Meta-Analysis Data Extraction

• Activity Number: 421
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Shemra Rizzo*+ and Robert E Weiss and Raj R. Makkar
• Companies: UCLA Biostatistics department and University of California, Los Angeles and Cedars-Sinai Heart Institute
• Keywords: censoring ; rounding ; uncertainty ; data input ; missing data ; reliability

Abstract

Meta-analyses are widely used in all sciences because they summarize the findings of multiple studies to provide stronger evidence about the efficacy of a treatment. A typical meta-analysis of odds-ratios requires a binomially distributed number of events in a treatment and control group. As published research papers often lack this information, the event probability is commonly extracted manually from a Kaplan-Meier survival plot. With this approach, the follow-up times are necessary but often missing or incomplete, and it is optimistically assumed that there is no loss to follow-up. Published meta-analyses typically do not provide details on how they addressed data extraction problems. In this study, we analyze summary statistics extracted from 11 published studies going into a meta-analysis comparing mortality between coronary artery bypass grafting (CABG) and percutaneous coronary intervention (PCI) in unprotected left main coronary artery stenosis. This meta-analysis will serve as our basis for developing appropriate Bayesian models to address the problems that arise during data extraction and to explain how to account for the added layer of uncertainty at the inference level.