Exposure-Focused Cognitive-Behavioral Therapy for Youth Anxiety Disorders: Superiority Over Relaxation-Based Comparator and Predictors of Response

Abstract

Objective

Evidence supports exposure-focused cognitive-behavioral therapy (EF-CBT) for child and adolescent anxiety disorders, but protocols comparing EF-CBT with other therapy modalities (eg, relaxation) are understudied, and little is known about predictors of EF-CBT response. Extending an interim report, this study compared EF-CBT with an active comparator, relaxation and mentorship training (RMT), and examined predictors of EF-CBT response.

Method

Participants were 139 youths with transdiagnostic anxiety disorders (age range 7-18 years, mean age 12.3). Sample demographics were male, 24.5%; female, 75.5%; Asian, 0.7%; Black, 5.8%; multiracial, 0.7%; other, 0.7%; other Pacific Islander, 11.5%; unknown, 2.9%; White, 77.7%; 5.8% Hispanic/Latino. Participants were randomized (2:1 ratio) to 12 sessions of EF-CBT (n = 94) or RMT (n = 45). Treatment response was determined by independent evaluator ratings on the Pediatric Anxiety Rating Scale and the Clinical Global Impressions–Improvement scale.

Results

EF-CBT was associated with 3.53 higher odds of treatment response based on Clinical Global Impressions–Improvement scale (EF-CBT 47.1%, RMT 20.1% response rate) whether delivered in person or via teletherapy. Notably, age moderated EF-CBT response, such that younger age predicted better outcomes (F_1,89 = 4.38, p = .04). By contrast, number of anxiety diagnoses, comorbid externalizing symptoms, and subclinical depressive symptoms did not predict EF-CBT outcomes.

Conclusion

As expected, EF-CBT was more effective than the RMT comparator. Greater improvement in younger patients receiving EF-CBT suggests that response to exposure may be developmentally sensitive.

Clinical Trial Registration Information

Dimensional Brain Behavior Predictors of CBT Outcomes in Pediatric Anxiety (Anxiety-CBT); https://clinicaltrials.gov/study/NCT02810171

Plain language summary

139 youths with clinical anxiety disorders received either exposure-focused cognitive-behavioral therapy (EF-CBT) or a relaxation-focused treatment (RMT). Youth who received EF-CBT were 3.5 times more likely to respond better to treatment than those who received RMT. Younger age was associated with better outcomes in EF-CBT, which may inform clinician use of exposure therapy.

Anxiety disorders are the most frequently diagnosed mental health concerns during childhood and adolescence.¹ Cognitive-behavioral therapy (CBT) has been shown to outperform nonactive control conditions (eg, waitlist, educational support) in the treatment of pediatric anxiety disorders, especially when considering treatment response and remission.2, 3, 4 However, fewer studies have compared CBT with active treatment conditions, and evidence that CBT is superior to other credible alternative treatments is lacking.⁴ Additionally, only a small number of studies have examined the unique contributions of CBT components (eg, exposure vs relaxation) to treatment outcome.⁵ Moreover, predictors of CBT response remain understudied,^3,6,7 with relatively small sample sizes in prior studies of CBT limiting examination of potentially relevant predictors, such as age, anxiety severity, and comorbidities.⁴

A recent review⁴ identified studies comparing CBT with waitlist and found that CBT was associated with greater remission of all anxiety diagnoses (odds ratio [OR] 4.43, 28 studies). However, compared with active treatments such as attentional controls (ie, educational interventions or interventions incorporating psychoeducation about anxiety along with other attentional control activities), treatment as usual, or alternative treatments (ie, protocolized therapy omitting CBT components), there was only a possible benefit of CBT (OR 2.75, 5 studies; OR 2.74, 5 studies; OR 0.89, 4 studies, respectively). Thus, CBT for pediatric anxiety disorders has strong support when considered compared with waitlist, but the literature is less robust or conclusive for active comparisons.⁴

In clinical trials studying CBT, choice of comparator condition has also been found to interact with how treatment-related change in anxiety severity is measured (eg, independent evaluator [IE], caregiver report, youth self-report). Youth and caregivers tend to report benefits for CBT over waitlist at medium effect sizes (mean difference between CBT and waitlist: 0.67 standardized mean difference, n = 45, and 0.70 standardized mean difference, n = 35, respectively).⁴ However, when comparing CBT with attentional controls, only small effect sizes on youth reports (mean difference between CBT and control: 0.31 standardized mean difference, n = 15) and no difference in caregiver reports (effect size crossed zero) emerged.⁴ No reliable differences were found for either youth or caregiver reports when comparing CBT with alternative therapies for anxiety.⁴ By contrast, across different comparator conditions, CBT for pediatric anxiety disorders is typically rated as more effective by IEs than patients or their parents.⁴ To understand the benefit of CBT for pediatric anxiety disorders, clinical trials can be optimized by measuring treatment response via trained IEs, in the context of a randomized controlled trial, using an active comparison treatment.

A growing body of evidence suggests that when considering the elements contained within CBT treatments for pediatric anxiety disorders (ie, relaxation, cognitive restructuring, exposure),⁸ exposure is the most effective treatment component. A review of more than 100 studies of youth found that higher doses of in-session exposure delivery were associated with larger outcome effect sizes, whereas inclusion of relaxation was associated with lower outcome effect sizes.⁵ Other work examining multicomponent CBT that began with relaxation therapy followed by cognitive restructuring and then exposure demonstrated little change in anxiety severity during relaxation sessions, initial improvement with cognitive restructuring, and accelerated improvement during exposure.⁹ This body of work suggests greater benefits of exposure therapy than relaxation-based treatment. Indeed, according to Society of Clinical Child and Adolescent Psychology guidelines, exposure has been identified as a level 1 treatment (ie, having the strongest research support), whereas relaxation alone is rated as level 2 (ie, probably efficacious, but less strong research support).¹⁰ These findings point to a potential for reducing emphasis on relaxation, moving rapidly through cognitive restructuring and increasing the dose of exposure to improve treatment response.^5,9,11, 12, 13, 14, 15

Despite the growing evidence supporting exposure for CBT, dismantling studies are rare, and exposure is seldom studied in isolation from other credible but possibly less effective skills, such as relaxation.^4,5,9 Active comparison treatments are important in clinical trials because such conditions are more credible to patients and clinicians alike, reducing confounding variables of credibility and expectancy of comparison with placebo or waitlist control. Moreover, active comparison conditions control for the other nonspecific effects of clinical trial participation (eg, structured measurement of outcomes) and meeting weekly with a supportive therapist to enable the isolation of “active ingredients” of the primary intervention being studied. Importantly, both patients and clinicians find relaxation to be a highly credible intervention,¹⁶ making it a useful comparator for isolating exposure-specific effects.

