Abstract
Treatment specificity and long-term recovery mediation of peer-involvement group cognitive behavioral therapy (GCBT) and parent-involvement CBT (PCBT) were investigated for youth anxiety disorders. Two hundred forty youths with primary anxiety diagnoses participated in a randomized controlled efficacy trial. Youth anxiety and peer variables/mediators (positive peer–youth relationships; social skills) and parent variables/mediators (psychological control; negative parent–youth relationships) were assessed. At posttreatment and 12-month follow-up, positive peer-youth relationships were significantly higher in GCBT than PCBT (specificity) according to youth and parent ratings. At posttreatment but not follow-up, parental psychological control was significantly lower in PCBT than GCBT (specificity) and findings varied by informants. Parental psychological control and positive peer–youth relationships were putative mediators. The two CBTs produced similar anxiety reductions through different mechanisms. CBT targets show specificity and mediation, providing insight into specific mechanisms through which GCBT and PCBT bring about anxiety reduction and guidance for streamlining these CBTs in practice.
Cognitive behavioral therapy (CBT) is the most well-researched and strongest evidence-based psychological treatment for anxiety disorders in children and adolescents (referred to as youth unless referring to a specific subgroup; e.g., Higa-McMillan, Francis, Rith-Najarian, & Chorpita, 2016; Silverman, Pina, & Viswesvaran, 2008). Since Kendall’s (1994) seminal study demonstrating the efficacy of youth-focused individual CBT (ICBT), peer-based group cognitive behavioral therapy (GCBT; e.g., Silverman et al., 1999) and parent-involvement cognitive behavioral therapy (PCBT; Kendall, Hudson, Gosch, Flannery-Schroeder, & Suveg, 2008; Silverman, Kurtines, Jaccard, & Pina, 2009) were developed, tested, and shown to be similarly efficacious (e.g., Breinholst, Esbjørn, Reinholdt-Dunne, & Stallard, 2012; Silverman et al., 2008; Thulin, Svirsky, Serlachius, Andersson, & Öst, 2014).
Unfortunately, randomized clinical efficacy trials that aim to advance understanding of mechanisms to explain how change is produced are scarce, despite calls for explanatory-efficacy trials for over two decades (Gaynor, 2017; Kazdin, Bass, Ayers, & Rodgers, 1990; Kendall, Olino, Carper, & Makover, 2017; Silverman & Kurtines, 1997; Weersing & Weisz, 2002). In this article, we present an explanatory-efficacy trial for youth anxiety disorders focusing on GCBT and PCBT. Both are well supported and yield similar medium effect sizes (e.g., Silverman et al., 2008). Thus, we hypothesized both CBTs will produce statistically significant changes on the study’s outcomes. We moved beyond efficacy outcome analysis to advance understanding about whether variables targeted in GCBT and PCBT are specific to each treatment and whether these variables in turn produce youth anxiety reduction, including long-term recovery mediation and patterns of directionality of change (e.g., La Greca, Silverman, & Lochman, 2009). The variables we targeted in GCBT and PCBT are emphasized in theoretical models of the development and maintenance of anxiety and its disorders in youths. These extensively studied models highlight the critical role of peer relationship functioning and problematic social skills and parental psychological control (i.e., behaviors that inhibit youths’ psychological and emotional development) and parent–youth relationships (e.g., Chorpita & Barlow, 1998; Ginsburg, Silverman, & Kurtines, 1995; La Greca & Landoll, 2011; Motoca, Williams, & Silverman, 2012). In the following, we present background literature, the study’s main research questions, and hypotheses.
Do GCBT and PCBT Produce Treatment-Specific Effects on Targeted Variables?
GCBT and PCBT emphasize graded exposures to anxiety-provoking situations and cognitive strategies (e.g., Kendall et al., 2008; Silverman et al., 1999, 2009). Each treatment targets distinct variables: peer variables in GCBT and parent variables in PCBT. Treatment-specific effects, or treatment specificity, address whether changes produced in targeted variables are indeed because of the specific therapeutic procedures contained/delivered in each treatment. Support for treatment specificity would suggest that mechanisms underlying therapeutic change can be explained by that specific procedure delivered in each treatment, opening critical new lines for further mechanism-based research and providing critically needed data to support the procedure’s inclusion/delivery. Data showing that specific variables change irrespective of targeting would suggest otherwise, supporting the notion that other mechanisms are likely involved in therapeutic outcome and raising questions about continuing the procedure’s inclusion/delivery.
Studies on treatment specificity are extremely rare in child and adolescent randomized clinical efficacy trials, including anxiety. Kolko, Brent, Baugher, Bridge, and Birmaher (2000), one of the few and earliest studies, found no evidence of treatment specificity for cognitive variables and family variables despite their respective targeting in CBT and family therapy. In youth anxiety, we found only one behavioral therapy efficacy trial. This was a comparison of Social Effectiveness Therapy for Children with social phobia (SET-C; N = 67) with Testbusters (control; Beidel, Turner, & Morris, 2000). SET-C targeted social skills and performance skills; Testbusters did not. Treatment specificity was found for observer-rated social skills. That is, children in SET-C showed higher social skills than children in Testbusters, from pretreatment to posttreatment.
We found only three youth anxiety efficacy trials that examined whether treatment specificity occurs when parent variables are targeted in PCBT but not ICBT (Jongerden & Bögels, 2015, N = 104; Kendall et al., 2008, N = 161; Silverman et al., 2009, N = 119). Parent variables targeted were parenting skills and parent-child relationships. Treatment specificity was not found in any of these trials: Parent variables significantly improved in PCBT when targeted but also in ICBT when not targeted, pretreatment to posttreatment.
