Statistics

Analyses of variance (ANOVAs) were employed to check differences in sociodemographic, clinical and biomarker data between diagnostic groups, and analyses of contingency tables (Χ² tests) to assess associations between nominal variables. Correlations were calculated using Pearson’s product moment correlation coefficients and stepwise automatic univariate regression analysis. We used automatic stepwise binary regression analysis to delineate the significant explanatory variables predicting major depression (versus controls) or melancholic depression (versus no melancholia) as dependent variables. We employed multivariate general linear model (GLM) analysis with the immune markers as dependent variables and diagnoses as explanatory variables, while adjusting for other background variables including age, sex, and drug status. Consequently, we used tests for between-subjects effects to assess which dependent variables were significantly associated with the significant explanatory variables. Estimated marginal mean values (standard error; SE) obtained by GLM analysis were used to interpret differences among the groups based on protected least significant differences (LSD). Principal component (PC) analysis was used as a data reduction method to create one or two composite factors that reflect use of psychotropic drugs (e.g., SSRIS, SNRIs, TCAs, and atypical antipsychotics), namely the first or first two interpretable PCs generated by PC analysis, whereby the number of PCs is based on eigenvalues > 1. For interpretation purposes, we employed varimax rotation and entered the varimax-rotated PC scores in statistical analyses including multivariate GLM and binary regression analyses as indexes of drug use. We employed Ln transformations of the biomarker data to normalize the data distribution (checked with Kolmogorov-Smirnov test). We used the IBM-SPSS windows version 22 to analyze all data. All tests were two-tailed and a p value of 0.05 was used for statistical significance.