A Randomized Controlled Trial of Attention Bias Modification Treatment in Youth With Treatment-Resistant Anxiety Disorders.

The researchers compared two computer-based attention training programs in children aged 7–16 who still had an anxiety disorder after completing 12–14 sessions of cognitive-behavioral therapy (CBT):

**Attention Bias Modification Treatment (ABMT):** A dot-probe task where a pair of faces (one angry, one neutral) appeared on screen, followed by a probe (a symbol the child had to identify). In ABMT, the probe always appeared where the neutral face had been — training the child's attention to shift away from threatening faces and toward neutral ones. This was designed to directly target the automatic, rapid attention bias toward threat that characterizes anxiety disorders.

**Attention Control Training (ACT):** The same dot-probe task, but the probe appeared equally often behind the angry face and the neutral face (50% each). This meant there was no contingency training attention away from threat. ACT was designed to control for nonspecific effects of the task itself — such as practicing focusing, sustaining, and shifting attention generally — without specifically training avoidance of threat.

Both interventions consisted of 8 sessions delivered over 4 weeks (2 sessions per week), with each session containing 160 dot-probe trials. The primary outcome was anxiety severity rated by an independent evaluator (a clinician who did not know which group the child was in) using the Pediatric Anxiety Rating Scale (PARS). Secondary outcomes included child self-reported anxiety (Screen for Child Anxiety Related Emotional Disorders – Child version, SCARED-C), parent-reported child anxiety (SCARED-P), and diagnostic recovery (whether the child still met criteria for their primary anxiety disorder). They also measured attention bias to threat (using the dot-probe task itself) and attention control (using the Attentional Control Scale for Children, ACS-C).

**Sample size:** 64 youths (34 boys, 30 girls)

**Age:** Mean 11.7 years (range 7–16, standard deviation 2.43)

**Ethnicity:** 85.9% Hispanic

**Setting:** Recruited from a university-based clinic in the United States (Florida)

**Inclusion criteria:** Had completed 12–14 sessions of manualized CBT for anxiety disorders in a prior clinical trial or general clinic services, AND continued to meet DSM-IV criteria for a primary anxiety disorder at the conclusion of CBT and at a 1-year follow-up evaluation (ensuring they were true nonresponders, not delayed responders)

**Primary diagnoses at baseline:** Social anxiety disorder (39.1%), generalized anxiety disorder (31.3%), specific phobia (14.1%), separation anxiety disorder (9.4%)

**Exclusion criteria:** Developmental disabilities, psychosis, or current involvement in another psychosocial treatment

**Recruitment:** 75 youths were eligible; 64 (85.3%) agreed to participate and enrolled

**Anxiety diagnoses:** Anxiety Disorder Interview Schedule for Children – IV: Child and Parent Versions (ADIS-IV-C/P). This is a semi-structured diagnostic interview administered separately to the child and parent. Interrater reliability in this study was excellent (kappa = 1.0 for social anxiety disorder and generalized anxiety disorder, 0.86 for separation anxiety disorder, 0.63 for specific phobia).

**Primary outcome – clinician-rated anxiety severity:** Pediatric Anxiety Rating Scale (PARS), 6-item version. An independent evaluator (masked to treatment assignment) interviewed the child and parent separately, then rated 6 dimensions of global severity and impairment on a 0–5 scale (total score range 0–30, higher = more severe). Internal consistency in this sample was good (alpha = 0.77).

**Secondary outcome – child self-reported anxiety:** Screen for Child Anxiety Related Emotional Disorders – Child Version (SCARED-C). 41 items, each rated 0–2 (total range 0–82, higher = more anxiety). Alpha = 0.93 in this sample.

**Secondary outcome – parent-reported child anxiety:** Screen for Child Anxiety Related Emotional Disorders – Parent Version (SCARED-P). Same 41 items, parent rates child's anxiety. Alpha = 0.92.

**Attention bias to threat:** Dot-probe task with angry and neutral faces. Bias scores were calculated as reaction time differences between congruent trials (probe behind angry face) and incongruent trials (probe behind neutral face). Positive scores indicate attention toward threat; negative scores indicate attention away from threat.

**Attention control:** Attentional Control Scale for Children (ACS-C). 20 items, child self-report, measuring ability to focus attention, shift attention, and sustain attention. Alpha = 0.70.

**Timing of measurements:** Pre-treatment (PRE), post-treatment (POST, after 4 weeks of training), and 2-month follow-up (FOLLOW-UP)

**Study design:** This was a randomized controlled trial (RCT) with two parallel arms: ABMT and ACT. Participants were randomized using a computer-generated random number sequence with a 1:1 allocation ratio. Randomization was stratified by age (7–11 vs. 12–16) and primary diagnosis (social anxiety disorder vs. other anxiety disorders).

