Client predictors of dropouts and outcomes in psychotherapy for generalized anxiety disorder: an exploratory study

ABSTRACT

This study is a secondary analysis exploring client predictors of dropouts and outcomes in the treatment of generalized anxiety disorder (GAD). Data was taken from a feasibility randomized control trial (RCT) exploring the relative efficacy of emotion-focused therapy (EFT) compared with cognitive-behavioral therapy (CBT). Relationships between client characteristics and dropouts were explored using logistic regression analysis. Relationships between client characteristics and psychotherapy outcomes were explored using hierarchical linear regression analyses. For the full dataset (n = 58), a comorbid personality disorder predicted a higher number of dropouts. Eleven participants dropped out of therapy. Having a disability predicted less improvement in symptoms of depression and general distress. Being male-identified predicted a greater reduction in symptoms of anxiety and general distress, and comorbid agoraphobia predicted a greater reduction in symptoms of depression for the CBT condition (n = 21 completers). For the EFT condition (n = 26 completers), being older predicted greater improvements in symptoms of anxiety and comorbid panic disorder predicted a lower reduction in symptoms of depression. Caution should be taken in interpreting these findings given the small number of participants. However, they provide useful insights for the generation of hypotheses in relation to predictors of dropouts and outcomes for future studies.

KEYWORDS:

• Client predictors
• generalized anxiety disorder
• emotion-focused therapy
• cognitive-behavioral therapy
• dropouts

Introduction

Generalized anxiety disorder (GAD) is a common yet debilitating condition, characterized by excessive anxiety across a range of events and activities (American Psychiatric Association, 2022). Estimates of the lifetime prevalence of GAD range from 4% to 6% (Kessler et al., 2005; Ruscio et al., 2017), and the level of comorbidity with other psychological disorders is high (Carter et al., 2001; Ruscio et al., 2017). Current treatment guidelines for GAD recommend medication and/or psychological therapy (National Institute for Health and Care Excellence [NICE], 2011).

In addition to identifying the relative efficacy of different psychological therapies for specific conditions, such as GAD, an important aspect of evidence-based practice lies in furthering our understanding of why two clients with the same presenting issue may have vastly different responses to the same intervention. Factors which explain these differences can be captured across client factors, therapist factors, therapy processes or relational factors, and contextual factors (Lutz et al., 2021). Client factors include demographic factors, clinical characteristics, beliefs and preferences in relation to treatment, intrapsychic factors, interpersonal factors, and contribution to the therapy process (Constantino et al., 2021). The present study will focus on client demographic and clinical characteristics although it is acknowledged that a combination of client factors, therapist factors, therapy process and contextual factors likely contribute to outcomes for a given client.

Dropout

Dropout can be described as occurring when a client ends therapy, of their own accord, before they have recovered from the issues that led them to seek help. High dropouts are associated with poorer client outcomes (Cahill et al., 2003; Delgadillo et al., 2014). Several reviews have estimated mean dropout rates of around 20% from cognitive-behavioral therapy (CBT) for anxiety disorders, including GAD (Carpenter et al., 2018; Fernandez et al., 2015). A review of GAD specific studies estimated the dropout rate as 17% (Gersh et al., 2017). The majority of studies included in the latter review focused on CBT and none focused on humanistic or experiential therapies, such as emotion-focused therapy (EFT). A wider review found higher dropouts amongst younger clients, clients with lower educational attainment and clients with a diagnosis of an eating disorder or personality disorder (Swift & Greenberg, 2012). Effect sizes for gender, race, marital status, and employment status were not significant. Another study similarly reported a positive relationship between personality disorders and dropouts (Cooper & Conklin, 2015), while others have found no evidence of such an association (Goddard et al., 2015; Snoek et al., 2021).

Demographic characteristics and outcomes

Evidence of associations between client age and psychotherapy outcomes has been largely inconsistent to date. Several studies have reported that psychotherapy is less effective for older adults (Carl et al., 2020; Gonçalves & Byrne, 2012; Gould et al., 2012; Wetherell et al., 2013), while others have found no evidence of such an association (Cuijpers et al., 2014; Lambert, 2010). One study found increasing age to be associated with improved outcomes (Catarino et al., 2018). No studies were identified which explored age as a predictor of change in the treatment of GAD specifically. This was similar for other potential predictors, where the number of studies focusing on GAD specifically were found to be limited.

