Clinical Psychology and Psychotherapy Clin. Psychol. Psychother. 23, 35–46 (2016) Published online 11 December 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/cpp.1936
Baseline Patient Characteristics Predicting Outcome and Attrition in Cognitive Therapy for Social Phobia: Results from a Large Multicentre Trial Juergen Hoyer,1* Joerg Wiltink,2 Wolfgang Hiller,3 Robert Miller,1 Simone Salzer,4 Stephan Sarnowsky,1 Ulrich Stangier,5 Bernhard Strauss,6 Ulrike Willutzki7 and Eric Leibing4 1
Clinical Psychology and Psychotherapy, Technische Universitaet Dresden, Dresden, Germany Department of Psychosomatic Medicine and Psychotherapy, University Medical Center, Johannes Gutenberg University Mainz, Mainz, Germany 3 Clinical Psychology and Psychotherapy, Johannes Gutenberg University Mainz, Mainz, Germany 4 Department of Psychosomatic Medicine and Psychotherapy, University Medicine, Georg-August University Goettingen, Goettingen, Germany 5 Clinical Psychology and Psychotherapy, Goethe University Frankfurt, Frankfurt, Germany 6 Institute of Psychosocial Medicine and Psychotherapy, Jena University Hospital, Jena, Germany 7 Clinical Psychology and Psychotherapy, Ruhr University Bochum, Bochum, Germany 2
We examined the role of baseline patient characteristics as predictors of outcome (end-state functioning, response and remission) and attrition for cognitive therapy (CT) in social anxiety disorder (SAD). Beyond socio-demographic and clinical variables such as symptom severity and comorbidity status, previously neglected patient characteristics (e.g., personality, self-esteem, shame, interpersonal problems and attachment style) were analysed. Method: Data came from the CT arm of a multicentre RCT with n = 244 patients having DSM-IV SAD. CT was conducted according to the manual by Clark and Wells. Severity of SAD was assessed at baseline and end of treatment with the Liebowitz Social Anxiety Scale (LSAS). Multiple linear regression analyses and logistic regression analyses were applied. Results: Up to 37% of the post-treatment variance (LSAS) could be explained by all pre-treatment variables combined. Symptom severity (baseline LSAS) was consistently negatively associated with end-state functioning and remission, but not with response. Number of comorbid diagnoses was negatively associated with end-state functioning and response, but not with remission. Self-esteem was positively associated with higher end-state functioning and more shame with better response. Attrition could not be significantly predicted. Conclusions: The results indicate that the initial probability for treatment success mainly depends on severity of disorder and comorbid conditions while other psychological variables are of minor importance, at least on a nomothetic level. This stands in contrast with efforts to arrive at an empirical-based foundation for differential indication and argues to search for more potent moderators of therapeutic change rather on the process level. Key Practitioner Message: • Personality, self-esteem, shame, attachment style and interpersonal problems do not or only marginally moderate the effects of interventions in CT of social phobia. • Symptom severity and comorbid diagnoses might affect treatment outcome negatively. • Beyond these two factors, most patients share a similar likelihood of treatment success when treated according to the manual by Clark and Wells. Copyright © 2014 John Wiley & Sons, Ltd. Keywords: Social Phobia, CBT, Treatment Outcome, Predictors, Comorbidity, Interpersonal Problems *Correspondence to: Prof. Dr. Jürgen Hoyer, Technische Universität Dresden, Klinische Psychologie und Psychotherapie, Hohe Str. 53, D-01187, Dresden, Germany. E-mail: [email protected]
Copyright © 2014 John Wiley & Sons, Ltd.
Cognitive behavioural therapy (CBT) has been shown to be highly effective in treating social phobia (Powers, Sigmarsson, & Emmelkamp, 2008), one of the most
36 frequent (Ruscio et al., 2008) and debilitating (Fehm, Pelissolo, Furmark, & Wittchen, 2005) anxiety disorders. Recent trials prove CBT or cognitive therapy (CT; Clark & Wells, 1995; which not only emphasises cognitive change, but also incorporates typical CBT interventions) to be at least as successful (Blanco et al., 2010; Davidson et al., 2004) or even more successful than pharmacological therapy (Clark et al., 2003). CT was also superior to interpersonal therapy (Stangier, Schramm, Heidenreich, Berger, & Clark, 2011), applied relaxation (Clark et al., 2006) and manualised short-term psychodynamic psychotherapy (Leichsenring et al., 2013). At the same time, the total number of patients who benefit from treatment in the largest multicentre trials of cognitive behavioural treatments is not higher than 60% (Leichsenring et al., 2013) or 51.7% when CBT is applied in groups (Davidson et al., 2004). Even the trials with the most favourable success rates (Clark et al., 2003; Clark et al., 2006) leave a considerable number of patients unchanged. In order to raise the rate of those benefiting from CBT/CT of social phobia, it is of cardinal importance to know more about the factors that moderate treatment effects (Durham, 2006; Hoefler, Gloster & Hoyer, 2010). A considerable number of studies on social phobia have addressed the issue of prediction of treatment attrition and success, but a clear and convincing picture has yet to be formed. So far, only a selection of the possible predictors for treatment outcome has been investigated in independent studies. Out of these, only very few have been consistently replicated. The studies differ with respect to which variables were investigated. In the majority of studies, only a limited range of predictors was considered (refer to Lincoln et al., 2005 and Mululo, De Menezes, Vigne, & Fontenelle, 2012 for overviews). Socio-demographic variables such as age, gender, marital status, education and occupation are typically unrelated to outcome. Severity of disorder, generalised subtype, comorbid axis I disorder (e.g., depression) and comorbid axis II disorder (e.g., avoidant personality disorder) are associated with worse response in most studies, although their results require replication (Lincoln et al., 2005; Mululo et al., 2012). Positive treatment expectations and homework compliance have been related to change and high end state but were investigated only in a few studies (Lincoln et al., 2005; Mululo et al., 2012); however, across different disorders (e.g., depression), those variables were found to impact treatment success in a similar manner (e.g., Tsai, Ogrodniczuk, Sichting, & Mirmiran, 2014). Overall, the practical significance, i.e., the effect size of the above-mentioned predictors of treatment outcome in social phobia, has been evaluated as low and ‘unsatisfying’ (Lincoln et al., 2005). Given the complexity of the therapeutic process and the number of potential effect modifiers involved, this finding may simply Copyright © 2014 John Wiley & Sons, Ltd.