In the present efficacy trial, we investigated whether GCBT, which targeted peer variables, changes those peer variables specifically but PCBT does not and whether PCBT, which targeted parent variables, changes those parent variables specifically but GCBT does not. Because the targets were clearly distinctive in the two CBTs, our first hypothesis was that GCBT but not PCBT would produce treatment specificity effects on positive peer–youth relationships and youths’ social skills at posttreatment because these were the variables targeted. Our second hypothesis was that PCBT but not GCBT would produce treatment specificity effects on perceived parental psychological control and negative parent–youth relationships at posttreatment because these were the variables targeted. We also explored whether these treatment specificity effects would be maintained at 12-month follow-up.
Do Peer Variables Targeted in GCBT and Parent Variables Targeted in PCBT Mediate Anxiety Treatment Outcome and/or Long-Term Recovery, and Is the Directional Pattern of Change From Targeted Variables to Anxiety Reduction?
Research on treatment mediation is rare, including in youth anxiety, despite calls for such research (e.g., Kazdin et al., 1990; La Greca et al., 2009; Weersing & Weisz, 2002; Weisz et al., 2017). Most trials focus on a single variable; in youth anxiety studies, cognitive variables are most often examined as mediators (Peris et al., 2015; Treadwell & Kendall, 1996), although a recent study examined somatic symptoms (Hale et al., 2018). Further, most studies focus on pre- to posttreatment analysis only. A rare exception is from the Child Anxiety Multimodal Study (Walkup et al., 2008), which focused again on cognitive variables as mediators of ICBT recovery at 3-month follow-up, with supportive results (Kendall et al., 2016).
For the variables of interest in our study, only one efficacy trial focused on peer variables (Alfano et al., 2009). Findings suggested that youths’ self-rated loneliness may be a mediator of SET-C outcome. Two efficacy trials examined parent variables’ mediational role in PCBT long-term recovery at 1-year follow-up (Settipani, O’Neil, Podell, Beidas, & Kendall, 2013; Silverman et al., 2009). In Silverman et al. (2009), no evidence was found that either youth or parent ratings of parent variables mediated anxiety recovery, although interesting patterns of directionality emerged (discussed in the following). Settipani et al. (2013) found youth ratings of parental psychological control and family affective involvement significantly mediated recovery.
In the present randomized clinical efficacy trial, we examined peer variables (positive peer–youth relationships, youth social skills) and parent variables (parental psychological control, negative parent–youth relationships) as mediators of long-term recovery at 12-month follow-up. All variables were assessed from perspectives of youths and parents; thus, in all instances, we are assessing sources’ perceptions of changes. We also extended past work in ways that aligned with the efficacy trials of Silverman et al. (2009) and Settipani et al. (2013). In almost all past trials, investigators made the plausible and reasonable assumption/hypothesis that the directional pattern of change is from targeted variable (i.e., hypothesized mediator) to outcome. Thus, this has been the only directional pattern of change modeled/analyzed. Exceptions are Silverman et al. and Settipani et al., who modeled/analyzed whether the directional pattern is from youth anxiety reduction at posttreatment to parent variables at 12-month follow-up. Findings revealed directional patterns varied by source and measure. Some showed change in the plausible, hypothesized direction: Targeted parent variables at posttreatment were associated with positive changes in youth anxiety outcome at 12-month follow-up. Other findings showed the direction was not as hypothesized but instead was from youth anxiety reduction at posttreatment to changes in parent variables at 12-month follow-up. This was true even in ICBT, which was especially unexpected because no parent variables were targeted in this arm.
Thus, in addition to treatment specificity, in the present study, we investigated contemporaneous mediation of treatment outcome and mediation of long-term recovery at 12-month follow-up for GCBT and PCBT and directional pattern of change. Investigating these issues is critically important to advance understanding of the complexities of therapeutic change and pave the way for future research to pursue further mechanism-based research as well as improve the design and outcomes of GCBTs and PCBTs. If the data provide evidence of mediation for long-term recovery and in the plausible/assumed direction, then continued targeting of the selected variables is supported. However, if the data suggest the directional pattern of change is from youth anxiety reduction to the peer and parent variables targeted, then greater scrutiny of the mechanisms underlying this change is needed, and it also raises questions about whether these variables ought to be targeted in future research and clinical work. Consistent with our expectation of treatment specificity effects, we hypothesized that the peer variables targeted in GCBT would have larger mediated effects on anxiety reduction at posttreatment and 12-month follow-up (long-term recovery) in GCBT than PCBT by virtue of larger mean differences on the peer variables in GCBT versus PCBT. We hypothesized that the parent variables targeted in PCBT would have larger mediated effects on anxiety reduction at posttreatment and 12-month follow-up (long-term recovery) in PCBT than GCBT by virtue of larger mean differences on the parent variables in PCBT versus GCBT.
Method
Participants
Two hundred forty youths (age range = 7–16 years; M = 9.81 years, SD = 2.28) and their parents (mothers) who presented to a youth anxiety disorders specialty research university clinic enrolled in this randomized efficacy clinical trial (ClinicalTrials.gov identifier: CT00073645). See Figure 1 for CONSORT diagram. Pediatricians, school psychologists, and other professionals were major referral sources. One hundred eighty-three (76.3%) completed treatment, comparable with most past efficacy trials (Silverman et al., 2008). To be eligible for inclusion in the trial, all youths were required to meet criteria for a primary fourth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM–IV; American Psychiatric Association, 1994) anxiety disorder using the Anxiety Disorders Interview Schedule for Children (Child and Parent Versions; ADIS-IV: C/P; Silverman & Albano, 1996). Primary diagnoses were separation anxiety disorder (41.3%), social phobia (25.4%), specific phobia (15.4%), generalized anxiety disorder (13.3%), and other anxiety disorders (4.6%). Exclusion criteria were developmental disabilities, psychosis, or current involvement in another psychosocial treatment. A small proportion was on a stable dose of serotonin reuptake inhibitors (10% GCBT; 6% PCBT). One hundred eighty (75%) were Hispanic/Latino, 47 (19.6%) European American, 7 African American (2.9%), and 6 (2.5%) reported “other” ethnicity or race. Fourteen percent had an annual family income of less than $20,000, 20% between $21,000 and $40,000, 15% between $41,000 and $60,000, 12% between $61,000 and $80,000, 10% between $81,000 and $100,000, 12% between $101,000 and $149,000, and 9% over $150,000. Eight percent of mothers did not report income. Seventy-four percent of mothers were married, 2% single, 12% divorced, 6% separated, 0.5% widowed, and 2.5% cohabitating with partner. Three percent of mothers did not report marital status.
