For polygenic scores, we first compared those with AD who had reported a suicide attempt to those with AD who had not reported a suicide attempt across all PGSs, independently, using logistic regression in R (version 4.2.1). Second, to ensure that results within those with AD were not biased by conditioning on AD [68], we also compared: 1) those with AD who had a reported suicide attempt, 2) those with AD who had not reported a suicide attempt, and 3) those without AD who had a reported suicide attempt to those who neither reported a suicide attempt nor meet criteria for AD (see Supplemental Table 1 for sample description) using a multinomial logistic regression model in the nnet package in R [69]. In both analyses, we included sex, age, the first three ancestral PCs (PC1-PC3), genotype array, and birth cohort as covariates. To adjust for familial clustering, we used cluster robust standard errors [70, 71]. We stratified analyses by ancestry and then meta-analyzed results (by PGS) within each of the analyses above. All analyses were corrected for multiple testing.
