A systematic review of treatments targeting cognitive biases in socially anxious adolescents

Abstract

Background

Adolescence is a period of brain plasticity that is affected by social and affective stimuli. Adaptive neurodevelopmental changes in the context of complex social situations may precipitate or exacerbate cognitive biases (i.e., attention and/or interpretation biases) and predispose at-risk individuals to symptoms of social anxiety.

Methods

This systematic review followed the PRISMA guidelines. Nine adolescent studies were examined including 3 studies using cognitive bias modification training (CBMT) to target attention biases (CBMT-A), 3 studies using CBMT to target interpretation biases (CBMT-I), and 3 aimed at reducing both attention and interpretation biases.

Results

The studies of CBMT-A alone did not find significant effects on cognitive and clinical outcomes. However, studies of CBMT-I alone showed some improvement in interpretation bias. The combination of CBMT-A and CBMT-I appeared promising in reducing both attentionl and interpretation biases.

Limitations

The paucity of studies and the heterogeneity across studies (e.g., format of CBMT, assessment measures) limit the calculation of overall effect sizes and the examination of predictors, moderators, and mediators of outcome.

Conclusions

Technology-driven interventions such as CBMT have the potential to extend treatments outside the clinic setting and to augment current therapies for social anxiety. Further research is needed to develop CBMT procedures that optimize learning in group and real-world settings and to identify predictors of treatment response. Understanding the neural correlates of response to CBMT may help identify future targets for intervention.

Introduction

Adolescence is a period of heightened experience-dependent learning and neural plasticity to socio-affective stimuli (Crone and Dahl, 2012). During this period, teenagers experience an increase in emotional reactivity and greater social understanding –functions that are necessary to successfully navigate a novel and complex social world. Adaptive changes in the experience and regulation of emotions in response to social cues and in the reasoning about the social world and others’ mental states are observed (Nelson et al., 2005). Additionally, these changes are driven by functional maturation in the neural networks underpinning social, affective, and cognitive processing (Blakemore, 2008).

However, these adaptive neurodevelopmental changes, in conjunction with the exposure to increasingly complex social contexts (Steinberg and Silverberg, 1986), can trigger the emergence of or exacerbate pre-existing cognitive biases, i.e. maladaptive systematic distortions in information processing (Haller et. al., 2015). These biases can predispose vulnerable individuals to several psychiatric illnesses, including social anxiety (SA), a condition that is characterized by a persistent fear of negative evaluation from others in social or performance situations and associated avoidance of anxiety-provoking social contexts (Hodson et al., 2008; Rapee and Spence, 2004).

SA affects up to 9% of adolescents (Burstein et al., 2011), with 75% of its extreme and persistent forms emerging by mid-adolescence with a median age of onset of 13 years (Kessler et al., 2005; Wittchen et al., 1999). If untreated, SA can follow a chronic course (Burstein et al., 2011) and lead to consequences such as depression (Beesdo et al., 2007), suicidality (Katzelnick et al., 2001), substance and alcohol dependence (Buckner et al., 2008), academic under-performance, and social isolation (Beidel et al., 1999). In addition, since peer interactions carry learning experiences, social avoidance may be especially disruptive during adolescence (Miers et al., 2014) and may result in impaired development of social skills (Stein and Kean, 2000). With SA contributing to substantial burden for patients and their families, and resulting in long-term societal costs, it is critical to intervene early (Acarturk et al., 2009; Patel et al., 2002).

Cognitive models of SA have identified two types of cognitive bias that are present at various stages of social information processing in adolescents: (i) attention bias, where attention is allocated toward threatening social cues, including increased orientation toward and affective response to negative social cues, like negative faces, fearful or angry voices (Pérez-Edgar et al., 2010); and (ii) interpretation bias, a tendency to interpret ambiguous social situations as threatening or negative, which can result in SA-driven maladaptive behaviors, including social avoidance and withdrawal (Haller et al., 2017, 2016).

Importantly, functional neuroimaging studies have shown that these cognitive biases are driven by dysregulated engagement of prefrontal-limbic-striatal-temporal circuits before, during, and after the processing of social stimuli (Bar-Haim et al., 2009, 2007; Battaglia et al., 2012; Guyer et al., 2008; Jarcho et al., 2013; Killgore and Yurgelun-Todd, 2005). These biases are hypothesized to be an etiological and maintaining factor for SA. For example, pre-existing attention biases to threatening social cues might be exaggerated as the circuits regulating reactivity to social-affective information mature. This could in turn affect how the brain functionally develops by biasing the nature of incoming information and the ways in which this information is processed and consolidated.

