Risk Profiles for Poor Treatment Response to Internet-delivered CBT in People with Social Anxiety Disorder

Page 1

Rp

Ma

b

c

d

e

a

ARRAA

KSSDGRC

1

iawFem&wt(aDpWLe

S

h0

Journal of Anxiety Disorders 33 (2015) 103–109

Contents lists available at ScienceDirect

Journal of Anxiety Disorders

isk profiles for poor treatment response to internet-delivered CBT ineople with social anxiety disorder

aria Tillforsa,∗, Tomas Furmarkb, Per Carlbringc, Gerhard Anderssond,e

Center for Health and Medical Psychology, JPS, Psychology, Örebro University, SwedenDepartment of Psychology, Uppsala University, SwedenDepartment of Psychology, Stockholm University, SwedenDepartment of Behavioural Sciences and Learning, Linköping University, SwedenDepartment of Clinical Neuroscience, Psychiatry Section, Karolinska Institutet, Stockholm, Sweden

r t i c l e i n f o

rticle history:eceived 18 February 2015eceived in revised form 20 May 2015ccepted 25 May 2015vailable online 1 June 2015

a b s t r a c t

In social anxiety disorder (SAD) co-morbid depressive symptoms as well as avoidance behaviors havebeen shown to predict insufficient treatment response. It is likely that subgroups of individuals withdifferent profiles of risk factors for poor treatment response exist. This study aimed to identify subgroupsof social avoidance and depressive symptoms in a clinical sample (N = 167) with SAD before and afterguided internet-delivered CBT, and to compare these groups on diagnostic status and social anxiety.We further examined individual movement between subgroups over time. Using cluster analysis we

eywords:ocial anxiety disorderocial avoidanceepressive symptomsuided internet-based CBTisk factors

identified four subgroups, including a high-problem cluster at both time-points. Individuals in this clustershowed less remission after treatment, exhibited higher levels of social anxiety at both assessments, andtypically remained in the high-problem cluster after treatment. Thus, in individuals with SAD, high levelsof social avoidance and depressive symptoms constitute a risk profile for poor treatment response.

© 2015 Elsevier Ltd. All rights reserved.

luster analysis

. Introduction

Randomized controlled trials (RCTs) show that cognitive behav-oral therapy (CBT) in various formats (individual, group, as wells guided internet-delivered self-help) is effective for peopleith social anxiety disorder, or SAD (e.g., Andersson, Carlbring, &

urmark, 2014; Clark et al., 2006; Heimberg, 2002; Mayo-Wilsont al., 2014). However, even with the best psychological treatments,ore than one in four do not improve sufficiently (e.g., Ponniah

Hollon, 2008), and this heterogeneity in treatment response isorthy of further investigation. It is a rule rather than an excep-

ion that people have several mental health and somatic problemsHarvey, Watkins, Mansell, & Shafran, 2004), and co-morbidity isn important factor to consider in relation to treatment response.epression and other anxiety disorders are common co-morbidroblems in people with SAD (Kessler, Chiu, Demler, Merikangas, &

alters, 2005; Rapee & Spence, 2004; Schneider, Johnson, Horning,

iebowitz, & Weissman, 1992). Moreover, use of dysfunctionalmotion regulation strategies like avoidance behaviors (both on

∗ Corresponding author at: JPS, Psychology, Örebro University, 701 82 Örebro,weden. Tel.: +46 19 30 39 59.

E-mail address: [email protected] (M. Tillfors).

ttp://dx.doi.org/10.1016/j.janxdis.2015.05.007887-6185/© 2015 Elsevier Ltd. All rights reserved.

an overt and a covert level) are common in SAD. Such behaviorsare positively related to clinical severity, and have been shown tomaintain the disorder (Harvey et al., 2004). Importantly, it is likelythe inflexible use of dysfunctional emotion regulation strategies,like avoidance behaviors, to manage intense anxiety in a range ofdifferent social situations that maintains SAD rather than the levelof anxiety per se (Harvey et al., 2004). Hence, both co-morbidityand avoidance behavior could underlie heterogeneity in treatmentresponse in individuals with SAD.

Indeed, co-morbid depressive symptoms as well as high lev-els of avoidance behavior have previously been shown to predictsuboptimal treatment response in people with SAD (e.g., Eskildsen,Hougaard, & Rosenberg, 2010; Hedman et al., 2012; Nordgreenet al., 2012; Rodebaugh, Holaway, & Heimberg, 2004), although theresults regarding depressive symptoms have been mixed (Eskildsenet al., 2010; Nordgreen et al., 2012; Rodebaugh et al., 2004). It canfurther be hypothesized that a combination of risk factors mayparticularly increase the risk for poor treatment outcome, possi-bly explaining the mixed findings regarding depressive symptomsas a treatment predictor. In other words, there may be subgroups

of individuals with different profiles of risk factors for poor treat-ment response. However, when examining such risk profiles wecannot rely only on variable-oriented methods such as regressionbased approaches, which are commonly used in analyses of RCTs.

Page 2

1 xiety

Fnpbwfiptis

sbgsifsit

2

2

ST22Tsw

2

dtn(airtt&ttaig(cslccoiwiyI

04 M. Tillfors et al. / Journal of An

irst, the relationship between the predictors and outcome mightot be linear. Second, there might be subgroups of people withrofiles consisting of a combination of risk factors which coulde hidden in variable-oriented methods (useful in understandinghat characteristics co-aggregate in a group of individuals) and

or which person-oriented methods (useful in finding subgroups ofndividuals) like cluster analysis are needed. Thus, the overall pur-ose of the current study was to use person-oriented, in additiono variable-oriented, methods to examine if treatment outcomen SAD is related to patterns of social avoidance and depressiveymptoms.