A consistent association between gender and outcomes has not been established (Constantino et al., 2021). Although, studies have found that women with GAD have lower rates of remission (Yonkers et al., 2003) and partial recovery (Rodriguez et al., 2006) than men. Research has found that for racial and ethnic minorities disparities exist in terms of both access to, and the quality of, mental health care (Cai & Robst, 2016; Jacoby et al., 2020; Menendez et al., 2020). However, two recent meta-analyses have found that race/racial and ethnic minority status did not moderate or predict psychotherapy outcomes (Carpenter et al., 2018; Kilmer et al., 2019).

Evidence in relation to relationship status is mixed, with a small number of studies finding a positive association between being married and treatment response (Dobkin et al., 2012; Hopko et al., 2015). Another study was identified which did not support such a relationship (Hedman et al., 2015). Some evidence suggests that lower socio-economic status (SES) may be associated with worse outcomes depending on how SES is defined and measured (Constantino et al., 2021). One meta-analysis found that unemployment was significantly associated with poorer outcomes, although the effect size was small (Finegan et al., 2018). Similarly, other studies have found that unemployed participants or participants with disabilities have worse outcomes (Delgadillo et al., 2016; Verbist et al., 2022).

Clinical characteristics and outcomes

Evidence generally supports a negative relationship between higher functional impairment (Beutler et al., 2001) and symptom chronicity (Boonstra et al., 2012; Bottlender et al., 2000) and poor outcomes for severe psychiatric illnesses. Contrastingly, one study of GAD found longer symptom duration to be associated with greater gains (Newman et al., 2019). Findings in relation to initial symptom severity and outcomes are also mixed (Constantino et al., 2021). One study reported that clients with higher pretreatment severity scores were less likely to achieve clinical recovery, but more likely to show clinically significant improvement on some measures (Catarino et al., 2018). The latter finding is consistent with other studies reporting that higher initial severity is associated with a greater reduction in symptoms, but not with greater recovery rates (e.g. Brown et al., 2001).

A meta-analysis on the influence of comorbidity on the treatment of anxiety disorders found that the presence of comorbid conditions alone did not influence treatment effect sizes (Olatunji et al., 2010). A study of CBT for GAD specifically found that the presence of comorbid mood and anxiety disorders did not have a negative impact on treatment outcomes (Newman et al., 2010). Other studies found no evidence of an association between comorbid personality disorders and outcomes (Dreessen et al., 1994; Mahoney et al., 2021).

Clinical applications of research on predictors

In addition to exploring factor-outcome associations, largely through correlational studies, an important emerging area of research explores practical applications of such findings within a clinical context. A developing area of patient-focused research explores evidence-based forms of personalized pretreatment recommendations (Lutz et al., 2021). Examples of this are studies using predictive analysis to inform treatment selections for a given individual (DeRubeis et al., 2014; Huibers et al., 2015), and studies using patient profiling and risk stratification to inform personalized treatment recommendations (Delgadillo et al., 2016). Another growing area of research focuses on tailoring treatments based on aspects of clients’ cultural identities, e.g. gender identity, race/ethnicity, or sexual orientation (Soto et al., 2018).

The present study

The data used in the present study was collected as part of a feasibility randomized control trial (RCT) exploring the relative efficacy of EFT for GAD when compared against an established treatment, namely CBT (Timulak et al., 2022). EFT is a research-informed humanistic-experiential treatment, which draws on person-centered, gestalt, experiential and existential therapies (Greenberg et al., 1993). Recently, EFT has been adapted for disorders such as GAD (Timulak & McElvaney, 2018; Watson et al., 2018; Watson & Greenberg, 2017). From an EFT perspective, it is assumed that individuals suffering from GAD are avoidant of, and worry about, specific triggers that might potentially evoke problematic emotional schemes centered around unbearable emotional experiences from the past (and associated feelings of sadness/loneliness, shame and fear/terror) which are chronically painful (Timulak & McElvaney, 2016, 2018). Individuals attempt to manage these painful experiences through various forms of problematic self-treatment, e.g. self-worrying, emotional avoidance and behavioral avoidance. Therapy then focuses on transforming the underlying vulnerability but also on letting go of, or setting a boundary to, symptom level processes such as self-worrying.

Results from two initial open trials of EFT for GAD have reported large effect sizes posttherapy (O’Connell Kent et al., 2021; Timulak et al., 2017). In addition to tentatively assessing the relative efficacy of EFT for GAD, the RCT aimed to assess recruitment processes, therapist adherence and participant retention rates to inform planning of a larger noninferiority trial.

