Statistical methods
Component analysis and cluster analysis were performed following the first study (12). The optimal number of clusters was based on the scree plot, and additional for visualization a “cluspot” was drawn, plotting subjects in two dimensions after multidimensional scaling. The variables SAE, TNFα, IL-17, and IFNɣ, were omitted from the cluster analysis because they were not or barely detected in the samples. We calculated differences between patients and controls and clusters and controls using the Mann-Whitney U test. Between-cluster differences of all parameters were tested by using the Kruskal-Wallis test with multiple group-comparisons. Overall, the p-values of ≤0.05 were set to be significant.