Background

Obsessive–compulsive disorder (OCD) is characterized by unwanted intrusive thoughts, images, or impulses that evoke anxiety or doubt, as well as compulsive actions or mental rituals whose primary purpose is to reduce or neutralize distress. Of the available treatments for OCD, cognitive-behavioral therapy (CBT), which includes exposure therapy with response prevention (ERP) with or without cognitive therapy (CT), has the greatest empirical support (Olatunji et al. 2013a, b; Öst et al. 2015). Recent studies also show that ERP with or without acceptance and commitment therapy (ACT) are equally, and highly effective treatments for OCD (Twohig et al. 2018). Randomized controlled trials indicate that, on average, patients achieve a 49% reduction in OCD symptoms from pre- to post-treatment with ERP, and 50–70% can expect clinically significant improvement (Öst et al. 2015). However, only 25% of patients are asymptomatic post-treatment (Fisher and Wells 2005). Thus, current treatments still need to be improved.

One way to optimize existing CBTs for OCD is to gain insights into potential mechanisms that are associated with improvement. Identifying these mechanisms and increasing the understanding of how they work can give us a better opportunity to tailor treatment to nonresponders. Over the past decades, there has been a growing interest in clinical studies aiming to elucidate the cognitive abnormalities that may cause the development and maintenance of OCD (Hezel and McNally 2016; Vogel and Solem 2017). Dysfunctional beliefs are the central elements in cognitive-behavioral conceptualizations of OCD, but the theoretical models differ in which beliefs and processes are considered to be most important in the development and persistence of the disorder. There are two main types of dysfunctional beliefs that have been examined in OCD: obsessive beliefs and metacognitive beliefs (Myers et al. 2009). Obsessive beliefs are derived from cognitive theories for OCD (e.g., Clark 2004; Frost and Steketee 1997; Rachman 1997, 1998; Salkovskis 1985), whereas metacognitive beliefs are concepts central to metacognitive theory (e.g., Wells 1997, 2000; Wells and Matthews 1994). One of the main distinctions between these models is that cognitive theory focuses on the content of thoughts and beliefs (e.g., “Having bad thoughts means I am weird”), whereas metacognition focuses on dysfunctional beliefs about thinking (e.g., “Worrying too much can be harmful to me”).

Cognitive theories differ with regard to what kinds of dysfunctional beliefs are thought to be most centrally associated with the development and maintenance of clinical obsessions. In an effort to develop and evaluate cognitive assessment strategies for OCD, an international research group extended theoretical work and identified three specific domains that are hypothesized to be central to the etiology and maintenance of OCD: (1) overestimates of threats and responsibility for harm, (2) perfectionism and intolerance for uncertainty, and (3) beliefs about the importance and need to control one’s thoughts (obsessive compulsive cognition working group: OCCWG 2005). There is considerable empirical support for the cognitive mediation hypothesis, which postulates that changes in dysfunctional beliefs cause a reduction in anxiety symptoms (Clark and Beck 2010). Research on mechanisms of change with repeated measurements in obsessive beliefs after CBT for OCD, however, has yielded mixed results. A few studies have reported that changes in one or more obsessive beliefs predicted changes in OCD symptoms to the next measurement point (Diedrich et al. 2016; Wilhelm et al. 2015), whereas other studies did not find that changes in obsessive beliefs precede changes in OCD symptoms (Levy et al. 2018; Olatunji et al. 2013a, b; Su et al. 2016). Thus, whether a reduction in obsessive beliefs mediates OCD symptom improvement is still an open question.

In contrast to cognitive models, the metacognitive model for OCD (Wells 1997, 2000; Wells and Matthews 1994) does not focus on the content of the appraisal but on individuals’ thinking about their own cognitive processes and metacognitive regulatory strategies. Intrusive thoughts or doubts are assumed to trigger underlying meta-beliefs, which in turn activate maladaptive thinking. According to Wells (1997, 2009), both dysfunctional cognitive and metacognitive beliefs are present in symptomatic OCD patients, but obsessive beliefs are considered byproducts of the metacognitive process and therefore add little to our understanding of the development of OCD.

Studies that have compared the relative contributions of obsessive beliefs and metacognition consistently find that metacognitive beliefs account for greater variance in OCD symptoms. This has been demonstrated in cross-sectional studies in non-clinical samples (Gwilliam et al. 2004; Myers and Wells 2005; Nance et al. 2018; Solem et al. 2010) and in prospective studies with both non-clinical (Myers et al. 2009) and clinical samples (Myers et al. 2017). In contrast to the numerous treatment studies that have examined mechanisms of change related to obsessive beliefs for OCD, only two studies have compared metacognitive beliefs with obsessive beliefs. Solem et al. (2009) found that changes in general metacognitive beliefs predicted post-treatment outcome over and above changes in obsessive beliefs in a sample of 83 OCD patients who received ERP. No further predictor analysis were conducted beyond post-treatment. However, there were high correlation among metacognitive beliefs at post-treatment and 12 month-follow-up, as well as among obsessive beliefs, indicating stability in the assumed mechanisms. This study was later replicated with a larger sample of 108 OCD patients who received a mixed treatment package consisting of behavioral, cognitive, and metacognitive components. Again, the result suggested that changes in metacognitive beliefs specific for OCD are better predictors of outcome than changes in obsessive beliefs (Grøtte et al. 2015).