J. Hoyer et al. indicate that other variables, such as therapist variables or treatment integrity, are more important. Alternatively, the low explanatory power of patient variables may also be due to methodological restrictions of previous studies. Many studies are seriously underpowered and risk overfitting (Babyak, 2004). Ratios of predictors to sample sizes of less than 1:20 are not recommendable, and as the total sample sizes often falls below the n = 100 margin, only few predictors can be analysed (Durham, 2006). Due to these methodological problems, several patient psychological characteristics have either rarely or never been studied with respect to their explanatory power for CBT/CT treatment outcome in social phobia, although a significant impact on treatment outcome and attrition can theoretically be expected. This is namely true for variables including personality, self-esteem, shame or interpersonal problems and attachment style, all of which might impact on the patient’s willingness to experiment with new behaviours both inside and outside the therapeutic setting. In the following, we will briefly describe why the above-mentioned variables can be expected to exhibit direct or indirect influences on the success of treatment. The role of comorbid personality disorders for treatment success in social phobia has been examined early (e.g., Van Velzen, Emmelkamp, & Scholing, 1997). To the best of our knowledge, however, the impact of personality traits has yet to be studied consistently. In this research, we will specifically focus on the personality dimensions of harm avoidance, novelty seeking and reward dependence (Cloninger, Przybeck, & Svrakic, 1991), all of which have been proven useful in describing socially anxious populations (e.g., Kashdan & Hofmann, 2008). We argue that especially persons with high tendency towards harm avoidance might encounter difficulties in experimenting with new behaviour, which would, in turn, require more effort to achieve a disconfirmation of beliefs, thus negatively affecting treatment outcome. Further, shame and low self-esteem may also reduce the likelihood of treatment success in social phobia: the increased habitual tendency to experience shame is not only closely associated with social anxiety (e.g., Gilbert, 2000; Matos, Pinto-Gouveia, & Gilbert, 2013) but might also block the kind of openness for new social experiences that is central for improvement in social anxiety disorder. As a further frequent concomitant of social phobia (Stangier et al., 2011), reduced self-esteem is associated with extremely negative assumptions about the self (‘I am worthless’) and may corrupt therapeutic progress. Specifically, those with a negative self-concept will tend to negatively distort the evaluation of behaviour experiments and engage in more post-event rumination (Fehm, Schneider & Hoyer, 2007). Clin. Psychol. Psychother. 23, 35–46 (2016)
Predictors of Outcome in Social Phobia Finally, interpersonal and attachment styles have been discussed as important effect moderators of treatment in social phobia (refer to, e.g., Alden & Taylor, 2004; Davies-Osterkamp, Strauss & Schmitz, 1996), e.g., by further impairing the patient’s ability to relate to others (Alden & Taylor, 2004), especially one’s own therapist. Recent data from McEvoy, Burgess, and Nathan (2014) demonstrated pre-treatment interpersonal problems to increase risk of dropout and predict poorer outcomes at least in cognitive behavioural group therapies. We argue that also in individual therapies, interpersonal problems (e.g., hostility) described to be heightened in social phobia (Stangier, Clark, & Ehlers, 2006; Uhmann, Beesdo, Becker, & Hoyer, 2010) may impede the implementation of certain interventions. In addition, the meta-analysis of Levy, Ellison, Scott, and Bernecker (2011) has shown that certain attachment styles (especially secure and anxious attachment) significantly predict therapy outcome. With regard to the definition of therapeutic outcome, we refer to the three types of indicators that have been typically used to assess treatment change: (a) the scores reached on a dimensional measure of symptom severity after treatment (end-state functioning); (b) the proportion of patients who improved but did not fully recover (response); and (c) the proportion of those who returned to normal, symptom-free, functioning (remission; Jacobson & Truax, 1991) (refer to operational definitions in the Methods section). Given these considerations, we explored the degree to which the above-mentioned variables (i.e., personality, self-esteem, shame, interpersonal problems and attachment style) predict outcome (i.e., end state, response, remission) and attrition/dropout rate after CT, even when known outcome predictors such as symptom severity or comorbidity status are being statistically controlled for. Data from the CT arm of a large multicentre comparative psychotherapy study conducted by the Social Phobia Research Network (SOPHONET; Leichsenring et al., 2009; Leichsenring et al., 2013) was analysed.
METHODS Design and Implementation Patient recruitment occurred at the outpatient clinics of the German universities of Bochum, Dresden, Goettingen, Jena and Mainz. The study protocol was approved by the institutional review boards and monitored by the Coordination Center for Clinical Trials (KKS Heidelberg), which is independent from the participating research centres. This study was conducted in accordance with the guidelines for good clinical practice and approved by the local ethic committees of all involved centres. Copyright © 2014 John Wiley & Sons, Ltd.
37
Subjects According to the main study of the overall trial (Leichsenring et al., 2013), the following inclusion criteria were applied: primary diagnosis of social anxiety disorder (SAD); Liebowitz Social Anxiety Scale (LSAS) >30; and age range of 18–70 years. The following conditions were excluded: psychotic and acute substance-related disorders; clusters A and B personality disorders; prominent risk of self-harm; organic mental disorders; severe medical conditions; concurrent psychotherapeutic or psychopharmacological treatments. Providing informed consent was required for inclusion.
Assessment and Masking Assessments were conducted at baseline, weeks 8 and 15 of treatment, post-treatment, and at follow-up. The Structured Clinical Interview (SCID I, II) for DSM-IV was used for diagnostic classification (German: Wittchen, Zaudig, & Fydrich, 1999). Additionally, using the respective rating item of the Anxiety Disorders Interview Schedule (ADIS-IV; Brown, diNardo, & Barlow, 1994), the primary (i.e., most severe) mental disorder was determined. Twenty-three specifically trained and independent assessors blind to the treatment conditions conducted and videotaped all interviews. Reliability was assessed by comparing the individual results of 23 diagnosticians to an expert’s rating of three videotaped interviews. Wellestablished self-report instruments were applied as secondary outcome measures (refer below).
