Studies in pharmacogenomics have identified many individual variants with sufficiently large effect sizes to have clinical utility, and many of these are now the subject of implementation studies at a variety of levels. Recent research on common diseases and complex traits have, however, raised the possibility that mixed models allowing separately for the contribution of variants with larger effect sizes and a polygenic background may yield improved prediction. As we medical centers routinely move to having large-scale genome data routinely available on patients, as opposed to one-off genotyping for the prescribing of specific drugs, the opportunity to build predictors of adverse events and efficacy using large scale genome data rather than individual (or small numbers of) variants becomes a real possibility. Using real examples from large-scale studies, we will contrast prediction based on individual or small numbers of variants with predictions based on large-scale information. We will also discuss efforts to implement these alternative approaches in EMR settings.