Although the research thus far provides better support for the metacognitive model than for cognitive models, the few existing treatment studies (Grøtte et al. 2015; Solem et al. 2009) are traditional pre-/post-treatment studies that cannot provide any inference about the causes and bi-directional relations among the assumed mechanisms and outcome. Prospective studies with repeated measures could provide the opportunity to establish temporal precedence, that is, to examine whether changes in obsessive beliefs precede changes in OCD symptom severity, which would provide stronger evidence for causal inference (Kazdin 2007). No studies have yet investigated whether changes in dysfunctional cognitive or metacognitive beliefs are also related to OCD symptom improvement many years after treatment. From a longitudinal perspective, to our knowledge, there are no published reports investigating the clinical long-term effects of changes in obsessive beliefs beyond 2 years (Whittal et al. 2008) and in metacognitions beyond a 1 year follow-up (Solem et al. 2009). Knowledge about whether changes in obsessive beliefs or metacognitions predict decline in OCD symptoms in the long run is important for at least three reasons: (1) it can help to identify the specific long-term mechanisms responsible for treatment change and the strength of each processes’ contribution to symptom reduction, (2) it can identify whether changes in obsessive beliefs and/or metacognitions protect against relapse in OCD in the long term, and if so, (3) it can help us to better understand what to focus on in treatment for non-responders in the long term.

Another benefit of repeated measurements, in contrast to datasets from single points in time, is the opportunity to differentiate relationships that hold within persons as well as relationships that hold across persons (Curran and Bauer 2011). Studies focusing on changes in obsessive beliefs and metacognition for OCD represent traditional between-person relationships, that is, whether the patients change more or less in a given variable compared to the average of others in the same sample. The majority of psychological theories and most therapists in clinical practice, however, primarily focus on within-person relationships that inform on how the patient changes between occasions relative to their own status. Disaggregation of within- and between-person relationships yields two sources for understanding variance in treatment outcomes. The clarification of mechanisms at the within-person level may in turn lead to better outcomes when treating the individual patient (Hoffart and Johnson 2017). Only one study has thus far differentiated between- from within-person changes among dysfunctional cognitive and metacognitive beliefs. Within-person main effects of dysfunctional metacognition and cognition have been reported on anxiety symptoms in a clinical sample with mixed anxiety disorder (Johnson et al. 2018).

In the absence of similar studies for OCD treatment and a lack of knowledge of the long-term relationships between obsessive beliefs, metacognitions, and OCD symptoms, the purpose of the current study is twofold. The first was to explore the relative contribution of changes in obsessive beliefs and metacognitive beliefs to symptom improvement over the long term for OCD patients treated with group ERP. The second purpose was to investigate this from both a between- and within-person perspective. Specifically, we examined the following three hypotheses: (1) obsessive beliefs and dysfunctional metacognitions will decrease over the course of OCD treatment and follow-up, (2) changes in obsessive beliefs and metacognitions will predict changes in OCD symptom severity at both the within-person and between-person levels over the course of OCD treatment and follow-up, and (3) changes in metacognitions are a better predictor for changes in OCD symptom severity than changes in obsessive beliefs, both over the course of OCD treatment and at follow-up. Beyond these specific hypotheses, we also wanted to examine the reciprocal within- and between-person relationships that changes in OCD symptom severity could predict time-specific changes in metacognitions and obsessive beliefs over the course of OCD treatment and follow-up.

Method

Participants and Procedure

Sixty-five treatment-seeking individuals with a primary diagnosis of OCD completed a manualized group treatment ERP program for OCD in a general Norwegian outpatient clinic between 2003 and 2009. Fifty of the 65 patients involved in the present study were included in the original study (Haaland et al. 2010). Fifteen more patients were included later and underwent the same treatment, and inclusion/exclusion and assessment procedures as Haaland et al. (2010). As a small summary of previously described procedures, the patients were mainly referred by general practitioners and by means of referrals from several outpatient clinics. In addition, approximately 20% were self-referrals from newspaper advertisements. The exclusion criteria (i.e., ongoing psychotic disorder, ongoing alcohol or drug abuse/dependence, use of benzodiazepines, and high risk of suicide) were designed to match diagnosis-specific group treatment in a community clinic. If the participants met the initial study criteria after the clinical screening, they were administered the structural clinical interview (SCID-1; First et al. 1995) for diagnostic and statistical manual of mental disorder, fourth edition (DSM-IV; American Psychiatric Association 1994), and The Yale-Brown obsessive compulsive scale (Y-BOCS), by one of two independent evaluators that were not part of the study. All subjects met criteria for current DSM-IV OCD.

The group ERP for OCD was from a manual developed by Himle et al. (2001). The groups consisted of 6 patients and 2 therapists who met weekly for 12 weeks, with each session lasting 2.5 h. Session 1–6 involved two main components of interventions; psycho-education and ERP (in-session and home based ERP exercise), while session 7–12 consisted of ERP only. In addition to ERP, some cognitive strategies were also included in the intervention; (1) use of coping self-statements (e.g., “it's not me, it's my OCD”), and (2) externalizing OCD (e.g., “fight back against OCD that is trying to scare me”). Based on patient access, and therapist resources in the clinic, there were usually two groups conducted annually which meant that there was no selection based on predominant OCD symptom type (e.g., obsessions about contamination, checking, or obsessional thoughts about sex, religion, and violence). During the group ERP nine of 71 patients (12.7%) dropped out (Haaland et al. 2011) which is similar to rates reported in a recent meta-analysis (14.7%: Ong et al. 2016).

