The nominal and reported significance level for the present study was set up to α=0.05. Regression calibration was used to correct for some of the dilution effects due to day-to-day variation in cotinine levels. We obtained repeat measurements one and 3 years apart for 502 individuals, including 96 current smokers who had not changed their smoking status, from the placebo arm of a randomized trial from Norway (WENBIT)(25) . The samples were analyzed in the same laboratory and using the same protocol as the EPIC samples. We used these measurements to estimate the within-individual variance of cotinine, assuming that the long term average was the ideal predictor of lung cancer. This allowed us to calculate regression dilution ratios (RDR) and obtain the adjusted ORs for the effect of cotinine on lung cancer risk by multiplying the observed regression coefficients with the RDR, as described by Clarke et al.(26). To account for the effect of regression dilution in the adjustment of the SNPs OR’s for lung cancer we applied the method described by Rosner et al.(27), modified to the fact that