As a consequence of these cognitive models, emphasis has been placed on the importance of developing experimental treatments that directly address the underlying neuropathology of SA and modify the cognitive biases associated with SA (MacLeod and Mathews, 2012). Cognitive Bias Modification Training (CBMT) is a treatment informed by cognitive neuroscience that leverages repetitive implicit learning mechanisms to reallocate attention away from threat-relevant information and/or to reduce negative interpretations of ambiguous social situations through repeated computerized training trials (Amir et al., 2010; Hakamata et al., 2010). In CBMT, more adaptive forms of information-processing styles are systematically reinforced through positive feedback, while maladaptive learning is discouraged through negative feedback.

Several clinical trials show that CBMT as a stand-alone treatment decreases negative affect, particularly anxiety, in adults with SA (Bar-Haim et al., 2007; Mobini et al., 2013), but these findings may not be generalizable to socially anxious adolescents given developmental differences. Multiple lines of evidence show that adaptive emotional and social learning is enhanced in adolescence and aided by a unique reactivity/receptiveness to certain learning signals and changing mechanisms by which these signals guide and influence behavior (Hauser et al., 2015; van Duijvenvoorde et al., 2008). Therefore, cognitive processing styles that are similar to the ones being targeted by CBMT may develop during childhood, stabilize and mature across adolescence (Lau et al., 2006; Lau and Eley, 2008), and therefore be more amenable to modification during adolescence than adulthood. Thus, the therapeutic benefits yielded by CBMT could be stronger and more long-lasting than those achievable at later developmental stages (Cohen Kadosh et al., 2013). This also suggests that intervening early with CBMT may not only translate into a timely improvement of symptoms to avoid long-term negative outcomes, but also radically change the trajectory of illness.

In this review, we examined published studies that delivered CBMT to adolescents with SA. We first investigated whether CBMT attenuated cognitive biases and reduced SA symptoms. Second, we studied whether the effect sizes yielded by CBMT in adolescents with SA were comparable to those reported in samples of adults (Heeren et al., 2015; Liu et al., 2017) and children (Cristea et al., 2015b) with SA.

Methods

1. Search Strategy

This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Moher et al., 2009). Peer-reviewed, English-language research articles were selected for the review; non-human, review and meta-analytic reports were excluded. We identified studies for inclusion through searching the electronic databases PubMed, PsycINFO, and EMBASE.

Three sets of keyword search algorithms were used, linked with the Boolean operator AND. The first was related to diagnosis: “social anxiety” OR “social phobia” OR “social anxiety disorder” OR “socially anxious”. The second was related to the intervention and included “attention bias modification” OR “interpretation bias modification” OR “cognitive bias modification” OR “cognitive training”. The third set of search terms was related to developmental period: “adolescent” OR “youth” OR “child” OR “teenage”. Using these criteria, all authors screened the titles and abstracts of search results. During this screening phase, we excluded study protocols that failed to clearly meet inclusion criteria (below). Whenever at least one author raised concerns about study inclusion, the full text was inspected and all authors discussed the study until a consensus was reached. For all search results that passed the first screening, we retrieved and reviewed the full texts. Additionally, at this stage we cross-referenced lists of included studies to gather any papers that the search terms had not identified.

2. Eligibility Criteria and Study Selection

We aimed to evaluate the effects of CBMT on cognitive biases and symptoms in adolescents with SA. Note that “cognitive bias modification training” is a broad-based term that can be used to refer to a range of approaches for the amelioration of cognitive biases. We did not specify any parameters regarding the type of paradigm employed by the studies, other than the following eligibility criteria. Studies were included if they: 1) presented findings from randomized controlled trials (RCTs), naturalistic, or single-arm studies focusing on CBMT; 2) recruited and reported data for patients ages 12–18; and 3) were peer-reviewed, English language original articles. Studies were excluded if they: 1) only provided data on feasibility, acceptability or engagement, or no data (e.g. published study protocols); 2) were single-case reports; 3) were exclusively conducted on child (mean age < 12) or adult (mean age > 18) samples; 4) were conducted on unselected samples of adolescents who did not endorse SA; 5) did not use a standardized and validated assessment measure of SA symptoms, and 6) were secondary analyses of original data previously reported.

