Self-efficacy effects in combined CBT-attention training for pediatric anxiety☆ Authors: Lili A. Massac, Meghan E. Byrne, Yair Bar-Haim, Daniel S. Pine, Krystal M. Lewis Published: 2026 ABSTRACT Augmenting cognitive behavioral therapy (CBT) with attention bias modification training (ABMT) has shown promising yet inconsistent results. Thus, research is needed to identify potential moderators of treatment outcomes. Self-efficacy beliefs have been linked to CBT treatment response and potentially to the cognitive mechanisms which influence engagement with ABMT. This study examined the relation between self-efficacy and treatment response among pediatric patients (N = 111) undergoing CBT for pediatric anxiety supplemented with either an active or sham ABMT. Moderated-mediation analyses revealed a significant interaction between child-reported self-efficacy and ABMT condition in predicting early treatment changes in parent-reported depressive symptoms (a3 = −0.18, SE = 0.05, p < .01), indicating that the association between active ABMT and early depressive symptom change was conditional on self-efficacy. This greater early treatment change related to better overall treatment outcomes (b = 0.09, SE = 0.02, t = 4.33, p < .001). No associations emerged for clinician, parent, or child-reported anxiety (ps > 0.4). Findings that self-efficacy prior to CBT exposures influences ABMT responsiveness highlights high self-efficacy as a potential correlate of change in mood symptoms. Current findings elucidate the complex, yet significant role of self-efficacy in mood-related treatment response and the need for further investigation into the relations between self-efficacy, attentional biases, and treatment outcomes among multiple reporters. Introduction Background Cognitive-behavioral therapy (CBT) is a well-supported, evidence-based psychological intervention for pediatric anxiety disorders [1], [2]. However, as many as 50% of youth do not exhibit clinically significant improvement following treatment [3], [4], and approximately 28% discontinue treatment before finishing [5]. Given the high dropout rates and suboptimal treatment responsivity, additional research is needed to improve the efficacy of CBT and to maximize therapeutic effect early in treatment. One frequent target of novel therapeutics - negative attentional bias - is thought to play a key role in both the development and maintenance of anxiety and predicts youth’s poorer response to CBT [6], [7]. Augmenting CBT with attention bias modification training (ABMT) may enhance therapeutic efficacy [8]. While CBT may target top-down processes through explicit instruction and practice [9], [10], ABMT aims to reduce patients’ implicit negative biases by training attention away from threat-related cues [11], [12]. Indeed, one study showed that as few as four, once-weekly sessions of dot-probe-based ABMT, can produce significant reductions in pediatric anxiety symptoms and severity relative to control conditions [13]. However, despite these promising results from early studies, several recent studies of CBT + ABMT for anxiety disorders report mixed efficacy [11], [14], [15] or lack of replication [16]. Moreover, several meta-analytic reviews have reported modest effect sizes or inconsistent results [17], [18]. Such findings highlight the need for clarification of the circumstances under which ABMT leads to therapeutic benefit. Identifying factors that contribute to individual differences in the efficacy of CBT + ABMT may help clarify for whom ABMT is most beneficial and reveal targets for supplementary interventions. Several studies have suggested that top-down cognitive mechanisms influence individual differences in ABMT response [19], [20]. However, less is known about socio-cognitive or motivational mechanisms, which are also thought to contribute to attentional processes and anxiety symptom severity [21], [22]. Motivational processes are particularly relevant for treatment effect given their role in treatment and task engagement [23], where engagement has been shown to influence the effectiveness of cognitive training [24]. One candidate socio-cognitive process is self-efficacy, which describes ones’ perceived ability to achieve goals and exert control over one's environment, emotions, and actions [25]. Self-efficacy has been shown to play a role in anxiety severity and treatment outcomes in both youth and adults, where lower trait self-efficacy has been associated with more severe symptoms of anxiety [26] and is predictive of treatment outcomes [27], [28]. Several studies have shown that changes in self-efficacy predicted changes in anxiety symptom severity [29], [30], [31]. More recently, Paersch, et al. [30] demonstrated that higher baseline general self-efficacy was associated with greater early symptom change during CBT for anxiety, extending its relevance to the initial phases of treatment response. Furthermore, CBT + ABMT aims to target both implicit and explicit mechanisms of anxiety; self-efficacy is a top-down process that influences both these underlying mechanisms [32]. For example, Karademas, et al. [33] demonstrated that, among adults, generalized self-efficacy influenced attentional biases in the processing of threatening information, such that lower self-efficacy was associated with a greater bias toward threat-related stimuli. Extending these findings to developmental populations, Cuder, et al. [34] found that among children with math anxiety, trait math self-efficacy was associated with distinct patterns of attentional bias, particularly avoidance of anxiety-evoking stimuli. Taken together, these studies indicate that trait self-efficacy can influence attentional biases and related behavioral responses across both adult and child populations. Converging evidence also suggests that task specific self-efficacy influences attentional control and may contribute to emotional learning processes [35], both of which are relevant to the proposed mechanisms