Epidemiology and biological plausibility in assessing causality.

Abstract

One of the well-accepted principles of epidemiology is the need to draw upon ancillary evidence from biological research in the selection of topics to pursue, design of studies, and especially, the interpretation of the results. In environmental epidemiology, understanding the biological pathways by which the exposure of concern may affect health often has a great value in framing research questions and guiding the studies that are done but calls for deeper reflection of how that ancillary biological information should (and should not) be used. Biological evidence of potential health harm may motivate epidemiologic studies and help to guide exposure assessment to maximize the likelihood of identifying an etiologic relationship if one is present. Decisions regarding exposure aggregation (lumping or splitting), duration of exposure, the timing of exposure in relation to disease occurrence, exploration of dose-response patterns (thresholds and ceilings), and other chemical and physical features of exposure to be examined in epidemiologic studies benefit from drawing on knowledge from pertinent biological research. Similarly, the choice of specific disease entities should be informed by knowledge of biological mechanisms in the analogous decisions regarding grouping, timing of onset, distinctive features of the disease (e.g., subsets of cancer with a shared etiology), and clinical manifestations. Markers of susceptibility that could result in effect-modification may be gleaned from biological research as well as candidate confounders. The product of this knowledge drawn from work done in other fields, if considered in advance, is an enhanced ability to design and conduct epidemiologic studies in which the measure of association is most likely to identify any causal effects that are present, that is, more valid studies. If positive or negative associations are found, they would be seen as concordant with expectations based on biology and if null associations are found, this would provide meaningful evidence that the plausible etiologic relationship is not likely to be present. But the dividing line between using biological evidence to optimize the design of studies and the use of biological evidence to render a verdict on the validity of the study calls for a closer look. The consideration of biological plausibility in the interpretation of the study results, as advocated by Sir Austin Bradford Hill1 raises some concerns when considering why measured associations may or may not reflect a causal effect. To the extent that the epidemiologic research is informed by sound biological insights, we will benefit from having focused on the most pertinent exposure and disease measures, minimizing exposure and disease misclassification, and isolating the most highly susceptible subgroups. Once we have gleaned all that we can from ancillary biological research on the topic, however, the epidemiologic study must stand on its merits in order to approximate the causal effect of interest, driven solely by freedom from biases. As elucidated below, the presence (or absence) of biological information does not independently influence the validity of the epidemiologic study that was conducted, which is determined solely by the usual methodological considerations. There is no validity bonus at the end for the epidemiologic studies because other lines of evidence point in the same direction nor are validity points deducted because of the lack of such support. If valuable information from biological research is not taken into account in the epidemiologic study due to lack of knowledge at the time the study was undertaken or because the biological research came later, validity will suffer. In that sense, the biological evidence does help to inform us as to whether misclassification is likely to have been present, for example, or whether underlying effect modification has not been taken into account. But this reconciliation of what is known from the varying lines of evidence can only take place when the research from multiple disciplines is compiled and then integrated for a global assessment of causality. With the comprehensive body of relevant research, we may find converging evidence from the varying lines of evidence supporting causality or that the epidemiologic research provides a flawed assessment of the etiologic process once evidence from other disciplines is taken into account. The aggregation of knowledge from the multiple contributing disciplines is needed for a fully informed assessment of causality, rather than looking to epidemiology alone or epidemiology interpreted through biology.