Group cohesion and alliance predict cognitive-behavioral group treatment outcomes for youth with anxiety disorders

ABSTRACT

Knowledge about how to enhance group cognitive behavioral therapy (GCBT) outcomes is needed. In a randomized controlled effectiveness trial, we examined group cohesion (the bond between group members) and the alliance (the client-clinician bond) as predictors of GCBT outcomes. The sample was 88 youth (M age 11.7 years, SD = 2.1; 54.5% girls; 90.7% White) with anxiety disorders. Observers rated group cohesion and alliance in 32 sessions from 16 groups. We examined early group cohesion and alliance (r = .50, p < .001) and group cohesion and alliance change from early to late in treatment in relation to outcomes using generalized estimation equations accounting for nesting within groups (ICCs .31 to .55). The outcomes were diagnostic recovery, clinical severity, and parent- and youth-reported anxiety symptoms, each at post-treatment, 12-months, and 4-years follow-up. There were more significant associations with 4-years follow-up than earlier outcomes. Clinical severity and parent-reported anxiety symptoms were more frequently predicted than diagnostic recovery. Clinician- and parent-reported outcomes were far more frequently significantly predicted by cohesion and alliance than youth-rated outcomes. We conclude that group cohesion and alliance are related but distinct variables, both associated with some GCBT outcomes for as long as 4 years after treatment.

KEYWORDS:

• Group cohesion
• alliance
• group cognitive-behavioral therapy
• anxiety
• youth

There is considerable evidence that cognitive behavioral therapy (CBT) is an efficacious treatment for youth anxiety disorders, also when delivered in a group format (Higa McMillan et al., 2016; Sigurvinsdottir et al., 2020). However, most of the evidence for group-based CBT (GCBT) derives from controlled university-clinic trials (i.e. efficacy trials; e.g. McKinnon et al., 2018; Silverman et al., 2019). In a review of trials conducted in community settings (i.e. effectiveness trials) for youth anxiety, Wergeland et al. (2021) found an overall remission rate of 59.7% post-GCBT (k = 10). This means that up to 40% of youth do not respond to GCBT in community trials. It is therefore important to examine predictors of GCBT outcomes, as such knowledge can be applied to improve recovery rates in GCBT.

Little is known about which factors influence GCBT outcomes. In the pursuit of GCBT outcome predictors, it is useful to focus on potential predictors that are unique to the group format. Groups represent several opportunities that differ from individual treatment, such as peer reinforcement, modeling, and normalization, as well as social support and exposure (Liber et al., 2008; Silverman et al., 1999). This is particularly relevant for anxiety disorders associated with impaired social skills and peer relationships (Motoca et al., 2012; Silverman et al., 2019). Some GCBT programs specifically emphasize peer-relationships and social skills enhancement in the group sessions (e.g. giving and accepting compliments) via role-plays and peer feedback (Silverman et al., 1999, 2019). Such processes can foster group cohesion, which is the focus of the current study. Various definitions of group cohesion exist, and the term concerns clients’ sense of attraction, belonging, and bonding towards other group members and the group as a whole (Stokes, 1983; Yalom & Leszcz, 2005). In the current study, we define group cohesion as the bond between treatment group members (G. Burlingame et al., 2002; Lerner et al., 2013).

Group cohesion has been found to be positively and significantly associated with clinical outcomes across 55 studies in a meta-analysis of group treatments for adults across a range of disorders (mean r = .26; G. M. Burlingame et al., 2018). A review of nine GCBT anxiety studies with adult clients showed that group cohesion explained up to 21% of the variance in anxiety outcomes (Luong et al., 2020). Of the few studies that have examined group cohesion in youth groups, results have shown that group cohesion is generally high when rated by youths (Kaufman et al., 2005; McGill et al., 2017). One study found that higher youth-rated group cohesion was associated with larger pre-post anxiety reduction in support groups in a non-clinical sample of youth exposed to traumatic events (Shechtman & Mor, 2010). A later study found that higher observer-rated group cohesion was associated with post-treatment variables in GCBT groups for youth anxiety disorders (Fjermestad et al., 2023). This was a psychometrically focused study of the Therapy Process Observational Coding System for Child Psychotherapy—Group Cohesion Scale (TPOCS-GC; Lerner et al., 2013). The study showed, for the first time with youth clients, that a four-item version of the TPOCS-GC could be reliably coded with non-adults. Further, the TPOCS-GC was associated with variables measured at pre-treatment (i.e. youth age and clinical severity), during treatment (child and clinician-rated alliance), and post-treatment (i.e. clinical severity and client satisfaction; Fjermestad et al., 2023).