The original sample (N = 65) was then invited to participate in the present long-term (mean 8.23 years, ranging from 5 to 11 years, SD = 1.86) follow-up outcome study of the group ERP. Forty (61.5%) of the eligible participants took part in the current study. All participants signed an informed consent form. The gains from pre- to post-treatment were maintained throughout the 3 follow-up sessions. From pre-treatment to the extended follow-up, the mean OCD symptom reduction was 38% and the effect size was 1.5 (Cohens d). At the extended follow-up, 40% of the patients were defined as having recovered, and 10% as having improved, while 47.5% were defined as unchanged, and 2.5% were classified as deteriorated (Sunde et al. 2017). The present study was approved by the Regional Committee for Medical and Health Research Ethics.Footnote 1

Of the 40 participants in the current study, 9 (22.5%) were men and 31 (77.5%) were women. The age ranged from 18 to 63 years (M = 33.35, SD = 12.43) at intake. At pre-treatment OCD symptom severity had a mean of 23.15 (SD = 3.63), according to the Yale–Brown obsessive compulsive scale (Y-BOCS; Goodman et al. 1989a). When examining pre-treatment Y-BOCS severity levels, 22 (55%) participants had moderate OCD (Y-BOCS scores from 16 to 23), and 18 (45%) had severe/extreme OCD (Y-BOCS ≥ 24). In the time between post-treatment and the extended follow-up, 65% of the participants had received one or more additional treatments for OCD (e.g., additional ERP and/or medication). Further demographic information at the extended follow-up and details about the treatment received during the follow-up period for these 40 participants have been described elsewhere (Sunde et al. 2017).

Measures

Measures of OCD Symptoms

The Yale-Brown obsessive compulsive scale (Y-BOCS; Goodman et al. 1989a, b) was utilized to assess OCD symptom severity. Y-BOCS is a 10-item, semistructured interview that provides 5 subscores for obsessions and 5 for compulsions and a total severity score ranging from 0 to 40. The Y-BOCS has been documented as a reliable and valid tool (Taylor 1995).

Measures of Metacognitive Beliefs

The short form of the metacognitions questionnaire (MCQ-30; Wells and Cartwright-Hatton 2004) is designed to measure agreement with beliefs about thinking and processes that are central to the metacognitive model. The MCQ-30 is a 30-item self-report scale developed from the original 65-item version (Cartwright-Hatton and Wells 1997). The shorter MCQ-30 version is more suitable for frequent measurement and reduces the burden for patients in clinical practice. The questionnaire uses a 4-point Likert scale ranging from 1 (“do not agree”) to 4 (“agree very much”). Higher scores indicate more maladaptive metacognitions. The 2 versions both yield a total score and 5 dimensions of metacognition. In the MCQ-30 version, the five dimensions consist of 6 items each: (1) positive beliefs about worry (POS) (e.g., “Worrying helps me to avoid problems in the future”), (2) negative beliefs about the uncontrollability and danger of worry (NEG) (e.g., “My worrying is dangerous for me”), (3) beliefs about cognitive confidence (CC) (e.g., “I have little confidence in my memory for words and names”), (4) beliefs about the need to control thoughts (NC) (e.g., “I should be in control with thoughts all the time”), and (5) cognitive self-consciousness (CSC) (e.g., “I think a lot about my thoughts”).

The participants completed the longer version of the MCQ, from pre-treatment to the 12 month follow-up, and the shorter version at the extended follow-up. The protocols for the long version were scored in accordance with the 30-item version. The MCQ-30 shows excellent internal consistency, test–retest reliability and convergent validity (Wells and Matthews 1994). The Norwegian version of the MCQ-30 has demonstrated adequate internal consistency, validity and reliability in a clinical OCD sample measured before and after treatment (Grøtte et al. 2016). In the current study, the whole MCQ-30 had excellent internal consistency, while the 5 subscale scores varied from moderate to high internal consistency across all time points (MCQ-30 total score, all αs ≥ 0.92; MCQ-30 POS, all αs ≥ 0.87; MCQ-30 NEG, all αs ≥ 0.77; MCQ-30 CC, all αs ≥ 0.85; MCQ-30 NC, all αs ≥ 0.65; MCQ-30 CSC, all αs ≥ 0.85).

Measures of Obsessive Beliefs

The obsessive beliefs questionnaire-44 (OBQ-44; OCCWG 2005) measures agreement with cognitive beliefs that have been proposed as central to the development and maintenance of OCD. The OBQ-44 is a 44-item self-report that was developed from the original 87-item OBQ (OCCWG 2001,2003). The participants completed the longer version of the OBQ, from pre-treatment to 12-month follow-up, and the shorter version at the extended follow-up. The OBQ-44 is designed to measure three empirically derived belief domains of OCD. These subscales are labeled as follows: (1) responsibility/threat estimation (RT) (e.g., “Even if harm is very unlikely, I should try to prevent it at any cost”), (2) perfectionism/certainty (PC) (e.g., “Things should be perfect according to my own standards”), and (3) importance/control of thoughts (ICT) (e.g., “Having intrusive thoughts means I’m out of control”). The participants’ answers were rated on a 7-point scale (1 = “disagree very much” to 7 = “agree very much”). The importance/control of thoughts (ICT) subscale has been labeled as a metacognitive construct (OCCWG 1997) and has items that overlap with questionnaires in the MCQ-30. As a result, this subscale has been omitted in several studies comparing the relative contribution of obsessive beliefs and metacognition. However, like Solem et al (2009), we wanted to compare metacognition with a complete version of the cognitive model. The protocols for the long version were scored in accordance with the 44-item version. The OBQ-44 has demonstrated good internal consistency, reliability, discriminant and convergent validity (OCCWG 2005). There are no published psychometric reports about the Norwegian version of the OBQ-44, but the instrument has been used in several studies (e.g., Grøtte et al. 2015; Solem et al. 2009, 2010). In the current study, Cronbach’s alphas for the whole OBQ-44 and the three subscales ranged from high to excellent internal consistency (OBQ-44 total score, all αs ≥ 0.96; OBQ-44 RT, all αs ≥ 0.92; OBQ-44 PC; OBQ-44 ICT, all αs ≥ 0.86).

