Classification, direction, and prevention of bias in epidemiologic research.

We describe an extension of an existing framework used by several investigators for classifying various types of bias. The framework consists of three categories of bias: selection, information, and confounding. The existing framework is expanded to include subclassification according to the type of study design: cross-sectional, case-control, retrospective cohort, and prospective cohort. Direction and method of prevention of biases within each category in the framework are discussed. This article provides a useful checklist for epidemiologists to determine possible sources and methods of reduction of bias that are specific to a particular type of study design.