We performed formal tests of interaction of rs1659258 by sex (women as compared to men) and age (stratified above/below age 60). Briefly, each study generated the interaction regression coefficient, its standard error and its p value through regression. For the sex interaction, we included, age, age-squared, smoking (yes/no), sex, and any principal components (and study center) that were used in the original discovery analysis. We additionally added rs1659258 and the cross-product rs1659258*sex. The age interaction analysis was performed in women only. Because the AGES and Health ABC studies only included participants older than 65 years, they did not contribute to this analysis. Model covariates included age, age-squared, smoking (yes/no), and any principal components (and study center) with the addition of rs1659258 and the cross-product of rs1659258 and the dichotomized age term. We meta-analyzed the interaction terms across all studies using the weighted z-score approach.