**Blinding:** Several layers of blinding were used:

Independent evaluators (IEs) who administered the PARS and ADIS were masked to treatment assignment. They were not present during training sessions and had no access to treatment condition information.

Participants and parents were not explicitly told which condition was hypothesized to be active, though the study did not formally test whether blinding was successful.

The researchers analyzing the data were also blinded to condition until analyses were complete.

**Duration:** The active intervention phase lasted 4 weeks (8 sessions, 2 per week). Each session took approximately 15–20 minutes. Follow-up assessment occurred 2 months after the last training session.

**Statistical approach:** The primary analyses used mixed-effects regression models (also called hierarchical linear modeling or multilevel modeling), which is appropriate for repeated measures data because it accounts for the correlation between measurements taken from the same person over time. They tested:

Between-group differences (ABMT vs. ACT) at POST and FOLLOW-UP, controlling for PRE scores

Within-group changes from PRE to POST and PRE to FOLLOW-UP (both groups combined)

Moderation effects of attention control on treatment response

They used an intent-to-treat approach, meaning all randomized participants were included in analyses regardless of how many sessions they completed. Missing data were handled using full information maximum likelihood estimation, which is a modern and generally acceptable method.

**What this design can and cannot prove:**

This design can prove that any differences between groups are due to the specific training contingency (ABMT vs. ACT) rather than to placebo effects, natural recovery, or passage of time — because both groups received the same number of sessions, same computer interface, same number of trials, and same face stimuli. The only difference was the contingency (100% vs. 50% probe behind neutral face). Therefore, if ABMT outperformed ACT, it would provide strong evidence that training attention away from threat specifically causes anxiety reduction.

However, this design cannot prove that ABMT or ACT caused the observed improvements compared to no treatment at all, because there was no waitlist control or placebo control group that received no intervention. Both groups received an active intervention. The improvements seen in both groups could theoretically be due to:

Natural recovery over time

Repeated exposure to faces (which might itself be therapeutic)

The therapeutic alliance or attention from study staff

Regression to the mean (since participants were selected for high anxiety)

The design also cannot determine whether the mechanism of improvement is the same in both groups. Both groups showed increases in attention control, but without a no-treatment control, we cannot know if this increase was caused by the training or would have happened anyway.

**Major methodological weaknesses:**

No waitlist or no-treatment control group, making it impossible to know if either intervention is better than doing nothing

Relatively small sample size (n=64), which limits statistical power to detect small between-group differences

The sample was predominantly Hispanic (85.9%) and from one geographic region, limiting generalizability

The study was not registered for this specific trial design (the clinical trial registration was from the prior CBT trial)

Attention bias measurement used the same dot-probe task as the training, which can create measurement confounds

No formal assessment of whether blinding was successful (i.e., they didn't ask participants or IEs which group they thought they were in)

**Primary outcome – clinician-rated anxiety severity (PARS):**

Both groups showed significant decreases from PRE to POST and from PRE to FOLLOW-UP

Mean PARS scores decreased from approximately 18 at PRE to approximately 12 at POST and approximately 11 at FOLLOW-UP (exact means not reported in abstract; full text needed for precise numbers)

The between-group difference (ABMT vs. ACT) was NOT statistically significant at POST or FOLLOW-UP

Effect size for within-group change: Not reported in abstract, but described as "significant"

**Secondary outcome – diagnostic recovery:**

Combined across both groups: 50% of youths no longer met criteria for their primary anxiety disorder at POST

At FOLLOW-UP: 58% no longer met criteria

No significant difference between ABMT and ACT in recovery rates

**Secondary outcome – child self-reported anxiety (SCARED-C):**

Significant decreases from PRE to POST and PRE to FOLLOW-UP in both groups

No significant between-group differences

**Secondary outcome – parent-reported child anxiety (SCARED-P):**

Significant decreases from PRE to POST and PRE to FOLLOW-UP in both groups

No significant between-group differences

**Attention bias to threat:**

No significant changes from PRE to POST or PRE to FOLLOW-UP in either group

No significant between-group differences

This is notable because ABMT was specifically designed to reduce attention bias to threat, but it did not do so

**Attention control (ACS-C):**

Significant increases from PRE to POST in both groups combined

No significant between-group differences

The increase was maintained at FOLLOW-UP

**Moderation analyses:**

Attention control at PRE did NOT significantly moderate treatment response (i.e., children with high vs. low attention control did not respond differently to ABMT vs. ACT)

**Attrition:**

64 enrolled, 56 completed POST assessment (87.5% retention)

53 completed FOLLOW-UP assessment (82.8% retention)

No significant differences in dropout between groups

The most clinically meaningful finding is that approximately half of these treatment-resistant youths no longer met criteria for their primary anxiety disorder after just 4 weeks of attention training (8 sessions, each 15–20 minutes). This is a substantial effect for a population that had already failed to respond to 12–14 sessions of CBT — the gold-standard psychological treatment.

