Regression to the truth: replication of association in pharmacogenetic studies.

Large, prospective trials demonstrating the value of genotyping in patient management will be required to support the introduction of pharmacogenetics into routine medical practice. However, such studies will be costly and can be justified only if there is a reproducible association between genotype and a clinically relevant phenotype. Unfortunately, non-replication is prevalent among genetic association studies. In some cases non-replication may reflect real population differences but multiple comparisons, biases and other design limitations suggest that many initial positive associations represent Type I errors. Successful detection of a true genetic effect requires not only an informed and careful selection of candidate genes but also the assiduous application of sound principles of study design. Most important, independent and, ideally, prospective confirmation of the hypothesized genetic effect in a population similar to the one originally studied is required. In selected situations, pharmacogenetic studies in healthy volunteers may support a decision to perform such prospective association studies. When the potential health or economic consequences of therapy are significant and the results of pharmacogenetic association studies are convincing, it is reasonable to consider a major clinical trials program to assess the usefulness of genetically targeted therapy.