Procedure
Data were derived from The Netherlands Obsessive Compulsive Disorder Association (NOCDA) study, an ongoing longitudinal cohort study investigating the naturalistic long-term course of OCD in patients referred to mental health care centres and to examine determinants in predicting the course of OCD. The NOCDA study design and baseline characteristics of the study sample are described in detail elsewhere [36]. The NOCDA study was accredited by the Medical Ethical Committee of the VU-university Medical Centre in 2005.
After intake at one of the contributing mental health clinics, 687 patients aged 18 years and over with a lifetime diagnosis of OCD, as determined by the administration of the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) [37], were asked to participate in the NOCDA study. Since NOCDA aims to follow a large representative sample of OCD subjects in different stages of the disease and with different degrees of illness severity, the only exclusion criterion was an inadequate understanding of the Dutch language for the purposes of the completion of interviews and self-report questionnaires. Comprehensive measurements were done at baseline and after 2 and 4 years.
Of the 687 patients who were asked to participate in the NOCDA study, 419 (60.9%) gave written informed consent and were enrolled in the study. A comparison on basic demographic characteristics between patients that did (n = 419) and did not (n = 268) agree to participate yielded no significant differences.
Baseline measurements took place between 2005 and 2009, and included validated semi-structured interviews and self-report questionnaires to gather information on a broad range of variables related to (amongst others) OCD, comorbidity, and psychosocial consequences. The baseline assessment took about 5 h. All included participants were contacted after 2 years and 4 years for follow-up, irrespectively of their treatment status. The 2-year and 4-year assessments took about 3 h, and in most cases (80%), they were done by the same research assistant. During the follow-up period, participants received treatment as usual that was based on Dutch multidisciplinary guidelines.
Participants in the present study
In the present study, only those patients who had a current OCD diagnosis at baseline were included, pertaining to 382 patients at baseline, 278 patients at 2-year follow-up (total dropout 27%), and 268 patients at 4-year follow-up (total dropout 30%). Of 239 patients, complete SCID data at baseline, 2, and 4 years were available. Patients with incomplete data did not differ significantly from those with complete data on sociodemographic and clinical characteristics except that they were younger (F(1, 380) = 4.16; p = 0.04) and less educated (F(1, 380) = 23.17; p < 0.01).
Patients were divided in three groups: chronic (current diagnosis of OCD at baseline, 2- and 4-year follow-up; n = 144), intermittent (current diagnosis of OCD at baseline and 4-year follow-up but not at 2-year follow-up; n = 22), and remitting (current diagnosis of OCD at baseline but not at 4-year follow-up, irrespective of diagnostic status at 2-year follow-up; n = 73). This study is reported conform the STROBE statement [38].
Primary outcome measure: QoL
The self-rated EuroQol five-dimensional questionnaire (EQ-5D) was used to assess QoL. It is a widely used, generic QoL instrument. It is applicable in many populations and can be used to compare QoL in various conditions. The EQ-5D contains five dimensions significant for QoL: mobility, self-care, daily activities, pain/discomfort, and depression/anxiety. Each dimension is rated at three levels: no problems, some problems, and major problems. These health states are converted into an index score—the EQ-5D—reflecting the generic overall QoL. The EQ-5D has a value between 1 (best possible health) and 0 (worst possible health). The EQ-5D was proven reliable, valid, and feasible [39,40,41,42,43].
Potential predictors of course of QoL in remitting patients
Repeatedly measured variables included severity of OCD, number of current comorbid mental disorders, comorbid anxiety and depressive symptoms, loneliness, need for affiliation, social support, and social network. These were assessed at baseline, 2- and 4-year follow-up. The variable time of remission was determined on the basis of the time-dependent variable current diagnosis of OCD, but is not itself a time-dependent variable. All other characteristics were assessed at baseline only: sociodemographics, age of onset, personality characteristics, attachment style, and perceived expressed emotion.
Sociodemographic characteristics included age (in years), gender, partner (yes, no), children (yes, no), education (number of years), and employment (yes, no).
The severity of OCD was assessed by the Yale Brown Obsessive Compulsive Scale for Severity (Y-BOCS) [44, 45]. The interrater reliability (ICC = 0.96) and test–retest reliability (ICC = 0.85) of the Y-BOCS are high [46].
Age of onset of OCD was assessed with the SCID-I as the earliest age at which patients fulfilled the criteria for OCD. In case of remission, time of remission was defined as early (remission at 2-year follow-up) or late (remission at 4-year follow-up). To assess the number of current comorbid mental disorders, the ascertained diagnoses on the SCID-I were counted. Interrater reliability of the SCID-I is fair-to-excellent, and the test–retest reliability and validity are substantial [47, 48]. Comorbid depressive symptoms were measured by the Beck Depression Inventory (BDI) [49, 50]. Comorbid anxiety symptoms were assessed by the Beck Anxiety Inventory (BAI) [51].
Furthermore, personality characteristics according to the Big Five were assessed with the Five-Factor Personality Inventory (FFPI) [52]. Subscales of the FFPI are: extraversion, agreeableness, conscientiousness, emotional stability, and autonomy. Attachment style was assessed with the General Attachment Style Questionnaire [53], with the subscales: dismissing, preoccupied, fearful, and secure. Loneliness was assessed with the Loneliness Scale [54], subscales: emotional loneliness and social loneliness. The need for affiliation was assessed with the Need for Affiliation Scale [35]. Social support was assessed with the Social Support Inventory [55]. Subscales of the SSI are: emotional support, informative support, social companionship, and instrumental support. Perceived expressed emotion of significant others was assessed with the Level of Expressed Emotion (LEE) [56]. Subscales of the LEE are: lack of emotional support, perceived intrusiveness, perceived irritation, and perceived criticism. Social network (number of friends) was assessed by an interview designed for the NOCDA study.
