Phenotype Harmonization: Balancing between psychometric ideals and empirical data reality in the ACTION consortium

Behavioral phenotypes are usually measured with different questionnaires across cohorts. Since different items tap into different aspects of a behavior, the aggregation of whatever items are available in each cohort can... [ view full abstract ]

Behavioral phenotypes are usually measured with different questionnaires across cohorts. Since different items tap into different aspects of a behavior, the aggregation of whatever items are available in each cohort can introduce considerable noise in the phenotype measure across cohorts. This in turn leads to loss of power in subsequent genetic analyses. Psychometric phenotype harmonization has been shown to improve power but faces the challenge of severe structural missingness at the item level since any given cohort usually only has item level data on one of the questionnaires used by the different cohorts. The missingness can be dealt with either by imputation or by the collection of an additional small phenotypic “reference panel” which provides data on all items. The various additional challenges of psychometric phenotype harmonization are discussed using our recent harmonization of the aggression phenotype in the ACTION consortium as an example.