For aims 3 and 4, we tested these associations in multivariable models overall, and stratified by sex, and found that associations were robust to adjustment for conventional risk factors in females, but not in males (Fig. 4a, b; polygenic score for loneliness by sex interaction P < 0.05). We next accounted for the genetic correlation between MDD and loneliness using the mtCOJO package [34] (Pearson correlations between polygenic scores are in Supplementary Table 5). The polygenic score for MDD|loneliness was not associated with CAD risk (Fig. 4c), whereas the score for loneliness|MDD remained associated with CAD risk in females, even in the fully adjusted model (Fig. 4d). The polygenic score for loneliness|MDD was associated with a 1.23-fold (95% CI, 1.02–1.47; P = 0.017) increased odds of CAD in females in the fully adjusted model that maximized patient inclusion by including categories for missing values. The association was equivalent, although less statistically significant (OR, 1.23 [95% CI, 0.94–1.60]; P = 0.126) in the fully adjusted model that excluded the 699 female BioVU patients with missing covariate data.