Data Analysis

The data were analyzed using hierarchic linear modeling (Fitzmaurice et al. 2004) since the data had a nested structure (repeated measures within-persons). Missing data were handled using maximum likelihood estimation (MIE). MIE is considered “state of the art” for handling missing data (Schafer and Graham 2002) and uses all of the available data, complete and incomplete, to identify the parameter values that have the highest probability of producing the sample data.

The data were disaggregated according to the procedures outlined by Wang and Maxwell (2015). Thus, the predictor variables, metacognition (MCQ-30 total score) and obsessive beliefs (OBQ-44 total score) were person-mean centered, thus giving a between-person score and a within-person score that was used in the subsequent analysis. Disaggregation for within- and between-effects is important in longitudinal data to avoid ecological fallacy (Curran and Bauer 2011), which is when results obtained for group-level data are misattributed to individuals. Time was not included as a fixed effect, as recommended by Falkenstrøm et al. (2017) in situations when the predictor is manipulated by treatment. Time was, however, used to see whether the results were robust to the removal of trends.

The models for OCD symptoms, metacognitions, and obsessive beliefs were developed from an “empty model”. The intraclass correlation (ICC) was calculated, giving an overview of within- and between-person variance. Random intercepts and slopes were then added. Different time structures such as quadratic and linear splines and linear time were tested. Visual inspection of obsessive beliefs and metacognition gave no indication of a linear spline model. Model fit was assessed using log likelihood tests, and the most parsimonious model was selected. A fixed linear time with a heteroscedastic residual variance over time was chosen, and the covariance structure of the residuals was modeled with AR1 for both the MCQ-30 total score and OBQ-44 total score. The MCQ-30 total score and OBQ-44 total score were modeled without random slopes to avoid convergence errors. A linear model was chosen for the Y-BOCS, so both the between- and within-person effects of metacognitions and obsessive beliefs could be investigated over the various assessment points.

First, we assessed whether metacognitions and obsessive beliefs changed through the different time points. Second, we investigated whether obsessive beliefs and metacognitions predicted OCD symptoms. To establish a temporal sequence between predictor and outcome, the predictor scores were lagged and thus related to the OCD symptoms at next time-point. Thus, lagged OBQ-44 total scores and MCQ-30 total scores predicting OCD symptoms at the within- and a between-person levels were investigated. The reciprocal relationship, investigating whether changes in MCQ-30 total scores or OBQ-44 total scores were driven by changes in OCD symptom severity, was also investigated. We entered the MCQ-30 total scores and OBQ-44 total scores into the equation to determine which of the predictors explained most of the variance in OCD symptoms. Correlation coefficients for OBQ-44 and MCQ-30 total scores and sub-scores was calculated for all five measurement occasions. Finally, differences between baseline characteristics, as well as OBQ-44 and MCQ-30 scores, among participants versus non-participants were analyzed. For continuous measures, the students t-test was used, for categorical variables with two levels Fishers exact test was used, for categorical variables with more than two levels, chi-square was used.

We wished to address several questions with the described multilevel framework. Given the exploratory nature of our inquiry, correction for multiple testing was not feasible, and therefore we proceeded fully aware of our low power.Footnote 2

Results

Follow-up Participants Compared with Non-participants, Characteristics, MCQ-30 and OBQ-44 Scores

It has previously been reported (Sunde et al. 2017) that the non-participant group (n = 25) had significantly higher Y-BOCS total score at pre-treatment (M = 25.83, SD = 5.02) compared to those who participated (n = 40) in the current study (M = 23.05, SD = 3.52). There were, however, no differences in Y-BOCS change scores after the acute group ERP or at 3 and 12 month follow-up. Further, there were no significant differences in gender and age at the extended follow-up between the groups.

Table 1 displays characteristics at pre-treatment for the participant and non-participant groups. The nonparticipant group had significantly more years with OCD before intake, compared with the participant group. The higher illness duration and baseline OCD symptom severity in the non-participant group might indicate that these patients were more impaired than the participant group since these 2 variables are frequently reported negative predictors for outcome (Knopp et al. 2013). There were, however, no other significant differences in age of OCD onset, years of education, employment status, marital status, and use of SSRI/SNRI. There were also no significant differences between the groups in relation to comorbidity. The pre-treatment comorbidity rate in the current study ranged between 55 and 63% that is in line with epidemiological studies for OCD (Brown et al. 2001; Ruscio et al. 2010). This indicates that the sample in the current study is representative of OCD patients in community clinics.

Table 1 Comparison of characteristics of follow-up participants vs. nonparticipants at intake
Full size table

A series of independent t tests was conducted to compare the pre- and post-treatment MCQ-30 and OBQ-44 total scores between the participants in the current long-term follow-up study and those who refused to participate. A missing value analysis showed that a total of 56.06% of the cases had incomplete data for the MCQ-30 and OBQ-44 at pre- and post-treatment. At the item level, this represented 11.27% missing data. Based on the observational method by Little and Rubin (2019) and Little’s MCAR test (p = 1.00) the data were assumed to be missing completely at random (MCAR). To reduce the effect of the missing data on the total scores, multiple imputation (MI) was applied using 50 imputed datasets, which were pooled after analysis (van Buuren 2018).