For articles that were rated as not eligible by at least one author, we held a discussion meeting where we analyzed any disagreements until a consensus about study inclusion was reached. All articles matching our eligibility criteria were reviewed in full by the authors. From each RCT, naturalistic, and single-arm primary study, we extracted demographics, clinical characteristics (diagnosis, age), and within-group effect sizes of measured clinical and cognitive outcomes. For every mismatch in extracted data, all authors discussed the trial until a consensus was reached. Given the heterogeneity of study designs and samples, we did not code variables related to medication.

Results

We conducted full database searches in August 2019, with the inclusion and exclusion criteria identified prior to the collection period. Figure 1 shows a PRISMA flowchart of each stage of the search process. The search strategy returned 142 unique results after duplicates were removed. Of these, 20 articles were examined further by reviewing the abstract. Of the 15 publications reviewed in full, 6 were excluded as not meeting criteria. Thus, 9 studies were included in the review. For each of the 9 studies, the study design and key findings are presented in 3 sections that follow (CBMT-I, CMBT-A, Combination) with key details about study methods presented in Table 1.

Figure 1.

Flow Diagram

Table 1:

Summary of the included studies.

Reference	Type of study	Groups	Specific CBMT program (duration)	Measures	Participants	Mean age (sd)	Main results*
Fu et al., 2013	RCT	positive CBMT-I vs ACTr	Single session (5 blocks, 12 items each: 10 scenarios resolving positively, 1 training-incongruent scenario resolving negatively, 1 scenario resolving neutrally)	MINI-KID, SCARED, IREC-T, IBQ	28 GAD (60.7%) and/or SAD (39.3%) (53.6% F)	14.5 (1.8)	IREC-T ηp2= 0.17, d (CBMT-I)= 2.64, d (ACTr)= 1.27
Sportel et al., 2013	RCT	Combined bias CBMT vs CBGT vs no training	20 sessions (40 minutes each), delivered twice a week via the internet	ADIS-C, RCADS, stIAT, VPF, VPW, AIBQ, IREC-T	240 SAD (73% F)	14.1 (.7)	RCADS Social Phobia: pretest to posttest (d= 0.42) – overall posttest to 6 month FU (d=0.41) CBGT> control stIAT: pretest to 12 months FU (d=0.61) CBMT > CBGT
Reuland & Teachman, 2014	RCT	CBMT-I child-only vs CBMT-I parent-only vs CBMT-I for child and parent	8 sessions every 3 days	ADIS-C/P, SAS-A IREC-T, BAT	18 SAD (72% F)	13.0 (1.6)	Similar efficacy across the three groups: pretest to posttest (d=1.14) pretest to follow-up (d= 1.38)
Pergamin-Hight et al., 2016	RCT	CBMT-A vs ACTr	8 training sessions in a 4-week period	EATQ-R, SPAI, ADIS-IV-C/P, DPT	67 SAD (57 % F)	12. 7 (3.1)	SPAI Cohen’s d=.48 (in youth 13 or older)
Fitzgerald et al., 2016	RCT	CBMT-A vs ACTr	1 CBMT-A session per week over 4 weeks.	DPT, SCARED, SPAI, BFNE-R	127 with SPAI score ≥24 (57% F)	15.9(.7)	No significant changes between groups
De Hullu et al., 2017	RCT	Combined bias CBMT vs CBGT vs no training	20 sessions (40 minutes each), delivered twice a week via the internet	ADIS-C, RCADS, stIAT, VPF, VPW, AIBQ, IREC-T	240 SAD (73% F)	14.1 (.7)	IREC-T: pretest to 24 months FU: d (CBMT) =−1.44 d(CBGT)=−.63
Klein et al., 2018	RCT	positive CBMT-I vs ACTr	5 sessions in a 3-week period	IREC-T, SASC-R, AST, SCARED	69 with SCARED social phobia subscale score >8 (NA)	14.4 (1.5)	IREC-T ηp2= =0.12 AST ηp2=0.13 SASC-R (10 weeks) ηp2=0.10
Lisk et al., 2018	Cases series	combined bias CBMT	8 sessions in a 2-week period	AIBT, DPT, SAS-A, SCARED	19 with SAS-A score >50 (99% F)	17.0 (N/A)	SAS-A ηp2= 0.29 AIBT ηp2 SCARED ηp =0.55
Ollendick et al., 2019	RCT	CBMT-A vs ACTr	10 sessions twice a week over a 5-week period	ADIS-IV-C/P SCARED, EATQ-R, DPT	58 SAD (70.7% F)	14.3 (1.3)	No significant changes between groups

