1.1 Environmental mixtures

Common approaches that have been used on a daily basis in environmental epidemiology might fail to capture the complexity of exposures in our world. For several years, despite recognizing that individuals are commonly exposed to multiple environmental factors, the “one-at-the-time” approach has remained the standard practice in most epidemiological research. To better understand what we mean by “one-at-the-time” approach, and its limitations, let’s think of a study where we want to evaluate the effects of parabens - endocrine disrupting chemicals commonly used in the production of personal care products and cosmetics - on diabetes in a population of 1000 individuals. Let’s assume that through urine samples analysis we were able to detect concentrations of three common parabens compounds (metylparaben, butylparaben, propylparaben) in most of our individuals. The “one-at-the-time” approach would build 3 independent statistical models (these could even be very sophisticated models that account for any level of data complexity), one for each parabens compound, adjusting for potential confounders of the associations but without taking into account the other 2 detected compounds. This approach is subject to three main limitations:

  • We know that individuals are exposed to multiple factors, and we might want to estimate the joint (also knows as cumulative) effects of these chemicals. A “one-at-the-time” approach does not allow responding to this question.

  • Is there any interaction between the three compounds in predicting diabetes? A “one-at-the-time” approach does not allow responding to this question.

  • Last but not least, this approach is making strong assumptions with regards to the causal structure underlying the data. Specifically, we are assuming that, very unrealistically, the association between each compound and the outcome is not confounded by the presence of any of the other compounds. The “one-at-the-time”is affected by confounding bias.

To overcome these 3 major limitations we need to evaluate exposure to parabens as a mixture of the three evaluated compounds, building a single statistical model that could jointly evaluate the three exposures and possibly accounting for co-confounding, interactions, and other specific features of the data. Obtaining such statistical model is not easy, and things would only get more complex if we wanted to account for a larger mixture of chemicals, or even to incorporate several groups of exposures in an exposome-wide analysis. Over the last decade or so, many researchers have focused their effort on developing statistical approaches for environmental mixtures, adapting techniques from other fields or developing new methodologies from scratch. The National Institute of Environmental Health Sciences (NIEHS) launched a specific initiative, called Powering Research Through Innovative Methods for Mixtures in Epidemiology (PRIME), to encourage methods developments in this direction, and organized workshops and symposiums on the topics. An important symposium in 2015 identified several available approaches and discussed advantages and limitations for each (Taylor et al. 2016).

Approaches discussed by NIEHS in 2015 (from Taylor et al. 2016)

Figure 1.2: Approaches discussed by NIEHS in 2015 (from Taylor et al. 2016)

Five years later the number of available approaches has multiplied, and several of the discussed methodologies have been extended, revised, and presented to the public. The field of environmental epidemiology is gradually moving to a multi-pollutants or multi-chemical framework as a default (Dominici et al. 2010), leading the ground in exposome research, and more and more papers are published every year within this topic.

The goal of this class (and of this book) is to present and discuss some of these approaches, presenting their advantages and limitations and, most importantly, discussing what research question they target and when they should be chosen to evaluate environmental mixtures. While it is impossible to cover all available techniques in a short time, references for alternative methodologies that are not discussed here will be provided. Most of the examples and discussion will focus on environmental exposures; it comes without saying that extension of these approaches into other fields of exposome research (e.g. evaluating multiple nutrients, multiple lifestyle factors ) is recommended and would provide enormous benefits.


Dominici, Francesca, Roger D Peng, Christopher D Barr, and Michelle L Bell. 2010. “Protecting Human Health from Air Pollution: Shifting from a Single-Pollutant to a Multi-Pollutant Approach.” Epidemiology (Cambridge, Mass.) 21 (2): 187.
Taylor, Kyla W, Bonnie R Joubert, Joe M Braun, Caroline Dilworth, Chris Gennings, Russ Hauser, Jerry J Heindel, Cynthia V Rider, Thomas F Webster, and Danielle J Carlin. 2016. “Statistical Approaches for Assessing Health Effects of Environmental Chemical Mixtures in Epidemiology: Lessons from an Innovative Workshop.” Environmental Health Perspectives 124 (12): A227–29.