A Cox proportional hazards regression model was employed to model age at onset as a function of genotype and—as above—of other relevant covariates (i.e., birth cohort, sex, three PCs and study specific covariates). We included this approach as it utilizes all available information on the age of initiation among those who have initiated. It is expected to show superior power relative to an analysis of the “ever-never” dichotomy or an analysis restricted to those who initiated (see e.g. Kiefer et al. 2013). The Cox proportional hazard regression analysis was performed genomewide by accessing the survival R-package (Therneau 2014) from Plink. In fitting the model, we used the cluster option to get sandwich corrected standard errors that are robust to possible misspecification of the familial covariance matrix.