underlying ABMT. Beyond its relationship with attention and emotional learning, self-efficacy may also influence treatment response through its effects on motivation, engagement, and performance [30], [36], [37], potentially enabling individuals with higher self-efficacy to engage more effectively with, and, hence, derive greater benefit from, the cognitively demanding components of the active ABMT condition. To our knowledge, no study to date has investigated the role of self-efficacy on interventions targeting implicit mechanisms of anxiety, such as ABMT. The present study The current study reports secondary exploratory findings from a larger randomized controlled trial investigating the efficacy of a gamified ABMT as an adjunct to CBT for anxiety among a pediatric sample [14], [38]. Specifically, this secondary analysis investigated the effects of general trait self-efficacy on anxiety symptom changes amongst pediatric patients receiving CBT augmented with computer-delivered gamified ABMT. Efficacy of the ABMT training is discussed in detail elsewhere [14]. In short, there were no significant main effects of the ABMT on treatment outcomes; specifically, active ABMT did not produce greater reductions than the control condition on the Pediatric Anxiety Rating Scale (PARS). Past studies have reported efficacy in as early as four training sessions [13], and a need exists to identify predictors of early clinical change. Because early symptom improvement is a robust predictor of subsequent treatment outcomes, understanding factors that contribute to early response may help identify youth most likely to ultimately benefit from CBT + ABMT. Of note, recent work has demonstrated that higher baseline self-efficacy predicts greater early symptom improvement during CBT for anxiety disorders, suggesting that self-efficacy may be particularly relevant for early treatment change [30]. Furthermore, because self-efficacy may facilitate engagement with the cognitively demanding components of the active condition of the ABMT, we expect that any moderating effects of self-efficacy could emerge during the early stages of treatment before contributing to overall clinical improvement. Accordingly, we examined whether early treatment change statistically mediated the association between ABMT condition (active vs. sham) and end-of-treatment symptom improvement and whether this pathway was moderated by baseline self-efficacy. Specifically, we tested the hypotheses that (1) early reductions in clinical outcome measures (henceforth referred to as early treatment change) would mediate the relation between ABMT condition (active vs. sham) and end of treatment reductions in clinical symptoms; and (2) the relation between ABMT and early treatment change would be moderated by baseline self-efficacy, such that, amongst those receiving the active form of ABMT, those with higher baseline self-efficacy exhibit greater child, parent, and clinician reported symptom change, both early in treatment (CBT session 8) and at the end of treatment (CBT session 12). Method Participants Data was collected as part of a larger randomized controlled trial examining the augmenting effect of ABMT combined with CBT on the treatment of anxiety symptoms among youth with anxiety disorders. One hundred and eleven youth with a primary anxiety disorder diagnosis (Generalized Anxiety Disorder, Social Anxiety Disorder, Specific Phobia, and/or Separation Anxiety Disorder) were randomized to receive either active (n = 55) or sham (n = 56) ABMT in addition to manualized CBT. Current major depressive disorder, obsessive-compulsive disorder, psychosis or bipolar disorder, or posttraumatic stress disorder were exclusionary. Psychiatric disorder diagnosis was established by a semi-structured clinical interview, the Kiddie Schedule for Affective Disorders and Schizophrenia for School-Aged Children-Present and Lifetime version (KSADS, [39]), administered by a trained clinician. All patients had an intelligence quotient (IQ) of > 70, as assessed by the Wechsler Abbreviated Intelligence Scale version 2 (WASI II; Wechsler & PsychCorp, 2011 [40]), and had no current use of psychoactive substances of medication. Recruitment, enrollment, and detailed exclusion criteria are outlined elsewhere [14], [41]. All study procedures were approved by the National Institutes of Mental Health (NIMH) Institutional Review Board. Data for those who consented to share are publicly available on OSF at https://doi.org/10.17605/OSF.IO/YRK95. Given not all participants consented to share their data, exact replication of analyses with the open dataset will not be possible. Procedure Measures Self-efficacy The Self-efficacy Questionnaire for Children (SEQ-C) is a 24-item self-reported measure of children’s perceptions of their own self-efficacy [42]. The SEQ-C assesses three domains of perceived self-efficacy: 1) emotional self-efficacy: one’s ability to self-regulate and manage unpleasant emotions (i.e. “How well can you control your feelings?”); 2) social self-efficacy: one’s ability to get along with other peers and manage social challenges (i.e. “How well can you work in harmony with your classmates?”); and 3) academic self-efficacy: one’s ability to succeed academically and fulfill academic expectations (i.e. “How well do you succeed in understanding all subjects in school?”). For each domain, there are eight items which are rated on a 5-point Likert-type scale ranging from 1 = not at all to 5 = very well. Scores are summed for each domain and can be summed across domains to yield a total self-efficacy score. Higher scores reflect higher levels of perceived self-efficacy. The instrument shows good internal consistency (Cronbach’s alpha =.891) in the current sample. The SEQ-C was administered prior to exposures at CBT week 3. Depression symptoms The Short Mood and Feeling Questionnaire Parent and Child (MFQ-P, MFQ-C, respectively) is a 13-item parent and self-reported measure of depressive symptoms