Editors' ChoiceBiostatistics

Taking Advantage of What You Know About Complex Diseases

See allHide authors and affiliations

Science Translational Medicine  14 Jul 2010:
Vol. 2, Issue 40, pp. 40ec112
DOI: 10.1126/scitranslmed.3001468

Knowledge is power. This is especially true for research projects in which incorporating existing knowledge into a new data analysis can increase the power of the study. Bayesian analysis is one approach that allows for adding what we already know into a new analysis. Not only is a model specified for the data, but a model is also specified for the measure of association. The model for the measure of association is called a prior distribution and can be specified to reflect information previously gathered about the association of interest. As discussed by Fridley et al., the major strength of a Bayesian approach is that it allows for transparent assessment of the influence of including prior information into the analysis. These authors propose the use of a mixture distribution to specify the prior knowledge on the measure of association between genetic effects and complex disease. A mixture distribution is just as it sounds—a mixture of different distributions (such as three normal distributions, each with a different mean). Their mixture prior has three distributions, one corresponding to each of the three possible outcomes: the genetic effect is protective (a negative association), a risk factor (a positive association), or has no effect on disease (a value of 0). The user then inputs weights and values for each of the distributions on the basis of the prior knowledge that exists about the association of interest. The values defining the prior distributions can be flexibly chosen to reflect the strength of the information available about the disease from previous research—such as linkage studies or biologic knowledge. A major caveat when incorporating existing knowledge into a new analysis is that the existing knowledge must be generalizable to the current study, otherwise the results from the current analysis can be biased. Although this report discusses genetic association studies, the approach is general enough to be applied to any study with relevant existing scientific data.

B. L. Fridley et al., Bayesian mixture models for the incorporation of prior knowledge to inform genetic association studies. Genet. Epidemiol. 34, 418–426 (2010). [Abstract]

Navigate This Article