To confirm which haplotypes are different enough from one another to justify including them as separate parameters in a predictive model, all CYP2A6 haplotypes were considered in terms of their difference from zero activity, defined by the *2 and *4 alleles, and from the most common haplotype, *1B, which is assumed to represent full normal activity (Table 6). Only two haplotypes, *9 and *1A(51A), were significantly different from both zero and the reference haplotype. Therefore a minimal predictive model was constructed using the calculated parameter estimates for these two alleles based on measurements at 30 min, and setting all other alleles as equivalent to either full activity (*1B, *1D, *1H, *14, *1A(51G), *1X2 and *1B7), or no activity (*2, *4, *12 and *1D-Y351H, n=28), with the exception of *1B12 which is considered equivalent to *1A(51A) because these haplotypes are identical at all relevant nucleotides (Table 1). This model requires seven CYP2A6 polymorphisms including the deletion variant, and predicts phenotype in our sample with an R2 of 0.719. Inclusion of gender and smoking status as variables reduces the adjusted R2 marginally