in children and adolescents (8–16 years old; [43]), evaluated in relation to the two weeks prior to assessment. Items, which include statements like “[My child (for parent) or I (for self)] didn't enjoy anything at all” or “[My child (for parent) or I (for self)] felt lonely,”] are rated on a 3-point scale with the following possible responses “Not True” (scores as 0), “Sometimes True” (scores as 1), “True” (scores as 2); the overall score ranges from 0 to 39. Higher scores reflect higher levels of depression. The instrument demonstrates good to excellent internal consistency for both parent and child reports in the current sample (Cronbach’s alphas =.849 −.913, respectively). The MFQ-P and MFQ-C were administered prior to exposures (CBT 3), at the early treatment time point (CBT 8), and at the end of treatment (CBT 12). Anxiety symptoms Both the Pediatric Anxiety Rating Scale (PARS) and Screen for Child Anxiety Related Emotional Disorders (SCARED) were administered to evaluate anxiety. The PARS is a 50-item clinician-rated assessment of anxiety symptom severity, based on parent and child report [44]. The presence or absence of symptoms are evaluated across six symptom domains: Social Interactions or Performance Situations, Separation, Generalized, Specific Phobia, Physical Signs and Symptoms, and Other. Symptoms are then rated on a 6-point scale on seven subscales of severity: (1) number of symptoms (none to more than 10), (2) frequency (none to several hours per day), (3) severity of distress associated with anxiety symptoms (none to extreme), (4) severity of physical symptoms (none to extreme), (5) avoidance (none to extreme), (6) interference at home (no interference to totally or almost totally unable to function at home), and (7) interference out of home (no interference to totally or almost totally unable to function out of home such as with peers or at school). Scores from subscales 2, 3, 5, 6 and 7 are then summed to create a composite score ranging from 0 to 25. Higher scores reflect greater levels of anxiety. The PARS demonstrates acceptable internal consistency (Cronbach’s alpha =.790) in the current sample. The SCARED is 41-item child and parent reported measure of anxiety-related symptoms and behaviors [45]. Each item is rated on a 3-point scale (ranging from 0 = “Not True or Hardly Ever True,” to 3 = “Very True or Often True”), with overall scores ranging from 0 to 82. The scale assesses anxiety across five different subscales: generalized anxiety, school avoidance, social anxiety, separation anxiety and panic disorder symptoms. Higher scores reflect greater anxiety severity. The SCARED Parent and Child measures demonstrate good to excellent internal consistency (Cronbach’s alphas =.886 −.934, respectively) in the current sample. The SCARED was administered prior to exposures (CBT 3), at the early treatment time point (CBT 8), and at the end of the treatment (CBT 12). Treatment Cognitive behavioral therapy (CBT) All participants received 12 weekly sessions of exposure-based CBT administered by a licensed clinical psychologist who specializes in the treatment of anxiety in youth. Each session lasted between 40 and 60 min, and youth were assigned additional home practice as an extension of the exposures completed during the sessions. The treatment course begins with an introduction to CBT, psychoeducation, and self-monitoring. After the third week of treatment, children are instructed to complete their first out-of-session exposures; starting at week four, patients begin in-session exposures and cognitive restructuring exercises. All therapists were trained on the CBT protocol during a one-day training by Dr. Wendy K. Silverman at the Yale Child Study Center. Clinicians also received bi-weekly reviews with the study team to discuss treatment progress as well as additional supervision by Dr. Wendy K. Silverman, as needed. Attention bias modification training Patients were randomized to receive nine weeks (one session per week) of either active or sham ABMT prior to the beginning of in-session exposures starting at CBT week 4. All training (both active and sham) was completed immediately prior to CBT sessions on a laptop (1440 ×900 display) and administered by trained research staff. Specific details related to the gamified attention bias modification training have been described in depth elsewhere [14], [41]. In short, the current ABMT utilized a modified dot-probe task and visual search task. The dot-probe component included 80-trial blocks where patients viewed face pairs with either a threatening and neutral face, or two neutral faces for 500 ms (ms). Following the presentation of the faces, a left or right facing arrow probe (< or >) was presented in the location of one of the faces, prompting patients to respond with either a left or right button press corresponding with the direction of the arrow presented. Between each trial, a fixation cross was presented for 500 ms. Importantly, in the active condition, the arrow appeared behind the neutral face for 100% of the trials; in the sham condition, arrow probes appeared equally between threatening and neutral faces. Faces were selected from the TAU-NIMH Attention Bias Measurement Toolbox-Picture Set A. The visual search component included two trials of a computerized game where participants were prompted to find hidden numbers within visual scenes in a limited timeframe. The scenes displayed six children engaged in school activities. In the active training, four of the children display neutral faces, while two display threatening faces, and the hidden numbers are placed closer to neutral faces. In the sham condition, all six characters display neutral emotions. ABMT was gamified for both conditions as patients earned points (up to 80) for each correct response in the dot probe task, with additional points for quicker responses. Points could be redeemed in the visual search component in the form of an additional 10 s of search time. Statistical