Using cluster analysis we sought, first, to identify subgroups ofocial avoidance and depressive symptoms in a clinical SAD sampleefore and after ICBT and, second, to compare the derived sub-roups on diagnostic status after treatment and on social anxietyymptom severity before and after ICBT. A final aim was to examinendividual stability and movement between subgroups (clusters)rom pre- to post-treatment. We hypothesized that a cluster of highocial avoidance and depressive symptoms would be possible todentify and that this would be particularly characterized by poorreatment outcome.

. Materials and methods

.1. Design

In the current study, a clinical sample of 167 people withAD who had received guided ICBT for SAD was examined.hese data were pooled from five earlier RCTs (Andersson et al.,006; Carlbring et al., 2007; Tillfors et al., 2008; Furmark et al.,009, where the latter study included reports from two RCTs).he design was prospective and data for the purpose of thistudy were obtained at pre- and at post-treatment after nineeeks.

.2. Participants and procedure

In general, the procedure was the same in all studies and will beescribed briefly below. Participants were recruited through adver-isements and/or via a research web page (http://www.studie.u). Applicants to the study had to fill out online questionnairesSocial Phobia Screening Questionnaire; SPSQ; Furmark et al., 1999,nd the self-rated version of Montgomery Åsberg Depression Rat-ng Scale; MADRS-SR; Svanborg & Åsberg, 1994), and questionsegarding past and ongoing treatments including medication. Ifhey passed this first step, they were interviewed with the Struc-ured Clinical Interview for DSM-IV (SCID-I; First, Gibbon, Spitzer,

Williams, 1997; including the SAD-module). The inclusion cri-eria were the following: (a) a DSM-IV diagnosis of SAD accordingo the SPSQ; (b) scoring <31 on the MADRS-SR depression scale,nd <4 on the suicide item of this scale to prevent the inclusion ofndividuals in strong need of specialist consultation; (c) not under-oing any other psychological treatment during the study period;d) if on prescribed drugs for anxiety/depression, dosage had to beonstant for 3 months before the treatment onset and kept con-tant throughout the study; (e) being at least 18 years old; (f)iving in Sweden; (g) having access to a computer with Internetonnection; (h) not reporting another serious disorder (e.g. psy-hosis, substance abuse) that could be expected to influence theutcome of the study; and (i) a primary diagnosis of SAD accord-ng to SCID-I. Those who did not meet the criteria were advised

here they could turn for help elsewhere. The clinical sample usedn the current study comprised 167 people (69% women; Mage = 34ears, SD = 9.22) diagnosed with SAD that all had received guidedCBT.

Disorders 33 (2015) 103–109

2.3. Measures

2.3.1. Liebowitz social anxiety scale, the self-report version(LSAS-SR)

LSAS is an instrument for the assessment of social fear/anxiety(LSAS-F) and social avoidance (LSAS-A; Baker, Heinrich, Kim,& Hofmann, 2002; Liebowitz, 1987) in 24 potentially anxiety-provoking social situations (13 performance and 11 interactionalsituations). Each social situation was rated on a four-point scalewith the response options ranging from: No fear or anxiety (0) toStrong fear or anxiety (3) on the subscale of social fear/anxiety.The total score of this subscale ranges from 0 to 72 where higherresponses indicate higher social fear/anxiety. For the subscale ofsocial avoidance, the response options range from: Never (0% of thetime) to Usually (67–100% of the time), yielding a total score from 0to 72, higher scores representing higher social avoidance. LSAS hasgood psychometric properties (Fresco et al., 2001).

2.3.2. Social phobia screening questionnaire (SPSQ)The first section of the SPSQ includes14 questions about distress

in different social situations like “Speaking or performing in frontof a group,” Expressing your own opinions in front of others,” “Call-ing someone you do not know very well” (Furmark et al., 1999).Each situation is rated on a five-point scale ranging from: Not atall distressing (0) to Extremely distressing (4). The total score of thissection ranges from 0 to 56, higher scores indicating higher socialanxiety.

The second part of the SPSQ contains diagnostic questions cov-ering the A criteria of SAD according to the DSM-IV-TR (APA, 2000),e.g. “In the following situation(s) I fear that others will notice thatI’m nervous”. Following each of these questions, the 14 potentiallyphobic situations from section one were listed and the respon-dent could indicate each situation that produced anxiety or choose“none of the situations”. The E-criterion was assessed with threeyes/no questions, i.e. the person was asked whether the social fearsseverely interfered with or severely bothered him/her in (a) occu-pational or academic activities, (b) leisure time activities, or (c)social activities. In sum, people were classified as having a diag-nosis of SAD if they rated at least one potentially phobic situation as3 or 4 (Very or extremely distressing) on the social anxiety scale in thefirst section. This situation should be endorsed regarding the A-, B-,C-, and D-criteria in the second section, and lastly the participantsshould have answered ‘yes’ to at least one of the items assessing theE-criterion. Otherwise they were classified as not having diagnosisof SAD.

