Early maladaptive schemas impact on long-term outcome in patients treated with group behavioral therapy for obsessive-compulsive disorder

A total of 65 patients were invited to participate in this observational long-term follow-up study. All potential participants were treatment-seeking before they completed a manualized group ERP program for OCD [34] in a general outpatient clinic 5 to 11 years prior to participating in this study. The therapy was conducted in groups of six patients with two therapists. The groups met weekly for 12 weeks with each session lasting 2.5 h. All participants had a primary diagnosis of OCD before starting treatment. This original sample has been described in two previously published reports [35, 36].

Forty (n = 40) of the 65 eligible patients agreed to participate in the current long-term follow-up study. The remaining 25 patients either refused to participate (n = 24) or were not traceable (n = 1). The Regional Committee for Medical and Health Research Ethics¹ approved this study. All participants gave written informed consent before taking part in the current long-term outcome study. All participants in the current study were previously assessed at pre-treatment, post-treatment and 12-months following the original group ERP program. For the present study, participants were assessed at mean of 8.23 years (SD = 1.86) after completing treatment (from now on referred to as “extended follow-up”). Twenty-four (61.5%) of these participants reported that they had received additional OCD treatment (i.e., additional ERP; therapy not involving ERP; anti-obsessional medication) between completion of group ERP and the extended follow-up assessment. The mean age of the participants at the extended follow-up was 43.6 years and 77.5% were female. The mean Y-BOCS total score at pre-treatment was 23.15 (SD = 3.63), indicating moderate- to severe OCD symptoms. More information about participant characteristics (marital- and employment-status) and additional study procedures are described elsewhere [36].

The Yale-Brown Obsessive Compulsive Scale (Y-BOCS: [37, 38]). Obsessive-compulsive symptoms were assessed with the Y-BOCS. The Y-BOCS is a 10-item interview that provides sub-scores for obsessions, compulsions and a total score of OCD severity. The total score ranges from 0 to 40, with higher scores indicating greater OCD severity. The Y-BOCS has been documented as a reliable and valid tool [39].

Young Schema Questionnaire – Short Form (YSQ-SF: [21]). The Young Schema Questionnaire (YSQ) is a self-report inventory developed to assess underlying EMSs [40]. In the current study we used the original short version that contains 75 items and assess 15 specific early maladaptive schemas developed from the original long version of YSQ, which consist of 205 items [40]. Similar psychometric properties, validity and levels of clinical utility are reported between YSQ-SF and the long version of YSQ [41]. EMSs are organized in five schema domains in YSQ. The five schema domains are; (1) Disconnection and Rejection (trouble getting stable and safe attachment to significant others), (2) Impaired Limits (difficulty in respecting the feelings and needs of others), (3) Other-Directedness (tendency to emphasize other’s needs and feelings on the expense of their own), (4) Overvigilance and Inhibition (strict control over one’s own feelings and unrealistic high demands on oneself) and (5) Impaired Autonomy and Performance (difficulty functioning independently of others at same age). Each item is rated using a 6-point Likert scale from 1 = “completely untrue of me” to 6 = “describes me perfectly”. There are five questions for each of the specific 15 EMSs (see Table 1). The total sum (YSQ-SF total score) is the addition of all raw item scores divided by 75. YSQ-SF is widely used across cultures and translations (e.g. Canada: [42], Belgium [43];, Spain [44], Britain [41], Australia and South-Korea [45]) and has shown good to excellent consistency both for the YSQ total score and the individual EMSs in Norwegian samples [46, 47]. The Norwegian version of YSQ-SF has been translated back to English with no substantial differences in meaning from the American version for the 75-item scale [46]. YSQ-scores have shown discriminant validity between clinical populations in Norway [33, 46, 48] and between clinical and non-clinical samples [33]. Test-retest reliability of YSQ-SF in a Norwegian study has been shown to be satisfactory, with a mean duration of 72 days [46]. Finally, in the current study, Cronbach’s alpha for YSQ-SF total scores ranged from 0.97 to 0.99 at the four measurement occasions, indicating excellent internal consistency [49].

