In each contributing study, linear regression was used to calculate per-allele associations of rs1051730–rs16969968 genotype on daily cigarette consumption (cigarettes per day) and cotinine levels (nmol/L) in current smokers. Analyses were conducted for both unadjusted and adjusted for cotinine levels (in the case of associations of genotype with daily cigarette consumption) and cigarette consumption (in the case of associations of genotype with cotinine levels). In the Midspan Family Study, which is one of the Midspan studies (see Supplementary Methods, available online), within-family clustering was adjusted for by including a family-level random intercept (42). We assumed an additive model of genetic action, and a linear relationship between cigarette consumption and cotinine level, consistent with previous reports (3,43). The rs1051730 and rs16969968 variants are in perfect linkage disequilibrium in HapMap3 samples of European ancestry (D′ = 1.0, R2 = 1.0) (http://hapmap.ncbi.nlm.nih.gov/), and therefore, these interchangeable SNPs were considered as a single marker in these analyses. These per-allele associations were pooled using a random effects method (44). Random effects models are typically more conservative than fixed-effects models, although in the absence of substantial
