In research addressing causal questions about relations between exposures and outcomes, confounding is an issue when effects of interrelated exposures on an outcome are confused. For making valid inferences about cause-and-effect relationships, the biasing influence of confounding must be controlled by design or eliminated during data analysis. Consequently, researchers require a sound understanding of the concept of confounding to adequately deal with this type of bias when setting up and conducting (clinical) epidemiological research. For explaining confounding on a conceptual level, the counterfactual framework for causal inference is invaluable but can be very complicated. In this article, therefore, a nontechnical explanation of the counterfactual definition of confounding is presented. When considering confounding in a counterfactual way, the principle of exchangeability plays a pivotal role. Causal effects of an exposure on an outcome can be evaluated only when different exposure groups have comparable background risks of the outcome. Then, exposure groups are exchangeable and thus unconfounded. By providing a simplified explanation of the counterfactual principles of exchangeability, and consequences of nonexchangeability, this article aims to increase understanding of confounding on a conceptual level as well as the rationale underlying design and analytic strategies for dealing with confounding in (clinical) epidemiological research.
Keywords: Bias; Causality; Confounding; Counterfactual theory; Exchangeability.
Copyright © 2020 Elsevier Inc. All rights reserved.