The family-based Generalized Disequilibrium Test (GDT), which employs information from all discordant relative pairs (Chen et al., 2009), was used to test for association with the qualitative traits: AD and each DSM-IV criteria. A linear mixed effects model was employed using the kinship library (lmekin) implemented in R (http://www.inside-r.org/packages/cran/kinship/docs/print.lmekin) to test for association with the quantitative traits reflecting probability of membership in the latent classes. The linear mixed effects model in the kinship function allows for the covariance matrix to be completely specified for the random effects. The result is that each family has a different covariance pattern based on the kinship coefficients, to model the familial genetic random effects. Sex, age at last interview and birth cohort as well as the first principal component (PC1) from the EIGENSTRAT analysis were evaluated as potential covariates. Any significantly associated covariate (p<0.05) was included in all final association models. As described in Wang et al (Wang et al., 2012) imputed data were generated using BEAGLE 3.3.1(Browning and Browning, 2009) in the chromosomal regions providing the greatest evidence of association with the genotyped SNPs.