CONSORT diagram flowchart.
Measures
Diagnostic instrument administered to youths and parents
ADIS-IV: C/P (Silverman & Albano, 1996)
Independent evaluators (IEs) administered the ADIS-IV: C/P to youths and parents, respectively. Before administering the ADIS-IV: C/P, IEs received extensive training in administration and scoring protocol and met 100% reliability criterion on five videotaped child-parent assessments. The ADIS-IV: C/P yields retest reliability kappas between .80 and .92 for combined diagnoses (e.g., Silverman, Saavedra, & Pina, 2001) and significant associations with youth anxiety ratings (e.g., Wood, Piacentini, Bergman, McCracken, & Barrios, 2002). We also collected 2-week retest reliability data in the present sample by administering the ADIS-IV: C/P twice over 2 weeks to 20% of participants; kappa coefficients were .77 or higher on the primary targeted diagnosis. The disorder most interfering/impairing was “primary” and was targeted in treatment.
Primary outcome measure completed by youths and parents
Revised Children’s Manifest Anxiety Scale
The Revised Children’s Manifest Anxiety Scale (RCMAS; Reynolds & Richmond, 1978), child and parent versions (RCMAS/P), are 37-item rating scales that assess anxiety symptom severity, with extensive reliability and validity data (Reynolds & Richmond, 1978), including extensive data showing it is sensitive to change in clinical trials (Silverman & Ollendick, 2005). All items are rated either yes (1) or no (0). Total Anxiety scores range from 0 to 28, with higher scores indicating higher levels of severity. The parent version (RCMAS/P) is reworded from “I . . . ” to “My child . . . ,” with the identical stems, as done in past studies (e.g., Kendall, 1994; Silverman et al., 1999). In this sample, the alpha coefficient was .84 for the RCMAS and .78 for the RCMAS/P.
Clinically significant change measures
ADIS-IV: C/P
Whether participants met diagnostic criteria for the primary targeted diagnosis at posttreatment (i.e., recovered, not recovered) was one measure of clinically significant improvement. IEs were masked to participants’ treatment arm and pretreatment diagnoses.
Child Behavior Checklist–Internalizing
A minimum criterion T-score of less than 65 on the parent-completed Child Behavior Checklist-Internalizing (CBCL-I; Achenbach, 1991) broadband scale, adjusted according to age norms (e.g., Kendall, 1994; Silverman et al., 1999), was used for normative comparisons.
Children’s Global Assessment Scale
The Children’s Global Assessment Scale (C-GAS; Shaffer et al., 1983) is a 1- to 100-point clinician rating scale that assesses functional impairment. Scores less than 67 are viewed as in the clinical range (Shaffer et al., 1983). All case conference attendees were masked to participants’ identity, status, and assessment point to reduce rater bias (Bird et al., 1993; Dyrborg et al., 2000). IEs’ ratings during case conference meetings were used to establish interrater reliability. Intraclass correlation coefficients in the current study ranged from .93 to .99 across pretreatment, posttreatment, and 12-month follow-up.
Peer measures completed by youths and parents
Friendship Questionnaire–Positive Interactions
The Friendship Questionnaire (FQ), child and parent versions (FQ/P; Bierman & McCauley, 1987), are 36-item rating scales designed to assess peer–youth relationships. All items are rated on a scale from 1 (none of the time, never) to 5 (all of the time, always). Scores on the Positive Interactions (PI) subscale of the FQ range from 0 to 80, with higher scores indicating higher levels of positive peer interactions. The parent version is reworded from “I . . . ” to “My child . . . ,” with the identical stems, as done in past research (e.g., Flannery-Schroeder & Kendall, 2000; Motoca et al., 2012). The FQ-PI discriminates between positive and rejected/neglected social status and correlates significantly with parent and teacher ratings of social competence (Bierman & McCauley, 1987). In this sample, the alpha coefficients for youth and parent versions were .85 and .89, respectively.
Social Skills Rating System
The Social Skills Rating System (SSRS; Gresham & Elliott, 1990) is a 34-item (youth), 38-item (parent) rating scale system that assesses levels of youth social skills. All items are rated on a scale from 0 (never) to 2 (very often). Scores range from 0 to 68 for the youth scale and 0 to 76 for the parent scale, with higher scores indicating higher levels of social skills. Extensive data support the SSRS’s reliability and validity (Gresham & Elliot, 1990). In this sample, the alpha coefficients for youth and parent versions were .86 and .89, respectively.
Parent measures completed by youths and parents
Parenting Behavior Inventory–Parental Psychological Control
The Parenting Behavior Inventory–Parental Psychological Control (CRPBI/PRPBI-PC; Schludermann & Schludermann, 1970) is a 30-item rating scale that assesses perceived parents’ behaviors toward the youths from the perspective of youth and parent, respectively. All items are rated on a scale from 0 (not like) to 2 (a lot like). The Psychological Control (PC) subscale consists of 10 items, and scores range from 0 to 20; higher scores indicate higher levels of perceived PC. Coefficient alphas between .76 and .82 have been reported for different informants (Schwarz, Barton-Henry, & Pruzinsky, 1985). In this sample, the alpha coefficients were .79 for youth PC ratings and .72 for parent PC ratings.