^*Effect sizes are reported only for significant group x time interactions only

(alphabetical order): ABMT: Attention Bias Modification Training, ACTr: Active Control Training, ADIS-IV-C/P: Anxiety Disorders Interview Schedule for DSM-IV-Child and Parent Versions, AIBT: Adolescent Interpretation Bias Task, AST: Ambiguous Scenarios Task, BAT: Behavioral Avoidance Task, BFNE-R: Brief Fear of Negative Evaluation-Revised, CBGT: Cognitive Behavioral Group Therapy, CBMT: Cognitive Bias Modification Training, CBMT-A: Cognitive Bias Modification Training-Attention, CBMT-I: Cognitive Bias Modification Training-Interpretation, DPT: Dot Probe Task, EATQ-R: Early Adolescent Temperament Questionnaire – Revised; EATQ-R: Early Adolescent Temperament Questionnaire-Revised Short Form, GAD: Generalized Anxiety Disorder, IBQ: Interpretation Bias Questionnaire, IREC-T : Interpretation Recognition Task, MINI-KID: Mini International Neuropsychiatric Interview for Children and Adolescents 5.0, RCADS: xxx, RCT: Randomized Controlled Trial, SAD: Social Anxiety Disorder, SAS-A: Social Anxiety Scale for Adolescents, SASC-R: Social Anxiety Scale for Children—Revised, SCARED: Screen for Child Anxiety Related Disorders, Child and Parent Versions, SPAI: Social Phobia and Anxiety Inventory, stIAT: Single Target Implicit Association Test, VPF: Visual Probe task with Faces, VPW: Visual Probe task with written Words.

CBMT Interpretation Bias Studies

We identified three studies that examined the effects of CBMT programs targeting interpretation biases (CBMT-I). CBMT-I training protocols involve presenting participants with emotionally ambiguous scenarios that are resolved when a word fragment at the end is completed to convey meaning. The three CBMT-I studies identified in our search all used this approach by presenting participants with trials involving vignettes that primarily described social scenarios.

In a study by Fu and colleagues (Fu et al., 2013), Chinese adolescents with social anxiety disorder (SAD) and/or generalized anxiety disorder received a single session of either positive CBMT-I or Active Control training (ACTr) (Fu et al., 2013). CBMT-I involved completing a word-fragment to resolve the outcomes of sixty ambiguous scenarios. During positive training, scenarios ended with benign/positive resolutions, but during ACTr (neutral training), half of the scenarios were resolved positively and half negatively. Participants who received CBMT-I interpreted new ambiguous scenarios less negatively than the ACTr group (with small effect sizes), although training effects were not observed on self-reported assessment of interpretation bias and emotional vulnerability. Negative mood symptoms reduced across both groups, and no effects were seen for positive mood. Authors did not examine possible effects on anxiety symptoms, making it difficult to ascertain whether the reported change in cognitive bias was clinically meaningful.

Reuland and Teachman (2014) assigned adolescents with SAD and their mothers to 1 of 3 online interventions: i) CBMT-I for the child only (n=6), targeting cognitive biases associated with adolescents’ maladaptive beliefs about social situations; ii) CBMT-I for the parent only (n=5), targeting biases known to drive intrusive parenting practices (e.g., my child cannot tolerate anxiety; I will be a bad parent if I don’t stop my child’s anxiety); and iii) CBMT-I targeting both youth and parents’ biases in combination (n=6). Larger effect sizes were seen for change in parent compared with child interpretation bias, with no substantial differences in effect sizes between positive and negative bias, and no significant differences across conditions. Additionally, analyses revealed mostly large effect sizes for all SA self-report measures, with no significant differences between conditions. Although the small sample warrants cautious interpretation, findings provide modest support for the efficacy of CBM-I, with no significant differences across conditions.