There are several advantages to using observer-rated measures of treatment processes. In fact, observer-rated measures are recommended when examining treatment factors without a unified and clear definition, such as group cohesion (Elvins & Green, 2008). Arguably, behavioral observations allow for clearer operationalizations that can be generalized across individuals compared to subjective measures. For example, an individual client’s estimation of their relation to other group members is likely to be (partly) based on that client’s general social perceptions. In the case of social anxiety, a client is likely to apply general preoccupations about what others think of them into their assessment of the group climate. An observer, on the other hand, can provide a less biased estimate of how group members interact with each other. The “objectivity” of these observations can be further enhanced by using multiple observers and checking their agreement. Albeit the subjective perceptions of each client are important and should not be undermined, in the context of operationalizing process terms, observations have added value. Additional advantages of using an observation-based measure include the use of masked raters (Lerner et al., 2013). Medium to strong associations have been identified for self-reported group cohesion and outcomes in adult studies, a higher correlation than for observer-rated group cohesion (Bonsaksen et al., 2013; Luong et al., 2021). This is not surprising since within-informant effects tend to be larger than cross-informant effects (Achenbach et al., 1987). If we can identify cross-informant effects of an observed process measure on self-, parent-, and/or clinically reported outcome(s), this is important information for the conceptualization and understanding of potentially effective change mechanisms (i.e. associations that occur beyond “just” within-informant effects). Another advantage of applying observer-measured data to youth clinical data is that measurement issues such as ceiling effects and differentiation problems have been identified in youth samples (Fjermestad et al., 2012).

In the current report, we use the same sample as Fjermestad et al. (2023), adding two original features to enhance further the field’s knowledge about the role of group cohesion. First, we included the variable observer-rated alliance. The alliance is traditionally understood as the emotional bond between the client and the clinician and their agreement on the tasks and goals of the treatment (Bordin, 1979). The current study defines the alliance as the collaborative bond between the client and the clinician (Elvins & Green, 2008). Including the alliance is essential because a potentially complicating factor when examining group cohesion is that each client forms a collaborative bond with the clinician and the other clients. In CBT for youth anxiety disorders, the alliance has been argued as crucial since treatment programs typically entail homework and potentially distressing exposure tasks that may be facilitated by a high-quality youth-clinician relationship (Kendall et al., 2009). There is considerable evidence that a strong alliance predicts improvements in clinical outcomes in individual treatment (M. S. Karver et al., 2018; McLeod, 2011). However, studies with adult clients in groups that have included both group cohesion and alliance measures have shown mixed findings. For example, Norton and Kazantzis (2016) found stronger and more consistent associations between alliance and anxiety symptoms than between group cohesion and anxiety symptoms in a trial of group-based psychotherapy. Contrary to this. Luong et al. (2021) demonstrated stronger associations between group cohesion and GCBT outcomes than between alliance and GCBT outcomes. The relative role of group cohesion versus alliance has not been examined in GCBT for youth clients. The youth-clinician relationship may be even more critical when the client “competes” with other clients for the clinician’s attention. However, the youth-clinician relation may be less important in GCBT than the interpersonal relationships among the youths in the group. The current study examined the combined effects of group cohesion and alliance on outcomes.

We added several outcomes up to four years post-GCBT as a second original feature. This allows us to examine the role of group cohesion and alliance in the durability of GCBT effects. Specifically, we examined full diagnostic recovery from all inclusion anxiety disorders, the related but continuous outcome of clinician-rated clinical severity, as well as parent- and youth-rated anxiety symptoms. Hence, we cover a dichotomous recovery outcome, and continuous clinician-rated severity indexes and parent and youth-self report, all examined against observer-rated effects. Evidence from efficacy trials shows that GCBT effects are enduring up to 13 years post-CBT (see Saavedra et al., 2010; Silverman et al., 2019). However, there is hardly any evidence for the effectiveness of GCBT for youth anxiety disorders beyond 12 months post-treatment (see Kodal et al., 2018a, for an exception). Data are also lacking on factors that predict youth anxiety GCBT long-term outcomes (Heiervang et al., 2018). A review of predictors of long-term clinical outcomes (i.e. >2 years) included two studies of GCBT for anxious youth and found that none of these studies identified significant predictors of long-term outcomes (Gibby et al., 2017). In a long-term follow-up study after an anxiety trial conducted in community clinics, Kodal et al. (2018b) found that lower family social class, primary diagnosis of social phobia, and lower treatment motivation predicted poorer clinical outcomes 4 years after treatment. However, GCBT was only one of the arms in this trial, and the predictor study did not distinguish between treatment formats.

In sum, there is a knowledge gap regarding predictors of clinical outcomes in GCBT, particularly long term. The current study examined whether group cohesion and the alliance predicted GCBT clinical outcomes at immediate-, 12-months-, and 4-years-post treatment in an effectiveness trial. An important issue when considering the role of treatment processes for outcomes is the timing of the measures. We examined group cohesion and alliance scores measured early in treatment. We did this because early treatment processes are less confounded with symptom improvement during treatment (Marker et al., 2013; McLeod, 2011). We also examined group cohesion and alliance change from early to late treatment. We used residualized change scores, which are computed by regressing posttest scores (here, late group cohesion/alliance) on pretest scores (here, early group cohesion/alliance) and then computing the difference between observed late scores and predicted late scores (Valente & MacKinnon, 2017). We chose residualized change scores rather than pure change scores because they compensate for regression to the mean and are recommended for variables with potential ceiling effects, such as the alliance (De Los Reyes & Prinstein, 2004; Twisk & Proper, 2004). Since higher age was associated with lower group cohesion in an earlier report from this sample (Fjermestad et al., 2023), we included youth age as a covariate.

