for each score was: GPSall (112 594), GPS0.5 (57 053), GPS0.1 (12 161), GPS0.05 (6268) and GPS0.01 (1402). The resulting GRS was used to predict AO-AD, as well as other measures (AO-R, AO-I, AD-SX and Maxdrinks) in those datasets. Associations between age of onset measures and GRS were conducted using Cox proportional hazards analysis. Logistic and linear regression was used for dichotomous (AD) and continuously distributed (AD-SX, Maxdrinks) measures, respectively. Similar to the analysis in the discovery sample, a robust sandwich variance estimator approach was used to account for the familial correlation among OZ-ALC families. For both studies, sex, age at last interview and study source (COGEND vs FSCD vs COGA; NAG vs EDAC vs BIGSIB) were included as covariates.
