Secondly, the genotype imputation accuracy could have had an impact on our results. Many top SNPs from the analyses of age at smoking initiation were imputed in all 4 cohorts; however, the association peaks of these regions also contained SNPs genotyped in the majority (or even all) of the cohorts, which likely makes these findings more reliable. Given the fact that different, and not fully overlapping, genotyping chips were used, this imputation was crucial to obtain a comprehensive overview of many genetic associations. In our study, this is of special importance for the nonsynonymous SNP rs1801272 in CYP2A6, which had to be imputed in all cohorts and has no known proxy SNPs. Assuming that the association of this SNP with CPD is a true positive and possibly causal, imputation was the only one way to detect it. Lastly, we must acknowledge that COPD diagnosis has a significant impact on smoking behaviors studied, and especially on current CPD and smoking cessation. We took into account the severity of COPD, reflected in the level of lung function, as a potential confounder when analyzing these two phenotypes. However, we hypothesize that additional factors such as frequency of exacerbations may also affect smoking behaviors