Conflict Behavior Questionnaire–Negative Parent-Youth Relationship
The Conflict Behavior Questionnaire (CBQ; Prinz, Foster, Kent, & O’Leary, 1979) is a 42-item rating scale that assesses conflict in the parent–youth relationship. All items are rated either true (1) or false (0). Scores on the Negative Parent-Youth Relationship (NPYR) subscale range from 0 to 16, with higher scores indicating higher levels of negativity in the parent–youth relationship. Reliability and validity data have been documented (e.g., Robin & Foster, 1989). In this sample, the alpha coefficients for youth and parent CBQ-NPYR ratings were .75 for both.
Procedure
The study received human ethics approval from the university’s institutional review board. All parents and youths provided informed consent/assent prior to participation in the trial’s procedures. All measures were completed at pretreatment, posttreatment, and 12-month follow-up. Youths who met study inclusion were randomly assigned to either GCBT or PCBT in blocks of seven to avoid delay in the formation of groups. Each treatment contained 12 to 14 weekly sessions of 60 min in duration. Treatment manuals were developed for each treatment to standardize all treatment sessions. Nineteen child groups constituted the GCBT arm, with each group consisting of 4 to 8 youths (with most having 5–6 youths). There were no significant differences in the proportion of primary anxiety diagnosis contained in GCBT and PCBT, χ2(7) = .98; p >.05.
CBT involving peer group
GCBT targeted youths’ anxious symptoms in a peer group format using in- and out-of-session exposures and CBT strategies. GCBT contained a heterogeneous set of anxiety diagnoses, with subgroups of 2 to 3 youths formed to conduct the in-session exposures (e.g., Lumpkin, Silverman, Weems, Markham, & Kurtines, 2002; Silverman et al., 1999). For example, children with social phobia in a subgroup would each be asked to read in front of the other subgroup members, give a speech, and so on; parents of children with separation anxiety disorder would each be asked to leave the clinic for a certain duration; dogs would be brought to sessions for dog exposures for those with specific phobia of dogs; and children with generalized anxiety disorder would practice cessation of reassurance seeking in their subgroup. GCBT also targeted increasing (a) positive peer–youth relationships and (b) youth social skills. To address positive peer–youth relationships, therapists assisted youth in identifying positive peer–youth behaviors (e.g., getting together with a peer) and then modeled and conducted role plays. To address youth social skills, therapists modeled and conducted role plays using specific social skills that were taught in session (e.g., increase eye contact, dispense and receive compliments). For both positive peer–youth relationships and youth social skills, youths were encouraged to participate and provide peer feedback, particularly after role plays. Of note, GCBT contained three brief parent-group meetings, about 15 min in duration, led by the same therapists of the youth groups. The purpose of the parent meetings was to provide information about GCBT’s goals and procedures and advise parents that on occasion they might need to transport their child to an exposure task (e.g., a peer’s home). No instructions or prescriptions were provided to parents in these meetings, especially about their child’s peer interactions or social skills. The parent meetings were held at GCBT start, middle, and end, as in Barrett (1998), Flannery-Schroeder and Kendall (2000), and Silverman et al. (1999). Importantly, GCBT involved no training or instruction with parents, particularly with respect to any of the PCBT’s targeted variables.
CBT involving parents
PCBT targeted youths’ anxious symptoms in a parent-youth dyadic format using in- and out-of-session exposures and CBT strategies. Similar exposures as those noted previously were used in PCBT but with the parent present, or parent would be asked to leave the session room and clinic in cases of separation anxiety disorder. PCBT also targeted decreasing (a) negativity/conflict in parent–youth relationships and (b) parental psychological control. To address negativity/conflict in parent–youth relationships, therapists assisted youths and parents in identifying negativity/conflict in the parent–youth relationship (e.g., youth noncompliance to parental requests). This was followed by communication and problem-solving skills training. To address parental psychological control, therapists assisted youths and parents in identifying parental psychological control behaviors (e.g., parent guilt induction) for targeting. Therapists modeled and conducted role plays to identify and decrease these parental behaviors. Importantly, PCBT involved no training or instruction with parents with respect to any of the GCBT’s targeted variables. For example, although some youths in PCBT were asked to conduct out-of-session exposures that might involve peers, such as giving a brief classroom presentation, parents were not instructed in how to improve their child’s skills to successfully complete this exposure task.
Treatment integrity
Treatment sessions were video recorded; therapists were unaware of the sessions that would be assessed. IEs not involved in the study rated a randomly selected 20% of videotapes on a checklist, derived from past studies (Silverman et al., 1999, 2009). Ratings for the targeting of peer variables yielded 100% present in GCBT and 0% present in PCBT. Ratings for the targeting of parent variables yielded 100% present in PCBT and 0% present in GCBT. There was little chance of cross-contamination across treatments given the highly distinctive nature of the group versus parent work. Further, both treatments were manualized and carefully supervised by the first author.
Therapists
Therapists were eight doctoral-level psychology graduate students. Therapists were randomly assigned to cases across arms because neither required differential skill level and background. The first author provided didactic and clinical training and weekly supervision. There were no statistically significant differences between any therapists on any outcome, peer, or parent variables at posttreatment and 12-month follow-up.
Treatment noncompletion
The current study was designed as a specificity- and mediation-based efficacy trial that focused only on treatment completers. Given this, differences between completers and noncompleters are important to document, especially as a function of treatment arm, which could undermine random assignment. There was no significant difference in noncompletion rates across study arms (GCBT = 22.4%; PCBT = 24.8%; z = .44, p > .05), and these noncompletion rates are similar to those reported in past efficacy trials (e.g., Kendall, 1994; Silverman et al., 1999). Treatment completers and noncompleters were compared using data at pretreatment on all primary outcome and mediator measures as well as sociodemographics (e.g., youth age, youth sex). There were no statistically significant differences between completers and noncompleters on any of these 14 variables, except for parent marital status, χ2(1) = 17.44, p < .05. More completers than noncompleters were from families in which the mothers were in intact marriages. There were no significant differences in marital status by treatment arm.