The study found that pre-post change was large for both therapies, which were similar across all outcome measures. Gains were maintained at six-month follow-up. Therapist effects were analyzed with no significant relationships found. Research attrition was similar to dropout in that any clients who dropped out of therapy also typically failed to provide further data to the study. This study will further expand on the main study by exploring client predictors of dropout and outcome. Namely, we wanted to look at the available demographic variables (including age, gender, disability status, relationship status, employment, and educational attainment) and clinical variables (including pretherapy symptoms, number of comorbid conditions, presence of comorbid mood/anxiety/personality disorders, and presence of specific comorbid disorders) collected as part of the original study and investigate their relationship to dropout from treatment and outcomes at end of treatment and at six-month follow-up.

Method

Design and participants

Data for the present study came from a feasibility RCT comparing the efficacy of EFT and CBT for the treatment of GAD. Detailed information on the original study and its procedures can be found in Timulak et al. (2018, 2022). Fifty-eight adult (≥18 years) participants were recruited through the Counselling in Primary Care (CIPC) service, within the Irish public health service – the Health Service Executive (HSE). CIPC offers short-term psychotherapy to adults eligible for free public health care (they must be holders of a general medical card).

All participants were required to have a primary diagnosis of GAD, which was initially screened for using the Generalized Anxiety Disorder 7 scale (GAD-7; Spitzer et al., 2006) and confirmed using the Structured Clinical Interview for DSM-5-Research Version (SCID-5-RV; First et al., 2015). Further inclusion criteria were consent to the study conditions (e.g. attendance at pre- and posttherapy assessments, audio recording of sessions), and stabilization on any psychotropic medication for a minimum of six weeks prior to starting therapy. Exclusion criteria included concurrent psychological treatment; substance abuse; psychosis; and organic brain syndrome, as determined through the clinical interview and assessment using the SCID-5-RV; and suicide risk, as defined by a score other than 0 on Item 16 (‘I have made plans to end my life’) on the Clinical Outcome in Routine Evaluation – Outcome Measure (CORE-OM; Evans et al., 2002); and risk of harm to others, as defined by a score other than 0 on Item 6 (‘I have been physically violent to others’) on the same measure.

Treatments

Both interventions were offered for between 16 and 20 sessions, with therapy ending as per the therapists’ clinical judgment. The EFT intervention was based on the model described by Timulak and McElvaney (2018) and discussed briefly in the introduction. The intervention focused on restructuring and transforming problematic emotional schemes by facilitating clients to tolerate painful emotions, to articulate unmet needs embedded within those emotions and to access adaptive emotions such as compassion and protective anger. In addition, the intervention focused on addressing symptom level processes such as self-worrying. The CBT intervention was based on the intolerance of uncertainty model (Dugas & Robichaud, 2007). This intervention focused on psychoeducation around the relationship between intolerance of uncertainty and excessive worry, increasing worry awareness, the reevaluation of worry beliefs and the development of problem-solving skills. Eight therapists employed by the CIPC service randomly delivered both interventions. All eight therapists (six qualified psychologists and two qualified psychotherapists) were trained in both EFT and CBT for the purpose of the study. For the study, therapists were advised that they could have up to 20 sessions with participants.

Procedure

Participants were recruited from clients presenting to CIPC with anxiety symptoms. Clients were given information about the study where they scored ≥ 11 on the GAD-7, and where the therapist, based on clinical judgment, believed that they may meet the criteria for a primary diagnosis of GAD. Where clients expressed an interest in the study, met the inclusion criteria (described above), and did not meet the exclusion criteria (also described above), the clients’ consent was sought before referring them on to the research team for assessment. This assessment included the collection of demographic data and data relating to the presenting issue, as well as the administration of the SCID-5-RV (First et al., 2015), the Structured Clinical Interview for DSM-5 Personality Disorders (SCID-5-PD; First et al., 2016), the Generalized Anxiety Disorder Severity Scale (GADSS; Shear et al., 2006), the Clinical Outcomes in Routine Evaluation Outcome Measure (CORE-OM; Barkham et al., 2001; Evans et al., 2002), and Patient Health Questionnaire − 9 (PHQ-9; Kroenke et al., 2001). For the present analysis, data from the end of therapy assessment and the six-month follow-up assessment were analyzed. Participants were considered to have dropped out of therapy where they attended less than eight sessions. The dropout rate from treatment was 27.5% (n = 8) for CBT and 10.3% (n = 3) for EFT. The difference in dropouts across interventions was not found to be significantly different. For details see Timulak et al. (2022).

Measures

GAD-7

The GAD-7 (Spitzer et al., 2006) is a seven-item self-report questionnaire which assesses symptoms of generalized anxiety in the prior two weeks. Cronbach’s alphas for the present sample were .62, .85 and .92 at pretherapy, posttherapy and follow-up, respectively.