Moderators of Treatment Response

Despite the evidence of benefits of CBT broadly and of exposure as a CBT component for pediatric anxiety disorders, many patients do not experience optimal response,¹⁷ and identification of treatment predictors is needed to optimize treatment selection, inform prognosis, and guide novel therapy design.^6,7 Existing research has been characterized by inconsistency, with few variables reliably predicting CBT outcomes.^3,6 Although it is mixed, there is some research to support baseline symptom severity (defined as severity of symptoms or number of anxiety diagnoses), comorbid symptoms and conditions, and older patient age as being associated with worse treatment outcomes,^9,12,13,18, 19, 20 whereas treatment delivery context has not been shown to be associated with outcomes.²¹

Interim Analyses

The current article reflects a follow-up from interim analyses in which 102 youths with pediatric anxiety disorders were randomized (2:1) to 1 of 2 therapies comprising 12 sessions: exposure-focused CBT (EF-CBT) (n = 70) or relaxation and mentorship training (RMT) (n = 32).²² Higher rates of treatment response and faster and greater reduction of anxiety severity by IEs were demonstrated for EF-CBT compared with RMT.²² However, this interim analysis was limited in power given that the control condition included only 32 participants, and groups of at least 35 participants are typically recommended to reliably detect a difference in response between 2 active treatment conditions.²³ Additionally, the initial report was not sufficiently powered to examine predictors of treatment outcome. Given these limitations of the interim analysis, we intended to reanalyze data on completion of study recruitment to confirm the superiority of EF-CBT to RMT and test for predictors of treatment outcomes.

Aims and Hypotheses

In an expanded sample (total sample N = 139; EF-CBT n = 94, RMT n = 45), we hypothesized that RMT would be as credible as EF-CBT to participants, but that EF-CBT, compared with RMT, would be associated with greater clinical improvement and higher treatment response rates as rated by an IE. We further hypothesized that among patients treated with EF-CBT, greater IE-rated anxiety reduction from before to after treatment would be associated with lower baseline anxiety (operationalized as number of baseline anxiety diagnoses), fewer co-occurring attention-deficit/hyperactivity disorder (ADHD) or depressive symptoms, and younger age. Finally, we hypothesized that there would be no difference in anxiety reduction related to EF-CBT among youth who received in-person therapy or teletherapy.

Method

Sample

This study was conducted in the Midwestern United States. Prospective participants were initially contacted by phone and/or e-mail. Additional details about study recruitment methods are reported in the interim report.²² Participants included 139 youths (age range 7.0-17.8 years, mean age 12.3 years). Participants met criteria for 1 or more anxiety disorders (ie, generalized [n = 121], separation [n = 45], social [n = 86], panic [n = 2], or specific phobia [n = 51]; range of 0-4 anxiety diagnoses per participant) and were enrolled in a neuroimaging randomized controlled trial.²⁴ Participants were predominantly female (75.5%; n = 105), White (77.0%; n = 108), and non-Hispanic/Latino (94.2%, n = 131) (Table 1). Inclusion and exclusion criteria are fully detailed in prior work.^22,24 Participants were required to meet diagnostic criteria for at least 1 primary anxiety disorder based on a structured clinical interview. Eligibility requirements also included anxiety symptoms of at least moderate severity (total score of ≥13 on the Pediatric Anxiety Rating Scale [PARS]). Youth with any of the following comorbid diagnoses were excluded: current major depressive disorder, posttraumatic stress disorder, intellectual disability, autism spectrum disorder, substance use disorder, psychotic disorders, and acute risk to self or others. Other comorbid diagnoses (eg, OCD, ADHD) or concerns (eg, oppositionality, relational concerns) were permitted if an anxiety diagnosis was the primary source of interference and distress. Participants were not taking any medications with the exception of ongoing stimulant or α-agonist treatment for ADHD. Regarding incentives, families could earn $50 for completing the baseline assessment, $15 for each treatment session they attended, and $50 for poststudy assessment. Participants also earned $50 for completing each of the 2 functional magnetic resonance imaging scans (neuroimaging data not reported here). The study was regulated by the Michigan Medicine Institutional Review Board, and all children and legal guardians completed informed assent/consent.

Table 1.