Because we designed our study as an efficacy trial to test theoretical mechanisms surrounding efficacy of GCBT and PCBT, the per protocol analyses used were focused only on participants who completed the treatment (e.g., Feinman, 2009; Meuret, Rosenfield, Seidel, Bhaskara, & Hofmann, 2010). Our study was not designed to estimate broader effectiveness of the CBTs or identify the effects of treatment adherence on outcomes, which is typically the focus of later stage effectiveness trials and entails systematic efforts to obtain posttreatment data for noncompleters. (Flay, 1986; Greenwald & Cullen, 1985; Mitka, 2012). Indeed, mediation analysis in an effectiveness study conflates mechanisms associated with treatment efficacy (Why is the treatment effective?) with mechanisms associated with treatment adherence (Why do people not adhere to or complete the treatment?), which are theoretically distinct and likely governed by different factors (Feinman, 2009). Both processes are optimally studied separately to avoid confounding. In addition, analyzing data that include treatment noncompleters (i.e., intent to treat) can yield biased estimates of efficacy-based mediation.
Results
Preliminary analyses
Non–model-based and model-based outlier analyses were undertaken; no outliers were found. As a result of non-normality in several variables, analyses used a maximum likelihood estimator with robust standard errors as implemented in the MPlus 7.0 statistical software (option MLR; Muthén & Muthén, 2012). Missing data bias was assessed by creating a dummy variable to indicate presence or absence of missing data on each variable. Associations between dummy variables and other study variables were examined. No significant associations were observed. There were no statistically significant differences at pretreatment between families in GCBT and families in PCBT on demographic variables, primary diagnosis, medication usage, and scores on any of the anxiety, peer, or parent variables. Missing data were accommodated using full information maximum likelihood (Enders, 2010). Because GCBT contained 19 separate treatment groups, modeling incorporated adjustments for clustering effects (Baldwin, Murray, & Shadish, 2005) using cluster algorithms in MPlus. Participants in PCBT were grouped as one cluster, yielding a total of 20 clusters across arms. As a check on specification error, exploratory analyses tested for interactions between the treatment arm and the baseline measures of the outcome and mediators; none of these were statistically significant and are not considered further.
Outcome analyses
Initial analyses compared outcome means to baseline means to determine whether anxiety reduction occurred in the treatment arms. The correlations between the parent and youth ratings were low. For example, the correlation between the parents’ ratings of youths’ anxiety and youths’ self-ratings of anxiety was .14 at pretreatment and .29 at posttreatment. These modest correlations between parent and child ratings are typical of past research (Silverman et al., 2009). We therefore treated each source’s ratings as separate primary outcome measures.
Primary outcomes
There was statistically significant pretreatment to posttreatment mean change on the RCMAS in GCBT (pretreatment mean = 12.93, SD = 6.59; posttreatment mean = 7.56, SD = 5.85; Cohen’s d = 0.73) and PCBT (pretreatment mean = 13.56, SD = 6.62; posttreatment mean = 7.33, SD = 5.85; Cohen’s d = 1.03). This finding also was true for the RCMAS/P (GCBT pretreatment mean = 12.96, SD = 5.71; posttreatment mean = 8.42, SD = 5.79; Cohen’s d = 0.85; and PCBT pretreatment mean = 12.69, SD = 5.54; posttreatment mean = 8.04, SD = 5.63; Cohen’s d = 0.85). The changes for GCBT were not significantly different from the changes for PCBT, suggesting the effects for the two treatments were comparable. However, parent ratings of youths’ anxiety at posttreatment were significantly lower in PCBT than GCBT for youths with a primary diagnosis of separation anxiety disorder (RCMAS/P mean difference = −1.91, margin of error [MOE] = ±1.19, critical ratio [CR] = −2.65, p < .01). There were no significant changes on the RCMAS from posttreatment to follow-up for either treatment arm. However, there were significant posttreatment to follow-up changes on the RCMAS/P for both GCBT (posttreatment mean = 8.42, SD = 5.79; follow-up mean = 6.24, SD = 4.92; Cohen’s d = 0.44) and PCBT (posttreatment mean = 8.04, SD = 5.63; follow-up mean = 6.09, SD = 5.68; Cohen’s d = 0.34), with no significant difference between the two treatment arms, again suggesting comparable effects for the two treatments. The differential treatment outcome for separation anxiety disorder at posttreatment was not maintained at follow-up. The magnitude of change in both treatments is comparable with prior efficacy trials and attests to the proper implementation of the two interventions.
Clinically significant change
Clinically significant improvement was found on diagnostic recovery rates and CGAS ratings from pretreatment to posttreatment and from posttreatment to follow-up as reflected by standard tests of proportions (see Table 1). Clinically significant improvement was also found on the CBCL-I, reflected by return to normative levels at posttreatment and follow-up. There were no significant differences in these rates as a function of treatment arm.
Clinically Significant Change at Posttreatment and 12-Month Follow-Up: Percentage of Youths Not in Clinical Range
Note: All treatment differences in rates between GCBT and PCBT at a given time point were statistically nonsignificant on the basis of logit regression adjusted for clustering. GCBT = peer-involvement group cognitive behavioral therapy; PCBT = parent-involvement CBT; ADIS: C/P = Anxiety Disorders Interview Schedule, Child and Parent Versions; CBCL-I = Child Behavior Checklist-Internalizing; C-GAS = Children’s Global Assessment Scale.