Quality aspects of NOCDA
The NOCDA study was coordinated by the academic department at VU Medical Centre/GGZ inGeest Amsterdam and included seven sites that were specialized OCD mental health clinics spread over the Netherlands. All research assistants had extensive experience with the assessment of OCD. In addition, they received a 2-day course, and regular follow-up 1-day training sessions in which videos of the SCID were rated, assessor rating scales were practiced and questions and problems raised by the research assistants could be addressed. The first two interviews of all research assistants were audiotaped and monitored by the fieldwork coordinator to address any misunderstandings or errors in performing the measurements. All subsequent interviews were audiotaped for future reference. The monitoring of these audiotapes was continuously performed randomly on about 10% of all taped interviews, as well as on the basis of questions raised by the research assistants and the fieldwork coordinator. Assessments were done by around 30 research assistants (profession: psychologist or research nurse).
Power considerations
Differences in QoL between patient groups versus the general population will be evaluated in terms of between-group effect sizes. We expect that the between-group effect sizes will range from medium, i.e., Cohen’s d = 0.5, when comparing patients with severe complaints to the general population, to small, i.e., d = 0.2, when comparing patients with mild severity to the general population. Differences in mean QoL scores of patient groups and the general population will be tested using one-sample t tests. Assuming a total sample size of n = 382, the minimal detectable effect size will be (Cohen’s) d = 0.15 for the total patient group and d = 0.19, d = 0.27 and d = 0.47 for patient subgroups that consist of 60%, 30%, and 10% of the total group, respectively. Restricting the sample to the 239 respondents with complete cases inflates minimal detectable effect sizes slightly to d = 0.18 for the total sample and d = 0.24, d = 0.33, and d = 0.59 for patient subgroups mentioned earlier.
Statistical analyses
Baseline characteristics of the total sample were summarized. Next, baseline characteristics of the three patient groups were compared using one-way ANOVAs for continuous variables and Chi-square statistics for categorical variables. Post hoc tests consisted of pairwise t tests with Bonferroni correction for continuous variables and column proportions z tests with Bonferroni correction for categorical variables.
Furthermore, the mean QoL of the total sample and the patient groups were compared to the mean QoL of the general Dutch population with one-sample t tests. We used data from Szende et al. [57] to determine the mean EQ-5D of the general Dutch population, which is 0.89. Effect sizes (Cohen’s d) within time and between groups were calculated using pooled standard deviations assuming SD = 0.20 for the general population. A Cohen’s d of 0.2 is indicative of a small-effect size, d = 0.5 of a medium-effect size, and d = 0.8 of a large-effect size. The correlation between QoL and Y-BOCS total score over all measurements was established with Pearson’s correlation coefficient r.
We examined the 4-year course of QoL using linear mixed models (LMM). LMM was used to correct for the correlation in the data due to the repeated measure design. In a second analysis, we examined the association between course of QoL and course of OCD. In this LMM analysis, time and group (remitting, intermittent, and chronic) were added as categorical variables. To examine whether course of QoL differed for the three groups, the group-by-time interaction terms were entered in the model. Effect sizes between group and time were calculated using pooled baseline standard deviations.
In a third analysis (using only respondents with remitting OCD), we examined whether change in QoL was associated with the possible predictor variables, using LMM’s allowing quadratic development over time. Next, all possible predictor variables were added to the basic model, one at a time. Y-BOCS, number of disorders, BAI, BDI, social network, loneliness (emotional and social), need for affiliation, and all subscales of SSI were treated as time-dependent (repeatedly measured) variables. All other variables were treated as time-independent (baseline only) variables. For time-dependent variables, it was investigated whether a random slope improved the model. Thereafter, multivariable analyses were conducted in four steps. In step one, all sociodemographic variables showing statistical significance (p < 0.05) in the univariable analyses were analysed together (model one). Model 2 included all clinical variables showing statistical significance in the univariable analyses and model 3 included all psychosocial variables showing statistical significance in the univariable analyses. The final model (model 4) included all variables showing statistical significance (p < 0.05) in model 1–3. We regarded correlations of 0.80 and above as a sign of multicollinearity [58].
First, statistical analyses were conducted on the complete data set, including 239 participants. Next, using multiple imputation techniques, a second data set was created (n = 382) allowing to investigate potential bias due to missing data. We describe the incompleteness of the data for variables at baseline, 2-year and 4-year follow-up separately (see online supplement). The appropriateness of the imputation method relies on the Missing at Random (MAR) assumption, which allows the missingness of data to depend on the observed variables. We applied MI by chained equations (MICE) using predictive mean matching with a single nearest neighbour for all variables used in the analyses to create 100 imputed datasets.
Data analysis was performed with IBM SPSS Statistics version 25 [59]. Multiple imputations and analyses on the multiple imputation data set were performed with Stata version 15.1 [60]. Since the analyses with complete data (n = 239) and imputed data (n = 382) yielded identical outcomes except for two minor results, we only report the complete data analyses here. Results that differed will be indicated.