A Randomized Controlled Trial of Cognitive-Behavioral Therapy for Generalized Anxiety Disorder with Integrated Techniques from Emotion-focused and Interpersonal Therapies

Cognitive-behavioral therapy (CBT) is the only psychotherapy currently meeting criteria as an empirically supported treatment for generalized anxiety disorder (GAD), yet many clients retain clinically significant symptoms or fail to maintain gains long term. Theoretical models implicate emotional processing avoidance and interpersonal problems in maintaining worry and GAD. Worry’s verbal-linguistic nature may dampen emotional learning, and GAD individuals often fear emotions and show heightened reactivity. Interpersonal difficulties stemming from nonadaptive attachment, biased cognitions, and skills deficits are common. Standard CBT may insufficiently address these emotional and interpersonal maintenance factors. The study examines whether adding an interpersonal and emotional processing (I/EP) module to CBT improves outcomes relative to adding supportive listening (SL), with a primary hypothesis that CBT + I/EP would yield superior outcomes, particularly at long-term follow-up.

Prior work suggests worry functions as avoidance of emotional processing, dampening physiological reactivity and impeding emotional learning (e.g., Borkovec et al., 2004; Borkovec & Hu, 1990). Individuals with GAD report using worry to control emotion and experiencing greater threat and discomfort from emotions, alongside heightened emotional intensity and reactivity (Mennin et al., 2005; 2007; Erickson & Newman, 2007). Interpersonal models of GAD highlight attachment-related difficulties, biased interpersonal cognitions, and skills deficits (Newman & Erickson, 2010; Salzer et al., 2008), with reports of inconsistent caregiver availability in childhood (Cassidy et al., 2009). Evidence indicates lower emotional processing in CBT compared to reflective listening and poorer CBT response among clients with dominant interpersonal problems, with untreated interpersonal issues predicting failure to maintain gains (Borkovec & Costello, 1993; Borkovec et al., 2002). Initial open trial data on integrated CBT with interpersonal and emotional processing techniques were promising (Newman et al., 2008), motivating a randomized test of whether adding I/EP to CBT improves efficacy beyond CBT + SL.

Design: Randomized controlled trial with an additive, sequential design. Participants received 14 sessions, each 2 hours comprising 50 minutes of CBT followed by 50 minutes of either I/EP or SL. Conditions: CBT + I/EP (n = 43) vs. CBT + SL (n = 40). Therapists: Three experienced doctoral-level psychologists; weekly supervision; adherence and quality checks performed. Recruitment and Eligibility: Community recruitment (media and clinician referrals) from 1998–2003; follow-ups through 2005. Inclusion criteria: principal DSM-IV GAD diagnosis confirmed by two independent ADIS-IV interviews; GAD Clinician’s Severity Rating (CSR) ≥ 4; stable psychotropic medication ≥ 6 weeks or medication-free with agreement to maintain stability; ages 18–65; no concurrent psychotherapy or prior adequate CBT dose; absence of medical causes of anxiety, substance abuse, psychosis, or organic brain syndrome. Sample: 83 participants; primarily White; mean age ~37; comorbidity rates reported; random assignment to therapist and condition; 13 early drop-outs (CBT-SL = 4, CBT-I/EP = 9). Blinding: Independent assessors blinded to condition conducted evaluations at pre, post (10–14 days after last session), 6-, 12-, and 24-month follow-ups. Measures: Primary composite outcome for GAD symptoms combining standardized z-scores of CSR for GAD, PSWQ, HARS, and STAI-T anxiety subscale; clinically significant change indices (endstate functioning, responder status, GAD diagnosis status); Expectancy/Credibility (first session). Secondary: HRSD (depression), IIP-C (interpersonal problems), additional treatment seeking (psychotherapy, psychotropic medications), RRAQ (fear of relaxation). Treatment Protocols: CBT segment targeted intrapersonal anxiety (self-monitoring, relaxation, self-controlled desensitization, cognitive restructuring), explicitly excluding interpersonal/developmental/affective deepening work. I/EP segment targeted identification of interpersonal needs and patterns, facilitation of emotional processing, use of therapeutic relationship, alliance rupture detection and repair, emotion-focused techniques, and interpersonal skills training, avoiding direct cognitive techniques to foster experiential processing. SL segment provided supportive, reflective listening without techniques to deepen affect or provide suggestions/advice. Adherence and Quality: 100% of three sessions per client coded for protocol adherence; zero nontrivial breaches among 11,453 therapist utterances; trivial breaches minimal. Independent quality ratings met a priori competence criteria (average ratings: CBT = 4.8/6; I/EP = 4.31/6; SL reflective listening average ≥ 3; SL mean ~4.7). Statistical Analysis: Linear mixed-effects, piecewise model with pre-to-post (Piece 1) and post-to-follow-up (Piece 2); random effects at intercept and slopes retained when significant; fixed effects included time pieces, condition, interactions; restricted maximum likelihood for random effects and full maximum likelihood for fixed effects. Missing data assessed via pattern-mixture models; no meaningful bias detected; full information maximum likelihood used; multiple imputation sensitivity matched results. Clinically significant change definitions: High endstate functioning = meeting nonanxious norms on ≥ 3 of 5 measures (CSR, PSWQ, HARS, STAI-T anxiety, RRAQ); responder = ≥ 20% improvement on ≥ 3 of 5 measures; IIP-C high endstate = within normal range on ≥ 6 of 8 subscales. Power analysis targeted 25 per condition (f = 0.32, α = .05, power .80), actual sample larger.