Overview of statistical models
To evaluate models of treatment specificity, mediation of long-term recovery, and patterns of directionality of change, the data were analyzed using structural equation modeling focusing on treatment arm, posttreatment measures, and follow-up measures. Pretreatment measures were represented as exogenous variables that served the role of statistical covariates (Rausch, Maxwell, & Kelley, 2003) A dummy variable for the two treatment arms (GCBT, PCBT) was defined (0 = GCBT, 1 = PCBT) and was assumed to directly affect youths’ anxiety and the mediators at posttreatment and 12-month follow-up. The model included autoregressive, lagged (to assess temporal relation between mediators at posttreatment and anxiety at follow-up and between anxiety at posttreatment and mediators at follow-up) and contemporaneous effects (see Figs. 2 and 3). Three additional baseline covariates were included in the analysis because of their past association with anxiety: (a) youth sex, (b) youth age, and (c) youth ethnicity. These covariates were included to increase statistical power for the outcome analyses and reduce the need for correlated error parameters across time. Paths were included from each of these variables to all endogenous variables. We also included cross-sectional correlated error between the mediators to reflect their likely shared unmeasured common determinants. In both models, we assumed that the contemporaneous effect of the mediator on the outcome was comparable at the posttest and the immediate follow-up and introduced equality constraints between the paths accordingly. Results for the SSRS and CBQ-NPYR mediators yielded nonsignificant results for the effects of the treatment arms on them and their relationships to the outcomes. They were therefore excluded from the models and are not reported on further.
Youths’ ratings structural equation model with unstandardized path coefficients and margin of error.
Parent ratings structural equation model with unstandardized path coefficients and margin of error.
The tested models and the estimated path coefficients (half-widths of 95% confidence intervals are presented as margins of error) appear in Figures 2 and 3. Figure 2 focuses on youth-rated anxiety with youth-rated mediators (i.e., positive peer interactions, social skills, psychological control, negative parent–youth relationship) and Figure 3 on parent-rated youth anxiety with parent-rated mediators (i.e., positive peer interactions, social skills, psychological control, negative parent-youth relationship). Both figures exclude sex, age, and ethnicity and baseline exogenous variables (except treatment arm) to avoid clutter, but these were included in all model tests. Residual terms across time were not correlated (when such correlations were included in the model during diagnostic checks, they were statistically nonsignificant). Although not expected, differential estimated effects (path coefficients) of each mediator on the outcome as a function of treatment arm and primary anxiety diagnosis were tested using interaction analysis with product terms. None of these effects were statistically significant and are not discussed further.
Initial analyses of the youth data revealed good model fit except for one omitted path, as reflected by modification indices. This path linked perceived parental psychological control to positive peer interactions such that the more controlling children thought their parents were, the lower they tended to rate the quality of their peer interactions. This path makes theoretical sense, so it was included in the model (see Fig. 2). To determine if the effect replicated with the parent data, we included it in the parent model (see Fig. 3).
A variety of fit indices were used to evaluate global model fit, including the comparative fit index (CFI), the root-mean-square error of approximation (RMSEA), the p value for close fit (pClose), and the standardized root-mean residual (SRMR). In addition, we examined focused tests of ill fit, including modification indices and standardized residuals. Generally speaking, the models provided good fit to the data, with the exception that a direct path from baseline anxiety to anxiety at the follow-up needed to be added to each model. The global fit statistics for the model in Figure 2 were χ2 = 23.2, df = 16, p > .05, RMSEA = .05, pClose = .43, CFI = .98, SRMR = .03. For the model in Figure 3, they were χ2 = 22.7, df = 16, p > .05, RMSEA = .05, pClose = .46, CFI = .99, SRMR = .02.
We orient presentation of results using the youth data, after which we highlight results that replicated and differed for the parent data.
Youth data: treatment specificity and mediation
Treatment specificity effects in Figure 2 are reflected by the paths from the treatment dummy variable to the respective mediators. As predicted, youth ratings of perceived parental psychological control at posttreatment were lower in PCBT than GCBT (mean difference = −1.07, MOE = ±0.80, CR = 2.64, p < .01), with lower levels of youths’ perceptions of parental psychological control being associated contemporaneously with lower levels of youth-rated anxiety (path coefficient from control to anxiety = 0.33, MOE = ±0.15, CR = 4.22, p < .01). Also as predicted, youths’ ratings of positive peer interactions at posttreatment were higher in GCBT than PCBT (mean difference = 3.51, MOE = ±3.00, CR = 2.29, p < .03), with higher levels of positive peer interactions being associated in turn with lower levels of youth-rated anxiety (path coefficient from positive peer interactions to anxiety = −0.08, MOE = ±0.04, CR = 3.69, p < .01). Both of these results are consistent with treatment specificity effects. Given the opposing effects of treatment arm on the two mediators, the net result of no difference between GCBT and PCBT on youth anxiety, noted earlier, is not unexpected.
There is an important qualification to the aforementioned result, however. As seen in Figure 2, we also found evidence for a possible causal relation between the two mediators. Specifically, we observed a statistically significant estimated inverse effect of youths’ perceptions of parental psychological control on positive peer interactions such that the more parents were perceived as controlling of their child, the less positively youths perceived their peer interactions (path coefficient = −0.99, MOE = ±0.31, CR = 6.19, p < .01). Because the PCBT treatment yielded lower levels of parental control than GCBT, this contributed affirmatively to peer–youth positive interactions, ultimately reducing somewhat the advantage of GCBT in directly affecting positive peer interactions relative to PCBT.