Beck Depression Inventory (BDI: [50]) is a self-report inventory for depression symptoms, consisting of 21 items rated on a four-point Likert scale, ranging from 0 (“not at all”) to 3 (“severe”). The BDI has good psychometric properties [51]. In the current sample, Cronbach’s alpha for BDI were excellent, ranging from 0.92 to 0.94.

To define the status as recovered, we used the same classification [52] as in the original study by Haaland et al. [35]. To be classified as recovered, participants must have a post-treatment Y-BOCS total score of 14 or less and must have improved at least 8 points from pre to post-treatment. Sixteen (40%) patients achieved recovered status at the extended follow-up whereas 24 patients (60%) did not [36].

We assumed that 50% of the invited participants (N = 65) would be classified as recovered at the extended follow-up based on data from the previously published 12-month follow-up study by Haaland et al. [35]. Moreover, we expected that most patients would be willing to participate in the extended follow-up, which would likely yield a sample of 25 participants in both the recovered and the non-recovered groups. Statistical power analysis performed with regard to the detection of differences between the groups on the potential predictor variables showed that with an α-level at 0.05 and a β-level at 0.20, 50 participants would be sufficient to detect differences of a moderate effect size (Cohens d = 0.80) with a two tailed t-test.

With respect to YSQ-SF total score, the legitimacy of these analyses was supported by data from Thiel et al. [31] who found a difference of .3 between responders and non-responders and a pooled SD of .75 at post-treatment. At the extended follow-up in our study, it is to be expected that the difference between recovered and non-recovered patients would be considerably larger, but we used a conservative estimate of .60 in the analyses. Using these numbers in a power calculation gave an effect size of .80 and a power of .80 as well. With a reduction from 50 to 40 participants, power diminished from .80 to .70, which we considered to be adequate to perform the predictor analyses since the difference of .60 is a very conservative estimate.

The patterns of missing values for YSQ-SF protocols were examined. Independent t-tests were used to compare pre-treatment measures between patients who were recovered versus those who did not recover at the various measurement points. In addition, independent t-tests were used to compare the pre-treatment YSQ-SF total score of those who relapsed or had a delayed remission with those who were either unchanged or met criteria as recovered across all four measurement points (pre- and post-treatment, 12-month and the extended-follow-up). Fisher’s exact tests and independent sample t-tests were used to investigate differences in demographic variables between participants who were recovered and participants who were not recovered at the extended follow-up. All tests of differences were performed as two-tailed, unless otherwise noted.

Multiple regression analysis was conducted for predicting outcome with the Y-BOCS at extended follow-up as the independent variable. In the regression analysis we explored whether pre-treatment YSQ-SF total score was related to the extended follow-up outcome measured by Y-BOCS. Residual plots and histograms for residuals were checked which showed normal distribution of the residuals.

The third research question was addressed by the mixed models procedures in SPSS. YSQ-SF total score was entered as the outcome (dependent) variable. “Time” and “group” were entered as covariates, as well as the interaction between “time” and “group”. Time was coded as 0, 1, 2, and 3, corresponding to the four measurement occasions (pre-treatment, post-treatment, 12-month follow-up and 8-year follow-up). “Group” is a dummy variable indicating patients who were recovered versus patients who were not recovered from their OCD. Log likelihood estimation (LLH) and Akaike Information Criterion (AIC) were used to evaluate model fit.

