Abstract
Genome-wide association studies (GWAS) based on data from large-scale genotyping arrays are sometimes undertaken in the Phase II/III drug development setting to probe for DNA polymorphisms predictive of patient response to investigational drug treatments. Typically, Phase II/III studies are not prospectively designed as pharmacogenomic trials with primary or secondary pharmacogenomics objectives; rather, a GWAS is applied retrospectively using a subset of the original randomized cohort and is only well-powered for very large pharmacogenetic (PGx) effects. In such a setting, the type of statistical modeling used, including the testing approach and auxiliary covariates in the model, is critical to flagging potential PGx biomarkers. Adjusting for ancestry is a well-known strategy used to avoid the confounding influence of population substructure in the context of nonrandomized case-control studies probing the genetic epidemiology of complex disease. However, little has been said about the role and impact of such adjustments in the context of discovering biomarkers for treatment response to a randomized therapy. Using a dataset of genotypes from a Phase II study we have explored these issues via simulation. Findings suggest that when ancestry influences response via an interaction with treatment including genetic principal components can reduce power to flag and replicate causal loci.
Acknowledgments
The authors acknowledge improvements in the article that resulted from the review of two anonymous referees and appreciate their time in reviewing and critically commenting on the original article.