Last update:
13-Nov-2007

Bias can creep into epidemiologic studies from many directions. Various types of specific bias have been described, but it is helpful to classify bias into three broad categories: selection bias, information bias, and confounding. Selection bias is a systematic bias in a study that stems from the procedures used to select subjects and from factors that influence study participation. It comes about when the association between the determinant under study and the outcome differs for those who participate and those who do not participate in the study. There are two kinds of selection bias, volunteer bias (or self-selection bias) and selection bias, often referred to as the "healthy worker effect". Information bias can arise because the information collected about or from study subjects is erroneous. A simple definition of confounding would be the confusion, or mixing, of effects: this definition implies that the effect of the determinant under study is mixed with the effect of another variable (extraneous determinant), leading to a bias. Extraneous determinants are referred to as "potential confounders of the conditional relation of interest" because they are predictors of the occurrence of the outcome at issue. When extraneous determinants have different distributions among the compared categories of the determinant under study they constitute actual confounders. Confounding is a systematic error that investigators aim either to prevent or to remove from a study. In study design, there are two common methods to prevent confounding, randomization and restriction. A third method to prevent confounding is matching, which prevents confounding in studies with closed populations, but in case-control studies, surprisingly, does not. Two methods that can be used to deal with confounding in the analysis of data are stratification and regression models.

Key words: Confounder, Confounding, Misclassification, Systematic information bias, Systematic selection bias.