RE: “A FRAMEWORK FOR DESCRIPTIVE EPIDEMIOLOGY”

In a recent article, Lesko et al. presented a detailed “framework for descriptive epidemiology” while also stating, “Many, if not all, of the considerations discussed in this framework apply to estimation of valid causal effects” (1, p. 2063). Although the article contains many highly reasonable ideas, its contribution to the advancement of the understanding of descriptive epidemiology vis-à-vis its nondescriptive counterpart may be hampered by the apparent conflation of the concept of epidemiology as the practice of community medicine with epidemiology as a genre of health research/science. Consequently, some statements in the article—in particular, those regarding descriptive topics, unconcerned with causality—refer to inquires in epidemiologic practice (e.g., community-level diagnostication), while others—in particular, those regarding causal topics—refer to epidemiologic research. However, certain similarities between them notwithstanding, these types of activities fundamentally differ in their essence, objects of inquiry, and theory. Notably, the authors state, “A well-defined research [sic] question (causal or descriptive) states: 1) the target population, characterized by person and place, and anchored in time…” (1, p. 2065); but unlike in inquiries in epidemiologic practice, there is no place- and time-specific target population in either causal or descriptive epidemiologic research, where the domain of inference is a particular theoretical/abstract (super-)population, infinite in size. Consequently, validity assurance generally requires representative sampling (in the selection of the sample of the target population) in survey-type inquires in epidemiologic practice, but not (in the selection of the study base/population from the source population) in epidemiologic research. On the other hand, assurance of applicability of (the knowledge produced from) the results of epidemiologic research to a multitude of place- and time-specific populations cared for by community-medicine practitioners requires thorough attention to modifiers of the magnitude(s) of the parameter(s) at issue, while no such imperative generally exists in inquiries in epidemiologic practice. Unfortunately, the difference of the practice-pertinent concept of target population from the research-pertinent concepts of study population and superpopulation is left unexplained. Several other ideas expressed in the article are problematic, in my opinion. According to the authors, there are not only causal and descriptive questions but also “prediction questions” in epidemiology (1, p. 2063). However, both causal and descriptive thinking can be directed not only to the present (or the past) but also to the future. Thus, rather than constituting a separate category of epidemiologic questions, prediction questions are either causal or descriptive. The authors state, “A causal question requires specifying … covariates that are thought to be confounders” (1, p. 2065). However, because they concern counterfactual contrasts, causal research questions do not require specifying confounders in their formulation (while studies addressing these questions commonly do). And regarding the above-quoted term “valid causal effects,” it should be noted that effects cannot be valid or invalid (while their measures’ estimates can). The authors state that there are “multiple measures of incidence,” of which they discuss “risks and rates” (1, p. 2067). However, risks are not measures of incidence. Rather, it is (rate of cumulative) incidence that can, under certain conditions, serve as a measure of risk. Further, the authors define risk as “the proportion of people free from disease at baseline who develop the outcome during the study period” (1, p. 2067). But risk is a probability whose theoretical/true value for a person does not fully manifest itself in the “proportion of people” in some finite study population. What the authors refer to as “risk” here is the empirical cumulative incidence rate. Finally, the authors state, “Measurement error can bias descriptive studies when we do not use, or there is no gold-standard measure of, the outcome” (1, p. 2069). However, gold-standard measures of outcomes may not reveal the truth regarding those outcomes, so measurement/classification errors may occur even when gold-standard measures are used. Conflict of interest: none declared.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.