Bias and confounding in molecular epidemiological studies: special considerations.

Studies in the field of ‘molecular epidemiology’ are becoming common in the literature. This use of molecular-biological laboratory techniques in epidemiological research has some consequences that deserve specific attention, including the potential for some particular forms of bias. Most of the bias in epidemiological research, at large, occurs because of imperfect sampling or classification procedures that cause the data-set to misrepresent the true relationship that is being studied. While there has been a long history of methodological discussions about bias in mainstream (for example, questionnaire-based) epidemiology, the corresponding issues in relation to molecular measurements in epidemiology have not been systematically addressed. Bias, or systematic error, occurs when there is a difference between what the estimator (e.g. the relative risk) is actually estimating within the acquired data and the true effect of interest within the source population. Systematic error is attributable to methodological aspects of the study design or analysis other than sampling variation, in particular the selection of subjects, the quality of information obtained, and the superimposed effects of variables other than the disease and study factor (1). In the following, we propose (with examples) a categorization of bias, which is certainly not exhaustive but which can be a basis for further discussion. 1. Selection bias