The research question was as follows: Does early group cohesion, early alliance, group cohesion change, and/or alliance change predict GCBT outcomes? Based on previous findings, we expected that higher early group cohesion and group cohesion change would predict better clinical outcomes (G. M. Burlingame et al., 2018; Fjermestad et al., 2023). We also expected that higher early alliance and alliance change would predict better clinical outcomes (M. S. Karver et al., 2018; McLeod, 2011). We included the group cohesion and alliance variables in the same models without an a priori expectation concerning whether or not one of these would outperform the other due to the originality of this feature.

Materials and methods

Youth sample

The sample represents the full GCBT arm of a randomized controlled trial (RCT) comparing GCBT and individual CBT to a waitlist in community youth mental health outpatient clinics in Norway. The sample comprised 88 youth (M age at pre-treatment = 11.7 years, SD = 2.1, age range 8 to 15, 54.5% girls, 45.5% boys). The youth identified as European, White (90.7%), and Asian (1.6%; 7.7% did not report race or ethnicity). One-fifth of the sample (19.8%) lived in single-parent households. Parents’ occupational status was classified into rank-ordered social classes with the Registrar General Social Class coding scheme (Currie et al., 2008). Family social class was high for 35.2%, medium for 51.1%, and low for 5.7% (not reported for the remaining 8.0%).

The inclusion criterion for the RCT was a primary DSM-IV (American Psychiatric Association, 1994) diagnosis of a separation anxiety disorder (SAD), social phobia (SoP), or generalized anxiety disorder (GAD). Exclusion criteria were the presence of pervasive developmental disorder, psychotic disorder, or mental retardation.

Clinicians and treatment

Clinicians (N = 15, M_age = 49.8 years; SD = 9.4; 93.3% female; 100% European White) volunteered to participate and were employed at one of the seven clinics. On average, clinicians had 12.0 years of experience (SD = 6.0). Two clinicians led each group, one designated as the group leader and the other as the facilitator. The sample comprised 16 treatment groups, with 4 to 7 youth each (median = 5).

Clinicians were trained to deliver both the group and the individual version of the Norwegian translation of the FRIENDS for Life program (FRIENDS; 4^th ed., Barrett, 2004), which targets emotional awareness and regulation, cognitive restructuring, and exposure tasks. The program comprised 10 weekly 60-min sessions. There are two separate parent group evenings, and parents join the last 15 min of each youth group for summary and information. There is evidence of both efficacy (Liber et al., 2008; Shortt et al., 2001) and effectiveness for FRIENDS to improve youth anxiety (Kodal et al., 2018a; Wergeland et al., 2014).

Clinician training included six workshops focusing on the treatment manual, CBT and youth anxiety disorders, and the completion of two pilot cases or groups. All clinicians received group supervision from one of two psychologists experienced in CBT for youth anxiety every 2 to 4 weeks, with an average of 78.7 (SD = 34.0) supervision hours per clinician. All treatment sessions were videotaped. Coding of 20% randomly selected videotapes showed adequate adherence and competence in delivering the GCBT program as measured by the Competence and Adherence Scale for Cognitive Behavioral Therapy (Bjaastad et al., 2016).

Original trial outcomes

In the original trial, there were no differences between individual CBT and GCBT, and both arms outperformed the waitlist (Wergeland et al., 2014). In the GCBT arm, the percentage of participants who no longer met the criteria for their primary anxiety diagnosis in the intent-to-treat sample was 35.2% at post-treatment, 46.6% at 12-month follow-up, and 59.4% at 4-year follow-up. Clinical severity scores (scored on a 0–8 scale) were significantly reduced with 2.3 points from pre- to post-treatment, 3.2 points from pre- to 12-month follow-up, and 4.4 points from pre- to 4-year follow-up (Kodal et al., 2018a).

Measures

Observer-rated measures

Therapy process observational coding system for child psychotherapy—group cohesion scale (TPOCS-GC; Fjermestad et al., 2023; Lerner et al., 2013)

A four-item version of the TPOCS-GC was used to measure group cohesion. The items cover the extent to which the group member demonstrates positive affect toward the other group members (affect), shares their experiences with the other group members (share), and to what extent the interaction between the group members and the other group members is lively and energetic (energy). One item is reverse coded, i.e. the extent to which the group member appears anxious or uncomfortable with the other group members (anxious). The items are rated on a 6-point scale ranging from 0 (not at all) to 5 (a great deal). Coders watch entire treatment sessions, shifting their main focus between each group member every 5 min. A study of parent groups for children with ADHD in the USA reported an internal consistency of α = .80 and mean interrater agreement of ICC (1,2) = .75 for the TPOCS-GC (Lerner et al., 2013). A study of groups of adults with anxiety disorders reported a mean interrater agreement of ICC (1,2) = .88 (Luong et al., 2021). The current study is based on the first psychometric evaluation of the TPOCG-GC used with youth. The original English version was used, and all coders were fluent in English. The average inter-rater agreement was ICC (2,1) = .61. The internal consistency across the four TPOCS-GC items was α = .72. We have previously demonstrated that the TPOCS-GC used with the current sample was associated with pre- and post-treatment clinical severity and post-treatment client satisfaction. Herein, we expand these findings with further outcomes measured across 4-years post-treatment and linked with an alliance measure.