However, the effect was identical in both groups. This means the specific training contingency (training attention away from threat) added nothing beyond the general attention training. The effect size for the within-group improvement was likely moderate to large (based on the 50% recovery rate), but the between-group effect size was essentially zero.

To put this in perspective: If you have a child who did not respond to CBT, adding 4 weeks of daily attention training (whether ABMT or ACT) gives them about a 50% chance of no longer meeting diagnostic criteria for their primary anxiety disorder. This is roughly comparable to the response rate you'd expect from a second course of CBT or from starting medication, though direct comparisons are not available from this study.

The fact that attention bias to threat did not change, but attention control did improve, suggests the mechanism is not about reducing threat vigilance but about improving general attentional abilities — the ability to focus, sustain, and flexibly shift attention. This is a different mechanism than originally hypothesized.

**Acknowledged by authors:**

Small sample size limited statistical power for detecting between-group differences and moderation effects

No waitlist control group, so cannot determine if improvements exceed natural history

The sample was predominantly Hispanic, limiting generalizability to other populations

The study was conducted at a single site

Attention bias measurement may have been unreliable (dot-probe tasks often have low test-retest reliability)

The 2-month follow-up is relatively short; longer-term durability is unknown

**Additional critical observations:**

The lack of a no-treatment control is a major limitation. Without it, we cannot rule out that improvements were due to spontaneous remission, regression to the mean, or the passage of time. Given that these youths had already failed to respond to CBT and still had diagnoses at 1-year follow-up, spontaneous remission seems less likely, but it cannot be ruled out.

The study did not report effect sizes with confidence intervals, making it difficult to assess the precision of estimates.

The training was relatively brief (8 sessions over 4 weeks). It's possible that longer training or more sessions would produce different results.

The study did not measure treatment fidelity or adherence to the training protocol (e.g., whether children were actually paying attention during trials).

There was no assessment of whether blinding was successful. If participants or IEs could guess which condition was "active," this could bias results.

The study combined data from two different recruitment sources (prior clinical trial and general clinic services), which may have introduced heterogeneity.

The primary outcome (PARS) was rated by an IE who interviewed both child and parent separately, but the IE's ratings could still be influenced by demand characteristics or the child's/parent's expectations.

The study did not report on adverse events or side effects, though the intervention is low-risk.

The sample size calculation was not reported, so it's unclear whether the study was adequately powered to detect clinically meaningful between-group differences.

For someone running their own n=1 experiment (or for a parent considering this approach for their child):

**What to test:**

A daily dot-probe attention training task using emotional faces (angry vs. neutral). You can find free versions online or use apps designed for attention bias modification.

The key decision is whether to use a contingency (probe always behind neutral face, like ABMT) or a non-contingent version (probe equally behind angry and neutral, like ACT). Based on this study, the specific contingency doesn't matter — both work equally well. So you could use either, or alternate between them.

The "dose" tested was 8 sessions over 4 weeks (2 per week), with 160 trials per session (approximately 15–20 minutes per session). This is a reasonable starting point.

**Minimum meaningful duration:**

4 weeks appears sufficient to see effects, with improvements maintained at 2-month follow-up

Some studies have used daily training for 2–4 weeks; the optimal frequency is unknown

A reasonable self-experiment would be 4 weeks of training, with assessment before and after

**What to measure:**

**Primary metric:** Anxiety severity. Use a validated self-report scale like the Generalized Anxiety Disorder 7-item scale (GAD-7) for adults, or the Screen for Child Anxiety Related Emotional Disorders (SCARED) for children. Measure weekly to track trajectory.

**Secondary metric:** Diagnostic status. For a more rigorous assessment, use a structured diagnostic interview or at minimum track whether you/your child still meets DSM criteria for the anxiety disorder.

**Process metric:** Attention control. The Attentional Control Scale (ACS) is a 20-item self-report measure. Track this weekly to see if improvements in attention control precede or coincide with anxiety reduction.

**Attention bias:** You could measure this using a dot-probe task, but be aware that test-retest reliability is low, and this study found no change in bias despite clinical improvement.

**Key confounds to control for:**

**Other treatments:** If you or your child are also in therapy or taking medication, these effects will be