For network parameters to be potentially informative markers or latent traits or vulnerability, they should be stable over time. Therefore, we first compared within patients their depressed state versus their recovered state. For this longitudinal comparison, all patients were included that reached K-SADS diagnostic criteria at baseline and no longer met criteria at follow-up (n=232, 49.9%). Second, for any stable latent trait to be informative, it needs to be associated with the phenotype (depression). Therefore, we compare acutely depressed adolescents (n=377) to their age- and sex matched never-depressed counterparts (n=377). Third, we are interested to see whether network parameters may be informative in terms of prognosis, as was suggested by Van Burkolo et al. in adults with depression. Therefore, we median split our sample in terms of good and poor responders to treatment (n=233 and n=232, respectively) and compared their baseline network parameters.

To compare two networks, we modified the network comparison test as provided by Claudia van Borkulo. By permutation testing, the NCT estimates the chances of two networks arising from the same (null hypothesis) or two different populations (alternative hypothesis). Briefly, the NCT merges both datasets, and then does 1000 permutations of random splits of the data in two. In permutations for cross-sectional comparisons, each subject is randomly assigned to one of two groups. In permutations for longitudinal comparisons, the order or time points is permuted within each subject. NCT then estimates network strength for both datasets, and compares network strengths and node strengths. By repeating this procedure 1000 times, NCT calculates a distribution of differences in network strengths if the split was based on chance, against which one can compare the observed difference between networks. We customised the NCT script such that for each permuted split of the data, it estimates the two thresholded binary networks in accordance with the procedures used to estimate the observed networks (as described above).
For each comparison, we report significant differences in global network strength as well as node strength for each of the individual MFQ items. For differences in global network strength, alpha is set at 0.05. For differences in node strength, alpha is conservatively set to 0.05/31=0.0016.