analysis Baseline characteristics Pre-exposure PARS, MFQ, SEQ, and SCARED differences between active and sham ABMT condition were examined using a t-test. Early treatment change Past literature is inconsistent in terms of the number of sessions of CBT considered to be “early” [30]. Studies have ranged anywhere from 2 to 8 weeks or later [46]. Aligning with past literature, we operationalized “early treatment” as CBT session 8 given that this time point allowed for participants to receive four sessions of ABMT. We utilize CBT 3 for the “baseline measures” as the first three sessions are dedicated to building rapport and psychoeducation. Previous studies have also shown that prior to exposures and early in treatment changes in self-efficacy are negligible [29]. To capture treatment change, we calculated change scores for each outcome measure by subtracting the scores reported at CBT 3 (prior to exposures) from those reported at CBT 8 (early treatment range). SEQ and treatment change Mediation and moderated mediation models with bias-corrected bootstrap confidence intervals (CI) were conducted using the Preacher and Hayes PROCESS macro v4.1 [47] for SPSS. Models were run separately for each outcome variable. All models tested treatment outcomes at CBT 12 as the dependent variable, and ABMT condition (active or sham) as the independent variable. Early treatment change was included as a mediator, and pre-exposure self-efficacy was included as a moderator of the pathway between ABMT Condition and early treatment change (i.e., a pathway). Early treatment change was selected as the mediator because ABMT-related symptom improvements have been observed after relatively few training sessions (13), allowing evaluation of whether early clinical response statistically accounted for the association between treatment condition and end-of-treatment outcomes. All continuous predictors were mean centered prior to analysis. All mediation and moderated mediation models were adjusted for age at baseline, sex, as well as baseline scores on each symptom measure (baseline scores on PARS, SCARED-Parent, SCARED-Child, MFQ-P, and MFQ-C). First, mediation analyses were conducted to test the individual pathways from ABMT Condition to early treatment change (a pathway), and from ABMT Condition to overall treatment change (b pathway), as well as the overall indirect mediation through early treatment change with 10,000 resamples to estimate the indirect effect (ab pathway) and the corresponding 95% bias-corrected confidence intervals (CI). Then, moderated mediation analyses were conducted testing whether self-efficacy moderated the association between ABMT Condition and early treatment change (a pathway). The index of moderated mediation was examined to determine whether the indirect effect of ABMT Condition on treatment outcomes through early treatment outcomes varied as a function of baseline self-efficacy. Results Participant characteristics Demographic, pre-exposure, and post-exposure clinical characteristics are summarized in Table 1. Baseline correlations between SEQ and all clinical variables and age are summarized in Fig. 1. There were no baseline differences between active and sham groups in pre-exposure SEQ (t(106.49) = −1.28, p = .20), SCARED-Child (t(104.46) = 0.21, p = .84), SCARED-Parent (t(104.1) = −1.00, p = .27), PARS (t(104) = −1.05, p = .14), MFQ-P (t(98.16) = −1.31, p = .20), and MFQ-C ratings (t(104.37) = 0.82, p = .45). There were no significant differences between conditions in pre-exposure age (t(106.5) = 0.91, p = .37) or sex (χ2 = 0, p = .99). Ninety of the 111 randomized participants completed treatment, defined as completing CBT session 12. However, only 8 or fewer participants (<7.2%) were missing any main end-of-treatment outcome measure. Independent-samples t-tests for each clinical measure indicated that there were no significant differences in early treatment response between participants who completed CBT session 12 and those who did not (ps >.13). Results did not change when using a multiple imputation technique for the 8 missing data points. As such, missing data were handled using listwise deletion. Independent-samples t-tests for each clinical measure indicated that there were no significant differences in early treatment response between participants who completed treatment and those who dropped out prior to completion (ps >.13). For all participants, early treatment PARS scores (i.e. CBT 8) significantly decreased relative to pre-exposure scores (change in PARS score: mean difference = −2.23, SD = 3.63; t = −6.27, df = 103, p < .001). There were no significant differences in PARS ((t(98.05) = 0.11, p = .91), MFQ-P (t(101.24) = 1.51, p = .13), MFQ-C (t(97.1) = −1.36, p = .18), SCARED-Parent (t(99.35) = −0.84, p = .40), or SCARED-Child ((t(93.82) = −0.97, p = .33) change scores across ABMT condition.Table 1Sample Characteristics and Descriptive Statistics of Clinical Variables.Total Anxiety Sample(N = 111)Active ABMT(N = 55)Sham ABMT(N = 56)Demographic VariablesAge (year) [M(SD)]12.86 (2.71)13.08 (2.65)12.65 (2.76)Sex, N (% Female)67 (60.36)33 (60)34 (60.71)Race, N (%)---American Indian/Alaskan Native1 (0.90)1 (1.82)0 (0.00)Asian5 (4.50)1 (1.82)4 (7.14)Black or African American3 (2.70)2 (3.64)1 (1.79)Multiple Races21 (18.92)11 (20.00)10 (17.86)White78 (70.27)38 (69.09)40 (71.43)Unknown3 (2.70)2 (3.64)1 (1.79)Ethnicity, N (% Hispanic/Latine)15 (13.51)9 (16.36)6 (10.71)Clinical VariablesPre-exposure SEQ-C [M(SD)]73.50 (14.23)71.76 (13.13)75.21 (15.15)Pre-exposure PARS [M(SD)]14.22 (3.95)13.74 (4.17)14.68 (3.71)Early-treatment PARS [M(SD)]11.90 (3.93)11.43 (4.10)12.38 (3.73)End-of-treatment PARS [M(SD)]9.88 (4.48)9.55 (4.92)10.24 (3.99)Pre-exposure MFQ-Parent [M(SD)]6.04 (4.82)5.44 (4.07)6.64 (5.44)Early-treatment MFQ-Parent [M(SD)]5.47 (4.52)5.35 (4.42)5.58 (4.64)End-of-treatment MFQ-Parent [M(SD)]4.99 (4.75)4.27 (4.29)5.75 (5.13)Pre-exposure MFQ-Child [M(SD)]7.76 (6.38)8.25 (6.40)7.24 (6.38)Early-treatment