Alternative Treatment to Reduce Chronicity in OCD: Research Into Brain Response and Adequacy of Treatment

Sample size calculation: for the primary study objective, the 95-95 approach to determine the non-inferiority margin was used. Results of a meta-analysis on the effectiveness of CBT for OCD indicated that the mean (95 percent Confidence Interval (1.04) generates a non-inferiority margin of .42. Expressed in scores on the primary outcome measure Yale Brown Obsessive Compulsive Scale (YBOCS), this would be a loss of about 2 points on this scale with a range of 0-40, which is clinically almost negligible. Applying an alpha of 0.05, a beta of 80%, and a .42 non-inferiority margin, a sample size of 71 per arm is adequate. Since analyses have to be run on the per protocol (P)- and the intention-to-treat (ITT) sample, 203 participants have to be recruited (anticipating 30 percent drop-out). Moreover, this sample size allows testing the secondary hypothesis: IBA is more effective than CBT for patients with OCD with poor insight (approximately 25 percent of the population). Applying a two-sided alpha of 0.05 and a beta of 80 percent and effect size of Cohen's d =1, a sample of 34 participants is needed (in a previous study a between groups effect of 1.7 was found), anticipating 30 percent drop-out, 48 participants with poor insight at baseline are needed.

Previous studies into the influence of psychotherapy for OCD on brain structure and functioning have used sample sizes of 9-45. Based on their findings, analysing data for a minimum of 30 participants per condition is planned to explore the working mechanisms of the treatment modalities. Anticipating an upper limit of 30 percent drop-out rate, 2 x 43 respondents will be included, as well as 43 control subjects.

Adherence to treatment protocol: the therapists will be extensively trained in adherence to the protocol and the use of the premade forms for exercises and homework assignments. All treatment sessions will be audiotaped. Each therapist will receive supervision. During supervision, the supervisor will listen to parts of the audiotapes to give feedback and to make sure that the therapist does not drift from the protocol.

Furthermore, after the study is completed, there will be an adherence check. A trained blinded assessor will sample-wise listen to audiotapes of the session to score whether it was an IBA of CBT session.

Nature and extent of the burden and risks associated with participation, benefit and group relatedness: patients will be assessed during five time points, including two follow ups. Further, patients will fill in two questionnaires after each treatment session. Participants of the additional imaging part will be subjected to two extra sessions for MRI scanning and cognitive testing. Healthy controls will be measured once.

Safety Committee: the current study has been classified as low risk. No risk is expected of administering IBA, compared to receiving standard clinical care (CBT). IBA treatment will be conducted by well-trained clinicians. Some psychological discomfort may occur with respect to discussing emotional symptoms, these will be constraint to the minimum. No extra discomfort regarding to standard clinical care is expected. MRI scanning, after extensive screening for eligibility, is classified as a non-significant to low risk.

Statistical analyses: statistical analyses will be performed on the intention-to-treat (ITT) as well as the per-protocol sample (PP). Missing data will be replaced using the multiple imputation method. For all analyses, corrected p-values of <0.05 will be regarded as statistically significant. Safety data (e.g. side effects) will be collected and summarized in the study results. A description of the analyses for the primary and secondary objectives is provided below.

Primary study parameter(s): the YBOCS, a validated questionnaire with 10 items and a maximum score of 40, will be used as the primary outcome parameter. To conclude non-inferiority, the upper margin of the 95 percent confidence interval, i.e. the maximum difference between groups within this 95 percent confidence interval, should not exceed the non-inferiority margin of 2 points on the YBOCS. To determine non-inferiority immediately following treatment and at follow-up, one-sided tests will be conducted at post-treatment, at six month follow-up and one year follow-up assessment. Linear mixed-effects models will be used, regressing the continuous outcome measures on intervention by time interaction terms, both in the ITT- and the PP-sample.