In a validation study using a diagnostic interview as a reference,the sensitivity of the SPSQ was found to be 100% and the speci-ficity 95% (Furmark et al., 1999). Lastly, this 14-item distress scalein the first section has been found to correlate highly with two well-established social phobia scales, the social phobia scale, r = 77, andthe social interaction anxiety scale, r = .79 (Furmark et al., 1999;Heimberg, Mueller, Holt, Hope, & Liebowitz, 1992) which indicatesan adequate concurrent validity for this scale.

2.3.3. Montgomery Åsberg depression rating scale, the self-ratedversion (MADRS-SR)

MADRS-SR is a questionnaire that measures the degree ofdepressive symptoms (Svanborg & Åsberg, 1994). It consists ofnine items that correspond to core symptoms of depression. Thetotal score ranges from 0 to 54 with higher responses indicatinghigher depressive symptoms. The instrument has good reliability

and validity. Cronbach’s alpha in earlier studies has ranged from .82to .90 (Svanborg & Åsberg, 2001). The outcome measures used havebeen shown to have good psychometric properties when adminis-tered via the Internet (Hedman et al., 2010; Thorndike et al., 2009).

Page 3

M. Tillfors et al. / Journal of Anxiety Disorders 33 (2015) 103–109 105

Table 1The content of the nine modules.

Module Content

1 Introduction and psychoeducation – education about SAD and thetreatment principles. Construing an anxiety hierarchy

2 Clark & Wells’ cognitive model for SAD – starting identifying andregister negative automatic thoughts (NAT)

3 Cognitive restructuring I – continuing working with NAT (evidencefor and against and alternative thoughts). Working withconstruing treatment goals

4 Cognitive restructuring II – continuing working with NAT andbeginning working with behavior experiments

5 Exposure I – education about exposure and starting doingexposure. Construing a more detailed anxiety hierarchy

6 Shift of focus – education and behavior experiments aboutself-focus attention.

7 Exposure II – continuing working with exposure8 Social skills – focus on social communication skills as for example

assertiveness training

N

2

cWf2eemwoctwccpau

2

U

2

dddte(SM

distaliC

Table 2Explained error sums of squares (EESS-values) and the change in percentage of thecluster coefficient to the next level for the cluster solutions between 2 and 6 at pre-and post-treatment.

Number of clusters

2 3 4 5 6

EESS (%; pre-treatment) 43 60 72 77 80Coefficient (%) 75 43 44 19 18

9 Relapse prevention

ote. SAD = social anxiety disorder; NAT = negative automatic thoughts.

.4. Treatment

The main treatment component was a self-help manual, whichonsisted of nine modules (see Table 1), adapted for use via the

orld Wide Web, and was based on established CBT-principlesor SAD (Furmark, Holmström, Sparthan, Carlbring, & Andersson,006). The participants were expected to complete one moduleach week. In general, each module consisted of information andxercises, and ended with a couple of essay questions. Each treat-ent week, participants were asked to summarize in their ownords a central section of the module in question, describe the

utcome of the exercises done, and answer an interactive multiple-hoice quiz. The rationale behind the home-work assignments waso promote learning and enable the Internet-therapists to decidehether the participants had assimilated the information and

ompleted their homework assignments. The internet-therapistsonsisted of either psychology students at MSc level or licensedsychologists. The therapist feedback mainly consisted of supportnd guidance. When the homework was completed, the next mod-le was made accessible.

.5. Ethical considerations

All RCTs were approved by the regional ethics committee atppsala, Sweden.

.6. Statistical analyses

The Statistical Package for Social Sciences, SPSS 21.0 for Win-ows was used to identify subgroups of social avoidance andepressive symptoms, and to compare the derived subgroups oniagnostic status after treatment as well as on social anxiety symp-om severity before and after ICBT (SPSS Inc., Chicago, IL). Toxamine individual stability and movement between subgroupsclusters) from pre- to post-treatment we used a statistical package,LEIPNER 21.0, developed for person-oriented analyses (Bergman,agnusson, & El-Khouri, 2003).More specifically, to examine subgroups of social avoidance and

epressive symptoms, we used Ward’s hierarchical cluster analysis,.e., a person-oriented analysis, with squared Euclidean distances asimilarity measure. The cluster variate consisted of the followingwo variables: (a) ratings of level of avoidance of 24 potentially

nxiety provoking social situations measured by LSAS-A, and (b)evel of depressive symptoms measured by MADRS-SR. Standard-zation in z-scores was performed prior to the cluster analysis.luster analysis groups participants within a sample according to

EESS (%; post-treatment) 49 66 75 80 83Coefficient (%) 95 50 33 26 20

their individual profiles, categorizing participants in order to createhomogeneity within each subgroup and heterogeneity between thesubgroups (Hair, Andersson, Tatham & Black, 2009). This approachwas used in the current study to identify subgroups, i.e. clusters,of people characterized by social avoidance and depressive symp-toms. There are several criteria and guidelines for the number ofclusters to be selected, but currently no selection procedure is gen-erally believed to be more beneficial than others. Our final choiceof cluster solution was based on a combination of the followingfour recommended criteria: (1) the cluster solution should be the-oretically meaningful (Bergman et al., 2003; Hair et al., 2009); (2)clusters are considered reasonably homogenous with explainederror sums of squares (EESS) values ideally around 67% and notless than 50% (Bergman et al., 2003); (3) the cluster coefficient per-centage change to the next level should preferably not be less than10% (Hair et al., 2009), and (4) each cluster should contain at least10 individuals (Hair et al., 2009).

