Comparing Harmonization Methods in ECHO

We propose a combination of real-data analyses, simulation studies, and methodological innovation to compare statistical methods for harmonization of person-reported outcomes (PROs) within ECHO. Item response data obtained from PRO administration can be harmonized in multiple ways, each with theoretical advantages. When items are shared across measures, item response theory can be used to directly model relationships between the latent trait(s) measured by PROs (item-level harmonization). Alternatively, scores on PROs can be harmonized directly by applying item-response-theory-based response pattern scoring and/or crosswalk conversions to place scores from different instruments on a common metric (score-level harmonization). Both item- and score-level analyses can be conducted in a multiple-group analysis which includes information from all participants from all included samples (mega-analysis) or conducted separately in each sample and followed by meta-analytic evaluation of the effect(s) of interest (coordinated analysis). An empirical comparison of these four methods of harmonization (item-level and score-level mega-analysis and coordinated analysis) is urgently needed, as is an improvement to score-level mega-analysis which accounts for differences in measurement error across participants and PROs.