Classification of epidemiological study designs

Abstract

In this article, I present a simple classification scheme for epidemiological study designs, a topic about which there has been considerable debate over several decades. I will argue that when the individual is the unit of analysis and the disease outcome under study is dichotomous, then epidemiological study designs can best be classified according to two criteria: (i) the type of outcome under study (incidence or prevalence) and (ii) whether there is sampling on the basis of the outcome. This classification system has previously been proposed by Greenland and Morgenstern (1988) and Morgenstern and Thomas (1993), all of whom followed previous authors in rejecting directionality (i.e. prospective/retrospective or from exposure to outcome vs from outcome to exposure) as a key feature for distinguishing study designs. Once this two-dimensional classification system has been adopted, then there are only four basic study designs (Table 1): (i) incidence studies; (ii) incidence case–control studies; (iii) prevalence studies; and (iv) prevalence case–control studies (Rothman et al. use the terms ‘incident case–control study’ and ‘prevalent case–control study’ where the adjective refers to the incident or prevalent cases). In this article, I will briefly illustrate these four different study designs for dichotomous outcomes; I then briefly consider the extension of this classification to include studies with continuous exposure or outcome measures and I briefly mention other possible axes of classification.