Intervention and Therapists CT was applied based on the cognitive model of social phobia proposed by Clark and Wells (1995) and on the therapeutic manual derived from the model, which is also available in a German version (Stangier, Ehlers, & Clark, 2006). A total of up to 30 individual 50-min sessions of CT (including diagnostics) were applied; up to six sessions were conducted with a 100-min duration, each counting as two 50-min sessions. Fifty-five CBT therapists collaborated in the trial (37 female, 18 male). The mean age of the therapists was 31.1 years (SD = 4.98). As the majority of the therapists still were in CBT training, the general professional CBT therapist experience was only 1.7 years (SD = 0.9). Before inclusion in the trial, therapists were specifically trained on the manual mentioned above. All treatment sessions were videotaped in order to control for treatment fidelity. Therapists received regular site-level and cross-site supervision. Details of the treatment implementation were described by Leichsenring et al. (2009; 2013; 2014). Clin. Psychol. Psychother. 23, 35–46 (2016)
J. Hoyer et al.
38
Potential Predictors (Independent Variables) (a) Socio-demographic variables. Age, gender and education were examined as potential predictors. The different educational levels were dichotomised into 1 ‘finished high school’ and 0 ‘not finished high school’. (b) Comorbid mental disorders. Mental comorbidity was measured using the number of mental disorders including personality disorders and by the Beck Depression Inventory (BDI; German version: Hautzinger, Bailer, Worall, & Keller, 1995). (c) Personality. Personality dimensions were assessed with the German version of the Tri-dimensional Personality Questionnaire (TPQ; Cloninger et al., 1991; Krebs, Weyers, & Janke, 1998). The internal consistencies of the scales novelty seeking (NS), harm avoidance (HA) and reward dependence (RD) have been reported to be high, and factor analysis of the TPQ supports Cloninger’s personality theory (Krebs et al., 1998). Given their trait-like nature, self-esteem and shame were also entered as variables into the regression equation within this group of predictors: self-esteem (FSSW) was measured with the respective 10-item subscale of the Frankfurter Selbstkonzeptskalen (FSKN, Frankfurt Self Concept Scales; Deusinger, 1986). As a measure of shame, we used the respective 15-item subscale of the German translation of the Test of SelfConscious Affects (TOSCA; Kocherscheidt, Fiedler, Kronmueller, Backenstrass, & Mundt, 2002). The subscale provided excellent internal consistency (Cronbach’s α = 0.90 in a clinical sample; Kocherscheidt et al., 2002). (d) Interpersonal problems and attachment style. The German version of the Inventory of Interpersonal Problems (IIP-D) measures self-reflected difficulties with other people. Sixty-four items are subsumed under eight scales (eight items each): domineering, vindictive, cold, socially avoidant, nonassertive, exploitable, overly nurturant and intrusive (Horowitz, Strauss, & Kordy, 1994; refer also to Salzer et al., 2008; Uhmann et al., 2010). In order to reduce variables as recommended (e.g., Babyak, 2004), only the two main dimensions of the interpersonal circle (Birtchnell, 2014) were analysed in this research. Behaviour incorporated in the horizontal axis (commonly referred to as love in IIP literature) ranges from excessively sacrificing one’s own needs in favour of others to lacking care for others and feeling interpersonally detached. However, the vertical axis represents problems with dominance (i.e., status and control) and extends from being overly controlling to being insufficiently assertive. Copyright © 2014 John Wiley & Sons, Ltd.
Romantic attachment was assessed with the two subscales anxiety and avoidance in the German version of the Experiences in Close Relationships—Revised Questionnaire (ECR-R; Ehrenthal, Dinger, Lamla, Funken, & Schauenburg, 2009).
Operational Definition of Outcome (Dependent Variables) As a measure of end-state functioning, we assessed the LSAS post-treatment, a score ≤30 defining remission (refer to Liebowitz, 1987; Mennin et al., 2002; German: Stangier & Heidenreich, 2005). Response was defined by a 31% reduction (or more) in the LSAS, which is comparable to a Clinical Global Impression Improvement Scale score ≤2 usually used to define response (Bandelow, Baldwin, Dolberg, Andersen, & Stein, 2006; Leichsenring et al., 2013). All patients who stopped treatment or assessment (treatment or study withdrawals) were rated as dropouts and not entered in the abovementioned analyses.
Statistical Analyses All statistical analyses were performed using SPSS Statistics 21 (IBM, Chicago, IL). In order to assess the predictive value of patient characteristics (grouped, as defined above, into socio-demographic variables, mental comorbidity, personality, interpersonal problems and attachment style) with regard to the defined outcome measures, a generalised linear modelling approach was chosen. The continuous criterion measure (LSAS post) was mapped by linear regressions, whereas the dichotomous criterion measures (response, remission and dropout) were modelled by logistic regressions. All patient characteristics were entered as independent variables in a blockwise fashion (thus, hereafter, we call the groups of variables entered in each step of the respective regression analysis blocks). Model 1 serves as a reference model assessing the dependence of end-state social anxiety (LSAS post) on initial symptom severity (LSAS pre). Models 2 to 5 serve to assess the incremental predictive value of the respective independent variable blocks. Please note that shame and self-esteem were integrated into the block personality, given the typically high retest-reliability of these variables (Deusinger, 1986; Kocherscheidt et al., 2002). In spite of the large number of predictors in model 5, we refrained from performing an automated variable selection. The commonly applied stepwise selection procedure (e.g., backward selection) depends heavily on slight differences in semi-partial correlation of variables, performing a large number of tests on a sample that is limited in size. Automated regression procedures were found Clin. Psychol. Psychother. 23, 35–46 (2016)
Predictors of Outcome in Social Phobia
39
to perform poorly under many conditions, including irrelevant and excluding relevant predictors (Babyak, 2004). All p values correspond to two-tailed tests. Since this is an explorative study, no adjustments were made for multiple comparisons (Bender & Lange, 2001). Due to the large number of tests applied in this study, p values need to be interpreted with caution and in connection with effect size estimates.