To compare subgroups (clusters) on diagnostic status aftertreatment and on social anxiety symptom severity, we usedvariable-oriented analyses consisting of chi-square analysis andone-way analysis of variance (ANOVA). Finally, to examine indi-vidual stability and movement between subgroups from pre- topost-treatment – we used the EXACON program (Bergman et al.,2003) which produces a contingency table of the cluster solutions(before and after treatment) and enables examination of individ-ual movement from one cluster to another over time. This is doneby calculating the �2 component for each cell and an exact prob-ability that the number of participants (observed frequencies) isgreater or fewer than would be expected by chance (expected fre-quencies). In that way, the program identifies significant types i.e.,predictor–outcome combinations for which there are more par-ticipants than would be expected by chance, and antitypes i.e.,predictor–outcome combinations for which there are fewer par-ticipants than would be expected by chance.

3. Results

3.1. Pre-treatment differences across trials

There was no significant difference regarding social anx-iety, F(4, 162) = 2.31, p > 05 (Andersson et al., 2006, N = 32,M = 36.88, SD = 12.09; Carlbring et al., 2007, N = 29, M = 36.00,SD = 11.66; Tillfors et al., 2008, N = 37, M = 30.57, SD = 11.48;Furmark et al., 2009, N = 40, M = 36.55, SD = 11.49, and N = 29,M = 38.31, SD = 11.28), gender, �2(4, N = 167) = 12.20, p > 05, andmarital status, �2(8, N = 167) = 4.31, p > 05, at pre-test between thefive pooled RCTs.

3.2. Subgroups of people based on social avoidance anddepressive symptoms

Table 2 shows the explained error sums of squares (EESS) val-ues and the change in percentage of the cluster coefficient to thenext level for the cluster solutions between 2 and 6 at pre- and

Page 4

106 M. Tillfors et al. / Journal of Anxiety Disorders 33 (2015) 103–109

Table 3Means (standard deviations) in z-scores and raw-scores for the 4-cluster solution atpre-treatment.

Cluster Socialavoidance

Depressivesymptoms

N (women), %

1. Low problem −1.14 (0.37) −0.76 (0.54) 47 (32), 2819.19 (4.43) 7.68 (4.13)

2. High problem 1.25 (0.58) 1.15 (0.63) 38 (26), 2347.97 (7.00) 22.21 (4.78)

3. Depressive symptoms −0.16 (0.36) 0.61 (0.64) 39 (30), 2331.00 (4.38) 18.08 (4.85)

pEtbtoeaa2sowT

a2c(olNslNetsa

3a

3

f

TMp

-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Low problem High problem Dep symptoms Socially avoidant

LSAS-A

MADRS

-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Low probl em High problem Dep symptom s Social ly avoid ant

LSAS-A

MADRS

Fig. 1. z-scores for the 4-cluster solution at pre-treatment (the upper figure) andat Post-treatment (the below figure). Note. Dep symptoms = depressive symptoms;

4. Socially avoidant 0.29 (0.65) −0.74 (0.41) 43 (27), 2636.40 (7.90) 7.84 (3.12)

ost-treatment. As can be seen in the table, and based on theESS criterion around 67%, all solutions between 4 and 6 clus-ers at pre-treatment and 3 to 6 clusters at post-treatment coulde used. Regarding cluster coefficient percentage change, all solu-ions between 2 and 6 clusters were above the recommended valuef 10% at both time-points. The criterion of ≥10 individuals inach cluster was obtained in solutions between 2 and 5 clusterst pre-treatment and between 2 and 4 clusters at post-treatment,s the smallest cluster comprised 17 people at pre-treatment and3 people at post-treatment. At post-treatment the 5 and 6 clusterolutions both contained 9 persons. Thus, based on our a pri-ri criteria and theoretical meaningfulness, the 4-cluster solutionas the final choice for further analyses at both time-points (see

ables 3 and 4 as well as Fig. 1).As expected, we found one cluster scoring high on social avoid-

nce and depressive symptoms, the High Problem cluster (Cluster; pre: N = 38, 23%; post: N = 30, 19%). We further identified oneluster showing the opposite pattern, the Low Problem clusterCluster 1; pre: N = 47, 28%; post: N = 59, 37%). Cluster 3 was highn depressive symptoms and low on social avoidance, thereforeabeled as the Depressive Symptoms cluster (pre: N = 39, 23%; post:

= 23, 14%). Lastly, cluster 4 was characterized by higher scores ofocial avoidance and lower scores of depressive symptoms, henceabeled as the Socially Avoidant cluster (pre: N = 43, 26%; post:

= 49, 30%). In summary, we observed subgroups consisting ofither high or low levels of social avoidance and depressive symp-oms (the High Problem and Low Problem clusters) as well as aubgroups scoring high on either depressive symptoms or socialvoidance.

.3. Subgroup profile analyses: diagnosis of SAD and socialnxiety symptoms

.3.1. Diagnosis of SADAfter treatment we found a significant relationship between the

our-cluster solution and diagnosis of SAD �2(3, N = 158) = 49.77,

able 4eans (standard deviations) in z-scores and raw-scores for the 4-cluster solution at

ost-treatment.