MFQ-Child [M(SD)]7.19 (6.12)7.18 (5.66)7.20 (6.57)End-of-treatment MFQ-Child [M(SD)]6.86 (6.11)6.65 (5.65)7.08 (6.63)Pre-exposure SCARED-Parent [M(SD)]29.80 (11.81)28.55 (10.32)31.02 (13.10)Early-treatment SCARED-Parent [M(SD)]26.39 (11.13)24.38 (9.58)28.36 (12.24)End-of-treatment SCARED-Parent [M(SD)]24.78 (11.36)22.49 (10.11)27.12 (12.16)Pre-exposure SCARED-Child [M(SD)]30.84 (15.71)31.15 (14.25)30.53 (17.17)Early-treatment SCARED-Child [M(SD)]28.41 (15.34)27.29 (15.28)29.43 (15.47)End-of-treatment SCARED-Child [M(SD)]25.81 (16.05)25.88 (15.06)25.74 (17.14)Note: ABMT = Attention Bias Modification Training; Pre-exposure = CBT 3; Early-treatment = CBT 8, HV = healthy volunteers; M = mean; SD = standard deviation; PARS = Pediatric Anxiety Rating Scale; SCARED = Screen for Anxiety and Related Disorders; MFQ = Mood and Feelings Questionnaire.Fig. 1Correlation matrix between pre-exposure self-efficacy (SEQ) and all pre-exposure clinical variables and participant age. *p < .05; **p < .01. Table 1Sample Characteristics and Descriptive Statistics of Clinical Variables.Total Anxiety Sample(N = 111)Active ABMT(N = 55)Sham ABMT(N = 56)Demographic VariablesAge (year) [M(SD)]12.86 (2.71)13.08 (2.65)12.65 (2.76)Sex, N (% Female)67 (60.36)33 (60)34 (60.71)Race, N (%)---American Indian/Alaskan Native1 (0.90)1 (1.82)0 (0.00)Asian5 (4.50)1 (1.82)4 (7.14)Black or African American3 (2.70)2 (3.64)1 (1.79)Multiple Races21 (18.92)11 (20.00)10 (17.86)White78 (70.27)38 (69.09)40 (71.43)Unknown3 (2.70)2 (3.64)1 (1.79)Ethnicity, N (% Hispanic/Latine)15 (13.51)9 (16.36)6 (10.71)Clinical VariablesPre-exposure SEQ-C [M(SD)]73.50 (14.23)71.76 (13.13)75.21 (15.15)Pre-exposure PARS [M(SD)]14.22 (3.95)13.74 (4.17)14.68 (3.71)Early-treatment PARS [M(SD)]11.90 (3.93)11.43 (4.10)12.38 (3.73)End-of-treatment PARS [M(SD)]9.88 (4.48)9.55 (4.92)10.24 (3.99)Pre-exposure MFQ-Parent [M(SD)]6.04 (4.82)5.44 (4.07)6.64 (5.44)Early-treatment MFQ-Parent [M(SD)]5.47 (4.52)5.35 (4.42)5.58 (4.64)End-of-treatment MFQ-Parent [M(SD)]4.99 (4.75)4.27 (4.29)5.75 (5.13)Pre-exposure MFQ-Child [M(SD)]7.76 (6.38)8.25 (6.40)7.24 (6.38)Early-treatment MFQ-Child [M(SD)]7.19 (6.12)7.18 (5.66)7.20 (6.57)End-of-treatment MFQ-Child [M(SD)]6.86 (6.11)6.65 (5.65)7.08 (6.63)Pre-exposure SCARED-Parent [M(SD)]29.80 (11.81)28.55 (10.32)31.02 (13.10)Early-treatment SCARED-Parent [M(SD)]26.39 (11.13)24.38 (9.58)28.36 (12.24)End-of-treatment SCARED-Parent [M(SD)]24.78 (11.36)22.49 (10.11)27.12 (12.16)Pre-exposure SCARED-Child [M(SD)]30.84 (15.71)31.15 (14.25)30.53 (17.17)Early-treatment SCARED-Child [M(SD)]28.41 (15.34)27.29 (15.28)29.43 (15.47)End-of-treatment SCARED-Child [M(SD)]25.81 (16.05)25.88 (15.06)25.74 (17.14)Note: ABMT = Attention Bias Modification Training; Pre-exposure = CBT 3; Early-treatment = CBT 8, HV = healthy volunteers; M = mean; SD = standard deviation; PARS = Pediatric Anxiety Rating Scale; SCARED = Screen for Anxiety and Related Disorders; MFQ = Mood and Feelings Questionnaire. Sample Characteristics and Descriptive Statistics of Clinical Variables. Note: ABMT = Attention Bias Modification Training; Pre-exposure = CBT 3; Early-treatment = CBT 8, HV = healthy volunteers; M = mean; SD = standard deviation; PARS = Pediatric Anxiety Rating Scale; SCARED = Screen for Anxiety and Related Disorders; MFQ = Mood and Feelings Questionnaire. Correlation matrix between pre-exposure self-efficacy (SEQ) and all pre-exposure clinical variables and participant age. *p < .05; **p < .01. Mediation models A mediation analysis showed that ABMT condition was not associated with early treatment change (a pathway, ps >.11) for any symptom measure including the SCARED-Child (a = −1.96, SE = 1.49, p = .19), SCARED-Parent (a = −2.29, SE = 1.40, p = .11), PARS (a = 0.49, SE = 0.66, p = .88), MFQ-P (a = 0.63, SE = 0.71, p = .38), and MFQ-C ratings (a = −1.12, SE = 0.81, p = .17). However, early treatment changes were significantly associated with overall treatment outcomes for all the clinical measures (b pathway, ps <.001). ABMT condition did not influence the level of overall treatment change through its effects on early treatment change for any of the clinical variables (Fig. 2). The indirect effect was not significant given that the CI contained zero for all clinical outcome variables (95% CIMFQ-C = [−0.22, 0.03]; 95% CIMFQ-P = −0.06, 0.19]; 95% CIPARS= [−1.12, 0.46]; 95% CISCARED-C= [−4.04, 0.91]; 95% CISCARED-P= [−3.31, 0.22]). No direct effects were observed for ABMT Condition on overall treatment (ps >.06). These analyses were controlled for sex, age, and baseline rating on the relevant clinical outcome measure.Fig. 2Mediation model examining early treatment change (MFQ-P) as a mediator of the association between ABMT condition and treatment outcomes. The a path represents the association between ABMT condition and early MFQ-P change, the b path represents the association between early change and treatment outcomes, and the c′ path represents the direct effect of ABMT condition on treatment outcomes controlling for the mediator. The ab path represents the indirect effect, and the c path represents the total effect. Early MFQ-P treatment change was significantly associated with MFQ-P treatment outcome (b path), but the indirect effect of ABMT condition on outcomes through early change was not significant. Models controlled for age, sex, and pre-exposure (CBT 3) baseline symptom severity. Mediation model examining early treatment change (MFQ-P) as a mediator of the association between ABMT condition and treatment outcomes. The a path represents the association between ABMT condition and early MFQ-P change, the b path represents the association between early change and treatment outcomes, and the c′ path represents the direct effect of ABMT condition on treatment outcomes controlling for the mediator. The ab path represents the indirect effect, and the c path represents the total effect. Early MFQ-P treatment change was significantly associated with MFQ-P treatment outcome (b path), but the indirect effect of ABMT condition on outcomes through early change was not significant. Models controlled for age, sex, and pre-exposure (CBT 3) baseline symptom severity. Moderated mediation In the moderated mediation model (overall ΔR2 = 0.26), ABMT was significantly associated with early treatment changes in the parent reported MFQ (a1 = 13.61, SE = 4.05, p = .001), as was the interaction effect of self-efficacy and ABMT condition (a3 = −0.18, SE =0.05, p = .002, ΔR2 = 0.09, f2 = 0.12) with a small-to-medium effect size (Fig. 3). Of note, this interaction effect remained significant after applying the Benjamini-Hochberg false discovery rate (FDR) family-wise correction for multiple comparisons, with an FDR threshold of 0.05 [48]. Greater early treatment change in MFQ-P was associated with better overall treatment outcomes (b = 0.09, SE = 0.02, t = 4.33, p < .001, ΔR2 = 0.45). The index of moderated mediation was significant given the CI did not contain zero (95% CI = [−0.03, −0.01]), suggesting that the proposed indirect effect (i.e. ABMT → Early Treatment Change → Overall Treatment Outcomes) was moderated by baseline self-efficacy. Specifically, conditional indirect effects were medium at high levels of SEQ (partially standardized indirect effect = −0.20). More specifically, post-hoc simple slopes analysis revealed that amongst those receiving the active form of ABMT (b = −0.13, SE = 0.04, p = .003), but not the control condition (b = 0.04, SE = 0.03, p = .18), higher baseline self-efficacy was associated with greater early treatment change. As with the simple mediation model, these analyses were controlled for sex, age, and baseline ratings on the MFQ-P. This significant moderated-mediation effect of self-efficacy and ABMT condition was not observed for any other parent, child, or clinician rated measure.Fig. 3Moderated mediation model testing baseline self-efficacy (SEQ) as a moderator of the indirect association between ABMT condition and treatment outcomes through early treatment change (MFQ-P). Paths a₁, a₂, and a₃ represent the effects of ABMT condition, SEQ at the pre-exposure baseline, and their interaction on early MFQ-P change, respectively. The b path represents the association between early MFQ-P change and MFQ-P treatment outcome, and c′ represents the direct effect of ABMT condition on treatment outcome controlling for the mediator. The index of moderated mediation was significant (95% CI [−0.03, −0.01]), indicating that the indirect effect varied as a function of baseline self-efficacy. Models controlled for age, sex, and MFQ-P scores at the pre-exposure baseline. Moderated mediation model testing baseline self-efficacy (SEQ) as a moderator of the indirect association between ABMT condition and treatment outcomes through early treatment change (MFQ-P). Paths a₁, a₂, and a₃ represent the effects of ABMT condition, SEQ at the pre-exposure baseline, and their interaction on early MFQ-P change, respectively. The b path represents the association between early MFQ-P change and MFQ-P treatment outcome, and c′ represents the direct effect of ABMT condition on treatment outcome controlling for the mediator. The index of moderated mediation was significant (95% CI [−0.03, −0.01]), indicating that the indirect effect varied as a function of baseline self-efficacy. Models controlled for age, sex, and MFQ-P scores at the pre-exposure baseline. Discussion The present study, conducted as a secondary exploratory analysis on a larger randomized controlled trial, evaluated self-efficacy effects on ABMT-related treatment change among youth receiving CBT with ABMT for the treatment of an anxiety disorder. We hypothesized that ABMT condition relates to early treatment outcomes such that those receiving active ABMT would have greater reductions in symptom severity early in treatment and that trait self-efficacy moderates this relationship. The findings provided only partial support for this hypothesis. More specifically, while early treatment change did not independently mediate the relationship between AMBT condition and overall treatment change, the association between ABMT condition and early treatment change differed because of baseline self-efficacy; evidence suggests that these associations may be relevant for parent reports of depressive symptom severity. Amongst those receiving the active form of ABMT, higher baseline self-efficacy was associated with greater early treatment change. Of note, because the mediator (early treatment change) and outcome (overall treatment change) reflected the same symptom construct assessed at different time points, these findings should be interpreted as describing the temporal progression of treatment response rather than a distinct psychological mechanism. Nevertheless, these results extend prior work on the influence of self-efficacy in learning and in cognitive task engagement and provide preliminary evidence for self-efficacy as one factor that may influence the effectiveness of ABMT. Self-efficacy effects on attentional processes and task engagement The finding that self-efficacy influences treatment-related change is consistent with prior experimental and theoretical work associated with self-efficacy’s effects on mechanisms of attention allocation and cognitive regulation. For instance, Bouffard-Bouchard [49] demonstrated that higher perceived self-efficacy enhances task performance through its association with improved task persistence and increased attention allocation toward task-relevant stimuli. While their study experimentally induced varying levels of task-specific self-efficacy, the findings may suggest that individuals with higher self-efficacy are better able to sustain effort and engage in goal-directed behavior, potentially allowing them to allocate greater attentional resources toward task-relevant processes. In other words, it is assumed that those with higher self-efficacy can dedicate more cognitive resources to achieve goal