Sample Baseline Characteristics and p Values for Tests of Covariate Balance Across Conditions

	RMT participants (n = 45)		EF-CBT participants (n = 94)		Total participants (N = 139)		p
	Mean	(SD)	Mean	(SD)	Mean	(SD)
Age, y	12.2	(3.3)	12.3	(3.1)	12.3	(3.1)	.863a
	Range		Range		Range
	7.0-17.2		7.2-17.8		7.0-17.8
	Mean	(SD)	Mean	(SD)	Mean	(SD)	p
Sessions attended	9.8	(3.2)	11.2	(2.0)	10.7	(2.5)	.001a
	Range		Range		Range
	1.0-12.0		3.0-12.0		1.0-12.0
	n	(%)	n	(%)	n	(%)
Gender							.998b
Male	11	(24.4)	23	(24.5)	34	(24.5)
Female	34	(75.6)	71	(75.5)	105	(75.5)
Race							.015b
Asian	3	(6.7)	5	(5.3)	8	(5.8)
Black	4	(8.9)	0	(0)	4	(2.9)
Multiracial	4	(8.9)	12	(12.8)	16	(11.5)
Other	1	(2.2)	0	(0)	1	(0.7)
Other Pacific Islander	1	(2.2)	0	(0)	1	(0.7)
Unknown	1	(2.2)	0	(0)	1	(0.7)
White	31	(68.9)	77	(81.9)	108	(77.7)
Ethnicity							.750b
Hispanic/Latino	3	(6.7)	5	(5.3)	8	(5.8)
Not Hispanic/Latino	42	(93.3)	89	(94.7)	131	(94.2)
Family incomed	n = 37		n = 81		n = 118		.536b
Less than $20,000	0	(0)	1	(1.1)	1	(0.7)
$20,000-$29,999	3	(6.8)	2	(2.1)	5	(3.6)
$30,000-$49,999	3	(6.8)	5	(5.3)	8	(5.8)
$50,000-$74,999	3	(6.8)	7	(7.4)	10	(7.2)
$75,000-$99,999	6	(13.6)	14	(14.9)	20	(14.5)
$100,000-$149,999	8	(18.2)	27	(28.7)	35	(25.4)
$150,000 or more	13	(29.5)	25	(26.6)	38	(27.5)
Unknown	1	(2.3)	0	(0)	1	(0.7)
Delivery context							.993b
In-person	33	(73.3)	69	(73.4)	102	(73.4)
Teletherapy	12	(26.7)	25	(26.6)	37	(26.6)
Hybrid	4	(8.9)	5	(5.3)	9	(6.5)
	Mean	(SD)	Mean	(SD)	Mean	(SD)	p
PARS baselinee	19.9	(3.8)	19.2	(3.2)	19.4	(3.4)	.389a
	Range		Range		Range
	13.0-27.0		13.0-26.0		13.0-27.0
	Mean	(SD)	Mean	(SD)	Mean	(SD)	p
MASC baselinee	58.0	(11.6)	58.6	(9.7)	58.4	(10.3)	.775a
	Range		Range		Range
	35.0-87.0		30.0-80.0		30.0-87.0
	Mean	(SD)	Mean	(SD)	Mean	(SD)	p
SCARED baselinee	27.3	(10.9)	27.4	(8.3)	27.4	(9.1)	.943a
	Range		Range		Range
	7.0-52.0		8.0-48.0		7.0-52.0
CEQ Credibility week 3	18.4	(4.77)	19.6	(5.14)	19.26	(5.04)	.200c
	Range		Range		Range
	8-27		3-27		3-27

Note: CEQ = Credibility/Expectancy Questionnaire; EF-CBT = exposure-focused cognitive-behavioral therapy; MASC = Multidimensional Anxiety Scale for Children; PARS = Pediatric Anxiety Rating Scale; RMT = relaxation mentorship training; SCARED = Screen for Child Anxiety and Related Emotional Disorders.

^aLinear model analysis of variance.

^bPearson χ² test.

^cIndependent t test.

^dThis question was added to the demographic questionnaire midway through the study.

^eBefore randomization.

Measures

Baseline (ie, prerandomization) clinical assessments were conducted approximately 2 to 4 weeks before therapy initiation. Additional clinical measures were collected within a week of therapy sessions 1, 3, 6, 9, and 12 of the 12-session, weekly treatment.

Schedule for Affective Disorders and Schizophrenia for School-Age Children–Present and Lifetime Version

The Schedule for Affective Disorders and Schizophrenia for School-Age Children–Present and Lifetime version (K-SADS-PL)²⁵ was administered to youth and caregivers at baseline to establish clinical diagnoses and confirm study inclusion/exclusion criteria. The K-SADS-PL was administered by master’s-level clinicians who achieved reliability with gold-standard raters (senior author K.D.F. and a senior clinical assessor trained and supervised by K.D.F.). Youth and caregivers were given the opportunity to be interviewed either together or separately.

Credibility/Expectancy Questionnaire

Child report of treatment credibility and expectancy was assessed using the Credibility/Expectancy Questionnaire (CEQ)²⁶ at session 3 after 2 initial sessions of psychoeducation during which youth learned about the EF-CBT or RMT to which they had been randomized. Aligned with previous work, the sum of the CEQ Credibility scale was used in these analyses.²⁷

Pediatric Anxiety Rating Scale

Trained IEs used the PARS,²⁸ a semistructured clinical interview, to assess youth anxiety symptoms and impairment by integrating interviews with both the affected youth and the caregiver(s). The PARS provides a total score of clinical severity specific to anxiety, ranging from 0 to 30 with higher scores indicating higher severity. To facilitate comparison with the Child/Adolescent Anxiety Multimodal Study (CAMS), the PARS was considered our primary continuous outcome measure.² It was administered at baseline and sessions 1, 6, 9, and 12.

Clinical Global Impressions–Improvement Scale

The Clinical Global Impressions–Improvement (CGI-I) scale²⁹ is a single-item clinician-administered measure of the improvement of clinical global functioning over the course of intervention,³⁰ collected at sessions 1, 6, 9, and 12 of treatment. The item was rated by trained IEs (ie, masked to study condition) on a 7-point scale ranging from 1 (“very much improved”) to 7 (“very much worse”). Youth receiving either a 1 (“very much improved”) or a 2 (“much improved”) rating were defined as treatment responders. To facilitate comparison with CAMS, the CGI-I at session 12 was considered the primary categorical outcome measure.²

Multidimensional Anxiety Scale for Children Self-Report

The Multidimensional Anxiety Scale for Children (MASC) self-report³¹ includes 39 questions about youth anxiety symptoms on a 0 (“never true about me”) to 3 (“often true about me”) scale. Youth report on the MASC was collected at baseline and sessions 1, 3, 6, 9, and 12.