Therapy process observational coding system for child psychotherapy–alliance scale (TPOCS-A; McLeod & Weisz, 2005)

The TPOCS–A was used to measure observer-rated alliance. The TPOCS-A comprises nine items. Six items cover affective elements of the group member–clinician relationship (bond items; e.g. to what extent does the group member demonstrate positive affect toward the clinician). Three items cover group member participation in therapeutic activities (task items; e.g. to what extent does the client engage/participate in therapeutic tasks). Items are rated on a 6-point scale ranging from 0 (not at all) to 5 (a great deal). Coders watch complete treatment sessions, shifting their main focus between each client and the lead clinician in the group every 5 min. When used with groups, the TPOCS-A has demonstrated item interrater reliability ICC(2,2) from .72 to .81, internal consistency from α = .80 to .92, and predictive validity with clinical outcomes (Kang et al., 2021; Liber et al., 2010). The current study used an approved Norwegian version of the TPOCS–A (Fjermestad et al., 2012). The average ICC (2,1) was = .65.

The internal consistency across the nine TPOCS-A items was α = .72.

Outcome measures

Anxiety disorders interview schedule–child and parent versions (ADIS-C/P; Silverman & Albano, 1996)

DSM-IV-based anxiety diagnoses were derived from the combined parent and youth report on the SAD, SoP, and GAD sections of the ADIS-C/P. The ADIS is a semi-structured interview with good reliability (ICC’s 0.78 to 0.95; Silverman et al., 2001) and concurrent validity with anxiety symptom scales (Wood et al., 2002). Clinicians administered the ADIS after a two-day onsite training with one of the interview developers (WKS). All interviews were videotaped. Based on a masked rescoring of 20%, the inter-rater agreement estimated by kappa (κ) for the presence of an inclusion anxiety diagnosis was 0.84 (ADIS-C) and 0.86 (ADIS-P).

For each ADIS-C/P section, a clinician’s severity rating (CSR) ranging from 0 to 8 was assigned based on a combined parent and child report (ADIS-CSR). A higher score reflects a higher impact on personal, academic, social, and family functioning. Intraclass correlations for the CSR were 0.82 (ADIS-C) and 0.82 (ADIS-P). ICCs for specific anxiety disorders were 0.72 (SAD), 0.88 (SoP), and 0.89 (GAD).

Spence children’s anxiety scale, child and parent versions (SCAS-C/P Spence, 1998)

SCAS-C/P was used to measure youth anxiety symptoms. The 38-item SCAS is rated on a 4-point Likert scale. The internal consistency was α = .88 for youth and α = .81 for parents.

Procedures

The regional review board approved the study for medical research ethics (Regional Committees for Medical and Health Research Ethics—West; # 2011/1004). Participants were informed they would receive treatment even if they declined to participate in the RCT. Parents and youth above 12 years provided written informed consent/assent, and youths below 12 provided verbal assent. No honorarium was given, except for the 4-years follow-up assessment, when participants received a gift certificate of USD50.

Coder training

We randomly selected one early videotape (sessions 2–5) and one late videotape (sessions 6–10) from each of the 16 GCBT groups, coding 32 tapes. The shortest time between the earliest and latest observed sessions was three, and the longest was eight sessions. The mean difference between early and late observed sessions was 5.9 (SD = 1.1). This means two sessions were coded for each of the groups conducted in the trial. The same coder team coded both the TPOCS-A and the TPOCS-GC. Coders were masked for trial clinical outcomes. The coder training comprised three main steps. First, coders jointly coded pilot tapes (not included in the data) while discussing their understanding of items. Second, coders independently coded pilot tapes until they reached 92% agreement, which is defined as a maximum 1-point difference from each other on each item. Third, the coders independently coded videotapes from the main trial (the current data) in a randomly pre-assigned sequence.

Participant flow and missing data

All 16 treatment groups in the GCBT arm were coded. Five of the original 88 participants had missing group cohesion and alliance data. Four participants with missing TPOCS-GC/A data dropped out of treatment, while the fifth was absent from the two randomly selected coded sessions. The amount of missing clinical outcome data for the various clinical outcome variables ranged from 0 to 22 cases depending on informant and outcome measure. Little’s Missing Completely at Random Test indicated that these values were missing completely at random (p = .115). The analyses were done based on participants with complete data. We examined whether there were significant differences in the group cohesion and alliance data between participants with clinical outcome data at the three post-measurement points (immediate post, 12-month follow-up, 4-year follow-up) and those with missing clinical outcome data. There were no such differences.