Klein and colleagues (Klein et al., 2018) randomly assigned socially anxious adolescents with mild intellectual disability to a positive CBMT-I or a neutral control training. In contrast to the control training group, adolescents in the positive training group showed a significant reduction in negative interpretation bias on two interpretation bias tasks after CBMT-I. Interestingly, these post-treatment reductions were observed in test tasks measuring both near transfer (i.e., test task is very similar to training task) and far transfer (i.e., different format from training task) of learning, suggesting gains from CBMT-I may generalize to untrained contexts. Gains continued to 10-week follow-up for near transfer of learning, and participants also showed a significant reduction in SA symptoms at 10 weeks post-training. Results suggest that it is possible to modify interpretation biases in youth with both SA and mild intellectual disability, though training protocols that are longer or that include booster sessions may be needed to produce durable gains that transfer to novel or untrained settings.

CBMT Attention Bias Studies

Three studies were identified that tested the effects of CBMT programs targeting attention biases (CBMT-A). CBMT-A protocols use a dot-probe task to systematically redirect attention away from threatening stimuli. The three studies identified as testing CBMT-A in adolescents with SA all used pictoral stimuli expressing angry or neutral faces, with the first two using faces of adults (Pergamin-Hight et al., 2016) (Fitzgerald et al., 2016), and the most recent study using adolescent faces (Ollendick et al., 2019).

Pergamin-Hight and colleagues (Pergamin-Hight et al., 2016) compared CBMT-A and AC and found that both conditions resulted in significant but similar reductions in threat biases and in clinician-rated and self-reported SA. Taken together, these results provided no evidence of CBMT-A efficacy, relative to ACTr, in the overall sample. Interestingly, age (but not baseline attentional bias score) was found to moderate the SA outcomes, with youth 13 or older showing a greater response to CMBT-A than ACTr on self-reported but not clinician ratings of SA.

In another comparison of CBMT-A and ACTr, Fitzgerald and colleagues (Fitzgerald et al., 2016) observed an overall reduction in SA over time. However, CBMT-A training did not alter attention bias to threat or SA symptoms more than ACTr. Given that the sessions were delivered once weekly for 4 weeks, authors hypothesized that CBMT-A sessions may have been too infrequent for participants to implicitly learn the contingencies between avoiding threat stimuli and directing attention to neutral stimuli. The inadequate training dose may explain the lack of reduction in attention bias (the proposed causal mechanism of change), and the related absence of impact on SA measures.

Ollendick and colleagues (Ollendick et al., 2019) also compared CBMT-A to ACTr. The CBMT-A consisted of a modified dot probe task that used a set of pictorial stimuli expressing angry or neutral adolescent faces in an attempt to increase the salience of the training stimuli for this population. A higher dose of the intervention was delivered than in the Fitzgerald et al. (2016) study, such that training was delivered twice a week over a 5-week period for a total of 10 sessions. In this way, the participants were provided with enough trials to facilitate change in attentional processes away from threat. No changes in threat biases on the dot-probe task or clinician ratings were observed in either condition. Small and comparable reduction of SA symptoms were found in both conditions based on parent and adolescent reports. Very few of the youths were diagnosis free following treatment (between 10 and 14%), and no significant differences were observed between the CBMT-A and ACTr conditions.

CBMT Combined Interpretation and Attention Bias Studies

Three additional studies tested the effects of CBMT programs targeting both attention and interpretation biases (Combination). Sportel and colleagues (Sportel et al., 2013) randomly assigned 240 participants with SAD to an internet-based combined CBMT package, a school-based cognitive behavioral group training (CBGT), and a control group (n = 70 per group). CBMT consisted of a 20-session, at home, internet-delivered training that included tasks to modify interpretation (9 sessions) and attention bias (8 sessions); CBGT consisted of 10 weekly 1.5-hour sessions that were delivered in small groups by a licensed psychologist after school hours in the school building, with homework assignments; and the control group received no training. Participants were assessed for threat-related automatic associations, interpretation bias, SA levels, and presence of SAD before and after the intervention and at 6- and 12-month follow-ups. Findings showed an overall decrease in self-reported SA symptom severity from pretest to posttest, and from pretest to 12-month follow-up. At 6-month follow-up, CBGT resulted in lower SA than the control condition, while for CBM, this effect was only trend-significant, with effect sizes in the small to moderate range. At 12-month follow-up, the control condition showed a similar reduction in SA symptoms compared with both active conditions. At post treatment, CBMT was demonstrated to modify both targeted attention and interpretation biases. From post-test to 12-month follow-up, the CBMT group showed a stronger decrease in negative automatic associations than both the CBGT and the no-intervention control group. However, between post-test and 12-month follow-up, CBMT did not have a more favorable effect on self-reported SA than the control condition. Since only a small fraction of the participants continued to meet diagnostic criteria for SAD at 12-month follow-up, changes over time or differences between groups could not be statistically tested.

