I have the simplest tastes—I am always satisfied with the best.”
The analytic choice in assessing gene–environment interaction can be both crucial and perplexing. Many epidemiologists likely would, with Oscar Wilde, simply choose the best. The best choice, however, may not be easy to spot. For example, as Noel Weiss discusses in this issue,1 the variation in disease risk by genotype may be more apparent in the presence of a strong environmental factor when the gene association is investigated among strata of the exposed and the unexposed, compared with looking at the effect of the environmental factor by genotype. Weiss advocates this approach particularly when the interest is on inference about causation, or the biology of disease. I think this is an astute point, and illustrates how one choice of stratification might obscure a possibly important pattern of interaction.
I should like then to broaden the discussion, initiated in the article's last sentences, and ask this question: Is there a better, simpler way—that is, a way one that avoids the risk of obscuring the patterns of interaction, and at the same time is general enough to be helpful in a broad range of epidemiologic questions?
I have in mind the 2-by-4 table approach, also briefly quoted by Weiss at the end of the discussion, and discussed in some detail elsewhere.2 I suggest that, if appropriately used, the concept behind the 2-by-4 table analysis may serve well as an initial scan of interaction patterns in a broad range of epidemiologic scenarios.
For illustration, let's imagine the admittedly simplistic setting of Table 1, in which a biologic interaction occurs between one genotype (G) and one environmental exposure (E), both of which are treated as being present or absent (plus or minus). Table 1 shows a typical setup, with fictitious data. The second-to-last column provides the data (in rearranged format) from the two 2-by-2 tables proposed by Weiss. Each table has its own referent. The analysis of the genetic data within the strata of exposure suggests that the exposure has its effect only among those carrying the gene variant.
In the 2-by-4 table analysis, a single reference group is used to derive all relative risks. This is shown in the right-most column in the Table (again, rearranged). One can see that the genotype does not confer risk in the presence of the exposure, but in fact the exposure has a unique protective effect that occurs only among those not carrying the gene variant. This potentially important finding might otherwise be missed.
While this is a concocted example, it illustrates the point that the full pattern of interaction appears clearly by using the 2-by-4 table approach. This clarity derives from using a single reference group—relative risks are commensurate, and each can be compared with each other. In Table 1, the patterns of relative risk show the strength and direction of the association of disease risk with and without the genotype, and with and without the exposure. The full pattern of interaction can be scanned at once.
A simple, general examination of a 2-by-4 table can help assess gene–environment interaction patterns relevant to both scientific discovery and public health—most epidemiologists like one of these 2 flavors, and many like both. It is sometimes a question of taste, but, at least initially, the simplest tastes may be best.
1. Weiss NS. Assessing the influence of a genetic characteristic on disease in the presence of a strong environmental etiology. Epidemiology
2. Botto LD, Khoury MJ. Facing the challenge of gene-environment interaction: the two-by-four table and beyond. Am J Epidemiol