The genome‐wide time‐to‐event analyses for smoking initiation, tolerance, and cessation were performed using the Cox proportional hazards (PH) model (Cox 1972) with random effects. We calculated the empirical kinship matrix based on the observed relationship of family members, and employed the coxme R package (Therneau 2012) which implements the Cox PH model with random effects of multivariate normal distribution by utilizing penalized partial likelihood (Ripatti and Palmgren 2000). The selected sample with only one co‐twin from each MZ pair and at most one parent in each family resulted in a kinship matrix in which individuals in a family share the same genetic correlation coefficient, dramatically reducing the computational time of the coxme function. For persistent smoking, we found that over a quarter of individuals (N = 450) became daily smokers within 1 year, and those who did not engage in daily smoking within 20 years since initiation were considered as long‐term survivors. In order to account for this, we first adopted and implemented a mixture cure model to analyze this transition in the context of survival framework (Yu and Peng
