Phenotype Harmonization – Challenges and achievements within a large international consortium

Large consortia collaborating on gene-finding studies are faced with the reality that cohorts may differ with respect to questionnaires that were collected, raters, and age of the subjects at which the data were obtained.... [ view full abstract ]

Large consortia collaborating on gene-finding studies are faced with the reality that cohorts may differ with respect to questionnaires that were collected, raters, and age of the subjects at which the data were obtained. Given the need to optimize total sample size and power, a practical strategy is to use whatever phenotype data are available and proceed with GWASs with or without corrections for differences in data characteristics. The increase in power due to larger sample size can be counterbalanced however by a loss induced by phenotypic heterogeneity. Psychometric modeling approaches aimed at phenotype harmonization can alleviate to some extent this loss of power, but require item level questionnaire data from all partners. Missing data patterns due to lack of overlap in phenotype items across partners are a major problem in phenotype harmonization approaches. In the ACTION consortium, partners agreed to upload item level phenotype data in a pre-structured format in R such that item level data can be easily combined across partners. Using the same format permits not only an overview of the available data but also easy selection of items, raters, age groups, etc., for any type of analysis with simple R scripts. Importantly, one of the ACTION partners (NTR) has collected a phenotype reference panel consisting of all aggression-related items collected by ACTION partners. Current simulations tailored to the data available in ACTION show that utilizing such a reference set during harmonization increases power in subsequent GWASs. Collecting phenotypic reference panel data in consortia therefore seems to be rather desirable from the perspective of a cost/benefit trade-off.