Screen for Child Anxiety Related Emotional Disorders Caregiver Report

The Screen for Child Anxiety Related Emotional Disorders (SCARED) caregiver report³² includes 41 items assessing symptoms of youth anxiety across domains and diagnoses (eg, school avoidance, panic, generalized anxiety) on a 0 (“not true or hardly ever true”) to 2 (“very true or often true”) scale, yielding total scores ranging from 0 to 82. Total scores greater than 24 indicate clinically elevated anxiety.³³ Caregiver report on the SCARED was collected at baseline and weeks 1, 3, 6, 9, and 12. Owing to significant missing data at other time points, only the following time points were examined in the current analyses: weeks 1, 6, and 12.

Child Behavior Checklist Attention Problems

Caregiver report on the Child Behavior Checklist (CBCL) Attention Problems³⁴ was administered at baseline to measure youth psychopathology across multiple domains, including the Attention Problems subscale. The CBCL consists of 113 items rated on a scale from 0 (“not true”) to 2 (“very true or often true”). The CBCL offers 11 subscales (eg, delinquent behavior, aggressive behavior, attention problems). The attention problems subscale consists of 11 items related to ADHD concerns (eg, “can’t concentrate” and “can’t sit still”) and has been used in previous work to assess ADHD symptoms across development.³⁵

Children’s Depression Inventory

Child report on depressive symptoms was measured using the Children’s Depression Inventory (CDI),³⁶ a 27-item self-report measure of depressive symptoms at baseline and weeks 1, 3, 6, 9, and 12. Each item has different anchors and can be scored from 0 to 2, with higher scores indicating higher levels of depression. CDI total scores range from 0 to 54. Only baseline CDI scores are included in the current analyses.

Demographic Questionnaire

A demographic questionnaire was created for this study to identify relevant demographic data for each participant. Caregivers of participants completed the form inquiring about youth gender, race, ethnicity, persons living in the home, parent demographics, child medical history, current medications, and other relevant health information. The current analyses include data from the caregiver’s report of the youth’s age, gender, race, and ethnicity.

Intervention

Participants were randomized (2:1 ratio) to EF-CBT or RMT. Intervention details are presented elsewhere.²² Briefly, both interventions (EF-CBT and RMT) were manualized and consisted of 12 weekly sessions. EF-CBT consisted of psychoeducation (2 sessions), cognitive restructuring and problem solving (1 session), and exposure practice (9 sessions, including 1 session with family education and exposure practice and 1 session with maintenance and relapse prevention with exposure practice). RMT consisted of brief psychoeducation, rapport building and identifying interests (2 sessions), family education (1 session), relaxation instruction and practice (5 sessions), additional engaging activities (3 sessions), and maintenance and relapse prevention (1 session). Therapists were trained in both interventions. Fidelity to intervention was maintained in the following ways: the clinical supervisor (a licensed PhD or MSW with expertise in pediatric anxiety disorders) listened to available recordings of a course of treatment for both conditions for each therapist, therapists participated in weekly clinical supervision, and a lead supervisor listened to portions of tapes for each additional study participant who consented to audio recording.

COVID-19 Procedures

All study procedures were conducted in person until March 13, 2020, at which time study procedures converted to teletherapy because of the COVID-19 pandemic. All participants who were enrolled after March 13, 2020, completed all study procedures remotely (with the exception of neuroimaging procedures, reported in,²⁴ via secure telehealth video conferencing. Nine participants (EF-CBT: n = 5; RMT: n = 4) were enrolled but had not yet completed study procedures as of March 13, 2020, thereby completing some of their study procedures in person and other procedures remotely.

Analytic Plan

Demographics

Linear model analysis of variance and Pearson χ² tests were conducted to compare baseline participant demographics across EF-CBT and RMT conditions. The following variables were examined: age, gender, race, ethnicity, family income, treatment delivery context, number of treatment sessions attended, and clinical measure baseline scores (prerandomization PARS, MASC, SCARED). Independent t tests were conducted to compare participant CEQ Credibility total scores across EF-CBT and RMT conditions at session 3. Baseline means and statistics for these variables by condition are reported in Table 1.

Primary Analyses

Primary analyses tested for differences in outcomes in youth randomized to EF-CBT compared with RMT and predictors of treatment response. Unless otherwise specified, the following variables were entered into all models as covariates: participant age (in years), participant gender, and number of study treatment sessions attended by the participant. Reference groups were defined as median group age (12 years), female gender, and 12 sessions attended.

Treatment Outcomes by Condition

Linear longitudinal mixed models were run to examine whether treatment condition (EF-CBT vs RMT) was associated with differential clinical response across the study period. Clinical response was defined as the participant’s PARS score at the following time points: sessions 1, 6, 9, and 12. Independent variables were the treatment condition (EF-CBT vs RMT), participant gender, time (sessions of study treatment completed), and treatment condition by time interaction.

For comparison with prior clinical trial literature (eg, CAMS),² the CGI-I was included as a secondary clinical measure of anxiety. Binomial logistic regression was conducted to examine whether treatment condition (EF-CBT vs RMT) was associated with differential rates of treatment response defined as responders or nonresponders based on CGI-I score at week 12 (responder: CGI-I = 1 (“very much improved”) or 2 (“much improved”); nonresponder: CGI-I = 3-7).

Linear longitudinal mixed models were run to examine whether treatment condition (EF-CBT vs RMT) was associated with differential change in symptom severity as reported by youth participants (MASC) and caregivers (SCARED-P) over the course of the study period. Independent variables were the treatment condition (EF-CBT vs RMT), participant gender, time (session of study treatment completed), and treatment condition by time interaction. The first model defined symptom severity response as MASC total raw score at sessions 1, 6, 9, and 12. The second model defined symptom severity response as caregiver-reported SCARED-P total raw score at sessions 1, 6, and 12.

