Editors' ChoiceBiostatistics

A Flexible Analysis of Gene-Environment Interactions

See allHide authors and affiliations

Science Translational Medicine  27 Oct 2010:
Vol. 2, Issue 55, pp. 55ec167
DOI: 10.1126/scitranslmed.3001826

Until recently, genetic and environmental causes of disease were studied independently. But now a body of research has revealed that for some diseases with genetic components, context is everything. In fact, certain diseases show genetic associations only in the presence of environmental exposures, such as smoking or a high-fat diet. For this reason, so-called gene-environment (GE) studies now focus on deciphering the complex interplay between genetic and environmental risk factors and their joint impact on disease. Past research has shown that when a genetic trait and an environmental exposure are independent—for example, those who smoke are not more likely to display the genetic trait of interest—precise information about how the gene is associated with the disease in both the presence and absence of the environmental exposure (that is, the GE interactions) can be determined by studying only those patients that have the disease (case-only studies). However, the resulting associations determined by using this approach can be quite misleading—indeed, incorrect—when the assumption of GE independence is violated. Thus, it is desirable to be able to relax the assumption of GE independence when appropriate, without resorting to a full case-control analysis—an epidemiological study design in which disease sufferers are compared with people who do not have the disease. Now, Mukherjee et al. have developed an analysis approach that strikes a balance between case-only and full case-control formats.

The authors designed a new fully Bayesian analysis approach that in essence places weights on whether the GE independence assumption is likely to be valid. This is done through the prior—the mathematical distribution that quantifies a priori information about the values of a parameter (such as the mean). If the information available from other studies indicates that the GE independence assumption is probably valid, a very informative prior (that is, a large weight on the GE independence assumption being valid) can be constructed, and the analysis is effectively a case-only one. If the existing information is unclear, the prior can be specified so that the result is effectively a hybrid of the case-only and case-control analyses. The benefit of this approach is that it offers a transparent and easy way to quantify the influence of GE independence on the result. The authors’ analysis platform is useful for any GE interaction study in which both the genetic and environmental effects are classified according to their presence or absence. This mode of scrutiny also is scalable to the study of multiple genetic regions of interest, which is a common occurrence in GE investigations.

B. Mukherjee et al., Case-control studies of gene-environment interaction: Bayesian design and analysis. Biometrics 66, 934–948 (2010). [Abstract]

Stay Connected to Science Translational Medicine

Navigate This Article