Clusters Socialavoidance

Depressivesymptoms

N (women), %

1. Low problem −0.89 (0.48) −0.94 (0.36) 59 (41), 3710.49 (5.50) 2.80 (2.23)

2. High problem 1.06 (0.76) 1.55 (0.67) 30 (21), 1932.73 (8.72) 19.10 (4.42)

3. Depressive symptoms −0.56 (0.38) 0.37 (0.44) 23 (17), 1414.22 (4.32) 11.39 (2.89)

4. Socially avoidant 0.68 (0.58) 0.00 (0.41) 49 (33), 3028.35 (6.66) 9.02 (2.72)

LSAS-A = Liebowitz social anxiety scale-the subscale for social avoidance; MADRS-SR = montgomery Åsberg Depression Rating Scale

p < .0001. To be able to examine which clusters contributed tothis relationship, we compared the standardized residuals foreach cluster. The portion of people who no longer fulfilled thecriteria for SAD diagnosis was underrepresented in the HighProblem cluster (z = −2.8) as well as in the Socially Avoidant clus-ter (z = −2.0), and overrepresented in the Low Problem cluster(z = 2.9).

Unfortunately the information regarding diagnosis of SADaccording to SPSQ was missing in the study of Carlbring and col-leagues (2007, N = 29). Therefore, people scoring below 20 on thesocial anxiety scale in the first section of SPSQ were classified as nothaving a diagnosis of SAD, in line with the cut-off reported for SADin the general Swedish population (Furmark et al., 1999). Impor-tant to note, we also ran our analyses on an independent clinicalsample in order to increase the external validity and we found thatusing diagnosis according to SPSQ gave the same result as above(see below under the heading of external validity).

3.3.2. Social anxietyWe compared participants with SAD in all of the clusters on

social anxiety at both time-points, using one-way ANOVAs withGames-Howell’s post hoc tests. The means and group differencesare shown in Table 5. The overall models were significant for bothassessments (F(3, 163) = 73.57, p < 0.001 for pre-treatment and F(3,157) = 64.74, p < .001 for post-treatment). Post hoc tests showedthat people in the High Problem cluster had higher levels of socialanxiety than all the other clusters both before (p < .01) and aftertreatment (p < .01; with the exception of people in the Socially

Avoidant cluster; p = .40) – see Table 5. Thus, people in the HighProblem cluster in general scored higher on social anxiety than theother clusters at both time-points.

Page 5

M. Tillfors et al. / Journal of Anxiety

Table 5Mean social anxiety symptoms by cluster from one-way ANOVAs with Games-Howell’s post hoc tests at both time-points.

Cluster Mean (SD)

Social anxiety at pre-treatmentLow problem (a) 23.28 (6.87)b,c,d

High problem (b) 47.82 (8.35)a,c,d

Depressive symptoms (c) 34.69 (6.27)a,b

Socially avoidant (d) 38.70 (9.19)a,b

Social anxiety at post-treatmentLow problem (a) 14.93 (5.96)b,c,d

High problem (b) 33.83 (9.55)a,c

Depressive symptoms (c) 20.87 (6.78)a,b,d

Socially avoidant (d) 30.65 (7.19)a,c

Nm

3s

udpF

3

iSottth(s

Frpa

ote. Superscript letters indicate which cluster means differed from the currentean at p < .01 or below according to Games-Howell’s post hoc tests.

.4. Examination of individual stability and change betweenubgroups from pre- to post-treatment

It was typical that people in the High Problem cluster contin-ed to belong to a profile with high levels of social avoidance andepressive symptoms after treatment while it was atypical thateople in this subgroup moved to the Low Problem cluster (seeig. 2).

.5. External validity

To evaluate the external validity we ran all analyses on anotherndependent clinical sample (N = 204, 60% women, Mage = 38 years,D = 11.10; Andersson, Carlbring, & Furmark, 2012). The patternf results in this independent clinical sample were in generalhe same, with a four-cluster solution as the final choice at bothime-points (pre-test: EESS = 68%; post-test: EESS = 73%). We fur-

her identified the same pattern of subgroups consisting of eitherigh or low levels of social avoidance and depressive symptomsthe High Problem and Low Problem clusters) as well as subgroupscoring high on either depressive symptoms or social avoidance. A

Pre-Treatment Post Treat ment

ig. 2. Descriptive representation of clusters at each time point. The first columnepresents clusters at pre-treatment and the second column represents clusters atost-treatment. Bold arrows show typical significant cluster pathways and dashedrrows show atypical significant pathways (p < .05).

Disorders 33 (2015) 103–109 107

significant relation between the four-cluster solution and diagnosisof SAD after treatment was found, �2(3, N = 183) = 66.78, p < .0001.People belonging to the High Problem and the Socially Avoidantclusters were underrepresented, z = −2.5 and z = −3.8, and the lowproblem cluster was overrepresented (z = 4.8) among those whono longer fulfilled the diagnostic criteria for SAD after treatment.Also in accordance with the main results the four clusters differedin social anxiety at both assessments (F(3, 200) = 82.57, p < 0.001for pre-treatment and F(3, 179) = 98.59, p < .001 for post-treatment)and post hoc tests showed that people in the High Problem clus-ter had higher levels of social anxiety than all the other clustersboth before (p < .001) as well as after treatment (p < .001). We couldneither, however, demonstrate statistically that it was typical forpeople in the High Problem cluster (p > .05) to continue to exhibithigh social avoidance and high depressive symptoms after treat-ment, nor that it was atypical for this group (p > .05) to move to theLow Problem cluster after treatment.