CORE-OM

The CORE-OM (Barkham et al., 2001; Evans et al., 2002) is a 34-item self-report questionnaire measuring global psychological distress across domains of subjective well-being, problems, life functioning, and risk. Cronbach’s alphas for the present sample were .90, .95 and .96 at pretherapy, posttherapy and follow-up, respectively.

GADSS

The GADSS (Shear et al., 2006) is a six-item interview-based assessment which measures the severity and frequency of GAD symptoms. Cronbach’s alphas for the present sample were .75, .88, and .88 at pretherapy, posttherapy and follow-up, respectively.

PHQ-9

The PHQ-9 (Kroenke et al., 2001) is a nine-item self-report questionnaire which measures symptoms of depression. Cronbach’s alphas for the present sample for pretherapy, posttherapy and follow-up were .83, .85 and .80, respectively.

Data analysis

The data analysis was conducted in IBM SPSS Statistics version 28.0 (IBM, 2021). Descriptive statistics, including means and standard deviations, were calculated for all variables at the start of treatment. Distributions were assessed using histograms and Q-Q plots. Chi-squared tests (for categorical independent variables) and independent t-tests (for continuous independent variables) were run to test the hypotheses that pretreatment client characteristics, including demographic and clinical characteristics, would be related to dropouts. Where significant relationships were found, a binary logistic regression analysis was then run for each relevant variable. Analyses were performed for the full sample, and for the two conditions – EFT and CBT – separately.

Correlations, independent t-tests and analysis of variance (ANOVAs) were used to explore relationships between pretreatment client characteristics and change scores. Separate analyses were performed for each of the primary (GAD-7) and secondary outcomes (CORE-OM, GADSS and PHQ-9), both posttherapy and at six-month follow-up. Change scores were calculated for each measure by subtracting the pretherapy scores from the scores posttherapy, and similarly by subtracting the pretherapy scores from the scores at six-month follow-up. Analyses were performed using the full sample and for each of the two conditions separately. Bootstrapping based on 2000 scenarios and a confidence level of 95% was used for all analyses to ensure accuracy of significance. In some cases, a number of these 2000 scenarios were excluded due to a lack of variation within certain variables. Testing was performed by increasing the number of intended scenarios to ensure that the lower number of scenarios included in the main analyses did not materially impact the results.

For the independent t-tests, Welch’s t and associated p-values are reported as these do not rely on the assumption that the variances of the two groups are equal. This was considered appropriate given the small sample size and unequal groups, particularly when the CBT and EFT condition were analyzed separately. Bootstrapped confidence intervals were calculated for all t-tests, and these were compared with the significance tests without bootstrapping. In cases where the p-value calculated without bootstrapping was not significant at the level of α = .05, but where the bootstrapped confidence intervals implied that there may be a significant relationship between two variables, the variables were included in the later linear regression models. The p-values from the correlation analyses and ANOVAs were bootstrapped based on 2000 scenarios. For the ANOVAs, Games-Howell post-hoc tests were used to analyze relationships between categorical variables as the procedure overcomes both differences in population variance and differences in sample size.

Where the previous analyses (i.e. correlations, t-tests or ANOVAs) found a significant relationship between demographic or clinical characteristics and outcomes, the variables were then included in the hierarchical linear regression analyses. This step was performed as linear regression analysis enables us to determine whether one or more variables predict another variable, rather than simply showing that a relationship exists between them. A separate hierarchical linear regression analysis was run for each outcome measure, posttherapy and at follow-up, for the full dataset and split by condition. In each case, the pretherapy scores were controlled for by entering them in the first block. All other variables which were significantly associated with change scores were then entered in separate blocks, in descending order from the strongest to weakest absolute relationship.

Ethics

The study received ethical approval from the Research Ethics Committee at the School of Psychology, Trinity College Dublin (TCD), and the HSE North East Area Research Ethics Committee.

Results

Demographic and clinical characteristics

Tables S1 and S2 in the Supplemental material summarize demographic and clinical characteristics for the full sample, and for participants who completed a minimum of eight sessions of therapy, respectively.

Pretherapy characteristics and dropouts

Table S3 of the Supplemental material summarizes the results of logistic regression analyses which found significant relationships between client characteristics and dropouts. Odds ratios imply that participants with a comorbid personality disorder (n = 22 completers, n = 9 dropouts) were 91% less likely to complete therapy than those with none (n = 25 completers, n = 1 dropout). Participants with a comorbid diagnosis of avoidant personality disorder (n = 19 completers, n = 9 dropouts), specifically, were 93% less likely to complete therapy, than those without such a diagnosis (n = 28 completers, n = 1 dropout).

