Contrasting Theories of Interaction in Epidemiology and Toxicology

Abstract

Background: Epidemiologists and toxicologists face similar problems when assessing interactions between exposures, yet they approach the question very differently. The epidemiologic definition of “interaction” leads to the additivity of risk differences (RDA) as the fundamental criterion for causal inference about biological interactions. Toxicologists define “interaction” as departure from a model based on mode of action: concentration addition (CA; for similarly acting compounds) or independent action (IA; for compounds that act differently).Objectives: We compared and contrasted theoretical frameworks for interaction in the two fields.Methods: The same simple thought experiment has been used in both both epidemiology and toxicology to develop the definition of “noninteraction,” with nearly opposite interpretations. In epidemiology, the “sham combination” leads to a requirement that noninteractive dose–response curves be linear, whereas in toxicology, it results in the model of CA. We applied epidemiologic tools to mathematical models of concentration-additive combinations to evaluate their utility.Results: RDA is equivalent to CA only for linear dose–response curves. Simple models demonstrate that concentration-additive combinations can result in strong synergy or antagonism in the epidemiologic framework at even the lowest exposure levels. For combinations acting through nonsimilar pathways, RDA approximates IA at low effect levels.Conclusions: Epidemiologists have argued for a single logically consistent definition of interaction, but the toxicologic perspective would consider this approach less biologically informative than a comparison with CA or IA. We suggest methods for analysis of concentration-additive epidemiologic data. The two fields can learn a great deal about interaction from each other.

Cumulative Health Risk Assessment: Finding New Ideas and Escaping from the Old Ones

“…But, paraphrasing John Maynard Keynes' observation on economic policy (1936)-the difficulty with conducting pertinent cumulative risk assessments lies not so much in developing new ideas as in escaping from the old ones. Toxicologists tend to approach CRA from a bottom-up, biological-mechanism perspective, which is often thwarted because of a lack of adequate data on important interaction mechanisms and related health outcomes (NRC 2009;Howard and Webster 2013). Epidemiologists, in contrast, tend to view CRA from a population-based, top-down Hum.…”

“…perspective, but are frequently frustrated by intrinsic problems associated with exposure evaluation, identification of confounders and co-factors, and validation of causal linkages (Levy 2008;Howard and Webster 2013). For CRA to be both realistic and relevant, it will be necessary to move beyond historical precedents and conventional thinking to conceive fresh analytical approaches, develop innovative methods for acquiring essential data, and implement appropriate science-policy procedures that ensure rational and defensible risk estimates.…”

A family of partial-linear single-index models for analyzing complex environmental exposures with continuous, categorical, time-to-event, and longitudinal health outcomes

“…For example, molar sums were used to show relationships between prenatal phenol and phthalate exposures and birth outcome (73), and a potency-weighted sum was used to calculate phthalates exposures among reproductive-aged women (74). The weights for environmental factors can be calculated from their expected potency relative to a reference factor, like the common cases in toxicology (75), or based on their percent contribution to the total mixture effect, like WQS (9). PLSI models can be considered as one of these weighting approaches, and their advantages from the semiparametric structure are In this study, the coherent algorithms for PLSI models are based on the 'gam' and 'gamm' functions from 'mgcv' package and 'qgam' function from 'qgam' package in R, which includes many of the generalized additive model (GAM) fitting techniques developed by Simon Wood et al (77).…”

Joint Action of Binary Factors in the Sufficient Causes Theory and Its Classification

“…We note, however, that at the current formalization stage, it is not possible to formulate concept of such a response function, which could be considered an analogue of the summing effects function known in epidemiology [9,[33][34][35]. With respect to this function, the observed dose-response dependences are compared, and on the basis of that comparison a conclusion is drawn about the manifestation of a sub-or superadditive joint action.…”