Thomas Webster, D.Sc.; Marc Weisskopf, Ph.D., Sc.D.
Boston University, Harvard School of Public Health
P42ES007381, R01ES028800, R01ES027813, R01ES024165, P30ES000002
NIEHS grantees found that certain approaches for analyzing mixtures of chemicals can, in some cases, amplify bias from unknown variables. This may be a concern when developing statistical approaches to identify the critical exposures from among the many that may contribute to health effects. Identifying these effects is a key goal of epidemiological analysis of complex chemical mixtures, which represent the multitude of chemicals that people are exposed to in everyday life.
The authors used directed acyclic graphs, a method used in epidemiology to display assumptions about causal relationships between variables in a specific context. When they assessed links between exposures and human health, certain exposures were often highly correlated. Unknown or unmeasured variables might also affect the estimates of health effects from mixtures. The study showed how including correlated exposures in a typical regression model could increase the effect, or bias, of unknown or unmeasured variables.
According to the authors, researchers must consider steps to minimize possible co-exposure amplification bias when designing and analyzing epidemiological studies of mixtures. They added that these studies would greatly benefit from interdisciplinary research. Options to remedy this issue include identification and control of the unknown variables, or other analytical approaches. The authors also discussed the importance of exposure assessment studies to understand the strength and basis for correlations between exposures, as well as physiological and toxicological information to improve interpretation of results. This may be particularly important when examining biological markers of exposure.
Citation: Weisskopf MG, Seals RM, Webster TF. 2018. Bias amplification in epidemiologic analysis of exposure to mixtures. Environ Health Perspect 126(4):047003.