4. Discussion

Our first aim of the current study was to investigate if a subgroupof people high on social avoidance and depressive symptoms can beidentified both before and after treatment. The second aim was toexamine the subgroups in relation to social anxiety and diagnosis ofSAD. Our final aim was to examine individual stability and changebetween subgroups from pre- to post-treatment. As expected, wecould distinguish a subgroup characterized by high levels of socialavoidance and depressive symptoms (the High Problem cluster) atpre- and post-treatment. People belonging to the High Problem clus-ter were underrepresented among those who were free of SADdiagnosis after treatment and they reported higher levels of socialanxiety at both time-points. These results held even when we ranthe analyses on an independent clinical sample. It was also typicalthat people in the High Problem cluster continued to belong to aprofile with high levels of social avoidance and depressive symp-toms after treatment. Hence, it seems that the combination of highavoidance and depressive symptoms constitutes a risk factor profilein people with SAD with relation to treatment outcome, and furtherthat individuals belonging to this profile show minimal individualmovements from pre to post-treatment.

The current study makes several unique contributions to theliterature on sample heterogeneity in treatment response. Oneis that it seems to be the combination of avoidance and depres-sion, rather than the individual risk factor of depressive symptoms,that predicts suboptimal treatment response in ICBT. This mayshed further light on the mixed results found in previous researchregarding depressive symptoms as a risk factor for treatment out-come in people with SAD (e.g., Eskildsen et al., 2010; Nordgreenet al., 2012; Rodebaugh et al., 2004). The current study was partlybased on the same data that Nordgreen and coworkers (2012) usedin their study. However, by using cluster analysis in relation totreatment response we have been able to examine subgroups ofpeople with risk factor profiles which otherwise have been hiddenin variable-oriented methods such as those that Nordgreen et al.(2012) had used. Indeed, ICBT seemed to be less effective for peoplein the High Problem cluster than for the Depressive Symptoms clus-ter having elevated levels of depressive symptoms only. The posttreatment scores of social anxiety in people in the High problemcluster in the current study were still in the same range as scores ofpeople in other clinical samples before entering treatment (LSAS-Fear = M = 34; Andersson et al., 2006; Fresco et al., 2001). This was

not the case for people belonging to the Depressive symptoms clusterwho reported scores of social anxiety at post-treatment in line withother clinical samples after treatment (LSAS-Fear = M = 21). How-ever, some studies (e.g., Hedman et al., 2012) have indicated that

Page 6

1 xiety

erpnts1(uitniop

tilimlo

FHtciOcttismstMtsbfupossusomoafsrthiuiidaT

08 M. Tillfors et al. / Journal of An

levated levels of depressive symptoms before treatment do notepresent a risk factor for suboptimal treatment response in peo-le with SAD in the sense that although depressive symptoms areot reduced after treatment, they are not disabling. We were ableo show that people in the High Problem cluster reported depres-ive symptoms at a clinical level (Snaith, Harrop, Newby, & Teale,986; Svanborg & Åsberg, 1994) both before and after treatmentMADRS: pre: M = 22; post: M = 19), and further that they werenderrepresented among those who did recover from SAD. Hence,

n people with SAD who reported co-morbid depressive symptoms,he elevated level of depressive symptoms before treatment wasot by itself a risk factor for suboptimal treatment response, rather

t was the combination of co-morbid depressive symptoms and usef avoidance as emotion regulation strategy that was a risk factorrofile for poor response to ICBT.

Importantly, people belonging to the High Problem cluster beforereatment tended to stay in this subgroup even after treatment, andt was atypical for people in this subgroup to move to the Low Prob-em cluster, indicating individual stability in this cluster. It is alsomportant to bear in mind that all participants were diagnosed pri-

arily with SAD, i.e., we reason about relatively higher and lowerevels of depressive symptoms and social avoidance in the contextf a clinical SAD population.

Some limitations of the current study should be mentioned.irst, we only used self-reports to measure our dependent variables.owever, we used questionnaires with good psychometric proper-

ies validated in clinical populations, Second, our clinical sampleonsisted of five pooled RCT’s. This is a threat to the internal valid-ty since the randomization did not take place at the same time.n the other hand, this may strengthen the external validity of theurrent study. What we can control for in terms of known threatso internal validity is the degree of social anxiety, and it turned outhat these five studies did not differ in terms of degree of social anx-ety before treatment. Third, since the participants only underwenttructural SCID-I interviews regarding SAD, we cannot evaluate co-orbid depression, or other co-morbid diagnoses, although we can

till comment on the degree of depressive symptoms and whetherhe symptoms are within the range for clinical levels according to

ADRS-SR. In addition, people were excluded if their scores onhe MADRS-SR indicated a severe form of depression. Fourth, ourelection of risk factors in the current study was limited. However,oth depressive symptoms and avoidance behavior are known riskactors, and the main aim was to identify risk profiles. Fifth, whensing the definition of diagnosis of SAD in the current pooled sam-le some people with SAD belonging to the subgroup ‘performancenly’ would theoretically have been able to still retain the diagno-is. However, we also ran our analyses on an independent clinicalample in order to increase the external validity and we found thatsing diagnosis according to SPSQ gave the same results, whichupported our results in the current pooled sample. Finally, person-riented methods like cluster analysis are based on mathematicalodels instead of inferential statistics like variable-oriented meth-

ds are (Bergman et al., 2003). Therefore, cluster analytic methodsre not taking into account for example sampling errors. It is there-ore important to base decisions of inclusion of variables on aound conceptual model and carefully address issues of sampleepresentatives to counterbalance the absence of the possibilityo statistically correct for sampling error. In the current study weave ran our analyses on an independent clinical sample, which

ncreases the likelihood that the current sample reflects the pop-lation from which it was extracted. Since these analyses resulted

n the same cluster-solution it indicates that the clusters obtained

n the current study were indeed “real” and did not arise at ran-om. In addition, the clusters could be meaningfully interpretednd showed discriminant validity on indices of external variables.his further strengthens their reliability and validity.