The model in Figure 2 suggests several additional mediational dynamics that are of theoretical interest. First, there was a statistically significant lagged effect of treatment arm on positive peer interactions measured at follow-up independent of the effects of treatment arm on peer interactions at posttreatment (path coefficient = −2.85, MOE = ±2.76, CR = 2.03, p < .05). This result suggests that it takes time (in this case, up to 12 months) for some of the differential effects of the two treatments on positive peer interactions to manifest themselves. Interestingly, no such lagged dynamics were apparent for perceptions of parental psychological control, likely because parents can become less controlling more rapidly. Second, although the direct paths from the posttreatment mediators to anxiety as measured at follow-up were not statistically significant, the mediators are still associated with follow-up anxiety by virtue of their estimated indirect effects through the contemporaneous and autoregressive effects. For example, youths’ perceptions of parental psychological control at posttreatment was statistically significantly associated with parental psychological control at follow-up, which in turn was linked to anxiety at the follow-up. Similarly, positive peer interactions at posttreatment were statistically significantly associated with positive peer interactions at follow-up, which in turn were inversely linked to anxiety at the follow-up.
Direction of change
The models in Figures 2 and 3 preliminarily presume the directional pattern of change is from the mediators (perceived parental psychological control and positive peer interactions) to anxiety, but it is possible that reverse causal dynamics also operate between these variables. Of interest in this regard are the lagged paths from the mediators at posttreatment to the outcome at follow-up and the lagged paths from the outcome at posttreatment to the mediators at follow-up. None of these paths were statistically significant, suggesting no reciprocal causality, at least over the 12 months spanning posttreatment to follow-up. This does not, however, rule out the possibility of contemporaneous reciprocal causality at a given time point. For example, the estimated effect of perceived parental psychological control on anxiety was 0.33 at posttreatment (see Fig. 2). However, this estimate presumes the causal direction is unidirectional. We added a contemporaneous reverse causal path between perceived parental psychological control and anxiety to the model in Figure 2 with an equality constraint for the path across the time intervals to make the model statistically identified. The path coefficient from perceived parental psychological control to anxiety remained statistically significant (path coefficient = 0.20, CR = 2.03, p < .05), as was the path coefficient from anxiety to perceived parental psychological control (path coefficient = 0.07, CR = 1.97, p < .05), suggesting the possible presence of contemporaneous reciprocal causality.
Parent data
There were some overlapping findings between the parent and youth data. As shown in Figure 3, parents’ ratings of positive peer interactions at posttreatment were higher in GCBT than PCBT (mean difference = 2.47, MOE = ±2.43, CR = 1.99, p < .05), which is consistent with treatment specificity. However, the treatment difference for the perceived parental psychological control mediator was not statistically significant. As with the youth data, the model in Figure 3 suggests additional mediational dynamics of theoretical interest. There was a statistically significant estimated lagged effect of treatment arm on positive peer interactions measured at follow-up independent of the effects of treatment arm on peer interactions at posttreatment (path coefficient = −3.03, MOE = ±2.69, CR = 2.21, p < .03). This replicates what was observed for the youth data. Although the path coefficient from treatment dummy variable to perceived parental psychological control at posttreatment was not significant (contrary to what occurred for the youth data), there was a statistically significant estimated lagged effect from the treatment dummy variable to perceived parental psychological control at 12-month follow-up. Unexpectedly, parents in GCBT rated themselves as less psychologically controlling compared with parents in PCBT at follow-up even though parental psychological control was not targeted in GCBT. Finally, the possible presence of contemporaneous reciprocal causality was not found, unlike the youth data.
Discussion
As expected, GCBT and PCBT were similarly efficacious across youth and parent anxiety rating scales, diagnostic recovery rates, and return to normative comparisons. Our main interest was to move beyond therapeutic outcome and investigate issues that shed light on mechanisms that produce change in these two CBTs. Thus, this is the first efficacy trial to examine treatment specificity of GCBT and PCBT. Our findings revealed GCBT had a larger effect on positive peer interactions at posttreatment relative to PCBT according to youth and parent ratings, supporting treatment specificity. Specificity was not found for GCBT’s other targeted peer variable, youth social skills. Although a small number of studies have examined treatment specificity of PCBT, the comparators in those studies were ICBT, not GCBT, and no evidence was found for PCBT specificity (i.e., parent variables changed even when not targeted in ICBT; Jongerden & Bögels, 2014; Kendall et al., 2008; Silverman et al., 2009). We found some PCBT specificity: PCBT had a larger effect on perceived parental psychological control relative to GCBT according to youth ratings. The other targeted variable, negative parent–youth relationships, did not show treatment specificity effects.
This is also the first study to report sustained treatment specificity effects. Sustained specificity effects were found again for positive peer interactions at 12-month follow-up on both youth and parent ratings. It is likely that these treatment specificity effects were found because GCBT and PCBT had clear and distinct targeted peer and parent variables, respectively.
Coupled with their statistically significant relationships to anxiety, finding positive peer interactions changed more when targeted in GCBT and perceived parental psychological control changed more when targeted in PCBT suggests these variables are worthwhile to target in these respective CBTs, and prioritizing their inclusion/delivery in GCBTs and PCBTs may render both more efficacious over a full course of treatment. The same cannot be said for targeting youth socials skills in GCBT and negative youth-parent relationships in PCBT, neither of which showed mediation potential.
The variables that showed treatment specificity, positive peer interactions, and perceived parental psychological control also mediated contemporaneous anxiety outcomes. These findings provide insight into mechanisms by which GCBT and PCBT reduce anxiety in youths. Further study with more extensive timelines will be important to gain more nuanced understanding of the temporal dynamics of these mediation effects (Gaynor, 2017).
Another important finding in this trial and one that has been neglected in the treatment efficacy research literature, including youth anxiety, is the intriguing possibility that structural relationships exist among the mediators themselves, thereby complicating the theoretical tracing of relative treatment effects through treatment-specific mediators. For example, we observed a statistically significant estimated effect of perceived parental psychological control on positive peer interactions such that the more controlling parents were of their child, the lower the quality of peer interactions the child tended to have. In this sense, the PCBT arm, which lowered parental psychological control relative to GCBT, indirectly promoted a positive effect on peer relationships relative to GCBT by this reduced perceived parental control. Having said this, there also was evidence that GCBT improved positive peer interactions over and above this indirect effect of PCBT vis-à-vis other mechanisms.
