A Positive or a Negative Confounding Variable? A Simple Teaching Aid for Clinicians and Students
Introduction
Confounding is a central concept in observational epidemiology, and anticipating the role of confounding variables, as positive or negative, on effect measures is important in interpreting results. In particular, we need to judge the direction of the effect of a confounder since failure to adjust for it can lead either to an over- or under-estimate of the primary association of interest. Earlier, Vander Stoep and colleagues (1) used a didactic visual device to help introductory students understand how a third variable will affect an association between an exposure and a binary outcome. In their recent book, Szklo and Javier-Nieto (2) summarize in a table (Table 5–8) the expectations of changes brought about by adjustment for a confounder based on the direction of association between the confounder and both exposure and outcome. However, both presentations are limited to a certain scenario where the relation between exposure and outcome is positive; and, hence, information provided is not applicable to the situation where exposure decreases the likelihood of the outcome. Using elementary rules of mathematics we describe below a more comprehensive and simpler instructional aid to be used by students and researchers, including non-epidemiologists, to ascertain the direction of the confounder. A heuristic mathematical justification of this analogy is also described.
Section snippets
Overview: Mathematical Analogy
Consider, in a relation between an exposure (X) and an outcome (Y), a covariate (Z). The three-way relation between these variables is symbolized by the following figure, with the confounder (Z) being a variable associated with the exposure and itself being an independent risk factor for the outcome.
Each of the variables (X, Y, Z) can either be positively (+ve, i.e., increases the likelihood of the other variable) or negatively
Conclusion
The table presented should provide teachers, students, and researchers a brief and straightforward derivation for predicting the direction of confounding bias (positive or negative), more comprehensive and simpler than those earlier presented in the literature. An understanding of how an uncontrolled potential confounder is likely to affect the primary association of interest is very crucial in cases where information on the confounding variable was not or could not be obtained (3). To this
References (3)
- et al.
A didactic device for teaching epidemiology students how to anticipate the effect of a third factor on an exposure–outcome relation
Am J Epidemiol
(1999)
Cited by (84)
Firearm-Related Lead Exposure and Child Lead Levels in the United States, 2012-2018
2024, Journal of PediatricsElevated circulating follistatin associates with increased risk of mortality and cardiometabolic disorders
2024, Nutrition, Metabolism and Cardiovascular DiseasesDietary per- and polyfluoroalkyl substance (PFAS) exposure in adolescents: The HOME study
2023, Environmental ResearchAssociation between male circumcision and human papillomavirus infection in males and females: a systematic review, meta-analysis, and meta-regression
2023, Clinical Microbiology and Infection