RESULTS Descriptive Statistics Table 1 displays the sample characteristics in all variables in study completers and those who dropped out from treatment or study and the internal consistency of the scales. Group comparisons using χ 2 tests or t-tests showed no significant differences on any of the variables (p ≥ 0.234). Additionally, the difference between baseline and posttreatment LSAS scores was analysed in the completer sample. Patients reported less symptoms of anxiety at post-treatment (M = 43.27, SD = 26.16) relative to baseline
(M = 72.20, SD = 22.08). This difference was significant (t(182) = 16.28, p < 0.001) and represented a large effect (d = 1.33). In Table 2, results of the bivariate associations between potential predictors and social anxiety (LSAS pre), diverse outcome indicators (end state, response and remission) and the probability to drop out from treatment or study are described. Interestingly, none of the possible predictor variables was significantly associated with dropout. The other associations can be described as follows: socio-demographic variables (except age) were uncorrelated with social anxiety and outcome indicators, while indicators of comorbidity were at least slightly and significantly correlated with social anxiety and outcome. Personality indicators, namely harm avoidance, self-esteem and shame, were consistently associated with social anxiety before and after (end state, remission) treatment but not with response. Novelty seeking and reward dependence were not related to any outcome variable. Of the IIP dimensions, love was unrelated to social anxiety and correlated significantly negatively with the LSAS post and positively with
Table 1. Internal consistencies, means and standard variations or percentages of variables at pre-assessment in completers and treatment or study dropouts Variable
Sample (n = 200–237)
LSAS Socio-demographic data Age Sex Female (%) Male (%) Education ‘Finished high school’ Comorbidity Number of comorbid diagnoses Depression (BDI) Personality Harm avoidance (TPQ) Novelty seeking (TPQ) Reward dependence (TPQ) Self-esteem (FSKN) Shame (TOSCA) Attachment and interpersonal problems Love (IIP) Dominance (IIP) Anxiety (ECR-R) Avoidance (ECR-R)
Completers (n = 165–183)
Dropouts (n = 46–60)
α
M
SD
M
SD
0.92
72.70
22.08
73.09
22.33
34.94
12.11
35.84
11.00
55.2% 44.8%
62.3% 37.7%
69.8%
67.9%
0.87
0.89 13.57
0.88 8.63
0.88 14.08
0.92 10.28
0.80 0.69 0.65 0.91 0.78
24.18 12.55 16.30 38.45 45.63
4.80 4.48 4.15 10.26 9.45
24.34 12.92 16.02 38.20 45.49
5.24 4.73 3.86 10.28 8.47
0.89 0.90 0.92 0.90
0.25 0.95 3.40 2.92
0.63 0.25 1.26 1.05
0.27 0.83 3.33 2.97
0.63 0.71 1.12 0.92
LSAS = Liebowitz Social Anxiety Scale; BDI = Beck Depression Inventory; TPQ = Tridimensional Personality Questionnaire; FSKN = Frankfurter Selbstkonzeptskalen (Frankfurt Self Concept Scales); TOSCA = Test of Self-Conscious Affects; IIP = Inventory of Interpersonal Problems; ECR-R = Experiences in Close Relationships—Revised.
Copyright © 2014 John Wiley & Sons, Ltd.
Clin. Psychol. Psychother. 23, 35–46 (2016)
J. Hoyer et al.
40
Table 2. Correlations between baseline patient characteristics, pre-treatment social anxiety and different outcomes (end state, response, remission and dropout) LSAS pre (n = 165–183)
Variable Socio-demographic data Age Sex Education Comorbidity Number of comorbid diagnoses Depression (BDI) Personality Harm avoidance (TPQ) Novelty seeking (TPQ) Reward dependence (TPQ) Self-esteem (FSKN) Shame (TOSCA) Attachment and interpersonal problems Love (IIP) Dominance (IIP) Anxiety (ECR-R) Avoidance (ECR-R)
End state (n = 165–183)
Response (n = 165–183)
Remission (n = 165–183)
Dropout (n = 212–244)
0.20** 0.00 0.03
0.08 0.07 0.01
0.04 0.09 0.01
0.07 0.08 0.04
0.03 0.05 0.00
0.31** 0.25**
0.37** 0.01
0.21** 0.06
0.21** 0.06
0.00 0.02
0.29** 0.06 0.01 0.36** 0.32**
0.22** 0.07 0.06 0.21** 0.15*
0.14 0.06 0.06 0.05 0.00
0.25** 0.06 0.01 0.18* 0.22**
0.01 0.04 0.03 0.01 0.00
0.10 0.31** 0.20** 0.22**
0.17* 0.16* 0.12 0.09
0.13 0.02 0.03 0.00
0.16* 0.07 0.09 0.09
0.01 0.08 0.03 0.02
LSAS pre = Liebowitz Social Anxiety Scale score, assessed at baseline; BDI = Beck Depression Inventory; TPQ = Tridimensional Personality Questionnaire; FSKN = Frankfurter Selbstkonzeptskalen (Frankfurt Self Concept Scales); TOSCA = Test of Self-Conscious Affects; IIP = Inventory of Interpersonal Problems; ECR-R = Experiences in Close Relationships–Revised. *p < 0.05. **p < 0.01.