Yet another important mediational dynamic observed and neglected in the literature was possible lagged effects on treatment-specific mediators at follow-up independent of any immediate posttreatment effects on the mediator. For example, the larger estimated effect of GCBT on positive peer interactions compared with PCBT became even more pronounced at 12 months than at immediate posttreatment. This suggests that among youths who complete treatment, it may take a longer period before the GCBT activities fully translate into more positive peer interactions, an intriguing temporal dynamic to consider in future evaluations of GCBT.
Outcomes for GCBT and PCBT did not vary across informants (youths, parents, clinicians), with the exception of parents of children with separation anxiety disorder rating significantly lower anxiety in PCBT than GCBT. Weisz et al. (2017) in their meta-analysis found no significant effect of informant on anxiety treatment outcomes, although their analysis did not focus on anxiety subcategories or different CBT approaches. In contrast, as found in past PCBT research (Settipani et al., 2013; Silverman et al., 2009), mediation dynamics for long-term recovery and patterns of directionality of change varied by informants. Although there was no specificity on targeted parent variables using parent ratings, specificity was found using youth ratings. On one hand, these findings may represent once again the common occurrence of informant discrepancy, suggesting the potential utility of collecting data on measures that rely on different sources. On the other hand, the findings may represent meaningful and key differences in how parents and their children perceive their worlds, especially about something as personally salient as parenting behavior. These two “hands” are not necessarily mutually exclusive, highlighting the need for further research to unravel the complexities of multi-informant assessment approaches (Silverman & Ollendick, 2005).
Overall, however, our findings suggest clinicians may have flexibility and achieve similar anxiety reduction outcomes by targeting peer interactions in GCBTs and parental psychological control in PCBTs. Similarly, addressing both of these mechanisms in the same intervention may result in additively superior effects than either one alone, a topic that should be addressed in future research. Importantly, our treatment specificity and mediation findings further suggest that clinicians might streamline CBTs for youth anxiety by removing procedures or components that do not contribute to outcome (social skills in GCBT, negative parent–youth relationships in PCBT) and focus more intensively on those that do contribute to outcome (positive peer interactions in GCBT, parental psychological control in PCBT).
Study limitations and future research
There are several study limitations. One is the correlational nature of some of the analyses and assumptions that were necessitated about the timing of dynamics between cause and effects. Future research could employ more intensive and frequent measurement (Kraemer, Wilson, Fairburn, & Agras, 2002). A second limitation is reliance on self-report data, although this was a reasonable place to begin this underdeveloped line of inquiry and given the informant data suggest intriguing possibilities, as noted previously. Third, although the efficacy of GCBT and PCBT are clearly established, history and maturational effects could be operating that are in part responsible for some of the changes observed across time. Subsequent studies that include a control arm would help clarify these issues. Fourth, although the high proportion of Hispanic/Latino participants is a study strength and past efficacy studies show that Hispanic/Latino samples respond to CBTs similarly to non-Hispanic/Latino samples (Pina, Silverman, Fuentes, Kurtines, & Weems, 2003; Vaclavik et al., 2017), it is important for research to investigate generalizability.
Also relating to generalizability, because our study was designed to address efficacy-based issues surrounding theoretical mechanisms of change, we cannot assume the findings will generalize to effectiveness trials conducted in community settings. Indeed, unlike the relatively uniform pattern of positive outcomes in youth anxiety efficacy trials in university-based research clinics, outcome effects in community clinics are lower and less consistent across studies (e.g., Bodden et al., 2008; Nauta, Scholing, Emmelkamp, & Minderaa, 2003). For example, a recent effectiveness trial found full recovery for only 23% of participants (Wergeland et al., 2014). Progress is being made to understand the conditions under which effectiveness CBT trials conducted in community clinic settings to reduce youth anxiety produce comparable outcomes to efficacy CBT trials conducted in clinical research settings (e.g., Southam-Gerow, Chorpita, Miller, & Gleacher, 2008). Such work requires an emphasis on understanding why participants do not complete treatment in community settings as well as issues of treatment fidelity. Note that youth anxiety efficacy trials have not revealed any consistent pattern of significant differences between treatment completers versus noncompleters (Kendall & Sugarman, 1997; Pina, Silverman, Weems, Kurtines, & Goldman, 2003). Further research on both efficacy and effectiveness processes (i.e., how is change produced, how can treatment adherence and completion be enhanced) is needed.
Finally, as in any study that employs complicated statistical modeling approaches, omitted variable bias, bias due to possible correlated error, unmodeled reciprocal causality, biasing effects of measurement error, and equivalent/plausible model alternatives warrant consideration. Nevertheless, as one of the first studies to focus on rarely studied issues in efficacy trials, our findings lay the groundwork for further mechanism-based research and carve out exciting new directions for designing, implementing, and analyzing trials that consider treatment specificity, mediation, directionality of change, lag effects in mediators, and causal and reciprocal influences among mediators.
Footnotes
Action Editor
Erin B. Tone served as action editor for this article.
Author Contributions
W. K. Silverman, W. M. Kurtines, and J. Jaccard developed the study concept. All authors contributed to the study design. Testing and data collection were performed by C. E. Marin and Y. Rey. C. E. Marin performed the data analysis and interpretation under the supervision of J. Jaccard. W. K. Silverman drafted the manuscript and C. E. Marin, Y. Rey, J. Jaccard, and J. W. Pettit provided critical revisions. All the authors approved the final manuscript for submission.
Declaration of Conflicting Interests
The author(s) declared that there were no conflicts of interest with respect to the authorship or the publication of this article.
Funding
This research was supported by National Institute of Mental Health Grant R01-MH063997 (to W. K. Silverman).