Data were analyzed using the Statistical Analysis System (SAS). Logistic regression was used to model dichotomous outcomes of smoking initiation, daily smoking, nicotine dependence, time to first cigarette, and cigarettes per day. In the primary analyses in COGA and COGEND, the continuous metabolism metric, sex, study site, and last interview age were included as predictor variables. In COGA, family structure was accounted for using generalized estimating equations via PROC GENMOD. Results from the COGEND replication sample were meta-analyzed with the primary COGA results (Table 2) using a publically available SAS macro (http://www.hsph.harvard.edu/donna-spiegelman/software/metaanal/). Meta-analyses results were based on fixed effect models to determine the evidence for association within the collected samples. In these analyses, we did not observe heterogeneity between the two studies based on the Q statistic (p>0.1).
