A0614
Title: Using Bayesian IRT for multi-cohort repeated measure design to extract latent change scores
Authors: Chun Wang - University of Washington (United States) [presenting]
Abstract: Repeated measure data design has been used extensively in a wide range of fields. Oftentimes, such data may be collected from multiple study cohorts and harmonized, with the intention of gaining higher statistical power and enhanced external validity. Traditional analysis may fit a unidimensional item response theory (IRT) model to data from one time point and one cohort to obtain item parameters and use such parameters throughout the analysis. Such a simplified approach ignores the item residual dependencies in the repeated measure design on one hand, and on the other hand, it does not exploit the accumulated information from different cohorts. Instead, we propose two approaches: an integrative approach using a two-tier bi-factor model via concurrent calibration, and if such calibration fails to converge, a Bayesian sequential calibration approach that uses highly informative priors on overlapping items across studies to establish a common scale. The three approaches are demonstrated using Alzheimer's Diseases Neuroimage Initiative cognitive battery data. Interestingly, the latent change scores extracted from the two proposed approaches are better separated from measurement errors, hence they contain more signals to identify people with mild cognitive impairment at the greatest risk of conversion to Alzheimer's disease.