Table 3. Patient characteristics predicting end-state functioning as measured by the LSAS post† in n = 156 patients treated with CBT for social phobia Model 1 2 3 4
5
Predictor variable‡ LSAS pre Age Sex Education Number of comorbid diagnoses Depression (BDI) Harm avoidance (TPQ) Novelty seeking (TPQ) Reward dependence (TPQ) Self-esteem (FSKN) Shame (TOSCA) Love (IIP) Dominance (IIP) Anxiety (ECR-R) Avoidance (ECR-R)
β§ 0.511 0.004 0.090 0.014 0.282 0.232 0.044 0.103 0.133 0.247 0.167 0.133 0.086 0.074 0.082
p(β)§
R2
Adj. R2
R2-change
F-change
p(F-change)
Baseline Patient Characteristics Predicting Outcome and Attrition in Cognitive Therapy for Social Phobia: Results from a Large Multicentre Trial Juergen Hoyer,1* Joerg Wiltink,2 Wolfgang Hiller,3 Robert Miller,1 Simone Salzer,4 Stephan Sarnowsky,1 Ulrich Stangier,5 Bernhard Strauss,6 Ulrike Willutzki7 and Eric Leibing4 1
Clinical Psychology and Psychotherapy, Technische Universitaet Dresden, Dresden, Germany Department of Psychosomatic Medicine and Psychotherapy, University Medical Center, Johannes Gutenberg University Mainz, Mainz, Germany 3 Clinical Psychology and Psychotherapy, Johannes Gutenberg University Mainz, Mainz, Germany 4 Department of Psychosomatic Medicine and Psychotherapy, University Medicine, Georg-August University Goettingen, Goettingen, Germany 5 Clinical Psychology and Psychotherapy, Goethe University Frankfurt, Frankfurt, Germany 6 Institute of Psychosocial Medicine and Psychotherapy, Jena University Hospital, Jena, Germany 7 Clinical Psychology and Psychotherapy, Ruhr University Bochum, Bochum, Germany 2
We examined the role of baseline patient characteristics as predictors of outcome (end-state functioning, response and remission) and attrition for cognitive therapy (CT) in social anxiety disorder (SAD). Beyond socio-demographic and clinical variables such as symptom severity and comorbidity status, previously neglected patient characteristics (e.g., personality, self-esteem, shame, interpersonal problems and attachment style) were analysed. Method: Data came from the CT arm of a multicentre RCT with n = 244 patients having DSM-IV SAD. CT was conducted according to the manual by Clark and Wells. Severity of SAD was assessed at baseline and end of treatment with the Liebowitz Social Anxiety Scale (LSAS). Multiple linear regression analyses and logistic regression analyses were applied. Results: Up to 37% of the post-treatment variance (LSAS) could be explained by all pre-treatment variables combined. Symptom severity (baseline LSAS) was consistently negatively associated with end-state functioning and remission, but not with response. Number of comorbid diagnoses was negatively associated with end-state functioning and response, but not with remission. Self-esteem was positively associated with higher end-state functioning and more shame with better response. Attrition could not be significantly predicted. Conclusions: The results indicate that the initial probability for treatment success mainly depends on severity of disorder and comorbid conditions while other psychological variables are of minor importance, at least on a nomothetic level. This stands in contrast with efforts to arrive at an empirical-based foundation for differential indication and argues to search for more potent moderators of therapeutic change rather on the process level. Key Practitioner Message: • Personality, self-esteem, shame, attachment style and interpersonal problems do not or only marginally moderate the effects of interventions in CT of social phobia. • Symptom severity and comorbid diagnoses might affect treatment outcome negatively. • Beyond these two factors, most patients share a similar likelihood of treatment success when treated according to the manual by Clark and Wells. Copyright © 2014 John Wiley & Sons, Ltd. Keywords: Social Phobia, CBT, Treatment Outcome, Predictors, Comorbidity, Interpersonal Problems *Correspondence to: Prof. Dr. Jürgen Hoyer, Technische Universität Dresden, Klinische Psychologie und Psychotherapie, Hohe Str. 53, D-01187, Dresden, Germany. E-mail: [email protected]
Copyright © 2014 John Wiley & Sons, Ltd.
Cognitive behavioural therapy (CBT) has been shown to be highly effective in treating social phobia (Powers, Sigmarsson, & Emmelkamp, 2008), one of the most
36 frequent (Ruscio et al., 2008) and debilitating (Fehm, Pelissolo, Furmark, & Wittchen, 2005) anxiety disorders. Recent trials prove CBT or cognitive therapy (CT; Clark & Wells, 1995; which not only emphasises cognitive change, but also incorporates typical CBT interventions) to be at least as successful (Blanco et al., 2010; Davidson et al., 2004) or even more successful than pharmacological therapy (Clark et al., 2003). CT was also superior to interpersonal therapy (Stangier, Schramm, Heidenreich, Berger, & Clark, 2011), applied relaxation (Clark et al., 2006) and manualised short-term psychodynamic psychotherapy (Leichsenring et al., 2013). At the same time, the total number of patients who benefit from treatment in the largest multicentre trials of cognitive behavioural treatments is not higher than 60% (Leichsenring et al., 2013) or 51.7% when CBT is applied in groups (Davidson et al., 2004). Even the trials with the most favourable success rates (Clark et al., 2003; Clark et al., 2006) leave a considerable number of patients unchanged. In order to raise the rate of those benefiting from CBT/CT of social phobia, it is of cardinal importance to know more about the factors that moderate treatment effects (Durham, 2006; Hoefler, Gloster & Hoyer, 2010). A considerable number of studies on social phobia have addressed the issue of prediction of treatment attrition and success, but a clear and convincing picture has yet to be formed. So far, only a selection of the possible predictors for treatment outcome has been investigated in independent studies. Out of these, only very few have been consistently replicated. The studies differ with respect to which variables were investigated. In the majority of studies, only a limited range of predictors was considered (refer to Lincoln et al., 2005 and Mululo, De Menezes, Vigne, & Fontenelle, 2012 for overviews). Socio-demographic variables such as age, gender, marital status, education and occupation are typically unrelated to outcome. Severity of disorder, generalised subtype, comorbid axis I disorder (e.g., depression) and comorbid axis II disorder (e.g., avoidant personality disorder) are associated with worse response in most studies, although their results require replication (Lincoln et al., 2005; Mululo et al., 2012). Positive treatment expectations and homework compliance have been related to change and high end state but were investigated only in a few studies (Lincoln et al., 2005; Mululo et al., 2012); however, across different disorders (e.g., depression), those variables were found to impact treatment success in a similar manner (e.g., Tsai, Ogrodniczuk, Sichting, & Mirmiran, 2014). Overall, the practical significance, i.e., the effect size of the above-mentioned predictors of treatment outcome in social phobia, has been evaluated as low and ‘unsatisfying’ (Lincoln et al., 2005). Given the complexity of the therapeutic process and the number of potential effect modifiers involved, this finding may simply Copyright © 2014 John Wiley & Sons, Ltd.