Disorders 33 (2015) 103–109

Despite these limitations, our study has several strengths. First,we analyzed our data both cross-sectionally and prospectivelyas well as with a combination of person-oriented and variable-oriented analysis methods. Hence, by using both variable-orientedand person-oriented methods give us new opportunities to studyheterogeneity in treatment response both within individuals andacross time. Second, we had access to another independent clini-cal sample, and could on a broad level replicate our results. This isimportant for the external validity of the current study. Third, sinceall participants in the five treatment samples were given the samestandardized treatment, that is the same written modules via theinternet, we have by definition a relatively high degree of conceptvalidity, although we could not control for individual differences intherapists’ email conversations.

Future research should investigate: (1) whether a combina-tion of risk factors will predict suboptimal response after guidedInternet-based CBT also in a clinical population of people withdepression, and (2) if covert avoidance like worry/rumination, thatare common emotion regulation strategies both for people withSAD and depression, contribute to sample heterogeneity of treat-ment response. The latter should be of importance to examinesince underlying common factors for both anxiety disorders anddepression among others are the personality dimension negativeaffect, which could imply an increased emotional reactivity in peo-ple that in turn may increase the risk wanting to get rid of internalunpleasant experiences. This in turn may increase the risk of usingdysfunctional emotion regulation strategies such as both overt andcovert avoidance. These emotion regulation strategies have alsobeen found to maintain different types of mental ill-health such asvarious anxiety disorders and conceptually these strategies couldbe thought of mediating the relationship between stressors andclinical levels of anxiety and/or depression. Together this gives sup-port for the importance of also targeting other types of emotionregulation strategies, in addition to overt avoidance, in relation toICBT and treatment response.

5. Conclusions

The present results show that the combination of high avoid-ance and depression in individuals with SAD entering guided ICBTconstitute a clear risk profile for poor treatment response. Thus,it is important to specifically target social avoidance and depres-sive symptoms in people with SAD in guided ICBT. This point tothe importance of measure both social avoidance and depressivesymptoms before, during and after ICBT. If a person exhibits thishigh-problem profile it might be particularly important to payattention to that homework assignments related to all forms ofexposure for social situations are not avoided. Furthermore, totargeting the depressive symptoms more specific a possibility tochoose a module consisting of components of behavioral activationmight be available.

References

Andersson, G., Carlbring, P., & Furmark, T. (2014). Internet-delivered treatments forsocial anxiety disorder. In: J. Weeks (Ed.), Handbook of social anxiety disorder (pp.569–587). New York: Wiley-Blackwell.

Andersson, A., Carlbring, P., Furmark, T., & on behalf of the Sofie research group.(2012). Therapist experience and knowledge acquisition in Internet-deliveredCBT for social anxiety disorder: a randomized controlled trial. PLoS ONE, 7,e37411. http://dx.doi.org/10.1371/journal.pone.0037411

Andersson, G., Carlbring, P., Holmström, A., Sparthan, E., Furmark, T., Nilsson-Ihrfelt,

E., et al. (2006). Internet-based self-help with therapist feedback and in-vivogroup exposure for social phobia: a randomized controlled trial. Journal of Con-sulting and Clinical Psychology, 74, 677–686.

American Psychiatric Association. (2000). Diagnostic and statistical manual of mentaldisorders, text revision ed. (4th ed.). American Psychiatric Press.

Page 7

xiety

B

B

C

C

E

M

F

F

F

F

F

H

H

H

items per webpage. Computers in Human Behavior, 25, 393–401.Tillfors, M., Carlbring, P., Furmark, T., Lewenhaupt, S., Spak, M., Eriksson, A., et al.

M. Tillfors et al. / Journal of An

aker, S. L., Heinrich, N., Kim, H-J., & Hofmann, S. G. (2002). The Liebowitz socialanxiety scale as a self-report instrument: a preliminary psychometric analysis.Behavior Research and Therapy, 40, 701–715.

ergman, L. R., Magnusson, D., & El-Khouri, B. M. (2003). Studying individual devel-opment in an interindividual context: a person-oriented approach. New York:Psychology Press.

arlbring, P., Gunnarsdottir, M., Hedensjö, L., Andersson, G., Ekselius, L., Furmark, T.,et al. (2007). Treatment of social phobia: randomized trial of internet-deliveredcognitive-behavior therapy with telephone support. British Journal of Psychiatry,190, 123–238.

lark, D. M., Ehlers, A., Hackmann, A., McManus, F., Fennell, M., Grey, N., et al. (2006).Cognitive therapy versus exposure and applied relaxation in social phobia: arandomized controlled trial. Journal of Consulting and Clinical Psychology, 74,568–578.

skildsen, A., Hougaard, E., & Rosenberg, N. K. (2010). Pre-treatment patient vari-ables as predictors of dropout and treatment outcome in cognitive behaviouraltherapy for social phobia: a systematic review. Nordic Journal of Psychiatry, 64,94–105.

