risk: a life-time mean may not be the ideal predictor even if we were able to estimate it with precision. In addition, our estimates of regression dilution were obtained from a distinct population, geographically unrepresentative of the EPIC Lung study, for which no genotype data was available. Our method also assumes that the extent of day-to-day variation in smoking is independent of genotype, which may not be correct. Taking these limitations together our regression-dilution corrections may be either an under or over correction, and the result should be interpreted with caution. Naturally, similar concerns of regression dilution apply to self-reported CPD, in this case we had repeated estimates from 5 time points. We used these to calculate an average CPD and this was the variable used in the analysis. Nevertheless, this analysis represents a first attempt to circumvent the limitation inherent in most observational studies using a single measurement. SNPs on 19q13 were also associated with cotinine, but the directions of the observed associations were opposite to those originally observed with CPD. Thus, the rs7937 SNP (T allele) on 19q13 was associated with decreased lung cancer risk, consistent with previous study showing an association with lower CPD(8), but increasing levels