J. Hoyer et al. indicate that other variables, such as therapist variables or treatment integrity, are more important. Alternatively, the low explanatory power of patient variables may also be due to methodological restrictions of previous studies. Many studies are seriously underpowered and risk overfitting (Babyak, 2004). Ratios of predictors to sample sizes of less than 1:20 are not recommendable, and as the total sample sizes often falls below the n = 100 margin, only few predictors can be analysed (Durham, 2006). Due to these methodological problems, several patient psychological characteristics have either rarely or never been studied with respect to their explanatory power for CBT/CT treatment outcome in social phobia, although a significant impact on treatment outcome and attrition can theoretically be expected. This is namely true for variables including personality, self-esteem, shame or interpersonal problems and attachment style, all of which might impact on the patient’s willingness to experiment with new behaviours both inside and outside the therapeutic setting. In the following, we will briefly describe why the above-mentioned variables can be expected to exhibit direct or indirect influences on the success of treatment. The role of comorbid personality disorders for treatment success in social phobia has been examined early (e.g., Van Velzen, Emmelkamp, & Scholing, 1997). To the best of our knowledge, however, the impact of personality traits has yet to be studied consistently. In this research, we will specifically focus on the personality dimensions of harm avoidance, novelty seeking and reward dependence (Cloninger, Przybeck, & Svrakic, 1991), all of which have been proven useful in describing socially anxious populations (e.g., Kashdan & Hofmann, 2008). We argue that especially persons with high tendency towards harm avoidance might encounter difficulties in experimenting with new behaviour, which would, in turn, require more effort to achieve a disconfirmation of beliefs, thus negatively affecting treatment outcome. Further, shame and low self-esteem may also reduce the likelihood of treatment success in social phobia: the increased habitual tendency to experience shame is not only closely associated with social anxiety (e.g., Gilbert, 2000; Matos, Pinto-Gouveia, & Gilbert, 2013) but might also block the kind of openness for new social experiences that is central for improvement in social anxiety disorder. As a further frequent concomitant of social phobia (Stangier et al., 2011), reduced self-esteem is associated with extremely negative assumptions about the self (‘I am worthless’) and may corrupt therapeutic progress. Specifically, those with a negative self-concept will tend to negatively distort the evaluation of behaviour experiments and engage in more post-event rumination (Fehm, Schneider & Hoyer, 2007). Clin. Psychol. Psychother. 23, 35–46 (2016)
Predictors of Outcome in Social Phobia Finally, interpersonal and attachment styles have been discussed as important effect moderators of treatment in social phobia (refer to, e.g., Alden & Taylor, 2004; Davies-Osterkamp, Strauss & Schmitz, 1996), e.g., by further impairing the patient’s ability to relate to others (Alden & Taylor, 2004), especially one’s own therapist. Recent data from McEvoy, Burgess, and Nathan (2014) demonstrated pre-treatment interpersonal problems to increase risk of dropout and predict poorer outcomes at least in cognitive behavioural group therapies. We argue that also in individual therapies, interpersonal problems (e.g., hostility) described to be heightened in social phobia (Stangier, Clark, & Ehlers, 2006; Uhmann, Beesdo, Becker, & Hoyer, 2010) may impede the implementation of certain interventions. In addition, the meta-analysis of Levy, Ellison, Scott, and Bernecker (2011) has shown that certain attachment styles (especially secure and anxious attachment) significantly predict therapy outcome. With regard to the definition of therapeutic outcome, we refer to the three types of indicators that have been typically used to assess treatment change: (a) the scores reached on a dimensional measure of symptom severity after treatment (end-state functioning); (b) the proportion of patients who improved but did not fully recover (response); and (c) the proportion of those who returned to normal, symptom-free, functioning (remission; Jacobson & Truax, 1991) (refer to operational definitions in the Methods section). Given these considerations, we explored the degree to which the above-mentioned variables (i.e., personality, self-esteem, shame, interpersonal problems and attachment style) predict outcome (i.e., end state, response, remission) and attrition/dropout rate after CT, even when known outcome predictors such as symptom severity or comorbidity status are being statistically controlled for. Data from the CT arm of a large multicentre comparative psychotherapy study conducted by the Social Phobia Research Network (SOPHONET; Leichsenring et al., 2009; Leichsenring et al., 2013) was analysed.
METHODS Design and Implementation Patient recruitment occurred at the outpatient clinics of the German universities of Bochum, Dresden, Goettingen, Jena and Mainz. The study protocol was approved by the institutional review boards and monitored by the Coordination Center for Clinical Trials (KKS Heidelberg), which is independent from the participating research centres. This study was conducted in accordance with the guidelines for good clinical practice and approved by the local ethic committees of all involved centres. Copyright © 2014 John Wiley & Sons, Ltd.
37
Subjects According to the main study of the overall trial (Leichsenring et al., 2013), the following inclusion criteria were applied: primary diagnosis of social anxiety disorder (SAD); Liebowitz Social Anxiety Scale (LSAS) >30; and age range of 18–70 years. The following conditions were excluded: psychotic and acute substance-related disorders; clusters A and B personality disorders; prominent risk of self-harm; organic mental disorders; severe medical conditions; concurrent psychotherapeutic or psychopharmacological treatments. Providing informed consent was required for inclusion.
Assessment and Masking Assessments were conducted at baseline, weeks 8 and 15 of treatment, post-treatment, and at follow-up. The Structured Clinical Interview (SCID I, II) for DSM-IV was used for diagnostic classification (German: Wittchen, Zaudig, & Fydrich, 1999). Additionally, using the respective rating item of the Anxiety Disorders Interview Schedule (ADIS-IV; Brown, diNardo, & Barlow, 1994), the primary (i.e., most severe) mental disorder was determined. Twenty-three specifically trained and independent assessors blind to the treatment conditions conducted and videotaped all interviews. Reliability was assessed by comparing the individual results of 23 diagnosticians to an expert’s rating of three videotaped interviews. Wellestablished self-report instruments were applied as secondary outcome measures (refer below).