No significant differences between non-participants and participants were observed for the total scores for these 2 measurements, either at pre-treatment [MCQ-30 total score: M = 65.76 vs. M = 64.73; 95% CI − 7.99, 10.05; t(53.71) = 0.23; p = 0.82], [OBQ-44 total score: M = 164.40 vs. M = 150.40; 95% CI − 13.66, 41.66; t(54.77) = 1.01; p = 0.31] or at post-post-treatment [MCQ-30 total score: M = 55.19 vs. M = 57.55; 95% CI − 10.48, 5.77; t(62.56) = − 0.58; p = 0.56], [OBQ-44 total score: M = 118.60 vs. M = 141.82; 95% CI − 54.87, 8.43; t(54.04) = − 1.47; p = 0.15]; furthermore, there was no significant difference in the pre- to post-treatment score changes [MCQ-30 total score: M = 5.08 vs. M = 6.50; 95% CI − 7.76, 4.92; t(48.46) = − 0.45, p = 0.65], [OBQ-44 total score: M = 30.80 vs. M = 16.48; 95% CI − 5.33, 33.98; t(46.93) = 1.47; p = 0.15].

Correlational Analyses

Correlations among OBQ-44 and MCQ-30 total scores and subscale scores for all 5 measurements points were analyzed. Pre-treatment correlations are displayed in Table 2. Large correlations were found between the five OBQ-44 and MCQ-30 total scores (r = 0.59–0.91). Moderate to high correlations were found among most of the scores from the three OBQ-44 subscales and the five MCQ-30 subscales. Correlation between the metacognitive OBC-44 subscale Importance/control of thoughts (ICT) and the five MCQ-30 subscales ranged from: (1) positive beliefs about worry, POS (r = 0.39–0.58), (2) negative beliefs about the uncontrollability and danger of worry, NEG (r = 0.60–0.67), (3) beliefs about cognitive confidence, CC (r = 0.34–0.65), (4) beliefs about the need to control thoughts, NC (r = 0.57–0.79), and (5) cognitive self-consciousness, CSC (r = 19–58). These results are not surprising given the suspected association between MCQ-30 and OBQ-44. To further investigate the direction of this association we performed MLM.

Table 2 Pearson’s correlations among OBQ-44 and MCQ-30 total scores and subscale scores, pre-treatment
Full size table

Changes in Metacognitions and Obsessive Beliefs

We first computed the ICC score for metacognitions using a random effects ANOVA model. An ICC of 0.75 indicated that 75% of the total variability in the metacognitions scores was due to between-person differences; thus, MLM was suitable. There was a significant random intercept, indicating that the patients varied significantly in their starting point. Over the course of treatment and follow-up, metacognitions were significantly reduced [β = − 2.81, SE = 0.49, t(33.6) = − 5.77, p ≤0.001].

The ICC score for obsessive beliefs was 0.72; thus, 72% of the total variability in obsessive belief was due to between-person differences. Thus, we progressed with a multilevel model. There was a significant random intercept, indicating that the patient varied significantly in their starting point. Over the course of treatment and follow-up, obsessive beliefs were significantly reduced [β = − 6.85, SE = 2.24, t(39.98) = − 3.06, p < 0.01].

Table 3 presents a descriptive overview of observed mean scores with standard deviations of OBQ-44 and MCQ-30 total scores and subscale scores at all five measurement points from pre-treatment to the extended follow-up. As can be seen from the total scores and subscale scores, there is a clear decline from pre- to post-treatment, which largely continues to decrease through the three subsequent follow-ups (3 and 12 month and the extended follow-up).

Table 3 Means and standard deviations: total scores and subscale scores for OBQ-44 and MCQ-30
Full size table

Obsessive Beliefs as Predictors of OCD Symptoms

Obsessive beliefs at both the between- and within-person levels were assessed to determine if they were predictors of OCD symptoms. There was a significant between-person effect of obsessive beliefs on OCD symptoms [β = 0.04, SE = 0.02, t(35.43) = 2.07, p = 0.046] but not a within-person effect [β = 0.03, SE = 0.02, t(39.28) = 1.81, p = 0.08]. Thus, higher average score of obsessive beliefs predicted higher OCD symptoms (between-person effect), but if a patient was particularly higher relative to its own average on obsessive beliefs (within-person effect), this did not predict higher OCD symptoms at the subsequent measurement-point.

We then investigated the reciprocal relationship to determine whether changes in OCD symptom severity were related to changes in obsessive beliefs at the between- and within-person level. OCD symptoms were related to obsessive beliefs at the between-person level [β = 4.01, SE = 1.46, t(37.98) = 2.74, p =  < 0.01] but not at the within-person level [β = 0.57, SE = 0.59, t(59.10) = 0.98, p = 0.33]. Thus, the results showed that there are reciprocal relationships between obsessive beliefs and OCD symptoms at the between-person level.

Subscales of Obsessive Beliefs

We progressed to investigate which subscales of the OBQ-44 responsible for changes in OCD symptoms by using the between-person score and within-person score of the subscales as predictors of OCD symptoms over the course of therapy and follow-up.

Perfection/certainty (PC) significantly predicted OCD-symptoms [β = 0.11, SE = 0.04, t(36.78) = 2.57, p = 0.014] but there were no within-person effects [β = 0.04, SE = 0.03, t(50.57) = 1.52, p = 0.14]. Thus, in general, higher beliefs about perfectionism were associated with a generally higher OCD symptom levels.

Importance/control of thoughts (ICT) did not significantly predict OCD-symptoms on a between-person level [β = 0.08, SE = 0.07, t(37.62) = 1.10, p = 0.28] or a within-person level [β = 0.07, SE = 0.05, t(50.91) = 1.46, p = 0.15].

Responsibility/threat estimation (RT) also did not significantly predict OCD-symptoms on a between person level [β = 0.08, SE = 0.04, t(39.68) = 1.94, p = 0.06] or a within-person level [β = 0.02, SE = 0.03, t(45.60) = 0.73, p = 0.47].

