J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

Contents lists available at ScienceDirect

Journal of Behavior Therapy and Experimental Psychiatry journal homepage: www.elsevier.com/locate/jbtep

Internet-based Cognitive Bias Modification of Interpretations in patients with anxiety disorders: A randomised controlled trial Elske Salemink a, *, Merel Kindt b, Henk Rienties c, Marcel van den Hout a a

Department of Clinical and Health Psychology, Utrecht University, Utrecht, The Netherlands Department of Clinical Psychology, University of Amsterdam, Amsterdam, The Netherlands c Altrecht Academic Anxiety Centre, Altrecht, Utrecht, The Netherlands b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 28 March 2013 Received in revised form 2 October 2013 Accepted 3 October 2013

Background and objectives: Previous research suggests that negative interpretation biases stimulate anxiety. As patients with an anxiety disorder tend to interpret ambiguous information negatively, it was hypothesised that training more positive interpretations reduces negative interpretation biases and emotional problems. Methods: In a randomised, double-blind placebo-controlled trial, patients with different anxiety disorders were trained online over eight days to either generate positive interpretations of ambiguous social scenarios (n ¼ 18) or to generate 50% positive and 50% negative interpretations in the placebo control condition (n ¼ 18) (Study 1). Results: Positively trained patients made more positive interpretations and less negative ones than control patients. This training was followed by a decrease in anxiety, depression, and general psychological distress, but this effect was also observed in the control group. To get a better understanding of these unexpected results, we tested a 100% neutral placebo control group (Study 2, n ¼ 19); now the scenarios described neutral, non-emotional situations and no valenced interpretations were generated. The results from this neutral group were comparable to the effects from the other control group. Limitations: An advantage, but potentially also a disadvantage of the study is that CBM-I training was performed online with less control over the procedures and setting. In addition, the scenarios were not matched to the specific concerns of each patient and the training sessions were performed in close proximity to one another. Conclusions: Compared to both control conditions, CBM-I had superior effects on interpretations, but not on emotions. The current findings showed the boundary conditions for CBM-I. Ó 2013 Elsevier Ltd. All rights reserved.

Keywords: Anxiety disorders Cognitive Bias Modification of Interpretations Internet-based

1. Introduction Cognitive theories argue that patients with an anxiety disorder interpret potential threatening information as much more threatening than they are and this biased interpretation is held to be the pathogenic nucleus of the disorder (Beck, Emery, & Greenberg, 1985; Williams, Watts, Macleod, & Mathews, 1988). It is hypothesised that biased interpretations are causally related to anxious feelings and behaviour and experimental evidence supports this causal claim (Mathews & Mackintosh, 2000). Interpretation bias was modified using a scenario-based Cognitive Bias Modification

* Corresponding author. Present address: Department of Developmental Psychology, University of Amsterdam, Weesperplein 4, 1018 XA Amsterdam, The Netherlands. Tel.: þ31 205258663; fax: þ31 206390279. E-mail address: [email protected] (E. Salemink). 0005-7916/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jbtep.2013.10.005

for Interpretations (CBM-I) training and anxiety was affected subsequently. In the meantime these findings have been replicated several times (see for an overview Hallion & Ruscio, 2011). There is overwhelming evidence that patients with an anxiety disorder have the tendency to interpret ambiguity in a threatrelated way. That is, patients with a social anxiety disorder (SAD) interpreted ambiguous social scenarios as more negative than a non-anxious control group (Amir, Foa, & Coles, 1998), patients with a panic disorder (PD) were more likely to interpret bodily sensations as signs of threat than other anxiety disorder patients (Clark et al., 1997), and patients with Generalised Anxiety Disorder (GAD) interpreted ambiguous scenarios as more threatening than non-anxious controls (Butler & Mathews, 1983). In a prospective study, it was shown that interpretation of initial posttrauma symptoms predicted Posttraumatic Stress Disorder (PTSD) symptoms at six and nine months follow-up (Dunmore, Clark, & Ehlers, 2001).

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

According to cognitive theories of psychopathology, it is important that treatments for anxiety disorders address these maladaptive processes. Cognitive Behavioural Therapy (CBT), for example, aims to alter those biased interpretations via cognitive restructuring and behavioural experiments. As biases in information processing are fast and more automatic processes, the use of verbal dialogue and explicit instructions might not be the optimal approach to change them (Beard, 2011). Furthermore, processing biases filter incoming information and may act as a barrier in CBT by hampering the activation of incongruent information (Baert, Koster, & De Raedt, 2011). A recent new development to change these processing biases in a different way is Cognitive Bias Modification (CBM). It is a computerised training consisting of “extensive practice on a cognitive task designed to encourage and facilitate the desired cognitive change” (p. 3, Koster, Fox, & MacLeod, 2009). In Cognitive Bias Modification of Interpretation (CBM-I), participants repeatedly practice more positive interpretations. Compared to CBT, CBM changes processing biases through practicing the desired process, and not through verbal instruction and explicitly challenging thoughts. Due to the different approach, CBM might have added value in the treatment of anxiety disorders. There is accumulating evidence that CBM-I training has effects in non-anxious and highly anxious analogue samples. For example, four sessions of positive CBM-I training using scenarios provided to highly anxious individuals resulted in stronger reductions in trait anxiety scores (d ¼ 0.58) compared to a testeretest control group (d ¼ 0.08) (Mathews, Ridgeway, Cook, & Yiend, 2007). These effects were replicated in a study where (scenario-based) CBM-I sessions were increased from four to eight in a sample of highly anxious students with stronger reductions in trait anxiety and levels of psychological distress in the CBM-I training condition (d ¼ 0.29 and 0.41 respectively) compared to a placebo control condition (scenarios had 50% positive and 50% negative outcomes, d ¼ 0.06 and d ¼ 0.04 respectively) (Salemink, van den Hout, & Kindt, 2009). Similar findings have been observed with different types of CBM-I paradigms (Amir, Bomyea, & Beard, 2010; Beard & Amir, 2008) and with different types of highly anxious analogues populations (Hirsch, Hayes, & Mathews, 2009; Steinman & Teachman, 2010; Teachman & Addison, 2008). To the best of our knowledge, two studies actually examined the effects of CBM-I in a clinically anxious sample (Amir & Taylor, 2012; Hayes, Hirsch, Krebs, & Mathews, 2010) and two other studies tested a combined training of both attention and interpretation bias training (Beard, Weisberg, & Amir, 2011; Brosan, Hoppitt, Shelfer, Sillence, & Mackintosh, 2011). Hayes et al. (2010) investigated the effects of a single CBM-I session in GAD. Forty patients were randomly allocated to a positive interpretation bias training (homograph and scenario-based training) or a placebo control condition (both training paradigms contained 50% threatening interpretations). CBM-I successfully modified interpretations and individuals who had followed the positive training had fewer selfreported negative intrusive thoughts. However, there were no significant differences between the groups in change in selfreported anxiety following a worry period (dpositive CBM ¼ 0.03 vs. deterioration in the placebo group dplacebo ¼ 0.19). The lack of effects on anxiety might be, among other reasons, due to the fact that individuals were only trained once, whereas multiple sessions of training might be necessary to obtain emotional effects in a clinical sample. Amir and Taylor (2012) provided a 12 session interpretation training (based on WordeSentence Association Paradigm, WSAP) to individuals with a generalised SAD. Training affected interpretations and, compared to a placebo control condition (receiving 50% positive and 50% negative feedback after endorsing threatening interpretations), participants who received the training were judged as less socially anxious by clinicians who

