Adjustment for smoking in lung cancer analyses in the EPIC cohort
Boshuizen, H.C.; Bueno-de-Mesquita, H.B.; Altenburg, H.P.; Agudo, A.; Le Marchand, L.; Berrino, F.; Janzon, L.; Rasmuson, T.; Vineis, P.; Lukanova, A.; Linseisen, J.; Riboli, E.; Miller, A.
IARC Scientific Publications 156: 59-61
2002
ISSN/ISBN: 0300-5038 PMID: 12484125 Document Number: 545815
The effect of different adjustment procedures for smoking in lung cancer analyses on the relative risk estimates for quartiles of fruit and vegetable consumption was examined. The EPIC (European Prospective Investigation into Cancer and Nutrition) follow-up data released on 25 May 2001, excluding Greece, were used. Dietary data were available for 407 131 persons, of which 499 were lung cancer cases. Three adjustment procedures were compared: the number of pack-years, the six-category index and the data-driven model. Results revealed that the data-driven approach yielded the best fitting model, whereas the number of pack-years approach fitted the worst. The data-driven model contained the following baseline smoking variables: current smoker, former smoker, number of cigarettes currently smoked, duration of smoking and a quadratic term for duration, inhaling and interaction term between gender and number of cigarettes currently smoked. Coefficients for quartiles of vegetable and fruit intake differed between adjustment methods with a magnitude up to the coefficient's standard error. Similar results were obtained when excluding the cases diagnosed within the first 2 years of follow-up from the analysis. It is suggested that outcomes from models using only pack-years as an adjusting variable may be affected by significant residual confounding compared to data-driven adjustment.