variables. The results for both models are presented. Further, unconditional logistic regression models were used to allow stratification by smoking status (current, former and never smokers), adjusting for matching variables (gender, date of birth, date of blood collection and country). To investigate if the risk effect of genotypes is constant across different levels of smoking exposure we tested for multiplicative interaction of genotype with quartiles of cotinine levels/CPD. Likelihood ratio tests, comparing the models with and without the interaction terms, were used to evaluate statistical significance.