Predictors of EF-CBT Response

Among patients who received EF-CBT, a series of general linear models was used to test baseline anxiety severity, comorbidities, age, and treatment delivery context as predictors of posttreatment PARS (also known as last EF-CBT session). These models covaried pretreatment PARS (also known as session 1), gender, and sessions attended to control for potential confounding effects of associations between predictor variables. For anxiety severity, the number of K-SADS-defined anxiety disorder diagnoses at baseline was the predictor. An auxiliary analysis examined baseline PARS as a predictor of CGI-I-defined treatment responder status. For analyses testing comorbidities as predictors, baseline T scores on the CBCL ADHD subscale and the CDI total scale were considered. A linear longitudinal mixed effects model was run to examine whether participant age in years predicted reduction in anxiety severity (PARS) across sessions 1, 6, 9, and 12. For treatment delivery context, the predictor variable was in-person vs telehealth delivery of EF-CBT. Participants who received hybrid care (some sessions in person, others via telehealth; n = 5) were omitted from these analyses.

Missing Data

For all primary analyses, missing data were handled using the machine learning imputation method missForest.³⁷ Additional information on missing data procedures can be found in Supplement 1, available online.

Results

Sample

For the current study, 262 subjects were assessed for eligibility as clinically anxious participants (Figure 1). Of these, 9 subjects did not meet criteria for an anxiety disorder diagnosis and were enrolled as healthy controls (healthy control participants were enrolled for neuroimaging analyses and are not included in this article), 5 were eligible to participate as healthy controls but declined, 83 did not meet study eligibility criteria, and 26 otherwise declined participation. The remaining 139 participants had at least 1 primary anxiety disorder diagnosis and were included in these analyses. Compared with our interim analysis,²² this sample has an additional 37 participants (36% more).

Figure 1.

CONSORT Diagram

Note: EF-CBT = exposure-focused cognitive behavioral therapy; fMRI = functional magnetic resonance imaging; PI = Principal Investigator; RMT = relaxation mentorship training. Please note color figures are available online.

The final sample included 94 EF-CBT participants and 45 RMT participants. Randomization was stratified by age and gender. Treatment completion was defined as attending at least 8 sessions of study therapy. Of the total sample, 127 participants met treatment completer status (EF-CBT: n = 88; RMT: n = 39). EF-CBT participants attended an average of 11.2 sessions, and RMT participants attended an average of 9.8 sessions (t₁₃₇ = −3.24, p < .001). There was no significant difference between EF-CBT and RMT groups in anxiety severity measured on the PARS at baseline (prerandomization) or at session 1 of therapy (postrandomization). At baseline, there was also no significant difference between EF-CBT and RMT groups in anxiety severity as measured by the MASC or SCARED. At session 3, there was no significant difference between groups in treatment credibility as measured by the CEQ- Credibility scale. Black participants were randomly assigned only to RMT, and a majority of multiracial participants were randomly assigned to EF-CBT, leading to a significant association between race and group assignment (p = .015) (Table 1). No other significant differences were observed between groups at baseline.

Treatment Outcomes

Independently Evaluated Anxiety Severity

Independently evaluated anxiety severity, measured on the PARS, improved across both treatment conditions (F_3,417 = 92.62, p < .001), with more pronounced improvement documented in the EF-CBT condition (F_3,417 = 7.65, p < .001) (Figure 2). Significant differences between conditions emerged, with EF-CBT outperforming RMT at weeks 9 (β = 1.28 ± .59, p = .03) and 12 (β = 2.81 ± .59, p < .001) (Table 2). Estimated marginal means of PARS scores at each week revealed a significant difference between conditions at session 12 (EF-CBT = 13.1; RMT = 14.9; p < .007) (Table 3). Hedges g comparing estimated PARS means at session 12 was 0.50, indicating a medium effect in favor of EF-CBT. Secondary analyses considering the last observation carried forward yielded similar findings (Supplement 2, available online; Table S1, available online; Figure S1, available online). Treatment response rates (CGI-I score of 1 or 2) were significantly greater among participants assigned to EF-CBT (47.1%) than RMT (20.1%; β = 1.26 ± .45, p = .005) (Table 3). EF-CBT was associated with 3.53 higher odds of treatment response.

Figure 2.

Estimated Marginal Means of Clinical Severity (Independent Evaluator Report) Across Treatment

Note:Estimated mean Pediatric Anxiety Rating Scale (PARS) scores across 12 weeks of treatment. EF-CBT = exposure-focused cognitive-behavioral therapy; RMT = relaxation mentorship training.

Table 2.

Longitudinal Model Examining Condition as a Predictor of the Pediatric Anxiety Rating Scale (PARS)

	β	SE	t	p
(Intercept)	15.79	0.36	43.74	<.001
Week 6 − week 1	−2.34	0.30	−7.90	<.001
Week 9 − week 1	−3.94	0.30	−13.29	<.001
Week 12 − week 1	−4.50	0.30	−15.19	<.001
Treatment (EF-CBT)	−.34	0.65	−0.52	.61
Age (years)	.67	0.30	2.26	.03
Gender (female)	.34	0.69	0.49	.63
Sessions completed	−.63	0.31	−2.05	.04
Week 6 − week 1 × treatment (EF-CBT)	−1.02	0.59	−1.72	.09
Week 9 − week 1 × treatment (EF-CBT)	−1.28	0.59	−2.15	.03
Week 12 − week 1 × treatment (EF-CBT)	−2.81	0.59	−4.73	<.001

Note: Fixed effects summaries from the longitudinal condition model with missForest imputations. Intercepts can be interpreted as the expected week 1 PARS scores for a 12-year-old boy who was assigned to the relaxation management training group and eventually completed 12 weeks of therapy. EF-CBT = exposure-focused cognitive-behavioral therapy.

Table 3.

