According to Box-Cox transformation 27 that identified approximately the best transformation of dependent variables, which could be applied to all cohorts, we studied log2-transformed age at smoking initiation and lifetime average CPD, and the square root of the current CPD. Regression models were run under an additive model coded SNPs, while adjusting for potential confounders (see online supplementary material for details). A fixed effect model was used for all meta-analyses. Effect allele was defined as the one associated with later age at smoking initiation, higher CPD or higher odds for smoking cessation. Meta-analytic p<5×10−8 was considered as genome-wide significant;28 5×10−8 ≤p<5×10−7 was interpreted as a suggestive association in the genome-wide panel with p<0.05 a suggestive association for candidate SNPs.
