2.7. Statistical analysis

As the MDD exhibit significantly higher levels of rumination compared to HC (t = 7.99(38), p < 1.2E−09), we tested the effect separately for each group to control for group effect. For each group, we conducted Pearson correlations to assess the association between the graph theory network features and the subjective rumination scores. Partial correlation was used to control for age, gender and region volume as covariates. All results were corrected for multiple comparisons using false discovery rate (FDR) (Benjamini and Hochberg, 1995) correction (q < 0.05) where global measures were corrected for the number of global network measures [i.e. 3], and local measures were corrected for number of nodes [i.e. 84].

To further explore the network ROIs that were depicted using the local network features we applied the D_EPNA. The ROIs D_EPNA dependency elements D were correlated to the RRS scores using Pearson correlations. Partial correlations were used to control for age, gender and region volume as covariates. The correlations results were then corrected for multiple comparisons using FDR, where each direction of influences (i.e. influences on, or influenced by all other regions) were corrected for the number of regions minus the ROI [i.e. 83].

For the entropy analysis, for each group we calculated the correlation between each subjects’ RRS scores and their DMN entropy measure using Pearson partial correlation controlling for age and gender.

In order to account for clinical specificity, for all results, comparison between the MDD and HC correlation coefficients were conducted using z-fisher transformation. According to statistical power analysis using Monte Carlo simulation (Gelman and Hill, 2006) (see details in Appendix 1), in order to achieve a statistical power of 80% given a sample size of 20 subjects the magnitude of the effect needed to be as high as r = 0.53 (Figure A1). Finally, to compare the network features between the MDD and HC we conducted a between-group t-test for each global feature and each region's local feature (see details in Appendix 1).