Participants in full-time employment (n = 10 completers, n = 1 dropout) were 1650% more likely to complete therapy than unemployed participants not looking for work (n = 4completers, n = 5 dropouts). Participants whose highest level of educational attainment was post-secondary education (n = 20 completers, n = 2 dropouts) were 727% more likely to complete therapy compared with those who stayed in education until the Junior Certificate (a national exam taken at age 15) (n = 5completers, n = 4 dropouts). No significant relationships were identified when the data was analyzed for CBT and EFT separately.

Pretherapy characteristics and outcomes

Table S4 of the Supplemental material summarizes the relationships between pretherapy characteristics and change scores for participants in both conditions, as identified from correlation analyses, t-tests and ANOVAs. These findings were used to determine the variables to include in the later linear regression analysis. For all outcome measures, correlation analyses found that higher pretherapy scores were significantly associated with a greater reduction in symptoms. Tables S5 and S6 of the Supplemental material summarize similar relationships for the CBT and EFT condition, respectively.

Table 1 shows the results of the hierarchical linear regression models used to determine predictors of outcomes for the full dataset. The final model for the PHQ-9 posttherapy, which accounted for 47% of the variance in the model, showed that after controlling for pretherapy symptoms, disability was significantly predictive of lower improvements, β = −0.35, t = −3.05, p = .02, 95% BS CI [−9.41, −1.40]. Disability was similarly predictive of outcomes for the PHQ-9 at follow-up, β = −0.32, t = −2.70, p = .04, 95% BS CI [−9.18, −0.78]. The final model for the CORE-OM posttherapy, which accounted for 40% of the variance in the model, showed that after controlling for pretherapy symptoms, disability was significantly predictive of lower improvements, β = −0.43, t = −3.55, p < .001, 95% BS CI [−1.05, −0.40]. Disability was similarly predictive of outcomes for the CORE-OM at follow-up, β = −0.27, t = −2.02, p = .05, 95% BS CI [−1.08, 0.01].

Table 1. Hierarchical linear regression models examining predictors of change in outcomes for full dataset.

	B (SE)	β	t	p	95% CI
Dependent variable: Change in PHQ-9 posttherapy
F(2,39) = 19.38, p < .001**; Adj. R^2 = 0.47
PHQ-9 pretherapy	−0.69 (0.13)	−0.63	−5.58	<.001**	[−0.91, −0.40]
Disability	−5.26 (2.05)	−0.35	−3.05	.02*	[−9.41, −1.40]
Dependent variable: Change in PHQ-9 at follow-up
F(2,38) = 18.03, p < .001**; Adj. R^2 = 0.46
PHQ-9 pretherapy	−0.66 (0.11)	−0.60	−5.11	<.001**	[−0.86, −0.45]
Disability	−4.75 (2.16)	−0.32	−2.70	.04*	[−9.18, −0.78]
Dependent variable: Change in CORE-OM posttherapy
F(2,39) = 14.52, p < .001**; Adj. R^2 = 0.40
CORE-OM pretherapy	−0.59 (0.10)	−0.49	−4.05	< .001**	[−0.79, −0.40]
Disability	−0.73 (0.16)	−0.43	−3.55	< .001**	[−1.05, −0.40]
Dependent variable: Change in CORE-OM at follow-up
F(2,38) = 8.80, p < .001**; Adj. R^2 = 0.28
CORE-OM pretherapy	−0.64 (0.17)	−0.47	−3.46	.001**	[−1.00, −0.32]
Disability	−0.54 (0.27)	−0.27	−2.02	.05*	[−1.08, 0.01]

SE = standard error; CI = confidence interval; PHQ-9 = Patient Health Questionnaire 9; CORE-OM = Clinical Outcomes in Routine Evaluation – Outcome Measure; * = p < .05; ** = p < .01.