Treatment Outcomes by Condition

	EF-CBT		RMT		pa	qb
CGI-I (95% CI)c	47.1%	(0.369-0.575)	20.1%	(0.106-0.349)	.003	.003∗∗
PARS (95% CI)d
Week 1	19	(18.1-19.8)	18	(16.8-19.3)	.70	.7
Week 6	16.1	(15.2-17)	16.2	(14.9-17.4)	.37	.4933
Week 9	14.4	(13.5-15.3)	14.7	(13.5-16)	.23	.46
Week 12	13.1	(12.2-13.9)	14.9	(13.7-16.2)	.007	.028∗

Note: This table presents the proportion of participants who were designated as treatment responders, as well as the estimated means for anxiety severity across the duration of the study for each condition.

CGI-I= Clinical Global Impressions–Improvement; EF-CBT = exposure-focused cognitive-behavioral therapy; PARS = Pediatric Anxiety Rating Scale; RMT = relaxation mentorship training.

^aMann-Whitney U test or Pearson χ² test.

^bFalse discovery rate correction for multiple testing.

^cTreatment responders defined as individuals receiving a 1 (very much improved) or 2 (much improved) on the CGI-I. Estimated proportions as determined by a logistic regression model are reported.

^dEstimated marginal means as determined by a linear mixed effects model.

^∗p < .05; ∗∗p < .01.

Youth- and caregiver-reported anxiety severity

Youth-reported anxiety symptoms improved across both treatment conditions (F_3,417 = 18.80, p < .001); however, no differences were observed between the 2 conditions (F_3,417 = 0.78, p = .50) (Figure 3). Caregiver-reported anxiety symptoms also improved over the course of the study across conditions (F_2,183 = 60.99, p < .001), but revealed no differences between conditions (F_2,183 = 1.50, p = .23) (Figure 4).

Figure 3.

Estimated Marginal Means of Anxiety Severity (Child Report) Across Treatment

Note:Estimated mean Multidimensional Anxiety Scale for Children (MASC) scores across 12 weeks of treatment. EF-CBT = exposure-focused cognitive-behavioral therapy; RMT = relaxation mentorship training.

Figure 4.

Estimated Marginal Means of Anxiety Severity (Caregiver Report) Across Treatment

Note:Estimated mean Screen for Child Anxiety and Related Emotional Disorders (SCARED) scores across 12 weeks of treatment. EF-CBT = exposure-focused cognitive-behavioral therapy; RMT = relaxation mentorship training.

EF-CBT Predictors

Anxiety Severity at Baseline

The number of anxiety diagnoses assigned by an IE at the initial evaluation did not predict EF-CBT response (PARS; β = .51 ± .45, p = .26). An auxiliary analysis examining PARS week 1 severity as a predictor of CGI-I-defined treatment response was also not significant (Supplement S4, available online).

Comorbidities

Neither ADHD (β = .03 ± .08, p = .66) nor depressive (β = .09 ± .04, p = .46) symptoms were associated with posttreatment PARS scores.

Age

The effect of time on PARS score reduction was significant (F_3,282 = 116.09, p < .001), indicating a consistent decrease in PARS scores throughout treatment across the age range (7-18 years). Although the main effect of age on PARS score reduction (F_1,94 = 3.37, p = .7) and the interaction between age and time (F_3,282 = 2.48, p = .06) were not significant, they suggested that age may have a potential influence on treatment outcomes. This was further evidenced by the parameter estimates from the fixed effects analysis, which revealed a significant interaction at week 12 when controlling for PARS scores at previous sessions (ie, session 12 − [session 1, session 6, session 9]). Additionally, there was a significant interaction of age and PARS reduction at session 12 when controlling for PARS at previous sessions (eg, session 12 − [session 1, session 6, session 9]; β = .557, p = .043) (Table 4). This finding suggests that the impact of age on the efficacy of the treatment becomes increasingly apparent over time, with younger participants exhibiting a greater reduction in PARS scores by the end of treatment compared with older participants.

Table 4.

Longitudinal Model Examining Age as a Predictor of Anxiety Severity

	β	SE	t	p
(Intercept)	15.441	0.376	41.081	<.001
Gender (female)	.633	0.762	0.83	.409
Sessions completed	−.203	0.33	−0.614	.54
Age (years)	.597	0.325	1.835	.07
Week 12 − (week 1, week 6, week 9)	−3.43	0.273	−12.549	<.001
Age × week 6 − (week 1)	−.439	0.335	−1.311	.191
Age × week 9 − (week 1, week 6)	.367	0.29	1.263	.208
Age × week 12 − (week 1, week 6, week 9)	.557	0.274	2.034	.043

Note: Fixed effects summaries from the longitudinal age model for the EF-CBT group with missForest imputations. Intercepts can be interpreted as the expected week 1 EF-CBT Pediatric Anxiety Rating Scale scores for a 12-year-old boy who eventually completed 12 weeks of therapy. EF-CBT = exposure-focused cognitive-behavioral therapy.

Treatment Delivery Context

The context of delivering EF-CBT (in-person vs telehealth) did not predict posttreatment PARS scores (β = .61 ± .96, p = .52).

Discussion

This clinical trial examined 139 patients with pediatric anxiety disorders who were randomized to receive 12 sessions of either EF-CBT or a relaxation-based control condition, RMT. Based on IE ratings, youth assigned to EF-CBT were more likely to be respond to treatment and experienced more pronounced and clinically significant improvement compared with youth who received RMT. Among patients who received EF-CBT, baseline number of anxiety disorders, depression and ADHD symptom comorbidity, participant age, and treatment delivery context were examined as predictors of treatment outcome. Notably, age emerged as the only significant predictor of EF-CBT outcome, with more clinical improvement experienced by younger participants.

