Mediation analysis in epidemiology: methods, interpretation and bias
01 Oct 2013-International Journal of Epidemiology (Oxford University Press)-Vol. 42, Iss: 5, pp 1511-1519
TL;DR: The impact of the three main sources of potential bias in the traditional approach to mediation analyses are reviewed and discussed: (i) mediator-outcome confounding; (ii) exposure-mediator interaction and (iii) mediATOR- outcome confounding affected by the exposure.
Abstract: In epidemiological studies it is often necessary to disentangle the pathways that link an exposure to an outcome. Typically the aim is to identify the total effect of the exposure on the outcome, the effect of the exposure that acts through a given set of mediators of interest (indirect effect) and the effect of the exposure unexplained by those same mediators (direct effect). The traditional approach to mediation analysis is based on adjusting for the mediator in standard regression models to estimate the direct effect. However, several methodological papers have shown that under a number of circumstances this traditional approach may produce flawed conclusions. Through a better understanding of the causal structure of the variables involved in the analysis, with a formal definition of direct and indirect effects in a counterfactual framework, alternative analytical methods have been introduced to improve the validity and interpretation of mediation analysis. In this paper, we review and discuss the impact of the three main sources of potential bias in the traditional approach to mediation analyses: (i) mediator-outcome confounding;(ii) exposure-mediator interaction and (iii) mediator-outcome confounding affected by the exposure. We provide examples and discuss the impact these sources have in terms of bias.
Citations
More filters
TL;DR: This review provides a survey and synthesis of four areas of active methodological research: (a) mediation analysis for longitudinal data, (b) causal inference for indirect effects, (c) mediationAnalysis for discrete and nonnormal variables, and (d) mediation assessment in multilevel designs.
Abstract: Mediation processes are fundamental to many classic and emerging theoretical paradigms within psychology. Innovative methods continue to be developed to address the diverse needs of researchers studying such indirect effects. This review provides a survey and synthesis of four areas of active methodological research: (a) mediation analysis for longitudinal data, (b) causal inference for indirect effects, (c) mediation analysis for discrete and nonnormal variables, and (d) mediation assessment in multilevel designs. The aim of this review is to aid in the dissemination of developments in these four areas and suggest directions for future research.
517 citations
TL;DR: Novel ways by which researchers can apply and interpret the Bradford Hill criteria when considering data gathered using modern molecular techniques, such as epigenetics, biomarkers, mechanistic toxicology, and genotoxicology are discussed.
Abstract: In 1965, Sir Austin Bradford Hill published nine “viewpoints” to help determine if observed epidemiologic associations are causal. Since then, the “Bradford Hill Criteria” have become the most frequently cited framework for causal inference in epidemiologic studies. However, when Hill published his causal guidelines—just 12 years after the double-helix model for DNA was first suggested and 25 years before the Human Genome Project began—disease causation was understood on a more elementary level than it is today. Advancements in genetics, molecular biology, toxicology, exposure science, and statistics have increased our analytical capabilities for exploring potential cause-and-effect relationships, and have resulted in a greater understanding of the complexity behind human disease onset and progression. These additional tools for causal inference necessitate a re-evaluation of how each Bradford Hill criterion should be interpreted when considering a variety of data types beyond classic epidemiology studies. Herein, we explore the implications of data integration on the interpretation and application of the criteria. Using examples of recently discovered exposure–response associations in human disease, we discuss novel ways by which researchers can apply and interpret the Bradford Hill criteria when considering data gathered using modern molecular techniques, such as epigenetics, biomarkers, mechanistic toxicology, and genotoxicology.
404 citations
Cites background from "Mediation analysis in epidemiology:..."
...For example, the biostatistical approach of mediation analysis allows for the disentanglement and decomposition of the various biological pathways of direct and indirect effects that play a role in filling the “black box” between exposures and observable outcomes [47]....
[...]
TL;DR: Challenges in interpreting Mendelian randomization analyses are described, including those from studies using genetic variants to assess causality of multiple traits; studies describing pleiotropic variants; and those investigating variants that disrupt normal function of an exposure.
