Bayesian Meta-Analysis of Epidemiological Data with Missing Covariates

• Activity Number: 598
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Lawrence McCandless*+
• Keywords: bias ; causal inference ; sensitivity analysis

Abstract

Meta-analysis is a statistical method that is used to combine the results of different studies in order draw conclusions about a body of research. For example, one might imagine extracting hazard ratios and odds ratios from a collection of different health research papers looking at the effectiveness and safety of a drug (e.g. antidepressants). An emerging area of innovation in statistics involves meta-analysis of epidemiological studies. Unlike randomized controlled trials, which are the gold standard for proving causation, epidemiological studies are prone to biases such as confounding. In this talk, I will give an overview of meta-analysis with special emphasis on the setting where there are missing covariates in the individual studies. I will draw parallels with sensitivity analysis techniques and Bayesian analysis. The discussion will highlight MCMC sampler convergence and choosing appropriate prior distributions in the face of model nonidentifiability.