase task engagement and improve attention bias change.
While the training task was generally acceptable, the feedback collected provides more insights into further features that could improve effectiveness and engagement. Participant feedback suggests that as the goal of the CBM-I training portion became clearer, it gradually gave the participant an understanding of not needing to view social situations so negatively. It may be that incorporating explicit instructions to practice the target bias may enhance CBM efficiency (MacLeod et al.
2009). This could be particularly true for the CBM-A tasks, as feedback suggests participants found these tasks ‘boring’ and ‘un-engaging’, partly due to not understanding why they were doing them. Feedback regarding task-specific elements of the CBM program suggests that, contrary to expectations, word-based social situations were in fact more successful in creating visual imagery than the picture based scenarios. Several participants found the unfamiliar visual stimuli harder to engage and immerse than the word-based descriptions. Use of more personalised picture stimuli may be of greater use perhaps with incorporation of media such as videos to improve immersion for the participant. This feedback is in line with recent research finding no difference in outcome when attempting to improve CBM-I effectiveness by incorporating visual imagery (de Voogd et al.
2017a). Whilst we have no way of quantitatively assessing whether this study benefited from multiple versus single sessions of training, the qualitative feedback suggests that after several sessions of CBM-I training some participants experienced increased insight, that they could ‘look at social situations less negatively’. This may point towards a combination of implicit and explicit processes—with implicit training effects on processing, extending to influence ongoing behaviour via increased insight. Finally, participant feedback does show some variation in specific aspects of the program that participants found helpful/engaging, which may partially explain individual differences found in symptom reduction.
Whilst the symptom changes on social anxiety are encouraging, the data also provide several challenges. That this study was a preliminary case series with a small sample size, the appropriateness of significance-testing of statistical comparisons is questionable with different approaches taken in prior studies, (Abeles et al.
2009; Bechor et al.
2014; Blackwell and Holmes
2010; Rozenman et al.
2011). However, we limited our statistical tests to key measures that related to a priori expectations. Second, although there are advantages to carrying out a A–B case series in the same participants (self-matching means that any potential confounders such as socioeconomic status, genetic risk, state of health etc., are automatically controlled for), the absence of an active control group or condition means we are unable to attribute symptom change directly to cognitive training procedures (over a placebo effect). Furthermore, this program was presented to participants as a new psychological training program designed to target cognitive biases, which may have increased demand effects and expectancy biases [see MacLoed et al. (
2009 ) for a more thorough discussion of this issue]. Use of questions to reveal expectancy beliefs (Schmidt et al.
2009) may be beneficial for future studies in assessing the possibility of demand effects. Additionally, as all participants completed the baseline phase prior to training and our design did not include a control group, we are unable to fully account for natural fluctuations in anxiety across time. However, as a first-step, such case series is important as performing a cross-over case series design and a randomised controlled clinical trial may be premature, and not an optimal strategy for investing research and patient resources. Third, the lack of bias effects might be to do with mixed training, as none of the training tasks were completed for more than two sessions. Fourth, the design could have benefited from a follow-up time point, with the possibility that all consequences of CBM may take a longer time to become evident. Previous CBM research has found that emotional outcomes continue post-CBM completion (Schmidt et al.
2009). Finally, the generalisability of our findings was affected by the strong gender disparity in our sample: Female pupils self-selecting into such studies have been a feature of school-based recruitment in many of our studies. As students were allowed to ‘opt out’ of the screening procedure, there was little we could do to change this. Despite these limitations, we find the study has provided some encouraging findings. The CBM program has demonstrated its potential as an easily accessible resource for adolescents with elevated social anxiety. The next step will be to test these tasks in a larger sample with a comparison condition or group.