Superiority of EF-CBT Over RMT

The current study replicated findings from an examination of an interim subset (73%) of the current sample.²² As in the initial analyses, youth assigned to EF-CBT exhibited lower independently measured anxiety severity at the end of treatment and were more likely to reach CGI-defined treatment responder status than youth assigned to RMT. Critically, the EF-CBT package excluded relaxation and moved to exposure sessions after only 3 sessions (2 sessions focused on psychoeducation and 1 session on cognitive restructuring). This is in contrast to some other commonly used CBT packages in which exposure does not begin until the ninth session, following 8 initial sessions devoted to psychoeducation, relaxation skill training, and cognitive restructuring.⁸ Overall, our findings support the conclusion of our interim report that CBT for pediatric anxiety disorders can be effective when patients move quickly to exposures and without the use of relaxation techniques.

Contextualizing in Pediatric Anxiety Research

The superiority of EF-CBT over RMT is especially important given that RMT was designed to be a credible and active control condition. This was confirmed by our finding that treatment was equally credible to participants across both conditions. RMT consisted of relaxation exercises, along with additional credible nonspecific control exercises, including mentorship and autobiographical writing. In a recent review,⁴ CBT was shown to outperform waitlist controls (OR 4.43) and to possibly outperform attention controls, treatment as usual, and alternative treatments (ORs 2.75, 2.74, and 0.89 respectively).⁴ Our current trial found an OR of 3.53 in favor of EF-CBT over RMT, providing evidence that EF-CBT outperforms a credible alternative treatment for pediatric anxiety. Further research is needed to examine whether EF-CBT could outperform CBT packages including relaxation along with exposure for pediatric anxiety.

Informant Reports

In contrast to more favorable outcomes for EF-CBT compared with RMT based on IE ratings of anxiety severity, response to these treatment conditions did not significantly differ when assessed based on youth or caregiver report. This is consistent with prior research on CBT for youth anxiety, which found that compared with IEs, youth and caregivers are less likely to report CBT outperforming control treatments for pediatric anxiety disorders.⁴

Discrepancy between IE- and participant-reported improvement after EF-CBT relative to RMT may result from interacting factors. For example, youth and caregivers may be less likely to report fine-grained changes in impairment and distress that occur gradually over time or to integrate information across sources (eg, child and caregiver) and may be less aware of normative development and behavior than a trained independent rater.³⁸ Additionally, the active control condition used in this study was evaluated as having the same credibility and expectation of improvement as EF-CBT, which can influence self-reported improvement.^39,40 It is also possible that the self-report measures chosen for this study were less reliable measures of treatment change than independent evaluation because measurement variance on self-report and caregiver report anxiety measures may be more variable over time. Some self-report measures, such as the Revised Children’s Anxiety and Depression Scale,⁴¹ have been shown to more reliably measure the same anxiety construct across treatment sessions^42,43 and should be prioritized in future research. Finally, it is possible that youth and caregivers were better raters than our IEs; these informants had access to more information and were perhaps able to rate the severity of illness more accurately than our clinical raters. Ultimately, more research is needed to tease out the factors driving differential outcomes by different reporters and help determine the best methodology for assessing outcomes in treatment research.

EF-CBT Predictor Analyses

Age

Participant age was found to be a significant predictor of EF-CBT outcome, such that younger patients experienced more improvement by the end of treatment. This novel finding is significant given that much of the prior literature has found no relation between age and treatment outcomes.^19,20,44 There is both empirical and theoretical reason to suspect that our result is more than an anomaly. Recent research has demonstrated that children experience steeper slopes of improvement during exposure sessions compared with adolescents.⁹ This difference may reflect neurodevelopmental changes that impair extinction learning during adolescence compared with preadolescent and postadolescent periods.⁴⁵ This finding may also be related to differential support offered across development; younger children may receive more support in completing and engaging in exposure tasks between sessions compared with adolescents, which could influence treatment response. Understanding the unique responses to treatment across development is critical as we aim to optimize treatment at each phase of development, across the life span.

More robust EF-CBT response at younger ages raises the question of how to optimize treatment by developmental stage. Our results suggest that children with anxiety disorders may benefit from moving quickly to exposure-based sessions of CBT. In older patients, it is possible that cognitive aspects of CBT may be more helpful. Indeed, prior work shows steeper slopes of anxiety reduction over the course of multiple cognitive restructuring sessions in adolescents compared with children,⁹ perhaps as a result of more mature cognitive capacity. Our EF-CBT package included only 1 session of cognitive restructuring with additional use of this technique incorporated in extra sessions per therapist discretion. Thus, it is possible that our older patients would have responded faster and more fully had more dedicated cognitive restructuring sessions been included during treatment. Alternatively, older youth may require more exposure therapy sessions and/or increased exposure intensity at each session to yield the same effect as in younger participants.

Ultimately, our results suggest that patient age could guide the selection of CBT components to improve treatment response among affected youth and justify future research to identify optimal treatment profiles across development. For example, the addition of cognitive testing or brain imaging to identify developmentally sensitive periods for individually guided selection of CBT components deserves consideration given the earlier development of putative neural substrate of exposure (ie, fear circuits) relative to cognitive restructuring (ie, prefrontal cortex) in children and adolescents, respectively.^46,47 Such work will need to acknowledge that older youth may have a longer duration of illness—a factor shown to reduce response to treatment⁴⁸ and that may impact cognitive and brain function. Disentangling age and duration of illness effects will require future clinical trial research that includes same-age participants with varying illness durations.

Nonsignificant Predictors

Contrary to study hypotheses, many of the variables examined in the current study were not significant predictors of EF-CBT improvement. Baseline severity of clinically measured anxiety symptoms—indexed as number of anxiety diagnoses—did not predict improvement. This was confirmed by our auxiliary analysis, which indicated that session 1 PARS scores did not predict session 12 CGI-I scores. This null finding is somewhat surprising given that baseline severity has been found to predict treatment response for youth with anxiety in prior work.⁶ However, previous literature has primarily looked at anxiety severity on treatment outcomes, rather than change in severity. Youth with higher baseline severity would require more improvement than less symptomatic youth to achieve the same treatment outcomes. The present study demonstrated that baseline anxiety (defined as number of anxiety diagnoses and in S4, available online, as pretreatment symptom severity) did not predict amount of improvement in EF-CBT. This is important because it suggests that youth with elevated anxiety may receive the same amount of benefit from EF-CBT as their less anxious peers; however, they may require more sessions of EF-CBT to reach the same treatment outcomes.