Data analytic plan

Because the 83 clients were nested across 16 groups, with a mean of five clients per group, we examined the level of variance in the measures of group cohesion and alliance measure at the group level, the group level variances were ICC = .55 for early group cohesion, ICC = .45 for group cohesion change, ICC = .31 for early alliance, and ICC = .44 for alliance change. Because these values indicate considerable nesting at the group level (Guo, 2005), multilevel analyses could be warranted. However, since the number of groups and participants per group in the current sample was too small for multilevel analyses (Wolf et al., 2013), we used generalized estimation equations (GEEs). GEEs represent a modification to the general linear model, which accounts for nesting within groups (Hanley et al., 2003). This means that the random effect (in this case, the specific treatment group in which the youth participated) was included in the model to account for its effect (McNeish et al., 2017). We used a model comparison approach to determine if unstructured, independent, or exchangeable correlation matrices indicated a better fit for each model using the quasi-likelihood under the independence model criterion (QIC; Pan, 2001). We included all four predictors (early group cohesion, group cohesion change, early alliance, alliance change) for each outcome. In all the models, youth age was included as a covariate. Due to the explorative nature of our study in examining multiple outcomes that group cohesion and/or alliance may be related to, we chose not to adjust the p-level for the number of outcomes. That is, a p-value < .05 was considered significant. We used the SPSS version 29 in all analyses.

Results

Descriptives

The mean early group cohesion score was 1.8 (SD = 0.6, range 0.5 to 3.5) on a 0 to 5 scale, indicating relatively low group cohesion scores. The mean residualized change from early to late was −0.4 (SD = 0.8), indicating overall group cohesion stability from early to late in treatment. The mean early alliance score was 2.8 (SD = 0.5, range 1.7 to 3.9) on a 0 to 5 scale, indicating medium alliance scores. The mean residualized change from early to late was −0.3 (SD = 0.5), indicating overall alliance stability from early to late in treatment. The correlation between early group cohesion and early alliance was r = .50 (p < .001).

Effects of group cohesion on clinical outcomes

Higher early cohesion and more positive group cohesion change were associated with a larger likelihood of diagnostic recovery at 4-year follow-up but not at the previous time points (See Table 1). Higher early group cohesion and more positive group cohesion change predicted a larger clinical severity reduction at 4-years follow-up (See Table 2). Higher early group cohesion predicted a larger decrease in parent-rated youth anxiety symptoms at 12 months follow-up. More positive group cohesion change predicted a larger reduction in parent-rated youth anxiety symptoms at all measurement times (post, 12 months, and 4-years follow-up) (See Table 3). None of the group cohesion or alliance variables predicted youth-rated anxiety symptom outcomes (See Table 4).

Table 1. GEE results of group cohesion and alliance scores on full diagnostic recovery.

Factors	Exp B	95% CI for Exp B	Wald χ2	p
Recovery post
Age	0.97	0.73,1.29	0.09	.827
Early alliance	1.12	0.29,4.38	0.03	.872
Δ alliance	2.88	0.90,9.21	3.17	.075
Early GC	1.05	0.37,2.96	0.01	.924
Δ GC	1.11	0.57,2.16	0.10	.751
Recovery 12 m post
Age	0.81	0.60,1.09	1.94	.164
Early alliance	4.70	1.01,21.91	3.88	.049
Δ alliance	2.78	0.93,8.29	3.37	.066
Early GC	0.53	0.15,1.86	0.99	.319
Δ GC	0.75	0.39,1.44	0.76	.385
Recovery 4y post
Age	1.05	0.77,1.45	0.09	.767
Early alliance	4.87	0.91,26.24	3.40	.065
Δ alliance	7.89	1.26,49.26	4.88	.027
Early GC	4.65	1.45,14.92	6.69	.010
Δ GC	5.68	1.91,16.90	9.73	.002

GEE = Generalized Estimation Equations. Recovery = loss of all anxiety diagnoses based on the Anxiety Disorders Interview Schedule (ADIS). GC = Group Cohesion. m = months. y = years. SE = Standard Error. CI = Confidence Interval. Δ = change.

Values in bold are significant at the p < .05-level.

Table 2. GEE results of group cohesion and alliance scores on clinical severity outcomes.

Factors	B (SE)	95% CI for B	Wald χ2	p
CSR post
Age	0.04 (0.11)	−0.18,0.25	0.11	.746
Early alliance	−0.68 (0.52)	−1.70,0.33	1.73	.188
Δ alliance	-1.16 (0.50)	−2.14,-0.18	5.36	.021
Early GC	-0.79 (0.29)	−1.36,-0.22	7.34	.007
Δ GC	-0.65 (0.23)	−1.10,-0.20	8.14	.004
CSR 12 m post
Age	0.15 (0.13)	−0.11,0.42	1.30	.254
Early alliance	-2.02 (0.64)	−3.27,-0.77	10.06	.002
Δ alliance	−1.09 (0.56)	−2.19,0.20	3.71	.054
Early GC	0.06 (0.44)	−0.80,0.92	0.02	.893
Δ GC	−0.31 (0.41)	−1.12,0.51	0.55	.461
CSR 4y post
Age	−0.12 (0.13)	−0.37,0.14	0.79	.376
Early alliance	-1.37 (0.46)	−2.28,-0.47	8.80	.003
Δ alliance	-1.67 (0.73)	−3.10,-0.25	5.28	.022
Early GC	-1.28 (0.46)	−2.18,-0.37	7.65	.006
Δ GC	-1.33 (0.15)	−1.63,-1.03	77.40	<.001

GEE = Generalized Estimation Equations. CSR = Clinical Severity Rating. GC = Group Cohesion. m = months. y = years. SE = Standard Error. CI = Confidence Interval. Δ = change.

