Assessing causal relationships in genomics: From bradfordhill criteria to complex geneenvironment interactions and directed acyclic graphs

© 2014 by Apple Academic Press, Inc. Observational studies of human health and disease (basic, clinical and epidemiological) are vulnerable to methodological problems-such as selection bias and confounding-that make causal inferences problematic. Gene-disease associations are no exception, as they are commonly investigated using observational designs. However, as compared to studies of environmental exposures, in genetic studies it is less likely that selection of subjects (e.g., cases and controls in a case-control study) is affected by genetic variants. Confounding is also less likely, with the exception of linkage disequilibrium (i.e., the attribution of a genetic effect to a specific gene rather than to an adjacent one) and population stratification (when cases and controls are drawn from different ethnic populations). There is in fact some empirical evidence suggesting that gene-disease associations are less prone to confounding (e.g., by socio-economic status) than associations between genes and environmental and lifestyle variables [1]. There are some well known methodological challenges in interpreting the causal significance of gene-disease associations; they include epistasis, linkage disequilibrium, and gene-environment interactions (GEI) [2].