Metacognition as a Predictor of OCD Symptoms

Metacognitive beliefs at both the between- and within-person levels were assessed to determine if they werepredictors of OCD symptom severity. There was a significant between-person effect of metacognitive beliefs on OCD symptoms [β = 0.19, SE = 0.06, t(38.91) = 3.4, p =  < 0.01] but not a within-person effect [β = 0.05, SE = 0.04, t(36.50) = 1.51, p = 0.14].Footnote 3 In the reversed analysis, OCD symptoms were predictive of metacognitions at the between-person level [β = 1.69, SE = 0.44, t(38.57) = 3.80, p ≤ 0.01] but not at the within-person level [β = 0.13, SE = 0.13, t(56.66) = 0.99, p = 0.33]. The results indicate a reciprocal relationship between metacognitions and OCD symptoms at the between-person level.

Obsessive Beliefs and Metacognitions in the Same Analysis

In the final analysis, obsessive beliefs and metacognitions at the between- and within-person levels were entered into the same analysis. In this analysis, only metacognitive beliefs at the between-person level emerged as a predictor [β = 0.25, SE = 0.09, t(36.94) = 2.9, p ≤ 0.01]. Thus, in the context of metacognitions, obsessive beliefs were not a significant predictor of OCD symptoms.

Subscales of the MCQ-30

We progressed to investigate which subscales of the MCQ-30 were responsible for changes in OCD symptoms by using the between-person score and within-person score of the subscales as predictors of OCD symptoms over the course of therapy and follow-up.

Positive beliefs about worry had a between-person effect on OCD symptoms [β = 0.79, SE = 0.21, t(38.04) = 3.63, p = 0.001] but not a within-person effect [β = − 0.14, SE = 0.21, t(67.81) = − 0.68, p = 0.50]. Thus, in general, higher positive beliefs about worry were associated with a general higher level of OCD symptoms.

Negative beliefs about the uncontrollability of thoughts and corresponding danger also predicted OCD symptoms at the between-person level [β = 0.99, SE = 0.21, t(36.84) = 4.77, p ≤ 0.001] but not at the within-person level [β = 0.09, SE = 0.12, t(40.40) = 0.77, p = 0.442]. In general, higher negative beliefs about the controllability of thoughts and corresponding danger were associated with higher OCD symptoms.

Beliefs about the need to control thoughts was a significant predictor at both the between-person level [β = 0.68, SE = 0.28, t(39.60) = 2.40, p = 0.021] and a within-person level [β = 0.29, SE = 0.12, t(39.30) = 2.49, p = 0.017]. Higher beliefs about the need to control thoughts were associated with higher OCD symptoms.

Beliefs about cognitive confidence were not a significant predictor at the between-person level [β = 0.24, SE = 0.22, t(39.34) = 1.08, p = 0.287] or at the within-person level [β = 0.10, SE = 0.16, t(60.31) = 0.66, p = 0.509].

Cognitive self-consciousness was not a significant predictor at the between-person level [β = 0.46, SE = 0.23, t(39.14) = 1.97, p = 0.055] or at the within-person level [β = 0.21, SE = 0.13, t(43.79) = 1.62, p = 0.113].

Discussion

The present study primarily aimed to explore whether changes in obsessive (OBQ-44) and metacognitive (MCQ-30) beliefs were associated with OCD symptom (Y-BOCS) reductions over the course of treatment and long-term follow-up. In addition, we investigated the unique contributions from between- and within-person effects on predictor variables, outcomes, and reciprocal relationships. The results of this study replicate earlier findings that both metacognitive and obsessive beliefs decrease throughout treatment and that changes in metacognitions are a better predictor of OCD symptom reduction than changes in obsessive beliefs (Grøtte et al. 2015; Solem et al. 2009). However, our results extend previous findings by showing that these observations continue many years after treatment. This is the first study to disaggregate the predictor variables, obsessive beliefs and metacognitive beliefs, which made it possible to separate the between- and within-person effects in the process variables for patients treated for OCD.

Our first objective was to investigate whether the mechanism variables, obsessive beliefs and metacognition, decreased over the course of treatment and at long-term follow-up. In line with previous studies, we found that both obsessive beliefs and metacognitions decreased during treatment and at follow-up. However, previous treatment studies reporting on changes in obsessive beliefs and metacognitions for OCD have only implemented follow-up periods of one (e.g., Olatunji et al. 2013a, b; Solem et al. 2009) or 2 years (Whittal et al. 2008). This is the first study to show that reductions in dysfunctional metacognitions and obsessive beliefs continue many years after group ERP for OCD. The results thus show that achieving reductions in maladaptive metacognitions and obsessive beliefs during treatment can also yield positive long-term change.

Our next objective was to investigate whether the between- and within-person relationships of obsessive beliefs and metacognition predicted changes in OCD symptom severity during treatment and at follow-up. The hypothesis was partially confirmed. Reduction in both maladaptive metacognition and obsessive belief was associated with subsequent improvement in OCD symptoms from pre- to post-treatment and at follow-up, but only at a between-person level. Furthermore, when metacognition and obsessive beliefs were entered in the same statistical model, only metacognition emerged as a predictor. This result fits with a series of studies that find metacognition to be a better predictor than obsessive beliefs (e.g., Grøtte et al. 2015; Gwilliam et al. 2004; Myers and Wells 2005; Solem et al. 2009). However, the current study extends previous findings to show that changes in obsessive beliefs and metacognitions experienced over short-term periods (3–12 months) are also associated with symptom reductions many years later (mean 8 years). These results support both cognitive and metacognitive theories of OCD. Reduction in obsessive beliefs not only predicts outcome during treatment but also after the end of treatment. There also seems to be an association between the maintained reductions in obsessive and metacognitive beliefs and OCD symptom severity in the long term. This may indicate that if metacognitive beliefs are not reduced during OCD treatment, they can have negative consequences in both the short and long term.