The same research group published two-year follow-up data from the same trial (de Hullu et al., 2017). In total, 121 of 240 participants (50%) completed the two-year follow-up assessment. At this follow-up, the CBMT intervention (but not the CBGT) showed a robust beneficial effect on interpretation bias as indexed with the task that was most similar to the modification procedure (i.e., the recognition task), meaning that adolescents in the CBMT condition reported larger changes during the recognition task than youth in the control condition. However, an equivalent beneficial effect was absent when interpretation bias was measured with a near-transfer measure that diverged more from the tasks used in the CBMT intervention. Additionally, both active treatment conditions were associated with significant decrease in SA observed after one year: CBMT and CBGT both showed large effects on SA (d=0.79 and 0.86, respectively), whereas changes in SA for the control condition were moderate (d=0.63). Since only a small fraction of the participants continued to meet diagnostic criteria for SAD at 2 years (5.9%), changes over time or differences between groups could not be statistically tested. The transience of improvements in automatic threat associations (no longer evident at two-year follow-up) and the absence of near-transfer effects on interpretation bias led the researchers to hypothesize that the training intervention was not sufficiently strong and intensive, which could explain the failure of CBMT to have a meaningful impact on adolescents’ vulnerability for SA after two years.

Lisk and colleagues (Lisk et al., 2018) enrolled 19 adolescents with elevated SA and measured symptoms and cognitive biases before and after 2 weeks of a multi-session, combined bias CBMT package. The protocol used a mix of training techniques and stimuli and was delivered in a school setting. Training sessions 1–4 were carried out on the same week and involved participants completing one interpretation training task per day from the training program. Training sessions 5–9 were carried out the second week and involved completing one attention training task per day. Significant reductions were observed in interpretation biases and SA levels. There was a significant correlation between interpretation bias change and SA symptom change.

Discussion

Summary of findings

The goal of this systematic review was to illustrate and evaluate the existing approaches to remediate cognitive biases in adolescents with SA.

Three studies evaluating CBMT-I in adolescents with SA reported improvements ininterpretation bias, with effect sizes ranging from small (Klein et al., 2018) to large (Fu et al., 2013; Reuland and Teachman, 2014). Two out of three of the studies reported an improvement in SA symptoms after CBMT-I (Klein et al., 2018; Reuland and Teachman, 2014), However, the small sample size (n = 18) in the Reuland and Teachman study and the inclusion of adolescents with mild intellectual disability in Klein and colleagues’s sample are factors that limit the generalizability of the findings.

On the other hand, three studies testing CBMT-A in adolescents with SA (Fitzgerald et al., 2016; Ollendick et al., 2019; Pergamin-Hight et al., 2016) failed to find significant effects on cognitive and clinical outcomes. Whether findings from these CBMT-A studies are truly comparable is debatable, due to the high degree of methodological heterogeneity (e.g. CBMT-A format including intensity and duration, participant age, assessment measures). In particular, patient age may be an important factor moderating the association between CBMT-A learning and treatment response, as suggested by the findings of Pergamin-Hight and colleagues (Pergamin-Hight et al., 2016) and others(Cristea et al., 2015a; Dudeney et al., 2015; Mogoaşe et al., 2014; Price et al., 2016). Although some meta-analyses indicate no effect of age on clinical outcomes (Hakamata et al., 2010; Heeren et al., 2015), it is notable that these meta-analyses are based on adult samples.

