Statistical analyses

Analyses were run in R version 3.6.0. General additive mixed models using the package mgcv version 1.8-28 were used to derive age functions with a random intercept term per participant. Hippocampal volumes were predicted from (1) a smooth function of age and a linear function of AD-PGS, (2) a smooth function of age and linear functions of AD-PGS and presence/absence of 1 or 2 APOE ε4 alleles, (3) smooth functions of age and AD-PGS and a tensor interaction term of age and AD-PGS, and (4) a linear function of the presence/absence of 1 or 2 APOE ε4 alleles and a smooth interaction between age and presence/absence of 1 or 2 APOE ε4 alleles. In all models, scanner, intracranial volume, sex, and the first 5 GAFs, in addition to genotyping batch (1; n = 1,014, 2; n = 166), were entered as covariates. DBP-PGS and SBP-PGS were added as additional covariates (repeating steps 2 and 4), to further investigate whether any effect was unique to AD-PGS and APOE. We did not have any hypothesis regarding differential hemispheric effects, and bilateral hippocampal volumes were entered in analyses. However, we additionally ran analyses for the left and right hippocampi separately (repeating steps 2 and 4), and to allow for comparison with others,⁶ results presented as e-Methods (links.lww.com/NXG/A316). We computed 95% CIs for the effect of PGS and APOE allelic variation. The reported effect sizes are mm³ difference in hippocampal volume per SD of PGS and when carrying 1 or more ε4 alleles vs not carrying any.