We first performed genome‐wide time‐to‐event analyses for each of the transitions with only sex as a covariate. As an exception, for smoking initiation we also included another genotype‐independent covariate, birth year, as the age of smoking initiation of an individual may be affected by the specific environment in his/her generation (Morabia et al. 2002). As a follow‐up analysis, we then conducted another set of genome‐wide time‐to‐event analyses with additional transition‐specific intermediate covariates in order to obtain comparable results for the following mediation analysis of significant SNPs, and to increase the chance of detecting SNPs associated with the transitions independently of the intermediate covariates. We evaluated biologically plausible covariates for each transition (selected based on literature and a priori knowledge of factors affecting smoking behavior), and included those significantly associated with the transition (P < 0.05) (Table 2), as only significantly associated covariates are eligible as candidate mediators.