Tables S7 and S8 of the Supplemental material show the initial hierarchical linear regression models for the CBT and EFT conditions, respectively. Tables 2 and 3 show the results of the final hierarchical linear regression models used to determine predictors of outcomes for the CBT and EFT conditions, respectively. For the CBT condition, the final model for the GADSS at follow-up, which accounted for 46% of the variance in the model, showed that after controlling for pretherapy symptoms, male-identified gender was significantly predictive of better outcomes, β = −0.45, t = −2.39, p = .02, 95% BS CI [−9.45, −2.03]. The final model for the PHQ-9 posttherapy accounted for 63% of the variance and showed that after controlling for pretherapy symptoms, comorbid agoraphobia was significantly predictive of greater improvements, β = 0.35, t = 2.16, p = .03, 95% BS CI [1.72, 11.45]. The final model for the CORE-OM posttherapy accounted for 69% of the variance and showed that after controlling for pretherapy symptoms, male-identified gender was significantly predictive of a greater improvement in symptoms, β = −0.50, t = −3.71, p = .01, 95% BS CI [−1.25, −0.34]. Having a disability was associated with a lower improvement in symptoms (compared with not having a disability) but fell just outside the range of significance, β = −0.18, t = −1.31, p = .054 95% BS CI [−0.78, −0.10].

Table 2. Final hierarchical linear regression models examining predictors of change in outcomes for CBT condition.

	B (SE)	β	t	p	95% CI
Dependent variable: Change in GADSS at follow-up
F(2,15) = 8.28, p = .004**; Adj. R^2 = 0.46
GADSS pretherapy	−0.88 (0.46)	−0.45	−2.38	.07	[−1.67, 0.10]
Gender	−6.20 (1.87)	−0.45	−2.39	.02*	[−9.54, −2.03]
Dependent variable: Change in PHQ-9 posttherapy
F(2,16) = 16.65, p < .001**; Adj. R^2 = 0.63
PHQ-9 pretherapy	−0.72 (0.21)	−0.59	−3.67	.01**	[−1.12, −0.30]
Agoraphobia	6.27 (2.45)	0.35	2.16	.03*	[1.72, 11.45]
Dependent variable: Change in CORE-OM posttherapy
F(3,15) = 14.53, p < .001**; Adj. R^2 = 0.69
CORE-OM pretherapy	−0.65 (0.13)	−0.59	−4.32	.001**	[−0.96, −0.43]
Gender	−0.81 (0.24)	−0.50	−3.71	.01**	[−1.25, −0.34]
Disability	−0.43 (0.18)	−0.18	−1.31	.05	[−0.78, −0.10]

SE = standard error; CI = confidence interval; GADSSS = Generalized Anxiety Disorder Severity Scale; PHQ-9 = Patient Health Questionnaire − 9; CORE-OM = Clinical Outcomes in Routine Evaluation – Outcome Measure; * = p < .05; ** = p < .01.

Table 3. Final hierarchical linear regression models examining predictors of change in outcomes for EFT condition.

	B (SE)	β	t	p	95% CI
Dependent variable: Change in GADSS posttherapy
F(2,20) = 4.98, p = .02*; Adj. R^2 = 0.27
GADSS pretherapy	−0.57 (0.30)	−0.31	−1.67	.08	[−1.22, −0.01]
Age	−0.21 (0.09)	−0.44	−2.39	.04*	[−0.45, −0.08]
Dependent variable: Change in PHQ-9 posttherapy
F(2,20) = 16.76, p < .001**; Adj. R^2 = 0.59
PHQ-9 pretherapy	−0.44 (0.13)	−0.50	−3.63	.002**	[−0.65, −0.14]
Panic disorder	−9.12 (2.07)	−0.70	−5.02	<.001**	[−13.63, −5.24]

SE = standard error; CI = confidence interval; GAD-7 = Generalized Anxiety Disorder − 7; GADSSS = Generalized Anxiety Disorder Severity Scale; PHQ-9 = Patient Health Questionnaire − 9; * = p < .05; ** = p < .01.

For the EFT condition, the final model for the GADSS posttherapy accounted for 27% of the variance and showed that after controlling for pretherapy symptoms, older age was significantly predictive of greater improvements, β = −0.44, t = −2.39, p = .04, 95% BS CI [−0.45, −0.08]. The final model for the PHQ-9 posttherapy accounted for 59% of the variance and showed that after controlling for pretherapy symptoms, comorbid panic disorder was significantly predictive of worse outcomes, β = −0.70, t = −5.02, p < .001, 95% BS CI [−13.63, −5.24].

Discussion

In this exploratory study, using RCT data collected within an Irish public primary care service, we identified several potential predictors of dropout and/or outcomes. Firstly, having a comorbid diagnosis of a personality disorder, alongside a primary diagnosis of GAD, was predictive of higher dropout from therapy. This finding is consistent with prior studies (Cooper & Conklin, 2015; Swift & Greenberg, 2012). Although the motivation for individual dropouts is unclear, the majority of clients with a personality disorder met the criteria for avoidant personality disorder. While research on avoidant personality disorder is limited relative to other personality disorders (Lampe & Malhi, 2018; Weinbrecht et al., 2016), the avoidant behaviors, emotional guardedness and perceived risk of intimacy associated with the condition (e.g. Marques et al., 2012) make this finding perhaps somewhat unsurprising. These findings suggest that for clients with a comorbid diagnosis of a personality disorder engagement strategies – such as discussing dropout from the start of treatment and exploring with the client what could potentially preempt such a dropout – may be beneficial for clinical practice.

