Directionality, timing, and sample selection in epidemiologic research design

Abstract

The foregoing commentaries demonstrate wide differences between their authors’ choice of terms describing epidemiologic research designs, and in the relative importance they attach to directionality, timing, and sample selection. Nowhere is this more evident than in their discrepant definitions of “prospective” and “retrospective” studies. According to Drs Greenland and Morgenstern, “Prospective studies are based on ascertainment of events (exposures and outcomes) at the time they actually occur; retrospective studies are based on purely historical determinations; thus the distinction is one of timing.” By contrast, Dr Abramson uses the same two terms to indicate directionality (prospective = “forwardlooking,” retrospective = “backward-looking”). Precisely because of this source of confusion, we prefer to avoid these terms entirely. We applaud Dr Abramson’s efforts at simplification, particularly when research design and other epidemiologic principles are presented to students and colleagues without a previous background in epidemiology. But we give an explicit place (as do Greenland and Morgenstern) to the timing of measurement of exposure and outcome in our classification scheme. In our view, timing is important because of both its potential for conceptual “confounding” with directionality and its relevance for apposite and valid measurement of exposure and outcome [ 11. In our original article, we defined directionality as “the order in which exposure and outcome are investigated.” [l]. As Greenland and Morgenstern surmise, we do indeed conceptualize directionality in terms of order of measurement. Perhaps our meaning would have been clearer had we used the term “ascertained” instead of “investigated.” But in separating directionality from timing, we think it is important to distinguish between when measurements are actually made and when those measurements are ascertained by the study investigators. For example, actual exposure measurements may have been carried out by a physician, employer, etc. at the time exposure actually occurred, but the investigators may seek access to the records containing those measurements only at the time of the epidemiologic study. When exposure information is obtained by interviews of the study subjects, the measurement and its ascertainment are of course simultaneous, even if the exposure inquired about occurred in the distant past. Greenland and Morgenstern appear to agree, then, that the order in which exposure and outcome are ascertained is an important aspect of research design. They also agree that its importance lies in the potential for biased measurement of exposure and outcome. We in no way wish to impugn Greenland’s and Morgenstern’s knowledge or writings about this source of bias; they have made a number of important contributions in this area. But we share Dr Abramson’s point of view that if directionality (order of ascertainment of measurement) is indeed important, as we all appear to agree that it is, then it is logical to include it in a classification system. Despite apparent overall consensus on the importance of order of ascertainment of exposure and outcome, there remain substantial differences in emphasis. Greenland and Morgenstern