The first step in the modelling procedures was to model the mean structure of the data [53], in which only “time” was included (LLH = 342, AIC = 354). The second step was to include a random intercept, which implies that each patient was assigned an individual estimate of the YSQ-SF total score at pre-treatment. This step also included the modelling of the covariance structure of residuals, comparing a diagonal covariance matrix with an autoregressive covariance matrix. The second step resulted in a considerably better model fit (LLH = 201, AIC = 211). In the third step, we compared a pure linear model (a straight regression line through all measurement occasions) with a linear spline model with the knot at post-treatment. This model implies that change trajectories were analyzed for two different time periods within the same model, i.e., from pre-treatment to post-treatment (where the knot was positioned) and from post-treatment to extended follow-up (through 12-month follow-up). The model fit did not change as compared with the second model (LLH = 200, AIC = 212). The fourth and last step was to include the interaction with “time” and “group” in the simple linear model as well as in the linear spline model to examine whether recovered patients had different change rates than non-recovered patients with respect to EMS. The inclusion of the “time by group” interaction in the simple linear model yielded a slight improvement of model fit (LLH = 192, AIC = 206). However, the linear spline model did not improve by including this interaction (LLH = 191, AIC = 209).

Effect sizes for the YSQ-SF total scores were calculated utilizing a formula derived from Cohen [54].² To reduce the risk of Type I errors due to multiple comparisons, a Benjamini-Hochberg procedure with false discovery rate was set at 10% for all analysis [55]. These methods are in line with the recommendation of the American Statistical Association (ASA) [56]. The Benjamini-Hocberg procedure was preferred above the Bonferroni correction because of risk for type II errors with the small sample sizes [57] in the present study. P-value in the current study was 0.045 corresponding to a p-value of 0.05 for a single comparison. All statistics were calculated using IBM SPSS version 23.0 [58] except Benjamini-Hocberg procedure [55] that was performed by hand.

Multicollinearity among predictor variables was statistically investigated by computing variance inflation factors (VIF). The highest VIF was 1.67 which is far below the suggested cut-off value of 10 that indicates a collinearity problem [59]. As a further part of the multicollinearity examination, correlation coefficients were conducted among predictor variables. A linear regression analyses assume that the degree of correlation should not exceed 0.90 due to problems with multicollinearity [60]. As shown in Table 2, a significant moderate degree of correlation was observed between the dependent variable and the pre-treatment YSQ-SF total score. A moderate to high degree of correlation was found between the YSQ-SF total score and the two other independent variables, BDI and Y-BOCS, indicating that problems with multicollinearity are not a serious problem in the current study.

Sixteen YSQ-SF protocols were totally missing and 3 more protocols were excluded due to missing items. Taken together; 3, 5, 6 and 5 of the YSQ-protocols were missing at respectively pre-test, post-test, 12-month and the extended follow-up. Little’s MCAR test was run on the remaining data, and it indicated that data was not missing completely at random (MCAR). However, when the pattern of missing variables was closely examined, using the method described by Little and Rubin [61], it was not possible to find any definitive pattern. Therefore, we proceeded with analysis treating the data as missing at random (MAR). Of the remaining YSQ-SF protocols, 0.3% of the items were missing for the four measurement points. One participant had 19 of 75 items that were missing due to data management issues. This participant was excluded, and missing data for YSQ-SF was by this reduced to 0.2%. To reduce missing data for the YSQ-SF total score, mean imputation was applied to individual scale items when fewer than 5% of items were unanswered [62]. Single EMSs were excluded if any items were missing.

Ten participants (62.5%) in the recovered group and 14 participants (58.33%) in the non-recovered group reported that they had received additional treatment between the initial group ERP and the extended follow-up. There were no significant differences between the recovered group and non-recovered group regarding whether they received additional treatment of any kind or whether they received a specific type of treatment. The largest difference between the groups was that nine of the 24 (37.5%) participants in the non-recovered group and two (12.5%) of the participants in the recovered group received additional ERP for OCD in the time period between the end of the original group CBT and the long term follow-up assessment point [M = 0.41 (0.50) vs. M = 0.14 (.36); t (34) = 1.84, p = .075].

There were no significant differences with respect to gender (p = 1.00), age [M = 43.56 (11.62) vs. M = 40.91 (13.52); t (37) = .637, p = .528], marital status (married and cohabiting versus single, divorced and separated) (p = .728) and employment status (employed, student, retired versus unemployed, homemaker and disabled) (p = .514) between the recovered and the non-recovered group.