Secondary study parameter(s): mixed-effects models will be used to test the treatment effects between the two treatment conditions for the secondary outcome measures level of insight, using the Overvalued Ideas Scale (OVIS); severity of depressive symptoms, using the Beck Depression Inventory (BDI); severity of anxiety symptoms, using the Beck Anxiety Inventory (BAI); social functioning, using the Sheehan Disability Scale (SDS); quality of life (using the Euroquol); relationship satisfaction, using the Relationship Satisfaction Scale (RSS) and tolerability of treatment, using the Treatment Acceptability/Adherence Scale (TAAS). Exploratively a multiple regression analysis will be used to determine which factors predict treatment outcome.

A linear mixed model will be used to determine whether insight improves more or earlier in patients who underwent IBA compared to those who underwent CBT. A time-lag model (a linear mixed model) will be used to determine whether change in insight is related to symptom reduction.

Neuroimage (pre)processing: fMRI data will be analysed using Statistical parametric mapping (SPM) software or FMRIB Software Library (FSL). Standard group comparisons will be conducted to analyse structural and task-related (symptom provocation) (f)MRI. Full factorial analyses with the factors group (IBA vs CBT) and time (pre- vs posttreatment) will be conducted to determine the effect of treatment on task-related brain activity. Independent component analysis (ICA; FSL MELODIC) and dual regression analyses will be conducted to investigate changes in functional connectivity within networks (e.g. Salience Network, Default Mode Network, CEN). Brain regions with fluctuations concurrent over time in blood-oxygen level dependent (BOLD; proxy for brain activity) will automatically per person be assigned to a component. After filtering and a clean-up of the components (components that seem to result from movement or scanner artefacts), the mean components per group will be determined. By means of non-parametric permutation tests (FSL randomise), functional connectivity of the networks will be compared between the intervention groups and over time. To conduct whole-brain network analyses, the structural MRI will be parcellated into 225 separate regions of interest (ROIs) based on the Brainnetome atlas. These parcellations will be transformed to the resting-state fMRI and the time series will be extracted per ROI and correlated to each other to arrive at a whole-brain network connectivity matrix per subject. Using a similar network approach, the DTI images will allow for tractography in order to measure structural connectivity. The Brain connectivity toolbox (BCT) will be used to calculate topological indices of the networks (e.g. modularity, betweenness centrality, clustering coefficient and efficiency) from these network connectivity matrices. Network topological indices reduce the abundant amount of information resulting from the brain scan to only a few neurobiologically meaningful measures. Global efficiency, for instance, provides a measure of how efficiently information can travel through a network and betweenness centrality of the relative importance of a brain region for the information flow within a network. These topological network measures will be calculated per assessment and condition and subsequently compared by means of permutation tests.

Statistical Neuroimage analysing: to investigate whether treatment modality (IBA/CBT) predicts changes in resting state and task-based functional networks, a multivariate regression analysis will be conducted with treatment modality as independent variable and pre-to posttreatment changes in functional networks as dependent variables. To investigate the association with changes in YBOCS scores and neurocognitive performance, these variables will be entered in the first block. Possible confounders (e.g. medication use, age, disease duration) will be controlled for.

Machine learning techniques will be conducted to investigate whether multimodal pre-treatment brain characteristics (using T-1, DTI and rs- and task-based fMRI) have predictive value for treatment response in the CBT and IBA group separately. The neuroimaging characteristics will be entered into a supervised multivariate classification procedure using a linear support vector machine (SVM) algorithm.

The procedure consists of a training, validation and testing phase. During the training phase, a hyperplane will be estimated that maximally separates the remitters from the non-remitters based on all available data points (features) that show a difference between the groups. The SVM validation phase will consist of a leave-one-per-group-out cross-validation. Then the accuracy will be tested with which the determined hyperplane could classify other patients during the classification stage with independent data (20%) not used for training. This procedure will be iterated 10.000 times for each network. This results in an accuracy measure, per subject, based on the number of times the subject was included in the test sample and correctly classified. The accuracy will be tested on significance. If predictors of response to IBA differ from those to CBT, a post hoc analysis will be carried out with predictors for both IBA and CBT response within the entire OCD group.