Weighing information from item data in GWAS meta-analyses

To increase statistical power, Genome-Wide Association Study (GWAS) results from multiple research groups are often combined in a meta-analysis. In the case of sum score phenotypes, phenotypes are not always directly comparable across samples (and sometimes not even within samples), due to for instance differences in measurement instrument. Item response theory (IRT) can be used to harmonize these phenotypes. In a recent GWAS effort for personality traits, IRTbased point estimates were used as phenotypes in a GWAS, after which in a meta-analysis all these point estimates were weighted equally, without any consideration of differences in the precision of these estimates. However, an estimate based on 48 items is much more reliable than one based on 8 items. Not only is the precision dependent on number and quality of the items, but also on the location of the scale; generally, point estimates are more precise for average scores than for extreme scores.
Here we address the question whether phenotypes should be weighed as a function of the amount of (Fisher) information about the phenotype. We contrast such a weighted approach with both the naive sum score approach, as well as the IRT-based approach from the Genetics of Personality Consortium, by determining the relative power to find QTLs associated with the phenotype. Generally, the IRT score estimates approach yielded best power.