- Expectancy and credibility: No significant between-condition differences after first session CBT segment or second segment (I/EP vs SL); expectancy ~66–68% and credibility ~20–22 across segments; MANOVA F(4,63) = 0.996, p = .993. - Primary composite GAD symptoms: Significant improvement pre-to-post across both treatments (Piece 1 B = −.34, SE = .03), t(77) = −11.86, p < .0001; 95% CI [−.40, −.28]; very large effect size d = 1.86. No between-condition differences at post (Piece 1 × Condition B = −.01, p = .76, d = 0.07) or across follow-up (Piece 2 B = .002, p = .70, d = 0.06; Condition B = −.06, p = .55, d = .13; Piece 2 × Condition B = −.004, p = .58, d = .12). Gains maintained through 24 months (nonsignificant Piece 2). - Clinically significant change at 24 months: CBT + I/EP: 68.8% high endstate (GAD measures), 71.9% responder, 75% no longer met GAD criteria, 83.9% high endstate on IIP-C. CBT + SL: 52.9%, 64.7%, 63.6%, and 66.7%, respectively. None differed significantly by condition (chi-square tests nonsignificant). - Depressive symptoms (HRSD): Significant pre-to-post decrease irrespective of condition (Piece 1 B = −.28, SE = .04), t(77) = −6.77, p < .0001; 95% CI [−.36, −.20]; d = 1.05. No condition differences; gains maintained (Piece 2 nonsignificant). - Interpersonal distress (IIP-C): Significant pre-to-post decrease (Piece 1 β = −.20, SE = .06), t(77) = −3.20, p = .002; 95% CI [−.32, −.08]; d = 0.50. No condition differences; gains maintained. - Additional treatment seeking: Subsequent psychotherapy rates low and similar between conditions across follow-ups. Among those not on medication at pretreatment, all individuals who sought psychotropic medication at 6- and 12-months were in CBT + SL; differences not present at 24 months. - Overall: Both treatments produced large improvements and maintained gains to 2 years; adding I/EP did not confer statistically significant advantages over adding supportive listening.

The study tested whether augmenting CBT for GAD with interpersonal and emotional processing techniques enhances outcomes, particularly long term. Both CBT + I/EP and CBT + SL yielded large, durable improvements in GAD symptoms, depression, and interpersonal distress, with no significant between-condition differences on primary or secondary outcomes or clinically significant change indices. Possible reasons include: the selected non-CBT techniques may have been insufficient or not optimally integrated; additive sequential design may have limited synergy relative to seamless integration; common factors controlled via SL may have matched incremental benefits; and sample size may have limited power to detect medium effects, as several clinically significant percentages favored CBT + I/EP. The augmentation may benefit specific subgroups (e.g., those with particular interpersonal problems or emotion regulation difficulties), suggesting aptitude-by-treatment interactions. Future work should examine process and mechanism (e.g., whether I/EP increased emotional experiencing/processing) and refine integration strategies to target emotional avoidance and interpersonal deficits more directly.

In a randomized additive design, CBT for GAD produced large, sustained benefits regardless of whether it was followed by interpersonal/emotional processing therapy or supportive listening. The addition of I/EP did not significantly augment outcomes across 2 years. Findings underscore the robustness of CBT and suggest that targeted augmentation may be warranted for specific patient profiles rather than universally. Future research should pursue seamless integrative protocols, larger samples to detect subgroup effects, and mechanism-focused assessments (e.g., emotional avoidance) to optimize personalized treatment.

- Recruitment via advertisements; participants largely White and educated, limiting generalizability. - Two-stage ADIS with second interview contingent on first may inflate interrater reliability estimates for GAD. - No reliability calculated for adherence raters; potential observer drift unassessed. - Additive, sequential design enhances internal validity but may reduce external validity and ecological fit for clinical practice. - Absence of a direct outcome measure of emotional avoidance limits assessment of targeted mechanism change. - Sample size, despite power analysis, may have been insufficient to detect medium between-condition effects; dropouts occurred early but were retained analytically via full information maximum likelihood.