Meta-Analysis Without Guessing: Bridging the Divide Between Ideal and Real Extracted Data

• Activity Number: 294
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Shemra Rizzo*+ and Robert E. Weiss
• Companies: University of California, Los Angeles and University of California, Los Angeles
• Keywords: Missing data ; uncertainty ; data input ; reliability

Abstract

Meta-analysis combines evidence from multiple studies to derive a stronger conclusion. Meta-analysis of aggregate data pools the effect sizes extracted from summary statistics of published studies. If the treatment effect of interest is an odds-ratio, the ideal extracted data are the number of subjects and events in the treatment and control groups from each study. If the treatment effect is a hazard ratio, the ideal extracted data are the log-hazard ratios and their variance. Often, some of the ideal data is not found in the published studies. The real extracted data may not contain the true number of events or variance for one or more studies. Meta-analysts use other information in the study to guess the missing ideal data. For example, readings from the Kaplan Meier plot may be manipulated into the desired data. We argue that treating best-guesses as observed summary statistics is a questionable practice, because nowhere in the current methodology of meta-analysis is the uncertainty of guessing accounted for. We propose Bayesian methods that model the unavailable data and incorporate the associated uncertainty in the meta-analysis.