To control for selection bias, the YSQ-SF total scores were compared between the group (n = 40) that agreed to participate in this long-term follow-up study and those who were contacted but refused to participate (n = 25). There were no significant differences at pre-treatment [M = 2.36 (0.84) vs. M = 2.41 (0.67); t(58) = 0.228, p = .821], post-treatment [M = 2.18 (0.82) vs. M = 2.10 (0.72); t(52) = 0.075, p = .941] and change scores at post-treatment [M = 0.19 (0.45) vs. M = 0.21 (0.49); t(49) = 0.122, p = .903] between these groups.

We investigated the factors that most influenced the OCD symptom severity at the extended follow-up. The dependent variable was the total Y-BOCS score at the extended follow-up time point. Due to the limited sample size (N = 40), only 3 independent factors measured at pre-treatment were included in the multiple regression analysis: 1) OCD symptom severity (Y-BOCS), 2) Depression symptoms (BDI), and 3) Maladaptive schema total score (YSQ-SF). In the first step, the three independent variables were entered one by one in a single linear regression analysis. Only the YSQ-SF total score was significantly related to the dependent variable (p = 0.029, adjusted R square = .105). To investigate whether the three independent variables adjusted for each other, thus giving better predictive ability, they were entered together in a backward elimination multiple-regression analysis (see Table 3). This analysis showed no change from the simple regression analysis. YSQ-SF total score at pre-treatment explained 10.5% of the obsessive-compulsive symptoms at the extended follow-up.

According to the best estimate of the regression line for Y-BOCS at the extended follow-up, an unstandardized B of 3.065 indicates that on average, an increase of 1 unit in YSQ-SF total score is associated with approximately a 3 unit increase in Y-BOCS at the extended follow-up.

Independent sample t-tests showed that the non-recovered group at the extended follow-up had significant higher pre-treatment YSQ-SF total score, as well as BDI scores, compared to the recovery group (Table 4). These differences were not significant between patients who were recovered versus not recovered at post-test and 12-month follow-up, respectively. There were no significant differences in pre-treatment OCD severity between the recovered and not-recovered groups neither at post-treatment, 12-month nor the extended follow-up time points.

A closer analysis showed that 19 (47.5%) of the participants had changed their status between recovered and non-recovered from post-test to the extended follow-up. Eleven participants (27.5%) had relapsed between post-test and the extended follow-up and had significantly higher YSQ-SF total score at pre-treatment compared to those who were recovered at the extended follow-up [M = 2.77 (0.85) vs. M = 1.91 (0.52); t(24) = 2.98, p = .009, Cohen’s d = 1.19]. In addition, eight participants who had remission later than post-test had a significant lower YSQ-SF total score at pre-treatment compared to those that were non- recovered [M = 2.05 (0.57) vs. M = 2.69 (0.89); t(27) = 2.21, p = .042, Cohen’s d = 0.86].

Figure 1 displays the YSQ-SF total score change profiles for recovered and non-recovered patients. It can be seen that the decrease in YSQ-SF total scores was most pronounced from pre-treatment to post-treatment for both groups. Moreover, the non-recovered group had higher YSQ-SF total scores at pre-treatment. This difference remained stable over time.

Indeed, the results of linear mixed models analyses indicated that there was a slight but significant decrease in YSQ-SF total score from pre-treatment to extended follow-up for the entire sample (decrease of .14 points per measurement occasion, t = 4.6, p = .000). Furthermore, the change in YSQ-SF total score was greater from pre- to post-treatment than from post-treatment to the extended follow-up period for the entire sample. A linear spline model with a knot at post-treatment showed that the decrease from pre-treatment to post-treatment was larger (.22 points, t = 3.0, p = .003) than the decrease from post-treatment through 12-months follow-up to extended follow-up (.11 points per interval, t = 2.5, p = .017).