187

were blind to treatment condition (d ¼ 1.95 vs. 0.61). However, both conditions did not differ on change in self-reported social anxiety symptoms; both the training condition (d ¼ 1.23) and the control condition (d ¼ 1.29) displayed large reductions in self-reported social anxiety. These first findings in clinical samples seem to suggest that CBM-I is successful in modifying interpretations, but that it does not outperform the control condition with respect to self-reported anxiety. Regarding the two studies that examined CBM-I in combination with attention training in a clinical sample; Brosan et al.’s study (2011) did not include a control group, thus it is unclear whether the observed reduction in self-reported anxiety (dstate anxiety ¼ 0.81, dtrait anxiety ¼ 1.12) is the result of the intervention. Beard et al., 2011 compared the combined CBM training (eight sessions WSAP training) with a placebo condition (attention: probe replaced neutral and disgust faces with equal frequency; interpretation: words were not related to social, but to superficial aspects). A stronger reduction in self-reported anxiety was observed in the CBM condition (d ¼ 1.04) compared to the placebo condition (d ¼ 0.20). Even though this result is promising, it is hard to evaluate the role of interpretive bias training, as the effects could also be due to the attentional bias training. Up to now, the effects of CBM-I on self-reported anxiety in clinical populations seem more mixed than the effects in non-anxious and highly anxious analogue samples. Furthermore, little is known regarding the longevity of the effects as only Amir and Taylor (2012) included follow-up measures, though only in the CBM-I condition. CBM-I training is a computerised training that has the possibility to be offered to participants via internet. Up to now, however, all published CBM-I studies with clinical samples have delivered the training in a laboratory or office setting with participants coming to that location to complete each training session (Amir & Taylor, 2012; Beard et al., 2011; Brosan et al., 2011; Hayes et al., 2010, but see Salemink et al., 2009 for an online CBM-I training in a highly anxious, but not clinical sample). This is surprising as repeatedly coming to a certain location might be a barrier for care. People living in remote areas, physically disabled patients with restricted mobility, or patients who are hesitant to seek face-to-face treatment might be better reached with internet-delivered therapies (Lange, van der Ven, Schrieken, & Emmelkamp, 2001). Thus internet-based treatment has the potential to increase availability, but also to facilitate dissemination. In addition, given the context sensitivity of training effects, online interventions are promising as they can be completed at home (Macleod, Koster, & Fox, 2009). Furthermore, E-mental health is a promising new area for treatment and studies have shown that internet-delivered CBT may be as effective as face-to-face CBT (e.g., Hedman et al., 2011). Finally, web-based interventions have the potential of being more costeffective. Computerised CBM training paradigms seem ideal interventions to be provided online and this has been tested for attentional bias training (CBM-A). The first findings are however mixed; while it has been shown that attentional bias can be modified using an internet-delivered training (Macleod, Soong, Rutherford, & Campbell, 2007), another study revealed no superior effects of CBM-A on self-reported anxiety symptoms (Carlbring et al., 2012). The findings of online interpretive bias training in a highly anxious, non-clinical sample were promising (Salemink et al., 2009). The aim of the present study was to examine the effectiveness of an internet-delivered CBM-I training in a clinical sample of patients with a broad range of anxiety disorders. In a randomised, doubleblind placebo-controlled trial, patients with anxiety disorders were trained online to either generate positive interpretations of ambiguous social scenarios or to generate 50% positive and 50% negative interpretations in the placebo control condition. This placebo-control condition was developed to control for the effects

188

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

Assessed for eligibility (n = 148) Excluded (n = 101) • Met exclusion criteria (n = 42) Receiving psychological treatment, n = 21 No pc / internet access, n = 14 Insufficient Dutch literacy, n = 7 • Declined to participate (n = 59) Randomized (n = 47)

Allocated to positive CBM-I condition (n = 24) • Received intervention (n = 24) • Did not receive intervention (n = 0)

Allocated to control condition (n = 23) • Received intervention (n = 22) • Did not receive intervention (n = 1) Refused after randomisation, n = 1

Lost during intervention (n = 4)

Lost during intervention (n = 3)

Completed intervention (n = 20) • Succesful (n = 18) • Unsuccessful; too many days no training (n = 2)

Completed intervention (n = 19) • Succesful (n = 18) • Unsuccessful; too many days no training (n = 1)

3 month follow-up assessment (n = 18) • Completed follow-up (n = 14) • Lost to follow-up (n = 4) Not interested anymore, n = 3 Untraceable, n = 1

3 month follow-up assessment (n = 18) • Completed follow-up (n = 16) • Lost to follow-up (n = 2) Untraceable, n = 2

Analysed • Direct training effects (n = 18) • 3 month follow-up (n = 14)

Analysed • Direct training effects (n = 18) • 3 month follow-up (n = 16)

Fig. 1. CONSORT flow diagram illustrating the flow of participants through the study.

of repeated exposure to emotional social material and non-specific treatment effects (cf. Hayes et al., 2010; Salemink et al., 2009). Furthermore, to examine the longevity of the effects, a three-month follow-up assessment was included for both the CBM-I training group as well as the placebo-control condition. During the eight day program, patients did not receive other treatments as they were on a waiting list to receive treatment. However, at follow-up assessment three months later, they received treatment as usual. It was predicted that positive CBM-I would modify the interpretation bias. That is, patients trained to interpret ambiguity in a positive way would interpret new ambiguous information less negatively and more positively compared to patients in the placebo-control condition. Secondly, based on the theoretical model and previous findings in high anxious individuals, it was predicted that patients in the CBM-I condition would display less anxiety, depressive mood, general psychological distress, and more positive mood following training relative to patients in the placebo control condition (primary outcome) and that those treatment effects would be maintained at follow-up. In order to clarify the unexpected results of the first study (see below), we performed a second study with a different placebo control condition. 2. Study 1 2.1. Participants The waiting list at Altrecht Academic Anxiety Centre or Mesos Medical Centre (Utrecht, the Netherlands) was searched to identify patients who had, based on the intake procedure, one of the following diagnoses: PD with or without agoraphobia; SAD; PTSD;

or GAD. Eligible participants received a detailed letter describing the study, the aim (to examine the effect of a computerised training for anxiety), randomisation, the assessments, the risks etc., and inviting them to participate. Patients were subsequently contacted by the research team (the first author or a trained research assistant) by telephone to answer any questions and, when interested, to arrange a face-to-face assessment to determine Axis I diagnosis using the Structured Clinical Interviews for the DSM-IV, version I (SCID; First, Spitzer, Gibbon, & Williams, 1996; van Groenestijn, Akkerhuis, Kupka, Schneider, & Nolen, 1999). All participants gave written informed consent to participate. Inclusion criteria were: 1) age above 18, and 2) one of the following diagnoses: PD with or without agoraphobia; SAD; PTSD; or GAD. Exclusion criteria were: 1) concurrent psychological treatment, 2) no access to internet, 3) insufficient Dutch literacy, 4) current psychosis, or 5) current substance/alcohol dependence. There were no exclusion criteria with respect to other Axis I or II diagnoses. Patients prescribed psychotropic medication needed to be taking a constant dose for at least two weeks before study entry and remaining stable throughout the study to reduce the likelihood that recent medication change could account for symptom improvement. The trial CONSORT flowchart (Fig. 1) describes the flow of potential participants assessed for eligibility through to randomisation either to the positive CBM-I (n ¼ 24) or the placebo control group (n ¼ 23). In total, 18 patients in the positive and 18 in the placebo-control condition successfully completed the eight-day procedure. As a main diagnosis, fourteen patients met diagnostic criteria for PD with or without agoraphobia, eight patients met GAD criteria, six SAD, six PTSD, and two agoraphobia without a history of

