Confounding: a bias in the estimation of the effects of exposure to health outcome

Abstract

An epidemiological study aims to investigate the relationship between an exposure and an outcome. The presence of another variable associated with the exposure, and with the disease outcome, may introduce a bias in that relationship. However, if a variable is part of the causal chain between exposure and outcome, it is not a confounding factor (or confounder). The presence of potential confounders must always be considered in the design and analysis of epidemiological studies. Controlling for confounders may be done both at the design stage and during analysis of data. The methods used for controlling for confounders at the design stage are restriction, matching, and randomization. These methods allow making similar distributions of confounders for each study group. To detect the presence of confounders during the analysis stage, the researcher may evaluate differences in the estimates of the association between the exposure and the outcome without and with the confounder (called crude and adjusted estimates, respectively). Stratification, standardization and multivariate analysis are also methods to control for the presence of confounders. (Epidemiology_statistics)