Abstract: Mendelian randomization (MR) is a burgeoning field that involves the use of genetic variants to assess causal relationships between exposures and outcomes MR studies can be straightforward; for example, genetic variants within or near the encoding locus that is associated with protein concentrations can help to assess their causal role in disease However, a more complex relationship between the genetic variants and an exposure can make findings from MR more difficult to interpret In this Review, we describe some of these challenges in interpreting MR analyses, including those from studies using genetic variants to assess causality of multiple traits (such as branched-chain amino acids and risk of diabetes mellitus); studies describing pleiotropic variants (for example, C-reactive protein and its contribution to coronary heart disease); and those investigating variants that disrupt normal function of an exposure (for example, HDL cholesterol or IL-6 and coronary heart disease) Furthermore, MR studies on variants that encode enzymes responsible for the metabolism of an exposure (such as alcohol) are discussed, in addition to those assessing the effects of variants on time-dependent exposures (extracellular superoxide dismutase), cumulative exposures (LDL cholesterol), and overlapping exposures (triglycerides and non-HDL cholesterol) We elaborate on the molecular features of each relationship, and provide explanations for the likely causal associations In doing so, we hope to contribute towards more reliable evaluations of MR findings
400 citations
TL;DR: Investigation of associations between prenatal exposure to maternal smoking and offspring DNA methylation at multiple time points in approximately 800 mother–offspring pairs found that the major contribution to altered methylation was attributed to a critical window of in utero exposure.
Abstract: Maternal smoking during pregnancy has been found to influence newborn DNA methylation in genes involved in fundamental developmental processes. It is pertinent to understand the degree to which the offspring methylome is sensitive to the intensity and duration of prenatal smoking. An investigation of the persistence of offspring methylation associated with maternal smoking and the relative roles of the intrauterine and postnatal environment is also warranted. In the Avon Longitudinal Study of Parents and Children, we investigated associations between prenatal exposure to maternal smoking and offspring DNA methylation at multiple time points in approximately 800 mother–offspring pairs. In cord blood, methylation at 15 CpG sites in seven gene regions (AHRR, MYO1G, GFI1, CYP1A1, CNTNAP2, KLF13 and ATP9A) was associated with maternal smoking, and a dose-dependent response was observed in relation to smoking duration and intensity. Longitudinal analysis of blood DNA methylation in serial samples at birth, age 7 and 17 years demonstrated that some CpG sites showed reversibility of methylation (GFI1, KLF13 and ATP9A), whereas others showed persistently perturbed patterns (AHRR, MYO1G, CYP1A1 and CNTNAP2). Of those showing persistence, we explored the effect of postnatal smoke exposure and found that the major contribution to altered methylation was attributed to a critical window of in utero exposure. A comparison of paternal and maternal smoking and offspring methylation showed consistently stronger maternal associations, providing further evidence for causal intrauterine mechanisms. These findings emphasize the sensitivity of the methylome to maternal smoking during early development and the long-term impact of such exposure.
316 citations
Cites methods from "Mediation analysis in epidemiology:..."
...However, the method of adjusting for a potential mediator in standard regression models to estimate the direct effect of an exposure may produce spurious conclusions (27,28)....
[...]
Posted Content•
TL;DR: It is shown that natural effects can be identified by methods that go beyond standard adjustment for confounders, applicable to observational studies in which treatment assignment remains confounded with the mediator or with the outcome.
Abstract: This paper reviews the foundations of causal mediation analysis and offers a general and transparent account of the conditions necessary for the identification of natural direct and indirect effects, thus facilitating a more informed judgment of the plausibility of these conditions in specific applications. We show that the conditions usually cited in the literature are overly restricted, and can be relaxed substantially, without compromising identification. In particular, we show that natural effects can be identified by methods that go beyond standard adjustment for confounders, applicable to observational studies in which treatment assignment remains confounded with the mediator or with the outcome. These identification conditions can be validated algorithmically from the diagramatic description of one's model, and are guaranteed to produce unbiased results whenever the description is correct. The identification conditions can be further relaxed in parametric models, possibly including interactions, and permit us to compare the relative importance of several pathways, mediated by interdependent variables.
298 citations
Cites background from "Mediation analysis in epidemiology:..."
...Recently, efforts have been made to interpret these conditions in more conceptually meaningful way, so as to enable researchers to judge whether the necessary assumptions are scientifically plausible (Coffman and Zhong, 2012; Imai et al., 2010a, 2011a; Muthén, 2011; Richiardi et al., 2013; Valeri and VanderWeele, 2013; VanderWeele, 2009)....
[...]
...…these conditions in more conceptually meaningful way, so as to enable researchers to judge whether the necessary assumptions are scientifically plausible (Coffman and Zhong, 2012; Imai et al., 2010a, 2011a; Muthén, 2011; Richiardi et al., 2013; Valeri and VanderWeele, 2013; VanderWeele, 2009)....
[...]
...…2009; Huber, 2012; Imai et al., 2010c, 2011b; Jo et al., 2011; Joffe et al., 2007; Kaufman, 2010; Mortensen et al., 2009; Petersen et al., 2006; Richiardi et al., 2013; Robins, 2003; Sobel, 2008; Ten Have et al., 2004; Valeri and VanderWeele, 2013; VanderWeele and Vansteelandt, 2009;…...