In summary, CBMT-A in its current form does not seem ready for implementation as a tool for SA treatment in clinical care. However, the presence of a possible efficacy signal warrants identifying the limitations of current CBMT administration formats and developing and integrating CBMT principles into more efficacious treatment approaches. Piloting of modified CBMT-A treatment protocols is needed. Further, while CBMT-I seems to engage better with its cognitive target than CBMT-A, more research is needed to determine if improvements in interpretation bias can serve as the mechanism of change to reduce SA symptoms in adolescents.

Integrated packages that delivered training exercises targeting both attention and interpretation biases (de Hullu et al., 2017; Lisk et al., 2018; Sportel et al., 2013) demonstrated significant reduction in attention and interpretation biases, and maintenance of these improvements up to two years after treatment (de Hullu et al., 2017). Findings from Lisk and colleagues suggest a possible effect on SA severity; however, the study design (case series) limits the generalizability of this finding. More studies are needed to elucidate if the concurrent delivery of exercises targeting attention and interpretation biases is associated with greater benefits compared to single-target approaches.

Limitations

The present review has some limitations. First, we reviewed studies that exclusively recruited patients with SA. Therefore, we may have missed studies that included a mixed diagnostic sample of patients with anxiety disorders, some having SA. At the same time, the dearth of RCTs studying CBMT in pure SAD samples may reflect the practical challenges inherent in the conduct of CBMT trials, which can be long and intensive in nature, making recruitment and retention of participants from a restrictive eligibility pool (e.g. a single diagnostic category) particularly difficult. Of note, the theoretical conceptualization of cognitive biases as a cross-diagnostic anxiety symptom dimension supported by considerable literature (Cristea et al., 2015b; Hallion and Ruscio, 2011) has contributed to the inclusion of SA patients with other anxiety disorders in treatment studies. However, some aspects of illness known to distinguish SA from other anxiety disorders could be potential mediators or moderators of response to CBMT(Bruch et al., 1993; Epkins, 1996). Additionally, given the heterogeneity in cognitive and clinical assessments used, and the fact that many studies did not report significance levels and effect sizes for non-significant outcomes, we are unable to compare outcomes across studies and compute an overall effect size across studies that includes both significant and non-significant outcomes. Future studies should consistently adopt the same set of behavioral assessments and repeat such tests at similar time points in order to facilitate comparisons across datasets. Fourth, because medication regimens, duration of illness, involvement in other therapies (e.g., CBT), and medical/psychiatric comorbidities were reported inconsistently across studies, we were not able to elucidate the role played by these critical factors on response to CBMT. Finally, systematic assessment of study quality was outside the scope of this brief review; however, it is possible that variability in study quality may have influenced findings.

Future directions

We have identified several potential areas of innovation for CBMT in SA. Existing CBMT for adolescent SA targets cognitive processes using fixed repetitive parameters, and training involves repetition of static computerized tasks. This approach does not take advantage of the neuroplastic potential of the circuits underlying social information processing to maximally promote adaptive learning and reinforce specific processing styles and social behaviors. This approach may also contribute to task disengagement via nonspecific factors such as levels of interest, boredom, and motivation. The current mode of CBMT tasks is akin to asking youths to repetitively play the same videogame without ever moving up in “levels.” This “leveling up” process could better foster both neuroplastic change and treatment engagement.

Existing CBMT exclusively targets cognitive biases, but it does not train cognitive control mechanisms. In other words, the focus is on correcting a dysfunction rather than on enhancing compensatory processes. Studies demonstrate that increasing general control processes with training is associated with reduction in anxiety and stress as well as improvement in mental health outcomes in various neuropsychiatric populations (Hoorelbeke et al., 2016, 2015; Wiers et al., 2013). We advocate for pairing the training of attention and interpretation biases with executive control training exercises that use emotional stimuli to target set switching, inhibitory control, and complex working memory.

