The diversity of individual data sets did not allow for a single reference group for age at onset of smoking across all studies. Accordingly, each site performed logistic regressions with heavy vs light smoking as the dependent variable (modeling the probability of heavy smoking) and sex and rs16969968 (coded additively) as the independent variables, stratified by age at onset of smoking (in individuals with age at onset of smoking ≤16 years and >16 years). Heterogeneity in the meta-analysis was assessed using the goodness-of-fit statistic Q′, detailed in eAppendix 2.