Values in bold are significant at the p < .05-level.

Table 3. GEE results of group cohesion and alliance scores on parent-rated anxiety symptoms.

Factors	B (SE)	95% CI for B	Wald χ2	p
SCAS-P post
Age	−1.25 (0.45)	−2.12,-0.37	7.84	.005
Early alliance	−1.44 (1.86)	−5.09,2.21	5.98	.493
Δ alliance	−2.42 (2.11)	−6.39,1.91	1.12	.290
Early GC	−3.16 (2.13)	−7.33,1.02	2.20	.138
Δ GC	-3.15 (1.07)	−5.26,-1.05	8.62	.003
SCAS-P 12 m post
Age	-2.03 (0.51)	−3.04-,1.02	15.56	<.001
Early alliance	-17.45 (1.16)	−19.73,-15.16	224.29	<.001
Δ alliance	-9.11 (1.92)	−12.87,-5.34	22.46	<.001
Early GC	-3.54 (1.78)	−7.03,-0.04	3.92	.047
Δ GC	-3.60 (1.06)	−5.67,-1.52	11.52	.<.001
SCAS-P 4y post
Age	−0.91 (0.60)	−2.08,0.26	2.34	.127
Early alliance	-14.08 (2.58)	−19.14,-9.02	29.73	<.001
Δ alliance	-6.22 (2.26)	−10.65,-1.79	7.56	.006
Early GC	2.63 (2.89)	−3.05,8.30	0.82	.364
Δ GC	-3.87 (1.66)	−7.12,-0.61	5.42	.020

GEE = Generalized Estimation Equations. SCAS = Spence Children’s Anxiety Scale. p = Parent-report. GC = Group Cohesion. m = months. y = years. SE = Standard Error. CI = Confidence Interval. Δ = change.

Values in bold are significant at the p < .05-level.

Table 4. GEE results of group cohesion and alliance scores on youth-rated anxiety symptoms.

Factors	B (SE)	95% CI for B	Wald χ2	p
SCAS-Y post
Age	−0.25 (0.99)	−2.19,1.68	0.07	.798
Early alliance	−1.55 (3.87)	−9.13,6.04	0.16	.689
Δ alliance	−1.77 (3.24)	−8.13,4.59	0.30	.586
Early GC	−0.50 (2.48)	−5.36,4.36	0.04	.839
Δ GC	−0.19 (1.58)	−3.28,2.90	0.01	.905
SCAS-Y 12 m post
Age	−0.51 (1.46)	−3.37,2.36	0.12	.729
Early alliance	−5.81 (7.24)	−20.01,8.39	6.43	.423
Δ alliance	−1.22 (3.11)	−7.31,4.87	0.16	.694
Early GC	2.53 (4.09)	−5.48,10.54	0.38	.535
Δ GC	−3.06 (3.66)	−10.24,4.12	0.70	.403
SCAS-Y 4y post
Age	1.24 (1.11)	−0.93,3.42	1.25	.264
Early alliance	-11.68 (5.86)	−23.15,-0.20	3.98	.046
Δ alliance	−1.40 (3.57)	−8.40,5.61	0.15	.696
Early GC	2.19 (5.57)	−8.72,13.10	0.16	.693
Δ GC	−2.34 (2.39)	−7.03,2.34	0.96	.327

GEE = Generalized Estimation Equations. SCAS = Spence Children’s Anxiety Scale. Y = Youth-report. GC = Group Cohesion. m = months. y = years. SE = Standard Error. CI = Confidence Interval. Δ = change.

Values in bold are significant at the p < .05-level.

Effects of alliance on clinical outcomes

The higher early alliance was associated with a larger likelihood of diagnostic recovery at 12-months and 4-year follow-up, but not at post-treatment. Alliance change did not predict diagnostic recovery (See Table 1). Higher early alliance predicted a larger clinical severity reduction at all measurement times (post, 12 months, and 4-years follow-up). More positive alliance change predicted a larger clinical severity reduction at 4-years follow-up, but not earlier measurement points (See Table 2). Higher early alliance and more positive alliance change predicted a larger decrease in parent-rated youth anxiety symptoms at 12-months and 4-years follow-up, but not at post-treatment (See Table 3). Higher early alliance predicted a larger reduction in youth self-rated anxiety symptoms at 4-year follow-up but not at the other measurement points (post-treatment, 12-month follow-up). Alliance change did not predict youth self-rated anxiety symptom outcomes (See Table 4).

Combined effects of group cohesion and alliance on clinical outcomes

The four main outcomes were all measured three times, leaving 12 potentially predicted outcome measurement points. Across the four outcomes × three times, there was one unique prediction path for group cohesion, i.e. an effect of group cohesion with no effect of the alliance. Specifically, more positive group cohesion change predicted lower parent-rated youth anxiety symptoms post-treatment, with no effects from the alliance. There were two unique prediction paths for alliance, i.e. an effect of alliance with no effect of group cohesion. Specifically, higher early alliance predicted diagnostic recovery at 12-months follow-up and lower youth self-rated anxiety symptoms at 4-year follow-up, with no effect on these outcomes from group cohesion. Five of the 12 outcomes were predicted by a combination of both group cohesion and alliance. Three of the 12 outcomes were not predicted either by group cohesion or alliance, i.e. post-treatment diagnostic recovery and youth self-rated anxiety symptoms at post-treatment and 12-month follow-up.