Participants who were unemployed and not looking for work were more likely to drop out of therapy than those in full time employment. Participants who ceased formal education midway through second level education were more likely to drop out of therapy than those who completed post-secondary education. However, caution should be taken in interpreting these findings, as given the limited number of participants in each category the findings may be artifacts. Nine participants were categorized as unemployed and not looking for work, and the same number ceased education following the Junior Certificate (an examination taken in or around age 15), four and five of whom completed therapy, respectively. While dropout analyses looking at the two treatment conditions separately yielded no significant results, one had to be aware that the analyses were underpowered.

Higher pretherapy scores were associated with a greater reduction in symptoms across all measures. Prior research in relation to associations between pretreatment symptom severity and outcomes have been somewhat mixed, but several studies have supported a relationship between higher initial severity and a greater reduction in symptoms, although not necessarily with greater recovery (Brown et al., 2001; Catarino et al., 2018). Here, recovery was defined as where an individual reliably moved from scoring above a specified clinical cutoff on a symptom-focused measure at the outset of treatment to scoring below the same cutoff at the end of treatment. A separate evaluation study of the CIPC service – the same service within which this study was conducted – found that higher initial severity was predictive of a lower likelihood of recovery (Brand et al., 2021). This CIPC service evaluation study only considered recovery as a binary outcome (where participants were considered recovered if their scores on a measure of general distress moved from the clinical to the non-clinical range). The present study builds on this by considering the magnitude of change in symptoms.

Having a disability was associated with a lower reduction in symptoms on outcome measures for general distress and depression, but not GAD, after controlling for pretherapy scores. Caution should be exercised when interpreting these, and other, findings given that the present study was likely underpowered, limiting our ability to draw dependable conclusions from the results. Disability was self-defined by participants and included several clients with learning disabilities, physical disabilities, and chronic health conditions, one participant with an acquired brain injury and another with a sensory disability. This finding was in line with studies using larger naturalistic samples from similar primary care services in England (Delgadillo et al., 2016; Verbist et al., 2022). Interestingly, the former study similarly found that having a disability predicted lower improvement in symptoms of depression, but not anxiety. A review of associations between depression and anxiety and physical disability in older adults found that when depression and physical disability are both present, they both tend to remain, worsen together or improve together (Lenze et al., 2001). It was suggested that this pattern reinforces the assertion that depression is a disabling condition. The same review found that anxiety was a predictor of disability, allowing for the same confounding variables (such as age, gender, education) as when analyzing depression but not necessarily independent of depression.

Although the finding that clients with disabilities improved less following therapy requires further replication, it could suggest that clients with disabilities may be particularly important to target through personalized pretreatment recommendations (e.g. Delgadillo et al., 2016; DeRubeis et al., 2014; Huibers et al., 2015) and adaptations in treatments, such as culturally adapted treatments (Soto et al., 2018). Clinical applications of findings in relation to factor-outcome associations are still significantly less common than correlation analyses, but they represent an important new direction in terms of identifying not only who is most likely to benefit from a given therapy, but how therapies can be tailored to the individual needs of the client to improve outcomes.

All the findings noted previously in relation to outcomes were based on the full dataset. Further analyses were performed for CBT and EFT separately. The present study was unique in that it specifically considered client predictors of change following EFT for GAD. EFT is a research-based treatment which has been recently adapted for GAD (O’Connell Kent et al., 2021; Timulak et al., 2017). For the CBT condition, after controlling for pretherapy symptoms, identifying as male-identified predicted greater improvements in general distress posttherapy, and anxiety (measured by the GADSS) at follow-up. This was consistent with other studies which have reported less improvement for women, relative to men, following psychological treatment for GAD (Rodriguez et al., 2006; Yonkers et al., 2003). CBT is a more practical and symptom-focused method of therapy, compared to treatments such as EFT. It is possible that this more practical style of therapy may have been preferred by some male-identified participants. However, the limited number of participants limits our ability to make such inferences. A comorbid diagnosis of agoraphobia predicted a greater reduction in depressive symptoms posttherapy than no comorbid diagnosis. Of the 21 completers in the CBT condition, three were male-identified and four had a comorbid diagnosis of agoraphobia.

