A multistage model examination: Arsenic exposure, smoking and lung cancer
Lung Cancer mortality associated with arsenic trioxide exposure at the Anaconda Smelter was studied in the context of the multistage model, using a risk difference formulation. Expected lung cancers were based upon a nonsmoking population selected for the contrast it provided for the smoking information. The Anaconda data supported the multistage model used reasonably well but several trends in the data were conflicting, given model predictions.
Lung cancer rates were higher as arsenic exposure, smoking habit and duration of employment increased. Mortality increased with increasing age at first exposure when controlled for duration, and increased with time lapsed since exposure ceased as well. These last two patterns indicate different stages of effect within a multistage framework. The observed latency period was long for a late stage carcinogen, though all of the analyses suggested that the late stage model was the best. The results obtained in this analysis are consistent with those of Brown and Chu (1983a, 1983b) even with the addition of smoking habit information.
Both the stratified and multivariate analyses of the smoking and arsenic data failed to rule out the presence of a multiplicative effect for the two exposures. The data were too sparse to make a judgment with confidence. Lung cancer rates among nonsmokers or light smokers seemed proportionally higher in this population than in others examined. There was evidence of possible confounding between arsenic and smoking in the Anaconda data.
The continued inconsistencies in the evidence suggest that a different formulation of a multistage model should be tried for arsenic-associated lung cancer mortality. Others have suggested that the lung retention time is important, and if this is true, it may help explain the continued rise in lung cancer rates despite the cessation of exposure.