Finally, current CBMT does not improve social skills and real-world functioning. To date, trials of CBMT in adolescents with SA have not demonstrated substantial improvements in social skills or functioning (Cristea et al., 2015b). The ability to “behave as desired” in social situations is perhaps the primary reason most individuals with SA seek treatment and is of critical importance to target social fears and social deficits with CBMT(Beidel et al., 2010). If the efficacy of CBMT as a stand-alone treatment for SA remains inconsistent or even ambiguous, an alternative approach is to explore the benefit of CBMT as an adjunct or augmentation strategy to enhance other evidence-based psychological treatments, such as cognitive-behavioral group therapy (CBGT). CBGT for SA involves learning specific cognitive strategies to set goals, engage in real-world activities, and navigate social situations. Pernicious cognitive biases may interfere with CBGT uptake, for example by preventing corrective learning of cognitive distortions in the context of exposure to social situations. Theoretically, CBMT may attenuate attention and interpretation biases, which could enable youths to more easily overcome avoidance and facilitate the learning and practice of cognitive and behavioral strategies. This pairing of CBMT with CBGT may therefore be more likely to be associated with meaningful improvements in aspects of psychosocial, interpersonal, and occupational functioning versus CBMT alone. While an augmentation study conducted in children with generalized, separation or SA disorders revealed no additional benefit from combining CBGT and CBMT in this population (Salum et al., 2018), the only RCT examining the augmenting effects of CBMT in adult patients with SA showed greater reductions in clinician-rated SA symptoms posttreatment, with effects maintained at 3-month follow-up (Lazarov et al., 2017). In summary, evidence supports the hypothesis that CBMT and CBGT could complement each other and produce synergistic effects by targeting distinct mechanisms underlying SA.

Conclusions

Novel treatment development is needed for teens with SA, the majority of whom do not receive evidence-based treatments or do not respond to these treatments (Ginsburg et al., 2014, 2011; Higa-McMillan et al., 2016). While treatment development research has traditionally focused on clinic-based interventions, technology-driven interventions such as CBMT may help extend the reach of treatment “beyond the clinic” and/or augment CBGT to boost response rates. Many adolescents indicate preference for technology-driven interventions over traditional clinic-based therapy (Spence et al., 2017). This appeal may be especially true for teenagers with SA who are anxious about participating in therapy in the clinic setting. It is crucial that novel treatments that are technology-driven are developed and rigorously evaluated.

Findings from this review provide some support for CBMT as a beneficial approach for targeting cognitive biases in youth with SA, especially if interpretation bias is targeted (Fu et al., 2013, Klein et al., 2018; Reuland and Teachman, 2014). Overall, we found some degree of domain specificity between the remediation target (attention vs interpretation biases) and improvements in corresponding outcomes. However, this tentative conclusion is based on a limited number of studies including some with methodological shortcomings. Therefore, further investigations with larger sample sizes and use of similar outcome measures across studies will determine the potential benefits of CBMT for treating children and adolescents with SA. Similarly to what has been reported in meta-analyses of CBMT for adults and children (Cristea et al., 2015a; Heeren et al., 2015), effect sizes varied considerably across individual studies, with an average falling in the moderate range, suggesting that adolescents with SA can respond to treatment at or above the level reported in children with SA. Although these findings should be interpreted with caution, given the few studies to date having examined CBMT in SA samples.

There are several potentially promising avenues for next-stage CBMT research in SA. Future research should attempt to develop adaptive CBMT procedures that optimize learning, integrate cognitive bias modification exercises in group-based and real-world settings, and identify predictors of treatment response, including potential subdiagnostic differences and cognitive endophenotypes within SA. Research is also needed to understand the neural correlates of response to CBMT, which may help identify future targets for intervention. Once potential predictors, mediators and moderators of response are identified, such elements will guide the refinement of mechanistically informed CBMT protocols and make it possible to personalize them to individual strengths and weaknesses, thus maximizing the efficacy of this promising therapeutic approach.

HIGHLIGHTS.

• Cognitive models of social anxiety (SA) have identified attention and interpretation biases that are present at various stages of social information processing in adolescents

• Cognitive Bias Modification Training (CBMT) is a treatment that leverages repetitive implicit learning mechanisms to reallocate attention away from threat-relevant information and/or to reduce negative interpretations of ambiguous social situations through repeated computerized training trials

• Our review of published studies that delivered CBMT to adolescents with SA found that studies of CBMT for attention bias did not find significant effects on cognitive and clinical outcomes. However, studies of CBMT for interpretation bias showed some improvement in interpretation bias. The combination of CBMT for attention and interpretation biases appeared promising in reducing both attention and interpretation biases.

• Further research is needed to develop CBMT procedures that optimize learning in group and real-world settings and to identify predictors of treatment response.