To mitigate the influence of several confounding factors, CT was residualized using a linear mixed effects model in which sex, age and intra-cranial volume (ICV) were modeled as fixed effects and site as a random effect. For each dataset, the recruitment site with the largest number of participants was set as the reference. Then PCA was performed on the mean-centered and scaled residuals using the “prcomp” function in R. Due to the sign ambiguity of PCA, column signs of the resulting rotation matrix (i.e., principal axes) were adjusted such that all diagonal elements of the matrix were positive [27]. Standard loadings were calculated by multiplying the rotation matrix and the square root of the variance explained by the PCs. The standard loadings can be interpreted as a Pearson’s correlation coefficient between the variables and a PC or as the contribution of individual variables to a PC [26]. The visualization of cortical statistics was implemented using the R package “ggseg” [28].