Discussion

We examined the relative role of group cohesion and alliance as predictors of GCBT clinical outcomes. The predictors were group cohesion and alliance measured early as well as change from early to late. We examined multiple outcomes at three different timepoints based on multiple informants (i.e. youth, parents, and clinicians). The patterns of findings varied by predictor type (i.e. group cohesion versus alliance; early versus change), timepoint, outcome variable, and informant. In terms of predictor type, early alliance predicted more outcomes than early group cohesion, whereas group cohesion change predicted more outcomes than alliance change. In terms of timepoint, there were most significant associations with outcomes measured at 4-years follow-up, followed by 12-months follow-up, and only one significant association at immediate post-treatment. In terms of outcomes, clinical severity ratings and parent-reported youth anxiety symptoms were equally frequently predicted, followed by diagnostic recovery, and only one significant association for youth self-report. In terms of informant, the clinician and parent reported outcomes were far more frequently significantly predicted than youth-rated outcomes. This reflects the complexity of associations when considering process variables in relation to outcomes in youth CBT—and may help explain why findings are tricky to synthesize across studies (e.g. Luong et al., 2020).

This is the first study to show that group cohesion predicted some, but far from all, diagnostic and clinical severity/symptom outcomes in youth anxiety GCBT. The positive group cohesion-outcome associations we hypothesized were partly confirmed in that half of the outcomes considered were significantly predicted. It is important to note that this means half of the outcomes were not predicted by group cohesion. Nevertheless, the finding of a particular association with outcomes corresponds with adult group therapy studies (G. M. Burlingame et al., 2018). Furthermore, the findings fill a knowledge gap in the literature and provide some direction for which treatment elements clinicians may target to enhance outcomes. Specifically, this concerns how clinicians can improve the group aspect in GCBT, e.g. to facilitate and reinforce interaction between group members, emphasize common challenges and solutions across the group members, and address and challenge negative or critical comments between group members.

The alliance also predicted some GCBT clinical outcomes, which means our second hypothesis was also partly supported. The alliance predicted a larger number of potential outcomes, with eight of 12 outcome measures × timepoint combinations predicted by an alliance variable. This finding partly aligns with studies showing that the alliance predicts clinical outcomes in youth and adult treatments (see Kang et al., 2021; M. S. Karver et al., 2018; McLeod, 2011). Notably, the finding also indicates that the alliance may predict some clinical outcomes in GCBT. Importantly, unlike group cohesion, there is an emerging evidence base of clinician behaviors and strategies that can be applied early in treatment to enhance alliance. Such strategies include attending to client experience, emphasizing collaboration, exploring clients’ motivation and cognitions, presenting a treatment model, and using CBT techniques (Fjermestad et al., 2020; M. Karver et al., 2008; Russell et al., 2008). Negative clinician-level predictors of the alliance have also been identified, including distorting from focus, overly structured, and pushing the youth to talk (Creed & Kendall, 2005; M. Karver et al., 2008). Whether these behaviors apply across formats remains to be determined, as the evidence rests on individual CBT, not GCBT studies.

Studies are needed that examine group cohesion and alliance building in groups. It may be more challenging for clinicians to perform alliance-building behaviors when dealing with a group rather than individual clients. For instance, much energy and effort may be directed towards unengaged/passive clients and disruptive/dominating group members. Such treatment aspects may prevent clinicians who deliver GCBT from evenly distributing their attention and relationship bonding between group members. Furthermore, evidence-based alliance-building techniques, such as explaining the treatment model and exploring motivation, may be more complex when multiple clients are in the room.

It is important to consider some differences in findings based on whether the group cohesion and alliance was measured at a single early time-point versus as a change score from early to late in treatment. We considered both because process variables measured late in treatment are often believed to be biased by recovery during treatment, i.e. the closer to the treatment endpoint a process variable is measured, the more likely this variable is to be associated with outcomes (McLeod, 2011). Interestingly, the pattern of findings for early versus late process measures in the current study was quite different for group cohesion and alliance. That is, group cohesion change was associated with far more outcomes than early group cohesion, whereas the opposite was the case for alliance—early alliance was associated with far more outcomes than alliance change. A possible explanation could be that group cohesion becomes more important as groups “settle” more over time. Whereas initially in treatment, youth may be more oriented towards the therapist (and the alliance with them), as CBT progresses they may start to orient more towards the other group members and the importance of group dimension increases. In the case of the alliance, our findings are trickier to interpret, since later alliance is typically more strongly associated with outcomes than early alliance (McLeod, 2011). It could be that due to the group format, the alliance to the therapist becomes less important over time, and “replaced” by the role of the relation to other group members.