For the EFT condition, pretherapy scores were not significantly associated with change scores for the majority of measures, unlike both the CBT condition and the full dataset. Although the majority of relationships were not significant, they did move in the expected direction, i.e. higher initial scores indicated greater reduction in symptoms. This finding suggests a possibly more complex picture for EFT, where other moderating variables are at play, although it is not yet fully clear what these may be. Being older was found to be predictive of greater improvements in symptoms of anxiety (measured by the GADSS) posttherapy in the EFT condition. This contrasts with other research suggesting that outcomes are better for younger participants in the treatment of anxiety disorders (Carl et al., 2020; Gonçalves & Byrne, 2012; Gould et al., 2012; Wetherell et al., 2013). EFT is a humanistic therapy which draws on aspects of existential theory and places greater emphasis on exploring meaning-making within the therapy. This could partially explain the finding that EFT was more beneficial for older participants, who may be more concerned with issues of meaning and understanding their life story relative to younger participants. A comorbid diagnosis of panic disorder predicted a lower reduction in symptoms of depression at the end of therapy than no comorbid diagnosis. Of the 26 completers in the EFT condition, four were diagnosed with panic disorder.

The above trends in relation to possible predictors of change for CBT and EFT specifically provide useful insights into possible differences between the two treatments in terms of which clients might be most likely to benefit from each. However, the small number of participants represents a significant limitation in terms of the current findings, and results will need to be replicated with larger datasets.

Limitations

It is important to further highlight some methodological issues that should be considered when interpreting these findings. This exploratory study aimed to identify trends in terms of client predictors of dropout and outcomes using data from a feasibility RCT. The sample size provided adequate power for the main feasibility RCT analysis, comparing two active treatments and pre-post within groups comparisons. The sample size was not sufficient to out-rule the possibility that the reported findings in terms of predictors could be spurious associations or artifacts of the data. These findings instead provide indicators of possible relationships of interest to be explored in future studies with larger samples. Larger samples enable more advanced statistical methods, such as cross-validation and resampling (Sauerbrei, 1999), to be used to validate the findings.

Given the exploratory nature of the study, the decision was made not to correct for multiple comparisons. Normally, where repeated testing is performed on a single measure a Bonferroni correction is applied to reduce the significance level and avoid Type I errors. However, this is a conservative approach which increases the risk of Type II errors. All analyses were run with bootstrapping to ensure that robust results were obtained. Analyses were set to produce 2000 scenarios, however, in some cases, several scenarios were excluded due to a lack of variation within certain variables. This was particularly the case where analyses included categorical variables and data from a single active condition only. Further, statistical analyses could have been improved by accounting for the non-independence of errors within person via linear mixed modeling.

Finally, our selection of predictor variables was limited to those collected as part of the original study. The use of other predictors which may have been more relevant from the perspective of case formulation (e.g. the levels of self-criticism or self-worrying) would have enabled the study to offer more theoretical approach-specific insights in relation to both CBT and EFT. This could be a useful direction for similar studies in the future.

Conclusion

This exploratory secondary analysis aimed to identify trends between client factors and outcomes in the treatment of GAD. The purpose of the analysis was both to provide information for the planning of future larger trials, and to provide indicative findings which could inform hypotheses for future studies of predictors using larger datasets. The findings suggest that clients with comorbid diagnoses of personality disorders may be more likely to dropout from therapy, and thus may benefit from targeted retention strategies. Clients with disabilities were found to improve less following therapy. These findings require replication. More also needs to be understood about which and how many pretreatment variables are needed to accurately predict outcomes. However, the finding suggests that a useful direction for future research may be the exploration of adaptations of treatment specifically for clients with disabilities, as well as further exploration of how disability status interacts with other client, therapist, relational, and contextual factors. Analyses of the CBT and EFT conditions suggest that there may be some differences in terms of the client groups likely to benefit most from each treatment, i.e. males vs females for CBT and older clients vs younger clients for EFT. These trends require further testing and replication.

Disclosure statement

Ladislav Timulak is one of the developers of an adaptation of emotion-focused therapy for the treatment of generalized anxiety. Craig Chigwedere is the director of the MSc in Cognitive-Behavioral Therapy at Trinity College Dublin.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/14779757.2024.2335383.

Additional information

Funding

The study was funded by an award from the Health Research Board: HRA-POR-2015-1052. Other than by funding the study, the HRB has no role in the design of the study; or in the collection, analysis, interpretation or reporting of data.