Comparison of Statistical Harmonization Methods in Individual Participant Data Meta-Analysis

• Activity Number: 253
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): Jinhui Ma*+ and Parminder Raina and Lauren Griffth
• Companies: McMaster University and McMaster University and McMaster University
• Keywords: individual participant data ; data harmonization ; meta-analysis ; simulation ; item response theory ; comparative analysis

Abstract

Individual participant data (IPD) meta-analysis has been increasingly used to combine unique patient data from randomized trials or observational studies. It may increase the power to detect the association between the risk factors and the health outcome, and allow adjustment for confounding factors. To achieve a proper comparative analysis, a certain level of clinical and methodological similarity across studies is required. When multiple methods are used to measure the same underlying construct, specialized statistical methods of data harmonization are needed for this construct before the IPD meta-analysis. Several methods of data harmonization have been proposed in the literature. They are recategorization, percentile conversion, z-score transformation, response conversion, linear factor analysis, two-parameter logistic item response theory model, moderated nonlinear factor analysis, extension analysis in exploratory factor analysis, item response theory combined with latent growth/decline curve modeling, and multiple imputation. The aim of this project is to compare the performance of these statistical harmonization methods through a simulation study.