Intervention and Therapists CT was applied based on the cognitive model of social phobia proposed by Clark and Wells (1995) and on the therapeutic manual derived from the model, which is also available in a German version (Stangier, Ehlers, & Clark, 2006). A total of up to 30 individual 50-min sessions of CT (including diagnostics) were applied; up to six sessions were conducted with a 100-min duration, each counting as two 50-min sessions. Fifty-five CBT therapists collaborated in the trial (37 female, 18 male). The mean age of the therapists was 31.1 years (SD = 4.98). As the majority of the therapists still were in CBT training, the general professional CBT therapist experience was only 1.7 years (SD = 0.9). Before inclusion in the trial, therapists were specifically trained on the manual mentioned above. All treatment sessions were videotaped in order to control for treatment fidelity. Therapists received regular site-level and cross-site supervision. Details of the treatment implementation were described by Leichsenring et al. (2009; 2013; 2014). Clin. Psychol. Psychother. 23, 35–46 (2016)
J. Hoyer et al.
38
Potential Predictors (Independent Variables) (a) Socio-demographic variables. Age, gender and education were examined as potential predictors. The different educational levels were dichotomised into 1 ‘finished high school’ and 0 ‘not finished high school’. (b) Comorbid mental disorders. Mental comorbidity was measured using the number of mental disorders including personality disorders and by the Beck Depression Inventory (BDI; German version: Hautzinger, Bailer, Worall, & Keller, 1995). (c) Personality. Personality dimensions were assessed with the German version of the Tri-dimensional Personality Questionnaire (TPQ; Cloninger et al., 1991; Krebs, Weyers, & Janke, 1998). The internal consistencies of the scales novelty seeking (NS), harm avoidance (HA) and reward dependence (RD) have been reported to be high, and factor analysis of the TPQ supports Cloninger’s personality theory (Krebs et al., 1998). Given their trait-like nature, self-esteem and shame were also entered as variables into the regression equation within this group of predictors: self-esteem (FSSW) was measured with the respective 10-item subscale of the Frankfurter Selbstkonzeptskalen (FSKN, Frankfurt Self Concept Scales; Deusinger, 1986). As a measure of shame, we used the respective 15-item subscale of the German translation of the Test of SelfConscious Affects (TOSCA; Kocherscheidt, Fiedler, Kronmueller, Backenstrass, & Mundt, 2002). The subscale provided excellent internal consistency (Cronbach’s α = 0.90 in a clinical sample; Kocherscheidt et al., 2002). (d) Interpersonal problems and attachment style. The German version of the Inventory of Interpersonal Problems (IIP-D) measures self-reflected difficulties with other people. Sixty-four items are subsumed under eight scales (eight items each): domineering, vindictive, cold, socially avoidant, nonassertive, exploitable, overly nurturant and intrusive (Horowitz, Strauss, & Kordy, 1994; refer also to Salzer et al., 2008; Uhmann et al., 2010). In order to reduce variables as recommended (e.g., Babyak, 2004), only the two main dimensions of the interpersonal circle (Birtchnell, 2014) were analysed in this research. Behaviour incorporated in the horizontal axis (commonly referred to as love in IIP literature) ranges from excessively sacrificing one’s own needs in favour of others to lacking care for others and feeling interpersonally detached. However, the vertical axis represents problems with dominance (i.e., status and control) and extends from being overly controlling to being insufficiently assertive. Copyright © 2014 John Wiley & Sons, Ltd.
Romantic attachment was assessed with the two subscales anxiety and avoidance in the German version of the Experiences in Close Relationships—Revised Questionnaire (ECR-R; Ehrenthal, Dinger, Lamla, Funken, & Schauenburg, 2009).
Operational Definition of Outcome (Dependent Variables) As a measure of end-state functioning, we assessed the LSAS post-treatment, a score ≤30 defining remission (refer to Liebowitz, 1987; Mennin et al., 2002; German: Stangier & Heidenreich, 2005). Response was defined by a 31% reduction (or more) in the LSAS, which is comparable to a Clinical Global Impression Improvement Scale score ≤2 usually used to define response (Bandelow, Baldwin, Dolberg, Andersen, & Stein, 2006; Leichsenring et al., 2013). All patients who stopped treatment or assessment (treatment or study withdrawals) were rated as dropouts and not entered in the abovementioned analyses.
Statistical Analyses All statistical analyses were performed using SPSS Statistics 21 (IBM, Chicago, IL). In order to assess the predictive value of patient characteristics (grouped, as defined above, into socio-demographic variables, mental comorbidity, personality, interpersonal problems and attachment style) with regard to the defined outcome measures, a generalised linear modelling approach was chosen. The continuous criterion measure (LSAS post) was mapped by linear regressions, whereas the dichotomous criterion measures (response, remission and dropout) were modelled by logistic regressions. All patient characteristics were entered as independent variables in a blockwise fashion (thus, hereafter, we call the groups of variables entered in each step of the respective regression analysis blocks). Model 1 serves as a reference model assessing the dependence of end-state social anxiety (LSAS post) on initial symptom severity (LSAS pre). Models 2 to 5 serve to assess the incremental predictive value of the respective independent variable blocks. Please note that shame and self-esteem were integrated into the block personality, given the typically high retest-reliability of these variables (Deusinger, 1986; Kocherscheidt et al., 2002). In spite of the large number of predictors in model 5, we refrained from performing an automated variable selection. The commonly applied stepwise selection procedure (e.g., backward selection) depends heavily on slight differences in semi-partial correlation of variables, performing a large number of tests on a sample that is limited in size. Automated regression procedures were found Clin. Psychol. Psychother. 23, 35–46 (2016)
Predictors of Outcome in Social Phobia
39
to perform poorly under many conditions, including irrelevant and excluding relevant predictors (Babyak, 2004). All p values correspond to two-tailed tests. Since this is an explorative study, no adjustments were made for multiple comparisons (Bender & Lange, 2001). Due to the large number of tests applied in this study, p values need to be interpreted with caution and in connection with effect size estimates.