ayo-Wilson, E., Dias, S., Mavranezouli, I. G., Kew, K., Clark, D. M., Ades, A. E.,et al. (2014). Psychological and pharmacological interventions for social anx-iety disorder in adults: a systematic review and network meta-analysis. LancetPsychiatry, 1, 368–376.

irst, M. B., Gibbon, M., Spitzer, R. L., & Williams, J. B. W. (1997). Structured clin-ical interview for DSM-IV Axis I Disorders (SCID-I). Washington, DC: AmericanPsychiatric Press.

resco, D. M., Cores, M. E., Heimberg, R. G., Liebowitz, M. R., Hami, S., Stein, M. B.,et al. (2001). The Liebowitz social anxiety scale: a comparison of the psychomet-ric properties of self-report and clinician administered formality. PsychologicalMedicine, 31, 1025–1035.

urmark, T., Carlbring, P., Hedman, E., Sonnenstein, A., Clevberger, P., Bohman, B.,et al. (2009). Guided and unguided self-help for social anxiety disorder: ran-domized controlled trial. British Journal of Psychiatry, 195, 440–447.

urmark, T., Holmström, A., Sparthan, E., Carlbring, P., & Andersson, G. (2006). Socialfobi – Effektiv hjälp med kognitiv beteendeterapi [Social phobia – effective help viaCBT]. Stockholm: Liber.

urmark, T., Tillfors, M., Everz, P.-O., Marteinsdottir, I., Gefvert, O., Fredrikson, M.,et al. (1999). Social phobia in the general population: prevalence and sociode-mographic profile. Social Psychiatry and Psychiatric Epidemiology, 34, 416–424.

air, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (2009). Multivariate dataanalysis with readings (7th ed.). Englewood Cliffs, NJ: Prentice-Hall.

arvey, A., Watkins, E., Mansell, W., & Shafran, R. (2004). Cognitive behavioral pro-cesses across psychological disorders: a transdiagnostic approach to research and

treatment. New York: Oxford University Press Inc.

edman, E., Andersson, E., Ljótsson, B., Andersson, G., Andersson, E. M., Schalling, M.,et al. (2012). Clinical and genetic outcome determinants of Internet- and group-based cognitive behavior therapy for social anxiety disorder. Acta PsychiatricaScandinavica, 126, 126–136.

Disorders 33 (2015) 103–109 109

Hedman, E., Ljotsson, B., Ruck, C., Furmark, T., Carlbring, P., Lindefors, N., et al. (2010).Internet administration of self-report measures commonly used in research onsocial anxiety disorder: a psychometric evaluation. Computers in Human Behav-ior, 26, 736–740.

Heimberg, R. G. (2002). Cognitive-behavioral therapy for social anxiety disorder:current status and future directions. Biological Psychiatry, 51, 101–108.

Heimberg, R. G., Mueller, G. P., Holt, C. S., Hope, D. A., & Liebowitz, M. R. (1992).Assessment of anxiety in social interactions and being observed by others: thesocial interaction anxiety scale and the social phobia scale. Behavior Therapy, 23,53–73.

Kessler, R. S., Chiu, W. T., Demler, O., Merikangas, K. R., & Walters, E. E. (2005).Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in theNational Comorbidity Survey Replication. Archives of General Psychiatry, 62,617–627.

Liebowitz, M. R. (1987). Social phobia. Modern Problems of Pharmacopsychiatry, 22,141–173.

Nordgreen, T., Havik, O. E., Öst, L.-G., Furmark, T., Carlbring, P., Andersson, G., et al.(2012). Outcome predictors in guided and unguided self-help for social anxietydisorder. Behaviour Research and Therapy, 50, 13–21.

Ponniah, K., & Hollon, S. D. (2008). Empirically supported psychological interven-tions for social phobia in adults: a qualitative review of randomized controlledtrials. Psychological Medicine, 38, 3–14.

Rapee, R. M., & Spence, S. H. (2004). The etiology of social phobia: empirical evidenceand an initial model. Clinical Psychology Review, 24, 737–767.

Rodebaugh, T. L., Holaway, R. M., & Heimberg, R. G. (2004). The treatment of socialanxiety disorder. Clinical Psychology Review, 24, 883–908.

Schneider, F. R., Johnson, J., Horning, C. D., Liebowitz, M. R., & Weissman, M. M.(1992). Social phobia. Comorbidity and mobidity in an epidemiologic sample.Archives of General Psychiatry, 49, 282–288.

Snaith, R. P., Harrop, F. M., Newby, D. A., & Teale, C. (1986). Grade scores of theMontgomery-Asberg-depression and the clinical anxiety scales. British Journalof Psychiatry, 148, 599–601.

Svanborg, P., & Åsberg, M. (1994). A new self-rating scale for depression and anx-iety states based on the comprehensive psychopathological rating scale. ActaPsychiatrica Scandinavica, 89, 21–28.

Svanborg, P., & Åsberg, M. (2001). A comparison between the beck depression inven-tory (BDI) and the self-rating version of the Montgomery Asberg DepressionRating Scale (MADRS). Journal of Affective Disorders, 64, 203–216.

Thorndike, F. P., Carlbring, P., Smyth, F. L., Magee, J. C., Gonder-Frederick, L., Ost, L. G.,et al. (2009). Web-based measurement: effect of completing single or multiple

(2008). Treating university students with social phobia and public speakingfears: internet delivered self-help with or without live group exposure sessions.Depression Anxiety, 25, 708–717.