Our findings should be viewed in the context of several limitations. In Phases 1 and 2 (but not Phase 3), many associated loci were imputed, albeit with excellent quality (Table 2). Although in absolute terms our sample was reasonably large (over 12,000 subjects considered overall), in the context of complex trait GWAS, it is still modest. This factor may have led to false negative findings at all phases of the study. In addition, 11 of the 27 top-ranked associations in Phase 1 were observed with infrequent alleles (<5%); as none of these results replicated in subsequent phases, they may be false positives. Finally, our findings are not adjusted for testing association in two populations and with three (albeit highly correlated) traits. However, a Bonferroni correction is too conservative given the high correlation among the traits and distinct hypotheses for EAs and AAs (different populations frequently have different common risk alleles). Future studies in large independent samples are necessary to address these concerns.
