Aggregated Versus Individual Participant Meta-Analysis to Identify Potential Moderator Factors for a Continuous Outcome

• Activity Number: 443
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Tania B. Huedo-Medina*+
• Companies: University of Connecticut
• Keywords: Individual data ; meta-analysis ; continuous outcome ; multilevel ; latent variable model ; moderator variable

Abstract

Traditional aggregated data meta-analysis (AD-MA) introduces potential ecological bias by assuming that participant-level variables are the same within trials or by ignoring possible relevant subgroups that are hidden within the study-level data. The reanalysis of studies' individual-level data has been recognized as the gold standard for combining evidence from existing studies. Therefore, many entities are moving toward evidence based on individual-participant-data meta-analysis (IPD-MA). Some studies in the literature have compared IPD- and AD-MA; these two strategies have agreed on the estimates of the effect size if the data were quite homogeneous but they have not agreed under heterogeneity. However, currently, no study has developed an IPD-MA model to identify possible individual moderators under significant heterogeneity of the continuous outcome. This develops a multilevel latent variable approach to be compared to AD-MA when an individual covariate is included in the model. The approaches are compared in terms of the bias, efficiency and confidence interval coverage of the covariate estimates. A statistical development and a simulation study will allow us to identify the possible bias and differences between the two approaches and to develop guidelines under a series of conditions where they can be combined when IPD- MA data are not available.