New methodological approaches also allow us to draw new conclusions and implications. To our knowledge, almost all prior clinical OCD studies have focused on between-person relationships, except for one study (Olatunji et al. 2013a, b) that focused only on the within-person effect. The intraclass correlation in the current study indicated that approximately three-quarters of the variance of both obsessive beliefs and metacognitions could be explained by the between-person effect. This gives a more precise picture of the proportion of the variation that can be attributed to between-person effects than the usually aggregated effect, which is a blend of between- and within-person effects (Curran and Bauer 2011; Raudenbush and Bryk 2002). The last quarter of the variance is left to individual-specific differences and any random measurement errors. The therapist focus is at the within-person level. Questions addressed in this paper are thus relevant for clinicians because the analyses are conducted at the within-person level.

We also explored whether there were reciprocal relationships between OCD symptom severity and metacognitive and obsessive beliefs. A decrease in OCD symptoms predicted a subsequent reduction in both obsessive and metacognitive beliefs at the between-person level. The result in terms of the OBQ-total score is in line with a prior study (Levy et al. 2018) and partly in line with another study that reported reciprocal relationships at the within-person level (Olatunji et al. 2013a, b). In contrast, one study reported that changes in OCD symptoms predicted reductions in obsessive beliefs, but the converse did not hold (Woody et al. 2011). This emphasizes the importance of reversing the analysis of mechanisms when examining possible causalities. Reciprocal relationships between MCQ-30 total scores and outcome measures have been reported at the within-person level in a clinical sample that included a variety of anxiety disorders (Johnson et al. 2018). The present findings indicate that there are circular relationships between OCD symptoms and obsessive and metacognitive beliefs. It could be that changes in OCD symptoms precede changes in dysfunctional beliefs, especially since the patients were treated with ERP. In contrast to metacognitive therapy (MCT) and CBT, ERP is not targeted at changing metacognitive or obsessive beliefs but is based on therapeutic change obtained from the Inhibitory Learning Model of Exposure to OCD (Craske et al. 2008,2014). From the inhibitory learning theory the aim is to strengthen inhibitory (safety) association relative to the original dreaded association, which in the short-term leads to increased tolerance and in the longer-term extinction of fear as a result of disconfirmations of expectation (i.e., opportunities for patients to learn that feared outcomes do not occur) (Craske et al 2008; Jacoby and Abramowitz 2016).

An extended analysis showed that 3 of the five subscales from the MCQ-30 separately predicted outcome. Of special interest was the finding that beliefs about the need to control thoughts was a significant predictor at the within-person level in addition to the between-person level. Thus, a reduction in metacognitive beliefs about the need to control thoughts in the individual patient might lead to symptom reduction at the next time point. In addition, the positive beliefs about worry subscale and the negative beliefs about the uncontrollability of thoughts and corresponding dangers subscale were significant negative predictors for OCD symptom severity at the between-person level. These 3 subscales are the most consistent predictors across eight studies that have used MCQ-30 subscales as predictors in regression analysis where the Y-BOCS score was the independent variable (Grøtte et al. 2016). The results are also partially in line with a prior study where changes in the positive beliefs about worry subscale and the beliefs about the need to control thoughts subscale from pre- to post-treatment independently predicted outcome after ERP for OCD (Solem et al. 2009).

The lack of a within-person effect related to obsessive beliefs regarding OCD-symptoms is not evidence that this process is not important. Previous research has found a within-person effect of cognitions on anxiety symptoms at the next time point (Johnson et al. 2018). Furthermore, the relatively small number of time-points per person, and the long spacing between the intervals in the present study indicates the need for future studies to include session by session assessments of obsessive beliefs and OCD symptoms.

The extended analyses for the OBQ-44 showed that change in one subscale, perfectionism/certainty, significantly predicted reduction in OCD symptoms at a between-person level. This result is partly in line with a prior study where perfectionism and certainty beliefs was the only OBQ-44 subscale that mediated (improved) treatment response for 36 adults with OCD who were assessed at 5 measurement points during 24 weeks of CT (Wilhelm et al. 2015). An elevated score on perfectionism and certainty beliefs is associated with unrealistic and rigid standards of success and low tolerance for uncertainty (Frost and Steketee 1997). The current result indicates that change in the obsessive belief about perfectionism and intolerance for uncertainty is central for OCD symptom reduction both during the acute treatment and in maintenance of gains in the long-term.

There may be several explanations regarding why the other two OBQ-44 subscales did not show significant between-effect. According to a recent meta-analysis (Steketee et al. 2019) were OCD patients with higher responsibility and overestimation of threat beliefs more likely to improve if they received CBT or CT. In contrast were behavior therapy (ERP) not affected by these beliefs. It is conceivable that the role of obsessive beliefs is more important in CT/CBT which has interventions that are directly aimed at moderating maladaptive beliefs (Steketee et al. 2019). Additionally, studies show that not everyone with clinical OCD has endorsed obsessive beliefs and that this in fact maybe related to specific subtypes of OCD (e.g., Taylor et al. 2006). There is a growing evidence that OCD is not a unitary syndrome, but rather a heterogeneous disorder consisting of at least four empirically derived and thematically different subtypes or dimensions (i.e., responsibility, contamination, unacceptable thoughts, and symmetry/arranging) (Leckman et al. 2009; McKay et al. 2004). Furthermore, there is empirical support that different theoretical cognitive models may apply to different subtypes of OCD (Taylor et al. 2006). For example, one study reported that the OBQ-44 subscale, Importance/Control of Thoughts, predicted unacceptable thoughts (e.g., sex, blasphemy), symmetry symptoms were predicted by perfectionism/certainty and contamination symptoms by responsibility/threat estimation beliefs (Wheaton et al. 2010). As other studies have speculated (e.g., Levy et al. 2018), it may also be that the current sample is not sufficiently representative of OCD subtypes that tend to have elevated obsessive beliefs (e.g., importance/control of thoughts).