Our findings also shed further light on the relative importance of group cohesion versus alliance in GCBT. Whereas both variables predicted some clinical outcomes, in most models that showed significant associations, group cohesion and alliance were significant predictors. Notably, group cohesion and alliance only showed moderate overlap. This means group cohesion and alliance are distinct, although related processes, with unique and combined influence on clinical outcomes. The current study results suggest that group cohesion aligns with the process concept of the alliance in GCBT in a potential “family” of factors tapping relation-based treatment engagement. Across adult CBT studies, a review of meta-analyses found support for the alliance as a predictor of clinical outcomes but not group cohesion (Kazantzis et al., 2018). Further youth studies are needed to disentangle the concept of group cohesion in GCBT.

A particular strength of the current study is the inclusion of long-term outcomes. The finding that group cohesion and alliance predicted some clinical outcomes as late as four years post-treatment is significant. These findings are essential to the field, considering so little research has been conducted on long-term GCBT clinical outcomes (Heiervang et al., 2018). The results indicate that some elements of the relational processes in GCBT, both between group members and between each group member and the clinician, somehow influence long-term outcome trajectories. It is particularly relevant to consider why more of the effects occurred at 4-year follow-up and to a much lesser extent sooner (e.g. only two of the post-treatment outcomes were predicted by at least one of the process variables). Those with the strongest bonds to the other group members and the clinician continue to work on the skills learned in GCBT beyond post-treatment. For example, participants who experienced trusting, collaborative relations in GCGT may adhere more to the main GCBT behavioral components, such as cognitive restructuring and exposure. This may be through social reinforcement, facilitation, and learning mechanisms (Silverman et al., 2019). As such, the fact that there were more long-term than short-term findings may reflect client factors that play a role during treatment. There is some indication from studies that youth remember process factors as important parts of treatment several years later. Kendall and Southam-Gerow (1996) found that youth identified “the therapeutic relationship” as the most important aspect of CBT for anxiety disorders two years later. Hauber et al. (2019) analyzed the farewell letters of youth with personality disorders who had participated in psychotherapy groups and found that group cohesion was mentioned by 97%. Further research is needed to determine the processes that bridge group cohesion and alliance to clinical outcomes.

We examined associations between observer-rated group cohesion and alliance and clinician-rated clinical outcomes. Although the clinician-rated outcomes are based on structured diagnostic interviews with youth patients and their parents (combined), the current results were not based on self-report, and effects were identified across informants (i.e. observers and clinicians). This adds validity to our data, given that within-informant effects tend to be inflated (De Los Reyes, 2013).

An additional strength of the current study is the use of an observation-based measurement approach in a clinically effective GCBT trial for youth anxiety. From that vantage point, our sample size is quite impressive. We also observed two randomly selected sessions per client and included change scores as predictors. Although the study’s primary purpose was to examine the effect of GCBT versus ICBT and waitlist (reported in Wergeland et al., 2014), the time conditions for examining variables measured during treatment as predictors of clinical outcomes measured at post-treatment follow-ups are present.

The current study has limitations. The study was underpowered to include multiple predictors that could have shed further light on the interplay between factors influencing long-term GCBT clinical outcomes. Unfortunately, we did not have outcome data before post-treatment (i.e. from during treatment), and it would have been advantageous to control for potential effects of early symptom change on group cohesion and alliance. However, focusing on early measures of group cohesion and alliance, we argue that potential early symptom change influences were small. It is also essential to consider the study findings because both observation measures focus on the client and not explicitly on the clinician. There are separate measures that assess the clinician’s behaviors for the alliance (see Creed & Kendall, 2005; Fjermestad et al., 2020; M. Karver et al., 2008). To our knowledge, no observation-based group cohesion measure focused on clinician behaviors. Regarding potential early change effects, it is also important to consider that the observers were masked to client background and outcomes when rating cohesion and alliance. We also lack data on factors that may have influenced post-treatment and 4-year follow-up (e.g. life events, other treatment). We coded a random early session and a random late session, which means the time interval representing the change variables differed between groups. The advantage of this approach is that we may have reduced bias from potential manual elements in certain sessions that may have influenced our findings if we always coded the same sessions, but it also makes the change variables trickier to interpret. Furthermore, the alliance or group cohesion trajectories across the 10-session treatment may be non-linear. Coding more than two sessions would have allowed us to examine this question. The validity of our findings may have been strengthened if different (independent) teams coded group cohesion and alliance. Finally, our sample is comprised predominantly of European White youths presenting with primary anxiety problems; further research is needed with more diverse populations. We must also determine if our findings apply to other disorders treated in different contexts (e.g. university-clinic settings).

The main conclusion from the current study is that group cohesion and alliance represent potential factors that can be targeted to enhance GCBT outcomes. Future studies need to examine if clinicians can successfully strengthen group cohesion and alliance to improve long-term GCBT outcomes. The field still needs knowledge about how this can be done, and studies focused on clinician behaviors that may influence group cohesion and alliance are needed. Studies identifying predictors of group cohesion and alliance are required, as well as larger studies that include further outcome predictors alongside group cohesion and alliance. There is a particular need to identify clinician-level strategies and other amenable predictors. In combination, such studies can help the field towards better outcomes for youth with anxiety disorders.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data are available from the first author upon request.

Additional information

Funding

This study was funded by the Western Norway Health Authorities, under grant numbers 911366, 911253, and 911840.