RESULTS Descriptive Statistics Table 1 displays the sample characteristics in all variables in study completers and those who dropped out from treatment or study and the internal consistency of the scales. Group comparisons using χ 2 tests or t-tests showed no significant differences on any of the variables (p ≥ 0.234). Additionally, the difference between baseline and posttreatment LSAS scores was analysed in the completer sample. Patients reported less symptoms of anxiety at post-treatment (M = 43.27, SD = 26.16) relative to baseline
(M = 72.20, SD = 22.08). This difference was significant (t(182) = 16.28, p < 0.001) and represented a large effect (d = 1.33). In Table 2, results of the bivariate associations between potential predictors and social anxiety (LSAS pre), diverse outcome indicators (end state, response and remission) and the probability to drop out from treatment or study are described. Interestingly, none of the possible predictor variables was significantly associated with dropout. The other associations can be described as follows: socio-demographic variables (except age) were uncorrelated with social anxiety and outcome indicators, while indicators of comorbidity were at least slightly and significantly correlated with social anxiety and outcome. Personality indicators, namely harm avoidance, self-esteem and shame, were consistently associated with social anxiety before and after (end state, remission) treatment but not with response. Novelty seeking and reward dependence were not related to any outcome variable. Of the IIP dimensions, love was unrelated to social anxiety and correlated significantly negatively with the LSAS post and positively with
Table 1. Internal consistencies, means and standard variations or percentages of variables at pre-assessment in completers and treatment or study dropouts Variable
Sample (n = 200–237)
LSAS Socio-demographic data Age Sex Female (%) Male (%) Education ‘Finished high school’ Comorbidity Number of comorbid diagnoses Depression (BDI) Personality Harm avoidance (TPQ) Novelty seeking (TPQ) Reward dependence (TPQ) Self-esteem (FSKN) Shame (TOSCA) Attachment and interpersonal problems Love (IIP) Dominance (IIP) Anxiety (ECR-R) Avoidance (ECR-R)
Completers (n = 165–183)
Dropouts (n = 46–60)
α
M
SD
M
SD
0.92
72.70
22.08
73.09
22.33
34.94
12.11
35.84
11.00
55.2% 44.8%
62.3% 37.7%
69.8%
67.9%
0.87
0.89 13.57
0.88 8.63
0.88 14.08
0.92 10.28
0.80 0.69 0.65 0.91 0.78
24.18 12.55 16.30 38.45 45.63
4.80 4.48 4.15 10.26 9.45
24.34 12.92 16.02 38.20 45.49
5.24 4.73 3.86 10.28 8.47
0.89 0.90 0.92 0.90
0.25 0.95 3.40 2.92
0.63 0.25 1.26 1.05
0.27 0.83 3.33 2.97
0.63 0.71 1.12 0.92
LSAS = Liebowitz Social Anxiety Scale; BDI = Beck Depression Inventory; TPQ = Tridimensional Personality Questionnaire; FSKN = Frankfurter Selbstkonzeptskalen (Frankfurt Self Concept Scales); TOSCA = Test of Self-Conscious Affects; IIP = Inventory of Interpersonal Problems; ECR-R = Experiences in Close Relationships—Revised.
Copyright © 2014 John Wiley & Sons, Ltd.
Clin. Psychol. Psychother. 23, 35–46 (2016)
J. Hoyer et al.
40
Table 2. Correlations between baseline patient characteristics, pre-treatment social anxiety and different outcomes (end state, response, remission and dropout) LSAS pre (n = 165–183)
Variable Socio-demographic data Age Sex Education Comorbidity Number of comorbid diagnoses Depression (BDI) Personality Harm avoidance (TPQ) Novelty seeking (TPQ) Reward dependence (TPQ) Self-esteem (FSKN) Shame (TOSCA) Attachment and interpersonal problems Love (IIP) Dominance (IIP) Anxiety (ECR-R) Avoidance (ECR-R)
End state (n = 165–183)
Response (n = 165–183)
Remission (n = 165–183)
Dropout (n = 212–244)
0.20** 0.00 0.03
0.08 0.07 0.01
0.04 0.09 0.01
0.07 0.08 0.04
0.03 0.05 0.00
0.31** 0.25**
0.37** 0.01
0.21** 0.06
0.21** 0.06
0.00 0.02
0.29** 0.06 0.01 0.36** 0.32**
0.22** 0.07 0.06 0.21** 0.15*
0.14 0.06 0.06 0.05 0.00
0.25** 0.06 0.01 0.18* 0.22**
0.01 0.04 0.03 0.01 0.00
0.10 0.31** 0.20** 0.22**
0.17* 0.16* 0.12 0.09
0.13 0.02 0.03 0.00
0.16* 0.07 0.09 0.09
0.01 0.08 0.03 0.02
LSAS pre = Liebowitz Social Anxiety Scale score, assessed at baseline; BDI = Beck Depression Inventory; TPQ = Tridimensional Personality Questionnaire; FSKN = Frankfurter Selbstkonzeptskalen (Frankfurt Self Concept Scales); TOSCA = Test of Self-Conscious Affects; IIP = Inventory of Interpersonal Problems; ECR-R = Experiences in Close Relationships–Revised. *p < 0.05. **p < 0.01.
Table 3. Patient characteristics predicting end-state functioning as measured by the LSAS post† in n = 156 patients treated with CBT for social phobia Model 1 2 3 4
5
Predictor variable‡ LSAS pre Age Sex Education Number of comorbid diagnoses Depression (BDI) Harm avoidance (TPQ) Novelty seeking (TPQ) Reward dependence (TPQ) Self-esteem (FSKN) Shame (TOSCA) Love (IIP) Dominance (IIP) Anxiety (ECR-R) Avoidance (ECR-R)
β§ 0.511 0.004 0.090 0.014 0.282 0.232 0.044 0.103 0.133 0.247 0.167 0.133 0.086 0.074 0.082
p(β)§
R2
Adj. R2
R2-change
F-change
p(F-change)
