Sparse Data in Meta-Analysis: Parametric and Semiparametric Models and Parameterization Issues

• Activity Number: 509
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Arpita Chatterjee*+ and Sanjib Basu
• Companies: Northern Illinois University and Northern Illinois University
• Keywords: Bayesian Analysis ; Adverse events ; Markov chain simulation

Abstract

Adverse events are of serious concern to drug manufacturers, regulatory agencies such as the Food and Drug Administration (FDA), patients and the general public. In this article, we investigate the methodological issues in Meta analysis with sparse data. Our results show that the parameterization and the choice of priors play crucial roles in the statistical analysis. We propose parametric as well as Dirichlet process based semi parametric models for analysis of such data. In extensive simulation studies, we find that our proposed Bayesian estimates, in particular the semi parametric model based estimates, perform significantly better than the continuity corrected and other estimates proposed in the literature. We illustrate the proposed methods in a motivating example of Meta analysis of suicidal tendencies in children, based on 24 studies on the use of antidepressants in children.