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

panic disorder. On average, participants had 2.3 (SD ¼ 1.0) Axis I diagnoses and co-morbid diagnoses consisted of another anxiety disorder, depressive disorder (current or in remission), hypochondriasis, body dysmorphic disorder, or binge-eating disorder. Due to technical difficulties, five participants did not receive the postprogram recognition test (CBM-I condition n ¼ 1 and control condition n ¼ 4). Six participants did not return the follow-up questionnaires (CBM-I condition n ¼ 4 and control condition n ¼ 2), resulting in follow-up data for 14 patients in the positive CBM-I condition and 16 patients in the placebo-control condition. 2.2. Design Participants meeting the entry criteria were randomly allocated 1:1 (carried out within the online computer program) to the positive CBM-I training versus the control condition. Patients and researchers were blind to condition (double blind design). Data were collected at baseline assessment before randomisation, during the training, directly after the training, and at three months after completion of the training. G*Power was used to calculate sample size. The study was approved by the medical ethics committee of the University Medical Centre Utrecht, the Netherlands. Data collection was carried out from November 2006 to January 2008. 2.3. Materials 2.3.1. Positive Cognitive Bias Modification of Interpretations To modify interpretation bias, patients were trained for eight sessions at home over the internet (Salemink et al., 2009). In each session, participants received 104 unique scenarios presented in eight blocks with optional rests between each block. Scenarios that were used successfully in previous studies (Mathews & Mackintosh, 2000; Salemink, van den Hout, & Kindt, 2007, 2009) were adapted to reflect situations relevant for a non-student, clinical population. Each block contained eight modification scenarios, three neutral filler scenarios, and two probe assessment scenarios (see Interpretation Bias Outcome Measurements), all presented in a fixed random order. The modification scenarios consisted of three lines that were ambiguous in terms of valence. A word fragment disambiguated the scenario in a positive way. Participants were asked to complete the fragments as quickly as possible and the complete word was presented subsequently. A comprehension question with relevant feedback appeared on the screen to consolidate the interpretation imposed by the word fragment. An example is: You have just moved to a new area and your neighbour asks if you would like to go to the local pub that evening. When you arrive, they are not yet there. After your earlier conversation, they probably thought you were . likeab-e (likeable) Did you make a bad impression on your new neighbour? The ‘neutral filler scenarios’ had no emotional content, nor did they contain ambiguity. They were inserted to make the training less obvious. 2.3.2. Placebo control condition The difference between the positive CBM-I condition and the placebo control condition is that the latter group received four positive and four negative scenarios instead of eight positive scenarios within each block. 2.3.3. Interpretation bias outcome measurements To check whether CBM-I was successful in modifying interpretation bias, two interpretation bias outcome measures were

189

obtained. The first is the latency to complete probe word fragments. Probe scenarios were comparable to the emotional scenarios, but they had a fixed positive and negative valence, irrespective of condition. They were inserted to measure the speed of resolving word fragments of positive and negative valences across the training phase (Mathews & Mackintosh, 2000; Salemink et al., 2007, 2009). A more positive interpretation style is expected to facilitate finding solutions for the positive word fragments and impede finding solutions for the negative word fragments. The second interpretation bias outcome measure was the recognition task that consisted of ten new ambiguous scenarios (Mathews & Mackintosh, 2000; Salemink & van den Hout, 2010). In this task, solution of the word fragments did not disambiguate the scenario; the scenario remained ambiguous. An example is presented here: The evening class You’ve just started going to an evening class. The instructor asks a question and no one in the group volunteers an answer, so he looks directly at you. You answer the question, aware of how your voice must sound to the . oth–s (others) Have you been going to the evening class for a long time? In the second part of the recognition test, participants saw the title of the ambiguous scenario, together with four sentences presented in a random order. These sentences represented a) a possible positive interpretation, b) a possible negative interpretation, c) a positive foil sentence, and d) a negative foil sentence. The four corresponding sentences of “The evening class”-scenario are presented here: a) You answer the question, aware of the others listening attentively. b) You answer the question, aware of how unsteady your voice sounds. c) You answer the question and then realise what a good answer it is. d) You answer the question, but realise that you have made a mistake. Participants rated each sentence for its similarity in meaning to the original scenario using a 4-point scale ranging from 1 (very different in meaning) to 4 (very similar in meaning). 2.3.4. Emotional outcome measurements Four self-report measures were administered three times: Assessment 1 (before the start of the training), Assessment 2 (after completing all eight sessions), and Assessment 3 (at follow-up: approximately three months after training). The Dutch version of the State-Trait Anxiety Inventory (STAI-ST, Spielberger, Gorsuch, Lushene, Vagg, & Jacobs, 1983; Van der Ploeg, Defares, & Spielberger, 2000) was administered to measure state and trait anxiety. This four-point scale self-report questionnaire contains 40 items and has adequate psychometric properties. To measure level of depressive symptoms, the widely used and well-validated Beck Depression Inventory (BDI, Beck, Rush, Shaw, & Emery, 1979) was used. It is a 21-item self-report questionnaire with good psychometric properties. A third measure was the Dutch translation of the Symptom CheckList-90 (SCL-90; Arrindell & Ettema, 1986; Derogatis, 1977). It comprises 90 items and answers are given on a five-point scale. The total score was used as a measure of general psychological distress. The internal consistency of this score is satisfactory. Fourth, positive and negative affect was measured with

190

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

Table 1 Socio-demographic and clinical characteristics of participants Study 1. Statistic 50%/50% Positive CBM-I group control group (n ¼ 18) (n ¼ 18) Age, mean (SD) Women Marital status Single Relationship, not cohabitating Cohabitating Married Divorced Education: exams passed University e graduate level University e undergraduate level High school Middle school Elementary school No. of Axis I diagnoses, mean (SD) Duration of diagnoses, mean (SD), in years Years of treatment, mean (SD) Psychotropic medication use

p-Value

t ¼ .97 .34 c2 ¼ 2.22 .26

41.9 (9.5) 15

38.6 (10.8) 11

2 0

2 3

6 4 6

5 6 2

2 3

2 5

1 10 2 2.0 (1.1)

0 8 3 2.5 (.9)

c2 ¼ 1.9

.75

t ¼ 1.53

.13

14.9 (12.7)

15.7 (10.8)

t ¼ .20

.84

1.7 (2.4) 8

1.3 (1.6) 14

t ¼ .70 .49 c2 ¼ 4.21 .09

c2 ¼ 5.49 .24

Note. Data is given as number of individuals, except where is indicated that data is given as means with standard deviation in parentheses.

the psychometrically sound Positive and Negative Affect Schedule (PANAS, Watson, Clark, & Tellegen, 1988). This scale consists of 20 words that describe different feelings and emotions and participants indicated on a five-point scale to what extent they felt that way. 2.3.5. Exit interview An exit interview was conducted to obtain participants’ ideas regarding their allocated condition (positive vs. placebo-control). They were asked to indicate which condition they thought they had followed. If they had no idea, they were invited to guess. 2.4. Procedure During the first assessment session, participants completed the self-report questionnaires. As a next step, participants were told to imagine themselves being in the situations described in the scenarios and completed ten practice trials in the presence of the experimenter. Afterwards, the experimenter left and the participant could complete the first session individually at the time that suited him/her best. Participants completed the subsequent sessions at home over the internet and the experimenter received an automatically generated e-mail confirmation when a session was successfully completed by a participant. Each session started with some instructions on creating a quiet and undisturbed test environment. Then participants could start with the scenarios.1 The duration of each session was approximately 45 min. Participants had to complete eight sessions in a period of 11 days and were allowed to include three training free days. Directly after the last block of training on the eighth day, participants received the recognition task. Assessment II was planned within three days following the

1 Participants’ daily positive and negative mood states were assessed before every session with three positive and three negative items from the PANAS. These daily measurements were not reported, as we focused on the pre-, post-, and follow-up assessments.

