Bootstrap based multiple imputation was used to manage missing survey data, according to the protocol by Honaker and King, which utilizes the longitudinal nature of the data to improve the imputation process [8]. Ten imputed sets of data were generated, and multiple logistic regression within a generalized estimating equation (GEE) framework was performed to examine the association between CYP2A6 metabolic group (independent variable) and cessation (dependent variable) while accounting for the correlation between measures within individuals. No covariates were included. The additive model assumed a linear relationship between CYP2A6 metabolic group, consistent with genotype group impact on metabolism [3], and the odds of quitting. The statistical programs used to complete the analysis were R (version 2.14.1, available online: http://www.r-project.org/), the Amelia II package for multiple imputation (version 1.5-5, available online: http://gking.harvard.edu/amelia/), and the Zelig package for GEE models (version 3.5.4, available online: http://gking.harvard.edu/zelig).