aligned efforts [50]. In the context of the active condition of the current ABMT, the goal, albeit implicit, is to focus attention away from negative and threatening stimuli. In the modified dot-probe component of the current gamified ABMT, youth gain points when they respond correctly and quickly to the probe [38]. Hence, youth with greater self-efficacy may likely be more motivated and effective at achieving the explicit objectives of the game, potentially facilitating therapeutic gains from the implicit targets of the training. This may align with current literature given that, as outlined in, Bandura & Locke [37], this concept - that self-efficacy adaptively contributes to motivation and performance - has been consistently shown across domains in studies investigating both trait and state (including task specific or experimentally altered) self-efficacy. In sum, self-efficacy, through its effects on task-relevant attention allocation, may lead to greater efficiency of ABMT learning and, therefore, greater early treatment response from CBT + ABMT. Consistent with the view that self-efficacy enhances task performance, Zlomuzica, et al. [51] demonstrated that experimentally increasing perceived self-efficacy facilitated the extinction of conditioned fear in healthy participants. Although this manipulation targeted state self-efficacy, unlike the present study’s focus on trait self-efficacy, these findings implicate self-efficacy in emotional learning. This is particularly relevant given that fear extinction is commonly conceptualized as a form of inhibitory learning and represents one of several learning processes targeted by exposure-based interventions (e.g., Craske, et al. [52], Bouton [53]). Taken together, these findings suggest that self-efficacy may also enhance ABMT responsiveness by facilitating adaptive inhibitory learning during threat-related processing. The present results also indicated that self-efficacy may influence both treatment and training engagement. The broader framework by Bandura [54] suggests that individuals with higher self-efficacy tend to persist in the face of challenges. While not specifically in the context of clinical treatment, this framework may explain, in part, why self-efficacy is particularly important in exposure-based therapies, in which the exposures themselves - the act of confronting feared stimuli - constitute a substantial challenge. Furthermore, the ability to persist in the face of this challenge provides a critical foundation for inhibitory learning and may be particularly relevant in the context of the gamified ABMT used in the current study. Gamification strategies are designed to increase motivation, adherence, and engagement by incorporating clear task objectives, performance-based rewards (i.e. gaining points for correct responses in the dot probe tasks, gaining time tokens for quick responses, redeeming points in the visual search task), and opportunities for progression (i.e., participants are able to see their score progression from session to session) [38], [55], [56]. Youth with higher baseline self-efficacy, who enter treatment with a stronger belief in their ability to complete challenging tasks, may be more likely to engage with game-like elements of the ABMT as these aspects give implicit and explicit opportunities to capitalize on their abilities to persist in the face of challenges. Furthermore, Boendermaker, Prins, & Weirs (2015), outlined that integrating game-like techniques in cognitive bias modification training should target both intrinsic and extrinsic motivation to increase engagement with training programs [57]. Indeed, one study on self-efficacy and intrinsic and extrinsic motivation in academic contexts showed that self-efficacy was correlated with intrinsic motivation and predicted cognitive engagement [58]. Importantly, past studies have shown that engagement and motivation moderate the efficacy of cognitive training [24]. Though, in the present study, task specific self-efficacy and motivation were not measured, the current findings fit within the broader theoretical framework in which self-efficacy may influence both the cognitive mechanisms targeted by ABMT and motivational processes that sustain effortful training engagement. Self-efficacy efficacy effects on CBT + ABMT associated change in parent reported subthreshold depression Unexpectedly, significant associations were only observed for parent reported depressive - but not anxiety - symptom severity. Lack of significant interactions across all anxiety measures highlights that the current findings should be interpreted with caution and require replication. Nevertheless, although the trial included youth with a primary anxiety disorder, this finding highlights the transdiagnostic nature of negative attentional biases, which have been mechanistically linked to both anxiety and depression. Given the high comorbidity of these conditions, it is not uncommon for youth with anxiety to also display depressive symptoms, and the mechanisms targeted by ABMT (i.e. negative attention biases) are relevant across both disorders. Similarly, Baert, et al. [59] showed that the efficacy of ABMT depended on the severity of depression, such that ABMT was effective for young adults with mild depressive symptoms, but not for individuals with more moderate or severe symptoms or a diagnosis of major depressive disorder. Consistent with this, in the current study, where a diagnosis of full depression was exclusionary, our findings may reflect a greater capacity for ABMT to alter symptoms in subthreshold or mild depression. Beyond the transdiagnostic effects of ABMT, the specificity of findings related to depressive symptoms may also reflect the key role of self-efficacy in motivational processes. As previously discussed, self-efficacy may influence treatment outcomes by enhancing participants' motivation, persistence, and engagement in goal aligned tasks. These processes may be particularly