cessation of the program. Again the experimenter visited the patient at home and they completed the questionnaires. Afterwards, an exit interview took place. Assessment III was by mail at approximately three months follow-up. By that time patients were receiving treatment as usual, thus limiting conclusions based on the follow-up assessments. Participants did not receive compensation for screening, assessments nor participating in the study; they only received a small financial reimbursement for their internet costs. 2.5. Results Study 1 2.5.1. Between-group differences at baseline Comparison of the two conditions at baseline (Table 1) revealed no significant group differences regarding age, sex, marital status, educational level, number and duration of diagnoses, duration of treatment, or psychotropic medication use. Furthermore, the groups did not differ significantly on the pre-assessment questionnaires (see Table 2), except for the STAI state questionnaire, t(34) ¼ 2.11, p < .05. That is, before starting the program, participants in the control group felt more anxious than participants in the positive condition. As these scores were related to the negative scale of the PANAS and the follow-up assessment (and not to changes from T1 to T2 in reaction time data, recognition task, trait anxiety, SCL-90, BDI, and PANAS positive scale), these preassessment state anxiety scores were added as a covariate to the PANAS negative scale and follow-up analyses. 2.5.2. Interpretation bias outcome measures Regarding the latency to complete probe fragments, reaction time trials where an incorrect response was given to the fragment completion (3.9%) or to the comprehension question (6.1%) were omitted from the analysis. In addition, latencies were set aside if the latency was less than 200 ms (0.04%) or greater than three SDs above the individual mean (1.7%). A mixed model ANOVA2 with CMB-I group (positive vs. control) as the between-subjects factor and probe (positive vs. negative) as the within-subject factor was conducted to test the effectiveness of the online CBM-I in changing interpretations. There was a significant main effect of probe, F(1, 34) ¼ 32.63, p < .001, h2p ¼ .49, that was qualified by a significant Group  Probe interaction effect, F(1, 34) ¼ 21.53, p < .001, h2p ¼ .39. While, there was no significant difference in responding to negative and positive probes in the control group, t(17) ¼ 0.72, p ¼ .48 (Mnegative probes ¼ 1470 ms, SD ¼ 557, Mpositive probes ¼ 1447 ms, SD ¼ 539), participants who had been exposed to the positive CBMI training displayed a marked slowing in reacting to negative (M ¼ 1852 ms, SD ¼ 507) as compared to positive probes (M ¼ 1635 ms, SD ¼ 433), t(17) ¼ 7.80, p < .001. Furthermore, directly comparing both groups using independent samples t-test indicated that positively trained patients were significantly slower in responding to negative probes, t(34) ¼ 2.15, p < .05. The second interpretation bias outcome measurement was the recognition test. A mixed model ANOVA2 with group as the between-subjects factor and valence (positive vs. negative) and target (possible interpretation vs. foil sentence) as the withinsubject factors was performed. In addition to a significant main effect of valence, F(1, 29) ¼ 50.40, p < .001, h2p ¼ .64, and target, F(1, 29) ¼ 55.47, p < .001, h2p ¼ .66, the analysis revealed a significant Group  Valence interaction effect, F(1, 29) ¼ 6.58, p < .05, h2p ¼ .19. Both the positively trained and control trained participants gave higher similarity ratings to positive than to negative

2 An intention-to-treat analysis with last observation carried forward revealed the same pattern of results.

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

191

Table 2 Mean scores of the outcome measures (SD in parentheses) for the positive and 50%/50% placebo control group (Study 1). Positive CBM-I group (n ¼ 18)

STAI state STAI trait BDI SCL-90 PANAS PA scale PANAS NA scaleb

50%/50% control group (n ¼ 18)

Pre-assessment

Post-assessment

Within-group effect sizea

FU (n ¼ 14)

Pre-assessment

Post-assessment

Within-group effect sizea

FU (n ¼ 16)

40.6 (13.5) 50.7 (12.5) 17.2 (9.4) 190.6 (61.4) 29.2 (9.3) 18.3 (8.1) 21.2 adj. M

41.1 (10.1) 46.7 (10.1) 12.9 (8.3) 177.8 (62.5) 28.4 (8.1) 17.1 (6.7) 19.0 adj. M

0.04 0.35 0.49 0.21 0.09 0.16

44.6 (12.2) 46.0 (10.1) 13.4 (10.6) 179.7 (70.9) 32.2 (8.5) 18.9 (8.0) 20.7 adj. M

48.8 (9.6) 54.3 (10.1) 20.0 (10.6) 207.5 (53.1) 24.8 (7.9) 21.7 (7.9) 20.1 adj. M

49.7 (12.3) 52.3 (9.9) 17.4 (12.2) 187.3 (56.9) 25.1 (9.1) 20.9 (6.7) 18.8 adj. M

0.08 0.20 0.23 0.37 0.04 0.11

51.8 (12.3) 51.7 (12.1) 14.9 (12.6) 187.3 (61.3) 23.3 (8.6) 20.3 (8.2) 18.6 adj. M

Note. FU ¼ follow up. PA ¼ positive affect and NA ¼ negative affect. a Within-group pre- to post-training effect sizes ¼ (pre-assessment mean - post-assessment mean)/pooled standard deviation. b As the difference in state anxiety before the eight day program was related to change in PANAS negative scale, we also report adjusted means (adj. M).

interpretations, t(16) ¼ 6.34, p < .001 and t(13) ¼ 3.81, p < .01 respectively, but this was more marked in the positive CBM-I group (Mpositive interpretations ¼ 2.58, SD ¼ 0.42 vs. Mnegative interpretations ¼ 1.71, SD ¼ 0.50) than in the control group (Mpositive interpretations ¼ 2.42, SD ¼ 0.42 vs. Mnegative interpretations ¼ 2.01, SD ¼ 0.33). Note that training effects did not interact with Target type. This suggests that more general mechanisms might be involved such as emotional priming with CBM-I increasing the accessibility of an entire category of emotionally positive material or that general positive responses were trained. 2.5.3. Clinical outcomes and mood Table 2 gives means and standard deviations of the clinical assessments for the positive CBM-I and placebo control group. A twoway mixed model ANOVA2 with group (positive CBM-I vs. control condition) as the between-subjects factor and time (before vs. directly after) as the within-subject factor was performed on the clinical outcome measures. To examine the stability of the effects on the clinical measures, a subsequent two-way ANCOVA2 was conducted with group as the between-subjects factor and time (postassessment vs. three-months follow-up) as the within-subject factor and pre-assessment state anxiety scores added as a covariate. Regarding effects on trait anxiety, a significant main effect of time appeared, F(1, 34) ¼ 6.08, p < .05, h2p ¼ .15. Irrespective of group assignment, trait anxiety scores decreased from pre- to postassessment. The predicted Group  Time interaction effect was not significant, F(1, 34) ¼ 0.63, p ¼ .43, h2p ¼ .02. Analysis of the stability of the reduction of trait anxiety revealed no significant effects regarding the factors Group and Time, F’s (1, 27) < 0.32, p’s > .58, h2p’s < .01, indicating that there was no significant change in trait anxiety from post-assessment to follow-up. The predicted interaction effect of group by time was also not significant for state anxiety scores, F(1, 34) ¼ 0.01, p ¼ .98, h2p < .001. That is, the positive and control condition did not differ significantly in change of state anxiety. A significant main effect of group was observed, F(1, 34) ¼ 6.38, p < .05, h2p ¼ .16 with the control group being more state anxious than the positive CBM-I group. No significant effects regarding the factors Group and Time were observed from postassessment to follow-up, F’s (1, 27) < 1.32, p’s > .26, h2p’s < .05. Regarding effects on depression, a significant main effect of time appeared, F(1, 34) ¼ 15.05, p < .001, h2p ¼ .31, indicating less depressive symptoms following the eight day program. This was not qualified by a significant Group  Time interaction effect, F(1, 34) ¼ 0.93, p ¼ .34, h2p ¼ .03. No significant effects regarding the factors Group and Time were observed from post-assessment to follow-up, F’s (1, 27) < 1.35, p’s > .26, h2p’s < .05, suggesting that the decrease in depressive symptoms was maintained at follow-up. For levels of psychological distress (SCL-90), analyses revealed a significant main effect of time, F(1, 34) ¼ 10.35, p < .01, h2p ¼ .23,

