Email updates

Keep up to date with the latest news and content from Environmental Health and BioMed Central.

Open Access Highly Accessed Commentary

Causal models in epidemiology: past inheritance and genetic future

Paolo Vineis1* and David Kriebel2

Author Affiliations

1 Division of Epidemiology, Public Health and Primary Care, Imperial College London, Faculty of Medicine, Imperial College London, St Mary's Campus, Norfolk Place, London, W2 1PG, UK

2 Department of Work Environment, School of Health and Environment, University of Massachusetts Lowell, Kitson Hall, Room 200 (UML North), 1 University Avenue, Lowell, MA 01854, USA

For all author emails, please log on.

Environmental Health: A Global Access Science Source 2006, 5:21  doi:10.1186/1476-069X-5-21

Published: 21 July 2006

Abstract

The eruption of genetic research presents a tremendous opportunity to epidemiologists to improve our ability to identify causes of ill health. Epidemiologists have enthusiastically embraced the new tools of genomics and proteomics to investigate gene-environment interactions. We argue that neither the full import nor limitations of such studies can be appreciated without clarifying underlying theoretical models of interaction, etiologic fraction, and the fundamental concept of causality. We therefore explore different models of causality in the epidemiology of disease arising out of genes, environments, and the interplay between environments and genes. We begin from Rothman's "pie" model of necessary and sufficient causes, and then discuss newer approaches, which provide additional insights into multifactorial causal processes. These include directed acyclic graphs and structural equation models. Caution is urged in the application of two essential and closely related concepts found in many studies: interaction (effect modification) and the etiologic or attributable fraction. We review these concepts and present four important limitations.

1. Interaction is a fundamental characteristic of any causal process involving a series of probabilistic steps, and not a second-order phenomenon identified after first accounting for "main effects".

2. Standard methods of assessing interaction do not adequately consider the life course, and the temporal dynamics through which an individual's sufficient cause is completed. Different individuals may be at different stages of development along the path to disease, but this is not usually measurable. Thus, for example, acquired susceptibility in children can be an important source of variation.

3. A distinction must be made between individual-based and population-level models. Most epidemiologic discussions of causality fail to make this distinction.

4. At the population level, there is additional uncertainty in quantifying interaction and assigning etiologic fractions to different necessary causes because of ignorance about the components of the sufficient cause.