The linear model including the “time” by “group” interaction showed that the non-recovered group had higher YSQ-SF total scores at pre-treatment than the recovered group (t = 2.9, p = .005). However, EMS scores decreased at the same rate in both groups (t = .635, p = .529). The most complex model, the one in which the “time” by “group” interaction was included for two different time periods, showed similar results finding no difference in YSQ-SF total score change rates across the two groups, not from pre-treatment to post-treatment, neither from post-treatment to extended follow-up.

We followed Cohen’s [54] proposal to classify effect sizes as small (0.20–0.49), medium (0.50–0.79) or large (0.80 and above) in improvement in the YSQ-total score. For the entire sample the effect size (Cohens d) was small from pre- to post-test (d = 0.30), and remained small through the 12-month follow-up (d = 0.41) and to the extended follow-up (d = 0.48). Both the non-recovered (d = 0.37) and recovered group (d = 0.34) also had a small effect size for pre- to post-test. However, for the recovered group the effect size was medium (d = 0.57) at 12-month follow-up and large (d = 0.83) at the extended follow-up. For the non-recovered group, the effect size was still small (d = 0.39) at 12-month follow-up and medium (d = 0.50) at the extended follow-up.

To obtain more detailed information of the relationship between specific pre-treatment EMSs and extended follow-up outcomes, independent sample t-tests were conducted for the 15 EMSs. Table 5 shows that 10 of the 15 schemas (Abandonment/Instability, Emotional Deprivation, Mistrust/Abuse, Social Isolation/Alienation, Defectiveness/Shame, Dependence/Incompetence, Vulnerability to Harm and Illness, Subjugation, Emotional inhibition and Entitlement/Grandiosity) at pre-treatment were significant lower in the recovered group at the extended follow-up comparing to those who were not recovered. Five of these 10 schemas belong to the domain Disconnection and Rejection while the remaining five EMSs were evenly distributed on the other four domains. In addition, there were also a trend toward lower values on the remaining five EMSs, but these were not significantly different between the groups.

This study has limitations. For example, there was no control group, making it difficult to draw conclusions about treatment effects, especially since many participants had received additional treatment during the follow-up period [36]. The study was not designed to find causal relationships between reduction in EMSs and OCD symptoms. In addition, patients were treated with ERP which is not designed to change EMSs, but is specifically targeted toward improving OCD symptoms. Further, other potentially clinically important variables that have been shown previously to be significant predictors of outcome for OCD treatment have not been examined. For example, because of limited statistical power, hoarding pathology, increased anxiety, certain OCD symptom subtypes, unemployment, and being single/not married, were not included in the analysis, which is relevant because they have all been shown, at least in some studies, to be negative predictors for OCD treatment outcome [8]. As mentioned above, a shortcoming is also that the sample size did not allow for the systematic investigation of individual EMSs on outcomes over time. Future long-term outcome studies with larger samples would be very useful in addressing these important issues. Additionally, there is a substantial length of time between the 12-month follow-up and the extended follow-up periods. It would have been of interest to know more about the relationship between EMSs and OCD symptoms annually throughout the extended follow-up period. Annual assessments are particularly relevant for OCD given that waxing and waning of symptoms over time is commonly observed [70]. Finally, we did not assess for PDs in this study. Given that patients with personality psychopathology generally exhibit high EMSs [48], it is desirable to assess and control for PD comorbidity when studying relation between EMSs and OCD symptoms.

These weaknesses notwithstanding, a key strength with the current study is that we explore follow-up data for EMS at a mean of 8 years after OCD treatment. No other studies of CBT for OCD provide any follow-up EMS data beyond the post-treatment data point. Furthermore, contrary to some CBT studies for OCD [7, 71, 72], patients with severe PDs and/or major depression were included in this community-based project. Therefore, the participants in this study may be seen as more representative of typical treatment-seeking patients in routine outpatient clinics.