reflecting a general decrease in psychopathological symptoms, irrespective of group. No other effects were significant, including the predicted Group  Time interaction effect, F(1, 34) ¼ 0.53, p ¼ .47, h2p ¼ .02. There were no significant effects regarding the factors Group and Time from post-assessment to follow-up, F’s (1, 27) < 0.25, p’s > .62, h2p’s < .009. Analyses of the PANAS positive subscale data revealed no significant effects, including the Group  Time interaction effect, F(1, 34) ¼ 0.15, p ¼ .70, h2p ¼ .004. Examination of change in positive mood from post-assessment to follow-up revealed no significant effects regarding the factors Group and Time, F’s (1, 27) < 02.66, p’s > .12, h2p’s < .09. The scores on PANAS negative subscale were entered into an ANCOVA with pre-assessment state anxiety as a covariate. The assumption of homogeneity of regression slopes was met and results revealed a significant main effect for the covariate, F(1, 33) ¼ 11.88, p < .01, h2p ¼ .27, reflecting a positive relationship between preassessment state anxiety and negative mood state. No other results were significant, including the Group  Time interaction effect, F(1, 33) ¼ 0.75, p ¼ .39, h2p ¼ .02, thus showing that the groups did not differ significantly in change in negative mood state. No significant changes were observed regarding the factors Group and Time from post-assessment to follow-up, F’s (1, 27) < 0.51, p’s > .48, h2p’s < .02. 2.5.4. Exit interview Analysis of the results from the exit interview revealed that there was no difference between the groups in their ideas about the treatment they had received, c2(1, N ¼ 35) ¼ 0.24, p ¼ .71. Five participants in the positive CBM-I and four in the control group thought they had followed the positive CBM-I condition. As a next step, it was examined whether their belief about the received training was related to the dependent variables by adding this variable as a covariate to the analyses. Results showed that such a belief was neither related to the interpretation bias measures (reaction times, F(1, 32) ¼ 0.02, p ¼ .90, h2p ¼ .001; recognition task, F(1, 27) ¼ 0.10, p ¼ .75, h2p ¼ .004), nor to the dependent mood variables, F’s (1, 32) < 1.72, p ¼ .20, h2p ¼ .05. 2.6. Discussion of Study 1 Study 1 revealed that after an online eight-day CBM-I training program to interpret ambiguous information in a more positive manner, patients interpreted new ambiguous information less negative than patients who had received the placebo-control condition. While the interpretation bias data revealed the predicted differences between the two groups, we observed no difference between the two groups in reported change in emotions. Both groups showed a decrease in trait anxiety, depressive mood, and psychological distress irrespective of the training. This decrease was maintained at follow-up. The finding that both groups showed significant

192

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

reductions in self-reported symptoms with no superiority for the CBM condition has also been observed in other CBM studies (Amir & Taylor, 2012; Boettcher, Berger, & Renneberg, 2012; Carlbring et al., 2012). A post-hoc explanation for the change in symptoms is the role of general non-specific factors (support from researchers, participating in an academic clinical study, regression to the mean, hope of improvement, etc.); non-specific factors account. An alternative possibility is that not only the positive CBM-I training, but also the 50%/50% placebo-control condition has led to more positive interpretations and as a consequence to reductions in anxiety, depression, and psychological distress. At post-assessment, both groups made more positive than negative interpretations (though this was more marked in the positive group). As patients with anxiety disorders have a tendency to interpret ambiguous situations negatively (Ouimet, Gawronski, & Dozois, 2009), it seems likely that they started the experiment with a negative interpretation bias. Suppose that they are naturally inclined to interpret 90% of the ambiguous scenarios negatively. The “control” condition consisted of 50% positive and 50% negative interpretations and might thus have served to disconfirm spontaneous negative interpretations and enhance positive interpretations. Patients might have learnt that, at least in the present context, positive outcomes of ambiguous scenarios are more common than assumed. Thus, an alternative explanation (diluted training account) is that the 50%/50% placebo-control condition might have functioned as a diluted version of training resulting in more positive and less negative interpretations and, as a consequence, in reduced levels of anxiety, depression, and psychological distress. To test this possibility and clarify the findings of Study 1, an additional study was conducted. 3. Study 2 3.1. Introduction Study 2 was designed to clarify the findings of Study 1 and to evaluate two alternative explanations (diluted training account versus non-specific effects account). This study consisted of a 100% neutral control condition that resembled the two previous conditions in that participants read short scenarios and completed word fragments and comprehension questions. However, this time, the scenarios described neutral, non-emotional situations and no valenced interpretations were to be generated. We therefore expected that interpretations would remain unaffected in this 100% neutral condition. Table 3 Sociodemographic and clinical characteristics of participants Study 2. Neutral placebo control group n ¼ 19 Age, mean (SD) Women Marital status Single Relationship, not cohabitating Cohabitating Married Divorced Education: exams passed University e graduate level University e undergraduate level High school Middle school Elementary school No. of Axis I diagnoses, mean (SD) Duration of diagnoses, mean (SD), in years Years of treatment, mean (SD) Psychotropic medication use

33.4 (9.3) 11 6 1 8 3 1 6 2 1 10 0 2.1 (1.0) 8.1 (6.6) 2.2 (2.9) 13

Note. Data is given as number of individuals, except where is indicated that data is given as means with standard deviation in parentheses.

When directly comparing the results from the neutral group to the positive CBM-I and 50%/50% control condition from Study 1, different predictions were formulated based on the two alternative accounts. According to the diluted training account, participants who received the neutral control condition would interpret ambiguity more negative and less positive compared to Study 1’s positive CBM-I and 50%/50% condition. Furthermore, participants from this neutral control group would feel more anxious, depressed, and distressed than participants from Study 1’s groups. Alternatively, according to the non-specific effect account, participants from the neutral control group would e just like the 50%/50% group e interpret ambiguity more negatively and less positively than participants from the positive CBM-I group and no differences between the groups could be expected on emotional outcomes. 3.2. Participants Patients were recruited at Altrecht Academic Anxiety Centre, Utrecht, the Netherlands. The inclusion and exclusion criteria for Study 2 were the same as for Study 1 as well as the criteria regarding psychotropic medication. In total, 23 patients were allocated to the neutral condition. Ten patients met diagnostic criteria for PD with or without agoraphobia, six met diagnostic criteria for SAD, two for GAD, and one for PTSD. On average, participants had 2.1 (SD ¼ 1.0) Axis 1 diagnoses and comorbid Axis 1 diagnoses consisted of other anxiety disorders, depressive disorders (either current or in remission), trichotillomania, and ADHD. One participant did not receive the intervention, as he/she decided to stop directly after the pre-training assessment. Three participants were lost during training, resulting in 19 patients who successfully completed the neutral condition. Two participants did not return the follow-up questionnaires, resulting in a sample of 17 patients for the follow-up data. 3.3. Design All participants were allocated to the neutral control condition. Data were collected at the same time points as in Study 1. The medical ethics committee approved this second study and data collection was carried out from May 2008 to January 2010. 3.4. Materials Participants in the neutral condition received the same number of scenarios as participants in Study 1. However, in this condition no positive or negative emotional scenarios were presented, except for the probes scenarios. The neutral scenarios consisted of the filler scenarios from Study 1 and newly created neutral scenarios. An example is: You arrange to meet a friend at your local pub one evening. As you arrive, you cannot help noticing that the sign in front has been . pa-nt-d (painted) Has your pub changed it appearance? The manipulation checks, outcome measurements, exit interview and experimental software were identical to the ones in Study 1. 3.5. Results Study 2 3.5.1. Between-group differences at baseline When comparing sociodemographic characteristics between the neutral control group from Study 2 and the positive CBM-I and

