CitationSavitz, David A. (1994). In Defense of Black Box Epidemiology. Epidemiology, 5(5), 550-552.
AbstractAssume that an epidemiologic study has been conducted reporting a positive association of an exposure and a disease. The investigators selected subjects in a manner free of selection bias, response was excellent, accurate data were collected concerning exposure, potential confounders were carefully measured and controlled as needed, and a clear positive association with a dose-response gradient was found. Given this epidemiologic study, let us now vary the biological context for these results. In Scenario A, assume that there are prior studies of the exposure in experimental animals
and in vitro research consistent with the reported association. In Scenario B, no relevant laboratory investigations have been conducted. In Scenario C, inspite of repeated attempts, researchers have been unable to identify a mechanism by which this exposure could affect the disease and have not found similar adverse effects in experimental animals. Clearly, a comprehensive evaluation of scientific data on health risks associated with the exposure would carefully consider the data drawn from outside of epidemiology. Should the interpretation of the epidemiologic study itself be different under these three scenarios?