Issues in the epidemiological analysis and interpretation of intermediate biomarkers

We critically reviewed the validity and interpretation of two analytical approaches that have been used in the molecular epidemiological literature to investigate the role of gene-environment (GxE) interactions in disease (D) causation. Several studies have attempted to use biomarkers of biologically effective dose (BBED) such as polycyclic aromatic hydrocarbon-DNA and alfatoxin-albumin adducts to assess possible GxE interactions. To truly determine whether BBED results from a GxE interaction that is causally implicated in disease development would require data on G, E, BBED, and D, and thus far, few studies have had data on each of these components. In the absence of data on an antecedent E, one approach has been to assess interactions between G and BBED on D and to interpret the results as providing information on the presence of GxE interactions. In the absence of data on G, another approach has been to control for E in analyses of BBED and D and to interpret nonnull risk estimates for BBED as reflecting the role of G. We show that neither approach is valid. Analyses of interactions between G and BBED cannot be used to draw conclusions about the presence or absence of GxE interactions. Similarly, analyses of BBED and D, controlling for E, do not provide insight into the role of G. We discuss how differences in the risk estimate for BBED, with and without control for E may be interpreted.