Sinnott and Kraft (2012) and Uh et al. (2012) recently have demonstrated that substantial false positive rates occur when imputation is used to create a common set of SNPs for cases and controls genotyped on different arrays (Affymetrix vs. Illumina), which is analogous to combining controls from multiple studies as investigated in this study. Attempts to address this bias by adjusting for array effects using principal components failed (Sinnott and Kraft 2012). Post-imputation filtering of imputed SNPs required extreme thresholds on quality measures (R2 and RT2 ≥ 0.98), which did not fully remove false positive associations and left only 30 % of SNPs for analysis, substantially reducing statistical power for subsequent analyses (Sinnott and Kraft 2012; Uh et al. 2012). Thus, if the promise of using composite controls is to be realized on a large scale, alternative approaches of stringently limiting imputation-induced bias need to be developed.
