Nonlinearity in the lung cancer dose-response for airborne arsenic: Apparent confounding by year of hire in evaluating lung cancer risks from arsenicexposure in Tacoma smelter workers

Most analytic studies of human epidemiologic data have affirmed the linear association between excess lung cancer risk and airborne arsenic exposure. Recent Canadian analyses, however, based on the mortality follow-up of Taco ma smelter workers from 1940 1976, provided strong evidence of a nonlinear dose-response when lung cancer risk was expressed in terms of the standardi zed mortality ratio. Using recently updated data covering ten additional ye ars of mortality experience among Tacoma workers (1940-1986), new analyses were undertaken to further explore nonlinearity in the lung cancer dose-res ponse in this worker cohort. Lung cancer risk was expressed in terms of bot h the standardized mortality ratio (SMR) and the excess mortality rate (EMR ). As in Canadian analyses, nonlinearity was assessed through a three param eter model containing both linear and negative exponential terms for dose. Dropping the negative-exponential dose-term lead to the standard suite of l inear dose-response models, with and without intercept, used for comparativ e purposes. Analyses were undertaken by subcohort as there was strong evide nce of confounding by year of initial hire, which largely explained the non linearity in the dose-response observed in Canadian analyses. Subcohort ana lyses based on initial employment, prior to 1940 or thereafter, showed that the nonlinearity in the dose-response was strongly influenced by date of i nitial hire. whether the cohort risk was measured by either the SMR or EMR, a nonlinear dose-response was evident only among workers hired prior to 19 40. This, however, was strongly related to the artifactually low lung cance r mortality seen among workers hired between 1930 and 1939. Among workers h ired after 1940, analyses showed that a linear dose-response provided a cle arly superior fit. While analyses showed comparable goodness of fit when mo dels were fitted to the SMR and EMR. Only those based on the EMR provided s trong evidence of a dose-response. Overall, nonlinearity as observed in Can adian analyses was likely the result of several sources of bias not taken i nto account by Canadian investigators. (C) 1999 Academic Press.