Alternatively, the findings might be due to relatively large intercorrelations between the total scores of OBQ-44 and MCQ-30. Since we included the metacognitive scale of the OBQ-44 (ICT) this could contribute to high shared variance in the current study. Several of the cognitive OCD models also have metacognitive aspects, because they are appraisals theories, focusing on the way people think and evaluate their thoughts (Corcoran and Segal 2008). In Rachman's theory (1997, 1998) negative appraisals of intrusive thoughts are central for developing and maintaining OCD. Clark and Purdon (1995) model is about meta-cognitive processes with a focus on people with OCD misinterpreting the consequences of not having control over their thoughts. However, metacognition was never emphasized as an important or explicit feature of these models (Fisher 2009; Purdon and Clark 1999). This study showed that changes in several metacognitive beliefs of the MCQ-30 predicted a reduction in OCD symptoms better than changes in the obsessive beliefs of the OBQ-44. Even the metacognitive OBQ-44 factor, beliefs about the importance and need to control one’s thoughts, was not predictive. In general, our findings indicate that the factors involved in Well’s metacognitive model for OCD are broader than those defined in the metacognitive subscale for OBQ-44 (Fisher 2009).

This study has potential clinical implications. The results indicate that metacognitions may be a crucial mechanism for symptom reduction in OCD patients, not only shortly after treatment but also many years later. Interestingly, beliefs about the need to control thoughts were a significant predictor of higher OCD symptoms at both the within- and between-person levels. This knowledge about the within-person relationship can directly informs clinicians that attending to the patient’s need to control thoughts may be a potentially valuable approach to achieve maximum improvement with ERP. The perceived inability to control unwanted, disturbing and intrusive thoughts is a cardinal feature of OCD (Purdon and Clark 2002). The more a patient tries to stop an intrusive thought, the more frequent it often becomes. According to a recent study, beliefs about the need to control thoughts as indexed by the MCQ-30 (Wells and Cartwright-Hatton 2004) predicted most OCD symptom dimensions (checking, cleaning and rumination) (Tumkaya et al. 2018). If metacognitions are a key mechanism compared to obsessive beliefs, targeting metacognitions directly may increase the efficacy of treatment for OCD. This may be particularly relevant in cases where metacognitive beliefs are severe and the patient does not initially respond to standard ERP or refuses the treatment. The need to control thoughts might be handled by a specific intervention in metacognitive therapy called detached mindfulness. Detached mindfulness (DM) has multiple components, requiring the activation of metacognitive knowledge, metacognitive monitoring, and control, suspension of conceptual processing, attentional flexibility, and a de-centered relationship with thoughts. The technique is intended to create meta-awareness (e.g., being aware of worry) combined with decentering (Wells 2005). Even though DM involves several components, it could be addressed in therapy using specific exercises (Wells 2005).

There are some limitations that need to be mentioned. One of the main limitations is the small sample size, which limits the generalization of our results. In particular, the post hoc analyses concerning the subscales of obsessive and metacognitive beliefs should be interpreted with caution, given the low statistical power and multiple comparisons performed. Since we had no control group, we cannot conclude that the improvement in the variables is due to group ERP. In particular, we cannot draw firm conclusions for the period of time from post-treatment to follow-up because of the long interval between the 12-month and the on-average 8-year follow-up time points. Future studies should investigate obsessive beliefs and metacognitions both during treatment and throughout follow-up using repeated assessments to obtain a better estimate of the within-person effects. The current study, as well as almost all prior studies examining the relative contributions of obsessive and metacognitive beliefs for OCD, has assessed OCD symptoms globally (i.e., Y-BOCS sum scores). To focus exclusively on OCD as a homogeneous condition may have obscured important pattern of predictors linked to the individual subscales of dysfunctional obsessive and metacognitive beliefs and specific OCD symptom themes (Nance et al. 2018). Unfortunately, few studies have sufficient power to break down the data into OCD subtypes (Foa 2010). However, two recent studies with larger samples (ranging from 110 to 153) have shown that factors from both cognitive and metacognitive models contribute to predicting various obsessive compulsive dimensions in both clinical (Kart et al. 2020) and non-clinical samples (Nance et al. 2018). To fully examine the role potential cognitive and metacognitive factors play in the reduction of OCD symptoms, future studies should also include theme-based OCD dimensional analyses to better understand differential patterns of predictor variables. Theoretically, findings derived from a dimensional approach could lead to a more specific conceptualization of OCD. Perhaps more importantly, research at the level of symptom dimensions could foster the development of more effective and tailored interventions for OCD (Nance et al. 2018; Wheaton et al. 2010). In addition, future large trials should also compare the efficacy of MCT and CBT, including so called third wave therapies of CBT (e.g., ACT) that in contrast to ERP are directly designed to targeting the assumed mechanisms.

Overall, our results support previous findings that changes in metacognition are a better predictor than changes in obsessive beliefs for OCD treatment. The current study extends the literature by showing that metacognition is also related to OCD symptom severity many years after treatment. The finding of a within-person relationship between the metacognitive belief about the need to control thoughts and subsequent OCD symptoms has the primary clinical implication that reduction in this belief may be important for treatment response both in the short and long term.