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

control group from Study 1, no significant differences in sex, marital status, and educational level appeared (Table 3). However, groups differed significantly in age, F(2, 55) ¼ 3.54, p < .05, h2p ¼ .12; participants who had followed the neutral control group were significantly younger than participants why followed the positive CBM-I training (the two control groups did not differ significantly). The three groups did not differ in number or duration of Axis 1 diagnoses, nor in years of treatment or psychotropic medication use. Furthermore, participants in the three groups did not differ significantly for scores on the pre-assessment questionnaires. 3.5.2. Interpretation bias outcome measures Latencies to complete probe trials were again set aside if the response to the fragment (1.7%) or the corresponding comprehension question (2.9%) was incorrect, or if the latency was less than 200 ms (0%) or greater than three SDs above the individual mean (1.9%). A paired samples t-test revealed that that individuals in the neutral control group did not differ significantly in their responses to positive and negative probes, t(18) ¼ 0.21, p ¼ .98 (Mpositive probes ¼ 1457, SD ¼ 569 vs. Mnegative probes ¼ 1457, SD ¼ 546). A 3 (Group)  2 (Probe) Repeated Measures ANOVA2 was conducted on the reaction time data to examine the CBM-I effects on interpretation bias. The significant main effect of Probe, F(1, 52) ¼ 23.9, p < .001, h2p ¼ .32, was qualified by a significant Group  Probe interaction effect, F(2, 52) ¼ 17.6, p < .001, h2p ¼ .40. While the two control conditions did not differ significantly from each other in responding to positive, t(35) ¼ 0.05, p ¼ .96, or negative probes, t(35) ¼ 0.07, p ¼ .94, both groups differed significantly from the positive CBM-I group in responses to negative probes, t’s > 2.15, p’s < .05. That is, the positive CBM-I group was disproportionately slow in responding to negative probes. A 3 (Group)  2 (Valence: positive vs. negative)  2 (Target: possible interpretation vs. foil sentence) Repeated Measures ANOVA2 was conducted to examine the effects on the second interpretation bias measure, the recognition task. A significant main effect of Valence, F(1, 47) ¼ 67.9, p < .001, h2p ¼ .59, and of Target, F(1, 47) ¼ 93.6, p < .001, h2p ¼ .67, was found. In addition, a significant Group  Valence interaction effect, F(2, 47) ¼ 5.9, p < .01, h2p ¼ .20, was observed. Follow-up analyses revealed that the two control conditions did not differ in their endorsement of positive, t(31) ¼ 0.17, nor negative interpretations, t(31) ¼ 0.07 (neutral condition: Mpositive interpretations ¼ 2.39, SD ¼ 0.38 vs. Mnegative interpretations ¼ 2.02, SD ¼ 0.31), while both groups differed from the positive CBM-I training in endorsement of negative interpretations (neutral group: t(34) ¼ 2.22, p < .05; 50%/50% group: t(29) ¼ 1.90, p ¼ .06). That is, patients who had followed the positive CBM-I training gave lower similarity ratings to the negative interpretations than patients from the two control groups. In sum, both control groups are comparable regarding their interpretations

Table 4 Mean scores of the outcome measures (SD in parentheses) for the neutral placebo control group (Study 2). Neutral placebo control group (n ¼ 19) Pre-assessment Post-assessment Within-group FU (n ¼ 17) effect sizea STAI state 48.9 (13.2) STAI trait 52.2 (7.7) BDI 18.7 (8.5) SCL-90 190.2 (42.8) PANAS PA scale 24.2 (6.5) PANAS NA scale 24.6 (7.2)

46.2 49.5 18.6 180.1 25.7 17.7

(12.4) (9.2) (10.5) (48.7) (7.2) (6.5)

0.21 0.32 0.01 0.22 0.22 1.01

46.6 47.9 15.3 174.0 26.6 20.2

(14.9) (12.2) (12.4) (72.5) (9.1) (8.0)

Note. FU ¼ follow up. PA ¼ positive affect and NA ¼ negative affect. a Within-group pre- to post-training effect sizes ¼ (pre-assessment mean - postassessment mean)/pooled standard deviation.

193

of ambiguity, and the positively trained CBM-I group interpreted ambiguity less negatively. 3.5.3. Clinical outcomes and mood To examine the effects on the clinical measures, 3 (Group)  2 (Time) Repeated Measures ANOVA2’s were conducted, first with the within-subject factor time consisting of pre- versus postassessment scores, and second with this factor consisting of postassessment versus three-months follow-up scores (Table 4). Analyses of trait anxiety scores revealed a significant main effect of time (T1 vs. T2), F(1, 50) ¼ 9.68, p < .01, h2p ¼ .16. Irrespective of group, there was a decline in trait anxiety scores. The Group  Time interaction effect was not significant, F(2, 50) ¼ 0.38, p ¼ .69, h2p ¼ .02, and there were no significant effects when analysing trait anxiety scores from post-assessment to follow-up, F’s < 1.83, p’s > .17, h2p’s < .08. Analyses of state anxiety scores revealed no significant effects, including the Group  Time (T1 vs. T2) interaction effect, F(2, 52) ¼ 0.67, p ¼ .52, h2p ¼ .03, and T2 vs. T3 analyses, F’s < 1.89, p’s > .16, h2p’s < .08. Depression scores decreased significantly from pre- to postassessment (main effect Time, F(1, 52) ¼ 9.19, p < .01, h2p ¼ .15) and this did not differ significantly between the groups (Group  Time interaction effect was not significant, F(2, 52) ¼ 2.54, p ¼ .09, h2p ¼ .09). No significant effects emerged when comparing post-assessment to follow-up depression scores, F’s < 2.0, p’s > .17, h2p’s < .04. For levels of psychological distress (SCL-90 scores), the pattern of results was comparable to the pattern in trait anxiety and depression. That is, a significant main effect of Time (T1 vs. T2) was observed, F(1, 50) ¼ 15.2, p < .001, h2p ¼ .23, indicating a decrease in level of distress, irrespective of group. This effect was not qualified by a significant Group  Time interaction effect, F(2, 50) ¼ 0.57, p ¼ .57, h2p ¼ .02. There were no significant effects from postassessment to follow-up, F’s < 1.23, p’s > .30, h2p’s < .06. Analyses of the pre- and post-assessment scores on the PANAS positive scale revealed no significant effects, including the Group  Time interaction effect, F(2, 52) ¼ 0.43, p ¼ .65, h2p ¼ .02. A significant main effect of group emerged when examining postassessment scores to follow-up scores, F(2, 44) ¼ 3.63, p < .05, h2p ¼ .14; patients who had followed the positive CBM-I training felt more positive at post and follow-up assessment than patients who had followed either control condition. Regarding negative mood (PANAS negative scale), a significant main effect of Time was observed, F(1, 52) ¼ 13.5, p ¼ .001, h2p ¼ .21, that was qualified by a significant Group  Time interaction effect, F(2, 52) ¼ 5.9, p < .01, h2p ¼ .19. There was a general decrease in negative mood and the strongest reduction occurred in the neutral control group. No significant effects emerged when analysing negative mood scores from post-assessment to follow-up, F’s < 0.47, p’s > .63, h2p’s < .02. 3.6. Discussion Study 2 consisted of a 100% neutral control group to clarify the unexpected findings of Study 1 and to test two alternative explanations. We clearly found that the neutral control condition from Study 2 resembled the control condition from Study 1 in both interpretation style and changes in emotionality, thus providing support for the non-specific factors account. 4. General discussion The aim of the current studies was to examine the therapeutic effectiveness of an internet-based CBM-I training in a clinical sample of patients with a broad range of anxiety disorders. Study 1

