Trajectory and Predictors of Alliance in Cognitive Behavioral Therapy for Youth Anxiety

Abstract

Multilevel growth analysis was used to establish the shape of change (mean growth trajectory) for youth- and therapist-rated alliance in cognitive behavioral therapy (CBT) for anxious youth and to identify between-youth predictors of alliance trajectory. Youth (N = 69; ages 7–17; 52.2% female) and their parents participated in an empirically supported CBT protocol. Therapists rated alliance each session and youth every four sessions. Data were fit to four growth models: linear, quadratic, a dual slope, and a novel “alliance rupture” model. Two-level models were estimated to examine the effect of youth age, sex, pretreatment symptom severity, diagnostic comorbidity, early treatment factors (use of Selective Serotonin Reuptake Inhibitors), and coping styles (engagement, disengagement, and involuntary coping). A dual slope model fit therapist data best, whereas youth data did not evidence systematic growth. Two-level growth models identified that pretreatment anxiety severity predicted higher initial alliance levels. Depressive symptoms predicted less linear growth and engagement coping predicted greater growth during exposure sessions. No variables predicted preexposure growth. In the therapist model, 22% of initial alliance, 50% of preexposure growth, and 75% of postexposure growth were accounted for by between youth variables (mood disorder, anxiety and depression symptoms, engagement and involuntary coping). Therapist-reported alliance ratings may grow over the course of manual-based CBT, even during exposure-focused sessions. Pretreatment youth factors and coping style may influence the absolute value and linear trajectory of alliance during CBT. Findings about alliance-influencing factors can help set expectations for, and enhance training in, empirically supported treatments.

Notes

¹Three-level models (time nested within youth and therapist) were examined, but models did not converge, likely because mode number of cases per therapist was two, limiting distinction between case and therapist. Thus, two-level models were used.

²Multilevel change models can accommodate varying numbers of data collection waves, but potential problems can result from severely unbalanced data sets, commonly resulting from participant attrition (Singer & Willet, 2003). Models using balanced data sets can be parameterized more easily, random effects can be estimated more precisely, and computer algorithms will converge more rapidly. The CBT protocol in this study permitted flexibility in number of sessions (16–20) creating expected, but potentially problematic, variation in number of data points toward the end of therapy. Thirteen youth received more than 18 sessions (range = 19–23). We limited data to the first 18 sessions to remove outliers and minimize balance issues. Very little data were cut to create this balance: 81% of youth received 18 sessions or fewer, and the first 18 sessions accounted for 96.6% of total possible Level 1 symptom data.

³ΔDev was compared to critical values of a chi-square distribution with df set to the difference in number of parameters between the two models compared. ΔAIC and ΔBIC of 0 to 2 is considered “weak” evidence for significant differences between models, 2 to 6 is “positive,” 6 to 10 is “strong,” and differences over 10 are “very strong,” where smaller AIC and BIC values represent better fit (Singer & Willet, 2003).

Note: ADIS = Anxiety Disorders Interview Schedule; CSR = Clinician Severity Rating; Dx = diagnosis; TASC/A = Therapeutic Alliance Rating Scale Child/Adolescent; STAIC = State-Trait Anxiety Inventory for Children; RCADS = Revised Child Anxiety and Depression Scale; MDD = Depression subscale; CBCL = Child Behavior Checklist; RSQ = Responses to Stress Questionnaire.

Note: AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion.

^a Slope 1 = linear session count in Models B–D; preexposure slope in Model E.

^b Slope 2 = quadratic parameter in Model C, exposure covariate in Model D, and postexposure slope in Model E.

†p < .10. *p < .05. **p < .01. ***p < .001.

Note: AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion.

^a Slope 1 = linear session count in Models B–D; preexposure slope in Model E.

^b Slope 2 = quadratic parameter in Model C, exposure covariate in Model D, and postexposure slope in Model E.

†p < .10. *p < .05. **p < .01. ***p < .001.

Note: RCADS = Revised Child Anxiety and Depression Scale; MDD = Depression subscale; RSQ = Responses to Stress Questionnaire.

†p < .10. *p < .05. **p < .01. ***p < .001.

Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/hcap.