Methods for testing interactions, with applications to occupational exposures, smoking, and lung cancer

Abstract

Various approaches to assessing the interaction between smoking and occupational exposures are described. The definition of interaction depends on the measure of association under consideration and can be expressed in terms of disease risk, time, or dose. Simple descriptive methods and maximum likelihood model fitting methods are presented for analyzing interactions in terms of joint effects on disease risk. Methods for assessing the influence of exposures on times to disease, using appropriate denominators, are also described. The various approaches are illustrated with published data on lung cancer in relation to asbestos, radon daughters, chloromethyl ethers, and arsenic. Some of the mechanisms that can be invoked to explain the observed patterns include initiation and promotion, lung clearance, transport across cell membranes, and mucous secretion.

Latency Models for Analyses of Protracted Exposures

“…Examples of such investigations have been reported previously, although the estimates obtained from such model fittings may be unstable unless there are substantial numbers of cases within each stratum of the potential modifying factor 8,13…”

“…A lognormal probability density function ( pdf ) for the latent period also can describe a rise and subsequent fall in the effect of exposure with time-since-exposure 6,8. This implies a weighting function, w ( t ) = pdf (Lognormal, t,θ , λ ), where θ and λ are the location and scale parameters for this distribution and the modal value of the latency function is given by exp( θ − λ 2 ).…”

Lung cancer in chrysotile asbestos workers: analyses based on the two-stage clonal expansion model

“…However, prior work suggests that the degree of confounding by lifestyle factors, such as cigarette smoking, will be small in occupational analyses based upon internal comparisons of production workers within a single facility [15,16]. Cigarette smoking may also be a potential effect modifier of the association between chrysotile asbestos exposures and lung cancer mortality [17]; prior work suggests that the joint effect of these exposures is close to that expected under a multiplicative model [18]. However, the lack of individual level information on smoking history limits the ability to assess such interactions or to evaluate these effects via the TSCE model.…”

Contrasting Theories of Interaction in Epidemiology and Toxicology

“…CA has rarely been mentioned by epidemiologists as a method for analyzing interaction (Cornfield 1975; Thomas and Whittemore 1988), yet it is widely used by toxicologists, particularly in the TEF form. Just as we expressed the total effect of the joint exposure ( A,B ) in terms of an isoeffective dose of B given by γ A + B , the TEF method allows toxicologists to express a mixture of similarly acting polychlorinated dioxins, polychlorinated dibenzofurans, and polychlorinated biphenyls by a single equipotent dose.…”