Second, to assess the overall effect of the variants at the gene level, the association results were analyzed using Multi‐marker Analysis of GenoMic Annotation gene‐based tests (MAGMA, version 2). 41 Because not all phenotypes were present in all cohorts, we conducted these analyses separately per cohort. SNPs were mapped to CADM2 using 1000Genomes phase 3 data. We used the snp‐wise = top procedure, which is more sensitive when only a small proportion of SNPs in the gene shows an association. To control for the number of phenotypes tested, we computed the Benjamini–Hochberg False Discovery Rate (FDR) 42 p‐values within each variable category, using R (version 3.6.2). 43 When reporting the results, we present uncorrected p‐values with an asterisk indicating if the FDR‐corrected p‐value was below p = .05.
