Meta-analysis: Can We Mix Apples and Oranges?

Abstract

Systematic reviews are an essential part of evidence-based medicine as rigorous methodology can identify all relevant papers that address specific research questions. Clinicians often need to know if an intervention is effective and metaanalysis synthesizes the results from individual trials to provide a summary effect estimate (1). Metaanalysis has become very popular with a dramatic rise in papers using this approach over the last 15 yr (1). There is little doubt that it is appropriate to combine the results when all of the randomized controlled trials give very similar results. What is more controversial is whether metaanalysis is sensible when the results of trials give very different results. Critics feel that under these circumstances metaanalysis is an exercise in statistical alchemy (2), and it is not possible to synthesize diverse results (mix apples and oranges) (3). This paper will review the statistical techniques used to combine trial results and approaches that can be taken when studies give heterogeneous outcomes.