All models were adjusted for gender (in non-stratified models), self-reported identification as Black and/or Hispanic, study site, age, age cohort (0: born from 1982 to 1993, 1: born 1973 or earlier), and in genomic models, ancestral PCs1-3. For EEG coherence analyses, age at EEG assessment was also included as a covariate. All cross-term interactions (PRS*gender, COVID-19 latent factors*gender, COVID-19 latent factors*age, etc.) were included. All models were adjusted for familial clustering. Models were run simultaneously (i.e., all intercorrelations among factors accounted for), thus estimates are adjusted for all parameters and standardized. Given multiple comparisons across models, a Bonferroni correction adjusting for nine correlated latent factors across four groups was applied (p < 0.001). While primary and stratified models had acceptable statistical power (alpha > 0.80), statistical power was weaker for the interaction models, including PRS and EEG variables. This was the primary reason we included these as ‘exploratory models’ that will require replication in larger samples.