194

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

compared an eight-day positive CBM-I training with a placebo control condition. As predicted, individuals who received the positive CBM-I training interpreted ambiguous situations less negative compared to individuals who received the control condition. This suggests that the online training was successful in modifying interpretations, however, as there was no pre-training assessment of interpretive bias, the actual change in interpretive bias cannot be determined in the current study. Though note, that the observed post-training difference in interpretive bias was consistent with lab-/office-based studies revealing the pre- to post-CBM-I training change in interpretive bias in clinical samples (Amir & Taylor, 2012; Brosan et al., 2011). The current study is the first to examine CBM-I effects on interpretations using online training and across a range of different anxiety disorders, and the findings are suggestive, but not conclusive, of CBM-I affecting interpretations. Furthermore, we observed reductions in anxiety, depression, and psychological distress. This was, however, irrespective of group assignment. Thus in contrast to our hypothesis regarding emotional outcomes, there was no superior improvement in emotionality in the positive CBM-I condition. While this finding was unexpected based on the theoretical model and findings in analogue samples, it is, however, consistent with findings from some other CBM studies that also observed improvements in self-reported emotions in the control condition and no difference between the CBM and control group (Amir & Taylor, 2012; Boettcher et al., 2012; Carlbring et al., 2012). We hypothesised that, in the current study, this could be due to general non-specific factors (non-specific factors account), but could also be due to the 50%/50% placebo control condition promoting more positive interpretations in the clinical sample (diluted training account). Therefore, a second study, consisting of a neutral control group, was designed to clarify Study 1’s findings and test both accounts. It revealed that the 100% neutral control condition resembled the pattern of data from Study 1’s control condition, thus ruling out the possibility that the emotional improvement in Study 1 was due to changes in interpretation bias (diluted training account). Instead, other factors might have produced the effects on emotionality. Such factors might be related to the study (e.g., positive expectations arising from participation in an academic research project, support and attention from the researcher, demand effects, placebo expectation effects), but might also be independent of the study (e.g., spontaneous remission, expectation of the coming treatment as usual, regression to the mean). To disentangle these two types of factors, future studies should include an assessment-only control condition as this would provide information regarding the effects of being on a waiting list. The fact that in the current study CBM-I training was offered online might have influenced the results as well. Note that it is unlikely that this fully explains the current findings, as our online CBM-I training seem to have affected interpretations, and there are off-line CBM studies that also failed to observe superior effects of CBM on self-reported anxiety and depression symptoms (Amir & Taylor, 2012; Hayes et al., 2010). While the advantage of online (CBM) treatments is reduced barriers for patients who are unable or unwilling to visit a clinic or university, the disadvantage is less control over the procedures and setting. As training is performed at home, it is difficult to ensure a standard test and training environment. However, when looking at the percentage of incorrect responses given to the comprehension question, there were no differences between the current study and a lab-based CBM-I training (see Hayes et al., 2010, if anything, the current study had higher accuracy rates). In the same vein, the present CBM-I manipulation was superior in changing interpretations. It appears, therefore, that the fact that the training was internetdelivered is an unlikely explanation for the lack of differences between the experimental and control groups.

The current pattern of findings is in contradiction with cognitive models of psychopathology (Beck et al., 1985; Williams et al., 1988). In these models, it is argued that the tendency to interpret ambiguity in a negative way is causally related to anxiety. The observed difference between the experimental condition and the control groups in interpretive bias following training should, based on these models, be associated with concomitant change in anxiety. However, in the current study, no effect on anxiety and other emotional measures were observed over and above the effects of the control groups. Evidently, future studies are necessary to examine the robustness of these findings. If they are replicated, this should lead to a critical evaluation of the explanatory value of those early cognitive models (Beck et al., 1985; Williams et al., 1988). This could be accompanied by examining other factors that might play a role. Note that more recent cognitive theories (Mathews & Mackintosh, 1998) argue that not only bottom-up emotional activation (threat-evaluation system), but also top-down regulatory control plays an important role in anxiety (for supporting empirical data, see Salemink, Friese, Drake, Mackintosh, & Hoppitt, 2013; Salemink & Wiers, 2012). Based on this, it could be argued that interpretive bias only results in anxiety when the activation of the affective system exceeds the capacity for control. In other words, CBM-I training might be more effective in individuals with low levels of regulatory control (see Salemink & Wiers). In summary, future studies are necessary to examine the replicability of the current findings and also test recent theoretical models regarding other cognitive processes, such as regulatory control, in anxiety. Some study limitations should be acknowledged. The sample size is relatively small and reduced our ability to draw clear inferences about the absence of emotional change in light of the cognitive change. However, it is comparable to sample sizes of previous CBM-I studies (Beard et al., 2011; Brosan et al., 2011; Hayes et al., 2010; Mathews & Mackintosh, 2000). A second limitation is the lack of a baseline assessment of interpretive bias. Future studies should include a pre-training interpretive bias assessment to be able to examine change in bias. In addition, the study lacks a measure of credibility and expectancy and including such a measure would be helpful to be able to examine placebo expectation effects. A third point is related to the stimuli types used (social) and the current sample (heterogeneous group of anxiety disorder patients). The scenarios were not matched to the specific concerns of all the patients. This might have affected the results given the content specificity of interpretive biases (Clark et al., 1988; Voncken, Bögels, & de Vries, 2003) and might explain why there was a change in interpretive bias, but not in emotionality (Mackintosh, Mathews, Eckstein, & Hoppitt, 2013). However, as earlier CBM-I studies in anxious individuals suggested that the CBM-I effects were rather general and independent of the precise concerns of the individual (Mathews et al., 2007; Salemink et al., 2009), general social scenarios were used in the current study. It might, however, be that in clinical populations, it is more important to have a match between content of the training and the concerns of the population and/or intended emotional change. A fourth point is related to the temporal separation of the training sessions (see also Macleod et al., 2009). The current study consisted of eight training sessions to be completed in an 11 day period. While such a massed training has shown to impact upon interpretations and emotions (Salemink et al., 2009), other CBM-I studies have used more spaced training where participants completed training sessions only once or twice per week (Amir & Taylor, 2012; Beard et al., 2011; Brosan et al., 2011). More spaced training might provide more opportunities to practice the new ways of interpreting information in real life. A possibility is that in a clinical sample, interpretations can be modified in eight days, as the current studies revealed, but that further or more spaced repetition over time is needed to affect

E. Salemink et al. / J. Behav. Ther. & Exp. Psychiat. 45 (2014) 186e195

symptoms and emotions. This is an empirical issue and future studies might compare massed versus spaced CBM training. Finally, a limitation of the current study is the variability in (not) receiving treatment as usual (and the type treatment and number of completed sessions) during the follow-up assessment. In sum, the current randomised controlled trial in patients with anxiety disorders revealed that internet-delivered positive CBM-I training was associated with less negative interpretations, however it did not outperform the control conditions regarding effects on anxiety, depression, and levels of psychological distress. Report of these null findings is important to increase knowledge regarding the effectiveness of online CBM-I training and reduce file-drawer problems that easily occur with novel and theoretically intriguing interventions. CBM-I has been described as a potential new treatment for anxiety disorders, however, the present findings suggest that more research is necessary before CBM-I training can be disseminated. Acknowledgements The work was carried out at the Department of Clinical and Health Psychology, Utrecht University, the Netherlands. We are very grateful to the patients of our study. We want to thank Nellie Buurman and Erline Rood for their valuable help with patient recruitment, assessments, and data entry. We thank Saskia Righart and Hellen Hornsveld for their help in patient recruitment at Altrecht Academic Anxiety Centre and Mesos Medical Centre, respectively. Martin Laverman is thanked for his help in developing the internet-based computer program. This study was not supported by a grant and none of the authors have financial interests to disclose. References Amir, N., Bomyea, J., & Beard, C. (2010). The effect of single-session interpretation modification on attention bias in socially anxious individuals. Journal of Anxiety Disorders, 24, 178e182. Amir, N., Foa, E. B., & Coles, M. E. (1998). Negative interpretation bias in social phobia. Behaviour Research and Therapy, 36, 945e957. Amir, N., & Taylor, C. T. (2012). Interpretation training in individuals with generalized social anxiety disorder: a randomized controlled trial. Journal of Consulting and Clinical Psychology, 80, 497e511. Arrindell, W. A., & Ettema, J. H. M. (1986). SCL-90: Handleiding bij een multidimensionele psychopathologie-indicator (SCL-90: Manual for a multidimensional indicator of psychopathology). Lisse: Swets & Zeitlinger. Baert, S., Koster, E. H. W., & De Raedt, R. (2011). Modification of information processing biases in emotional disorders: clinically relevant developments in experimental psychopathology. International Journal of Cognitive Therapy, 4, 208e222. Beard, C. (2011). Cognitive bias modification for anxiety: current evidence and future directions. Expert Review of Neurotherapeutics, 11, 299e311. Beard, C., & Amir, N. (2008). A multi-session interpretation modification program: changes in interpretation and social anxiety symptoms. Behaviour Research and Therapy, 46, 1135e1141. Beard, C., Weisberg, R. B., & Amir, N. (2011). Combined cognitive bias modification treatment for social anxiety disorder: a pilot trial. Depression and Anxiety, 28, 981e988. Beck, A. T., Emery, G., & Greenberg, R. L. (1985). Anxiety disorders and phobias: A cognitive perspective. New York: Basic Books. Beck, A. T., Rush, A. J., Shaw, B. F., & Emery, G. (1979). Cognitive therapy of depression. New York: Guilford Press. Boettcher, J., Berger, T., & Renneberg, B. (2012). Internet-based attention training for social anxiety: a randomized controlled trial. Cognitive Therapy and Research, 36, 522e536. Brosan, L., Hoppitt, L., Shelfer, L., Sillence, A., & Mackintosh, B. (2011). Cognitive bias modification for attention and interpretation reduces trait and state anxiety in anxious patients referred to an out-patient service: results from a pilot study. Journal of Behavior Therapy and Experimental Psychiatry, 42, 258e264. Butler, G., & Mathews, A. (1983). Cognitive processes in anxiety. Advances in Behaviour Research and Therapy, 5, 51e62. Carlbring, P., Apelstrand, M., Sehlin, H., Amir, N., Rousseau, A., Hofmann, S. G., et al. (2012). Internet-delivered attention bias modification training in individuals with social anxiety disorder-a double blind randomized controlled trial. BMC Psychiatry, 12, 66. Clark, D. M., Salkovskis, P. M., Gelder, M. G., Koehler, C., Martin, M., Anastasiades, P., et al. (1988). Tests of a cognitive theory of panic. In I. Hand, & H. U. Wittchen (Eds.), Panic and phobias II. New York: Springer-Verlag.

