Computational protocol: Cognitive Behavioral Therapy Is Associated With Enhanced Cognitive Control Network Activity in Major Depression and Posttraumatic Stress Disorder

Similar protocols

Protocol publication

[…] To study the association between the severity of depressive symptoms and the level of brain dysfunction, we fit voxelwise linear mixed-effects (LME) models (using the nlme package in R) () on task activation data across all participants (control, MDD, and PTSD). We included all participants instead of patients only because MADRS scores were continuous across healthy and patient populations (), although for depressed participants there were no MADRS scores < 18. This dimensional approach allowed us to characterize a more complete spectrum of depressive symptoms. Thus, our LME model included fixed-effect terms for MADRS, task condition (a categorical variable with four levels), MADRS by condition interaction, age, gender, education, and head motion (mean relative displacement) to estimate their associations with brain activation. A random intercept term was included in the model to account for within-subject correlations among the repeated measures. (See for anxiety analyses.)For the analysis of the longitudinal impact of CBT, we used voxelwise LME modeling and conducted voxelwise analyses within the whole brain. We first tested for a group by time interaction and then assessed activation in patients to identify brain regions that changed significantly after treatment. Specifically, fixed effects included treatment (as an indicator for pre-CBT vs. post-CBT), task condition, task condition by treatment interaction, age, gender, and head motion; a random effect for subject was also included. For clusters identified in this voxelwise analysis, we extracted the region of interest (ROI) means in patients and control subjects at baseline and at 12 weeks. For patients, a post hoc comparison was performed to illustrate the treatment effect divided by diagnosis (MDD vs. PTSD). For control subjects, paired-sample t tests were performed within these a priori defined areas to test for time and practice effects. (See for within-subject analyses in control subjects.) For brain regions showing significant changes in patients following CBT (but not in control subjects), we computed partial correlations (controlling for age, sex, and motion) between change scores of activation and behavioral measures, either depressive symptoms or task reaction time (RT). Both change in symptoms and task RT were quantified using a normalized score defined as 100 × (postCBT − preCBT)/preCBT.For voxelwise analyses, multiple comparisons were corrected using Gaussian random field theory () (easythresh command implemented in FSL). We used a cluster-defining threshold of Z > 3.09 because recent work has emphasized that lower thresholds are subject to a higher risk of type I error. Ecklund et al. () showed that a cluster threshold of Z > 3.09 produced type I error rates that were close to the nominal familywise error rate of alpha = .05, whereas a more liberal threshold of Z > 2.33 resulted in familywise error rates of 10% to 50%. […]

Pipeline specifications

Software tools lme4, nlme
Application Mathematical modeling
Organisms Homo sapiens