Much Ado About Avandia: The meta-analysis of rare events in the service of health policy
John E Cornell, Dept of Epidemiology and Biostatistics, University of Texas Health Science Center, San Antonio
*Russell Localio, Dept of Biostatistics and Epidemiology, University of Pennsylvania School of Medicine
Cynthia D Mulrow, Dept of Medicine, University of Texas Health Science Center, San Antonio
Keywords: meta-analysis, Bayesian, continuity corrections
In a June article, Nissen et al concluded from a fixed-effects meta analysis of rare outcomes assembled from 42 trials that Avandia, a widely-used treatment for type 2 diabetes, significantly increased the risk of serious adverse cardiovascular events. The authors used Peto’s method, citing Sweeting (2004), Bradburn (2007), and Sutton (2002) for support. This widely reported finding prompted an immediate counter-analysis from Diamond et al (2007) as well as an accelerated FDA hearing on the continued marketing of Avandia. The talk will discuss issues of unbalanced designs, zero events in one and/or both study arms, and heterogeneous sample sizes and populations as they relate to alternative methods such as Peto, Mantel-Haenszel, random effects, Bayesian MCMC, and zero-cell continuity corrections when the goal is estimation of both relative risks and risk differences of adverse events.