195

Clark, D. M., Salkovskis, P. M., Ost, L. G., Breitholtz, E., Koehler, K. A., Westling, B. E., et al. (1997). Misinterpretation of body sensations in panic disorder. Journal of Consulting and Clinical Psychology, 65, 203e213. Derogatis, L. R. (1977). SCL-90: Administration, scoring and procedures manual-I for the r(evised) version. Baltimore: Johns Hopkins University School of Medicine, Clinical Psychometrics Research Unit. Dunmore, E., Clark, D. M., & Ehlers, A. (2001). A prospective investigation of the role of cognitive factors in persistent Posttraumatic Stress Disorder (PTSD) after physical or sexual assault. Behaviour Research and Therapy, 39, 1063e1084. First, M. B., Spitzer, R. L., Gibbon, M., & Williams, J. B. W. (1996). Structured Clinical Interview for DSM-IV Axis I Disorders-Patient Edition (SCID-I/P). Version 2.0. New York: Biometrics Research Department, New York State Psychiatric Institute. van Groenestijn, M. A. C., Akkerhuis, G. W., Kupka, R. W., Schneider, N., & Nolen, W. A. (1999). Gestructureerd Klinisch Interview voor de vaststelling van DSM-IV As 1 Stoornissen [Dutch translation of the SCID-I/P]. Lisse, the Netherlands: Zwets & Zeitlinger. Hallion, L. S., & Ruscio, A. M. (2011). A meta-analysis of the effect of cognitive bias modification on anxiety and depression. Psychological Bulletin, 137, 940e958. Hayes, S., Hirsch, C. R., Krebs, G., & Mathews, A. (2010). The effects of modifying interpretation bias on worry in generalized anxiety disorder. Behaviour Research and Therapy, 48, 171e178. Hedman, E., Andersson, G., Ljótsson, B., Andersson, E., Rück, C., Mörtberg, E., et al. (2011). Internet-based cognitive behavior therapy vs. cognitive behavioral group therapy for social anxiety disorder: a randomized controlled noninferiority trial. PLoS One, 6, e18001. Hirsch, C. R., Hayes, S., & Mathews, A. (2009). Looking on the bright side: accessing benign meanings reduces worry. Journal of Abnormal Psychology, 118, 44e54. Koster, E. H., Fox, E., & MacLeod, C. (2009). Introduction to the special section on cognitive bias modification in emotional disorders. Journal of Abnormal Psychology, 118, 1e4. Lange, A., van der Ven, J., Schrieken, B., & Emmelkamp, M. G. (2001). Interapy. Treatment of posttraumatic stress through the Internet: a controlled trial. Journal of Behavior Therapy and Experimental Psychiatry, 32, 73e90. Mackintosh, B., Mathews, A., Eckstein, D., & Hoppitt, L. (2013). Specificity effects in the modification of interpretation bias and stress reactivity. Journal of Experimental Psychopathology, 4, 133e147. Macleod, C., Koster, E. H. W., & Fox, E. (2009). Whither cognitive bias modification research? Commentary on the special section articles. Journal of Abnormal Psychology, 118, 89e99. Macleod, C., Soong, L. Y., Rutherford, E. M., & Campbell, L. W. (2007). Internetdelivered assessment and manipulation of anxiety-linked attentional bias: validation of a free-access attentional probe software package. Behavior Research Methods, 39, 533e538. Mathews, A., & Mackintosh, B. (1998). A cognitive model of selective processing in anxiety. Cognitive Therapy and Research, 22, 539e560. Mathews, A., & Mackintosh, B. (2000). Induced emotional interpretation bias and anxiety. Journal of Abnormal Psychology, 109, 602e615. Mathews, A., Ridgeway, V., Cook, E., & Yiend, J. (2007). Inducing a benign interpretational bias reduces trait anxiety. Journal of Behavior Therapy and Experimental Psychiatry, 38, 225e236. Ouimet, A. J., Gawronski, B., & Dozois, D. J. A. (2009). Cognitive vulnerability to anxiety: a review and an integrative model. Clinical Psychology Review, 29, 459e 470. Salemink, E., Friese, M., Drake, E., Mackintosh, B., & Hoppitt, L. (2013). Indicators of implicit and explicit social anxiety influence threat-related interpretive bias as a function of working memory capacity. Frontiers in Human Neuroscience, 7, 220. Salemink, E., & van den Hout, M. A. (2010). Validation of the “recognition task” used in training of interpretation biases. Journal of Behavior Therapy and Experimental Psychiatry, 41, 140e144. Salemink, E., van den Hout, M. A., & Kindt, M. (2007). Trained interpretive bias and anxiety. Behaviour Research and Therapy, 45, 329e340. Salemink, E., van den Hout, M. A., & Kindt, M. (2009). Effects of positive interpretive bias modification in highly anxious individuals. Journal of Anxiety Disorders, 23, 676e683. Salemink, E., & Wiers, R. (2012). Adolescent threat-related interpretive bias and its modification: the moderating role of regulatory control. Behaviour Research and Therapy, 50, 40e46. Spielberger, C. D., Gorsuch, R. L., Lushene, R., Vagg, P. R., & Jacobs, G. A. (1983). Manual for the State-Trait Anxiety Inventory. Palo Alto: CA: Consulting Psychologists Press. Steinman, S. A., & Teachman, B. A. (2010). Modifying interpretations among individuals high in anxiety sensitivity. Journal of Anxiety Disorders, 24, 71e78. Teachman, B. A., & Addison, L. M. (2008). Training non-threatening interpretations in spider fear. Cognitive Therapy and Research, 32, 448e459. Van der Ploeg, H. M., Defares, P. B., & Spielberger, C. D. (2000). Handleiding bij de Zelf-Beoordelings Vragenlijst. Een Nederlandstalige bewerking van de Spielberger State-Trait Anxiety Inventory [Manual for the State-Trait Anxiety Inventory]. Lisse, The Netherlands: Swets and Zeitlinger. Voncken, M. J., Bögels, S. M., & de Vries, K. (2003). Interpretation and judgemental biases in social phobia. Behaviour Research and Therapy, 41, 1481e1488. Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive and negative affect: the PANAS scales. Journal of Personality and Social Psychology, 54, 1063e1070. Williams, J. M., Watts, F. N., Macleod, C., & Mathews, A. (1988). Cognitive psychology and emotional disorders. Chichester, U.K.: John Wiley & Sons.