[...]
...1) as the gold standards for analysis (e.g., Albert and Nelson, 2011; Coffman and Zhong, 2012; Hafeman and Schwartz, 2009; Huber, 2012; Imai et al., 2010c, 2011b; Jo et al., 2011; Joffe et al., 2007; Kaufman, 2010; Mortensen et al., 2009; Petersen et al., 2006; Richiardi et al., 2013; Robins, 2003; Sobel, 2008; Ten Have et al., 2004; Valeri and VanderWeele, 2013; VanderWeele and Vansteelandt, 2009; Vansteelandt et al., 2012)....
[...]
References
More filters
TL;DR: This article seeks to make theorists and researchers aware of the importance of not using the terms moderator and mediator interchangeably by carefully elaborating the many ways in which moderators and mediators differ, and delineates the conceptual and strategic implications of making use of such distinctions with regard to a wide range of phenomena.
Abstract: In this article, we attempt to distinguish between the properties of moderator and mediator variables at a number of levels. First, we seek to make theorists and researchers aware of the importance of not using the terms moderator and mediator interchangeably by carefully elaborating, both conceptually and strategically, the many ways in which moderators and mediators differ. We then go beyond this largely pedagogical function and delineate the conceptual and strategic implications of making use of such distinctions with regard to a wide range of phenomena, including control and stress, attitudes, and personality traits. We also provide a specific compendium of analytic procedures appropriate for making the most effective use of the moderator and mediator distinction, both separately and in terms of a broader causal system that includes both moderators and mediators.
80,095 citations
TL;DR: A discussion of matching, randomization, random sampling, and other methods of controlling extraneous variation is presented in this paper, where the objective is to specify the benefits of randomization in estimating causal effects of treatments.
Abstract: A discussion of matching, randomization, random sampling, and other methods of controlling extraneous variation is presented. The objective is to specify the benefits of randomization in estimating causal effects of treatments. The basic conclusion is that randomization should be employed whenever possible but that the use of carefully controlled nonrandomized data to estimate causal effects is a reasonable and necessary procedure in many cases. Recent psychological and educational literature has included extensive criticism of the use of nonrandomized studies to estimate causal effects of treatments (e.g., Campbell & Erlebacher, 1970). The implication in much of this literature is that only properly randomized experiments can lead to useful estimates of causal effects. If taken as applying to all fields of study, this position is untenable. Since the extensive use of randomized experiments is limited to the last half century,8 and in fact is not used in much scientific investigation today,4 one is led to the conclusion that most scientific "truths" have been established without using randomized experiments. In addition, most of us successfully determine the causal effects of many of our everyday actions, even interpersonal behaviors, without the benefit of randomization. Even if the position that causal effects of treatments can only be well established from randomized experiments is taken as applying only to the social sciences in which
8,377 citations
TL;DR: In this paper, the authors introduce marginal structural models, a new class of causal models that allow for improved adjustment of confounding in observational studies with exposures or treatments that vary over time, when there exist time-dependent confounders that are also affected by previous treatment.
Abstract: In observational studies with exposures or treatments that vary over time, standard approaches for adjustment of confounding are biased when there exist time-dependent confounders that are also affected by previous treatment. This paper introduces marginal structural models, a new class of causal models that allow for improved adjustment of confounding in those situations. The parameters of a marginal structural model can be consistently estimated using a new class of estimators, the inverse-probability-of-treatment weighted estimators.
4,655 citations
TL;DR: Causal diagrams can provide a starting point for identifying variables that must be measured and controlled to obtain unconfounded effect estimates and provide a method for critical evaluation of traditional epidemiologic criteria for confounding.
Abstract: Causal diagrams have a long history of informal use and, more recently, have undergone formal development for applications in expert systems and robotics. We provide an introduction to these developments and their use in epidemiologic research. Causal diagrams can provide a starting point for identifying variables that must be measured and controlled to obtain unconfounded effect estimates. They also provide a method for critical evaluation of traditional epidemiologic criteria for confounding. In particular, they reveal certain heretofore unnoticed shortcomings of those criteria when used in considering multiple potential confounders. We show how to modify the traditional criteria to correct those shortcomings.
2,983 citations
TL;DR: A graphical approach to the identification and computation of causal parameters in mortality studies with sustained exposure periods is offered and an adverse effect of arsenic exposure on all-cause and lung cancer mortality which standard methods fail to detect is found.
2,223 citations