As with the number of independently evaluated anxiety disorders, ADHD and depressive symptoms at baseline were not significant predictors of EF-CBT improvement, suggesting that EF-CBT need not be reserved for youth with “clean” presentations of anxiety. This finding contrasts with some prior work showing a relation between comorbid concerns and poorer response to CBT in patients with pediatric anxiety disorders.^12,13 However, it should be noted that prior research on these relations is limited⁶ and somewhat mixed. In particular, some studies demonstrate that comorbid symptoms (ie, including subclinical levels) may be a less significant predictor of treatment response than comorbid diagnoses.⁴⁹ This study included only 15 youths meeting criteria for ADHD and excluded youth with major depressive disorder, posttraumatic stress disorder, psychotic symptoms, active risk to self or others, and autism spectrum disorder. These eligibility criteria were chosen owing to our focus on anxiety as the primary source of interference and target of treatment and for consistency with prior CBT clinical trial research (CAMS). However, to understand the full potential of EF-CBT for anxiety, future work should examine predictors of treatment response with broader inclusion criteria, as many of the youth who experience clinically significant anxiety also experience comorbid diagnoses.⁵⁰

The present analyses found no difference in treatment outcomes for youth receiving EF-CBT via in-person care compared with youth who received it via telehealth, consistent with prior research demonstrating the equivalence of teletherapy for youth with pediatric anxiety disorders dating before COVID-19.²¹ Equal outcomes across in-person therapy and teletherapy for EF-CBT are promising, as teletherapy provides increased access to care for youth experiencing barriers to attending in-person sessions (eg, transportation, distance to specialty care, childcare for siblings). Although this equivalence comes with many benefits, there is room for caution. The nascency of teletherapy research means that less is known about significant predictors of differential treatment outcomes in virtual vs in-person formats. As we await data indicating the optimal candidates for teletherapy, expert clinical recommendations suggest that consideration of characteristics such as patient age, motivation, and preferences be considered when determining whether teletherapy may be an appropriate treatment.⁵¹

Notably, the interim sample included only youth receiving therapy in person; all additional participants reported here participated after the onset of the COVID-19 pandemic and received care via teletherapy. This suggests that there was also no cohort effect of the pandemic on treatment outcomes for EF-CBT in this sample.

This study has several notable limitations. First, youth assigned to RMT completed fewer sessions on average and were more likely to not complete treatment. Nonetheless, analyses controlled for number of sessions attended, supporting the superiority of EF-CBT over RMT despite this difference between conditions. Second, related to sample characteristics, more than 75% of the sample identified as White and less than 10% identified as Hispanic/Latino, and thus this sample does not accurately reflect the national youth population.⁵² This limits the generalizability of our findings and perpetuates a historical centering of White participants in mental health research.⁵³ Additionally, the current study did not stratify randomization by race or ethnicity. This, combined with the underrepresentation of non-White participants across categories, led to imbalances in our assignment to condition. In the future we propose to stratify by race and ethnicity, as well as to examine whether race/ethnicity and principal anxiety diagnosis predict EF-CBT response.

Third, we were unable to directly examine baseline anxiety severity as a predictor of anxiety improvement due to methodological limitations. Our primary outcome variable was the PARS, and for each moderator analysis we controlled for baseline PARS severity to examine whether the moderator was associated with anxiety improvement, rather than posttreatment anxiety severity. As a result, we had to choose a different variable as a proxy for anxiety severity at baseline (ie, number of anxiety diagnoses). It is possible that a person with fewer anxiety diagnoses has more severe illness than someone with more diagnoses, so we also conducted an analysis of baseline anxiety severity predicting global improvement (reported in auxiliary analyses), which yielded similar findings.

In this article we opted to examine number of comorbid anxiety diagnoses and continuous nonanxiety comorbid predictors of EF-CBT (ie, symptoms of ADHD, symptoms of depression) to maximize power and interpretability. However, this prevented us from examining diagnostic comorbidities as predictors of treatment response. Additionally, major depressive disorder was an exclusion criterion for this study, which likely restricted the range of depressive symptoms that were reported, thus limiting our ability to fully test this hypothesis. Future studies with larger samples sizes and expanded eligibility criteria are needed to examine how specific anxiety disorders and comorbid nonanxiety diagnoses influence youth response to EF-CBT. For example, some research indicates that social anxiety disorder may result in worse outcomes for youth receiving CBT,^19,54 but to our knowledge this finding has not been extended in CBT with an emphasis on exposure. Additional variables that may also be worth examining as predictors of treatment outcome include family socioeconomic status, family accommodation, homework adherence, and exposure task difficulty.

This randomized controlled trial supports exposure-focused treatment for pediatric anxiety disorders, bolstering prior findings that relaxation is not necessary for youth engagement with or even response to CBT for anxiety^5,22,55 and adding important new insights that younger children may especially benefit from exposure-based models and that teletherapy is as effective as in-person care for this intervention. Whereas there remain many opportunities for additional research to further our understanding and optimization of psychotherapy for youth anxiety, these findings can guide clinicians and affected families alike in choosing exposure-based CBT practices for the treatment of pediatric anxiety disorders.

CRediT authorship contribution statement

Emily L. Bilek: Writing – original draft, Supervision, Investigation. Ann M. Iturra-Mena: Writing – review & editing, Writing – original draft, Formal analysis. Hannah C. Becker: Writing – review & editing, Formal analysis, Data curation, Conceptualization. K. Luan Phan: Writing – review & editing, Investigation, Funding acquisition, Conceptualization. Christopher S. Monk: Writing – review & editing, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization. Kate D. Fitzgerald: Writing – review & editing, Writing – original draft, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization.

Supplemental Material

Supplementary Figure S1.