relevant for depressive symptoms, which have been linked to diminished perceived control, motivational deficits, and impairments in cognitive control [60], [61]. Such characteristics may be especially important in the context of the cognitively demanding and goal-oriented gamified ABMT utilized in the present study. Current findings may suggest that higher baseline self-efficacy may be associated with greater responsiveness to ABMT; it is possible that such effects may be more readily observed in depressive symptoms, given the relationships among self-efficacy, perceived control, motivation, and depression-related cognitive and behavioral processes. Although this interpretation offers one potential explanation for why significant associations emerged for depressive, but not anxiety symptoms, findings should be interpreted cautiously, particularly given evidence suggesting that ABMT effects tend to be more robust and consistent for anxiety than depression [62]. Future research should therefore examine whether self-efficacy plays different roles across diagnostic groups and symptom severities. A further consideration for depressive symptom findings is the pattern of results across multiple informants. In the present study, we collected clinician-rated, child-rated, and parent-rated measures of outcome. However, significant findings were observed only for parent report. While this raises concerns about the lack of informant convergence, this pattern does manifest on occasion. Prior research has consistently shown discrepancies between child and parent reports, particularly early in treatment [55], [63]. Moreover, parent perspectives may capture clinically meaningful change that is less apparent to clinicians in structured assessments or to youth themselves, especially among younger participants [64]. Furthermore, parent reports have been found to reliably align with diagnostic outcomes and to predict both post-treatment diagnostic status and symptom changes over time [64], [65]. Parents may also be uniquely attuned to symptom severity and functioning in settings relevant to the context in which they interact with their child [66]. Therefore, our findings may suggest that while multi-informant data remain critical for a comprehensive understanding of treatment response, parent report may serve as a particularly sensitive early indicator of change in the context of ABMT. Nevertheless, the lack of significant findings across clinician and child ratings highlights the importance of replication and further investigation. Future studies should examine whether self-efficacy associated changes in treatment outcomes are initially most detectable by parents and whether such changes are later observed by other informants as well. Strengths and limitations Importantly, our study is limited by several factors. First, the generalizability of our findings is limited in that our sample was restricted to youth with primary anxiety diagnoses and excluded cases with comorbid psychological disorders outside anxiety (see methods and Byrne, et al. [14] for more detailed exclusion criteria). This is particularly relevant given high rates of comorbidities amongst children and adults with anxiety disorders [67]. Furthermore, our sample was predominantly White, and cultural and/or variation in self-efficacy functions may moderate effects [68]. Third, in the present study, self-efficacy was assessed only at baseline. Consequently, the limited number of assessment points prevented examination of how self-efficacy may fluctuate throughout treatment and how these fluctuations may influence treatment response. Repeated self-efficacy assessments would allow examination of dynamic changes in self-efficacy and show whether these changes temporally precede changes in treatment outcomes [29]. Because the present analyses relied solely on baseline self-efficacy, findings should be interpreted as informing the predictive role of baseline self-efficacy rather than the dynamic or mechanistic role of changes in self-efficacy during ABMT in influencing treatment outcomes. Future research using more frequent assessments across treatment phases is needed to better characterize these longitudinal processes. Fourth, Monte Carlo simulation indicated that the current study was underpowered for complex moderated mediation analysis, and a larger sample size is a recommended target for future replication studies. Despite reduced power, the significant conditional indirect effect of SEQ by ABMT condition suggests the moderated mediation pattern for parent-reported depressive symptoms is worthy of further investigation. Finally, hypotheses were not pre-registered, and the findings arose from secondary analyses. Despite these limitations, the present work has several important strengths. This study is among the first large, double-blinded clinical ABMT trials to incorporate self-efficacy as a potential moderator, and it provides preliminary empirical support for self-efficacy as a meaningful potentially modifiable target for ABMT optimization. Furthermore, we include several well-validated measures drawn from multiple reporters and collected prospectively rather than longitudinally, to capture changes in symptoms and behaviors over time. These findings lay the foundation to support future work investigating whether bolstering self-efficacy prior to or early in treatment enhances ABMT engagement, and therefore response. Conclusion In sum, the present study demonstrates that self-efficacy influences treatment response among patients receiving CBT + ABMT. Future studies should continue to evaluate the role of self-efficacy on attentional allocation and treatment engagement, and its role in predicting ABMT treatment outcomes. Integrated with past literature, our findings highlight self-efficacy as a potentially modifiable target in improving outcomes for cognitive training paradigms for child internalizing symptoms. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.