Meta-analysis: when and how.

Systematic reviews have a central role in evidence-based medicine. The quantitative systematic review, also known as meta-analysis, provides a logical structure for quantifying evidence and for exploring bias and diversity in research systematically. It is essential that clinicians, educators, and researchers understand the methods that comprise this research tool, particularly the basic step-by-step process, and know when numerical pooling of data is appropriate. The essay describes how systematic reviews are best conducted and when statistical pooling of data is appropriate. Systematic reviews are scientific investigations with planned methods that use original studies as subjects and synthesize the results of multiple studies using strategies to limit bias and random error. This process requires judgments to be made explicit, and should be question driven, protocol based, reproducible, and comprehensive in scope. Meta-analysis provides a framework for research synthesis, increases power and precision, provides an overall estimate and range of effect, and identifies greater-than-expected variability among study results (heterogeneity). Meta-analysis does not remove subjectivity from the process of synthesis, identify sources of variability among studies, or obviate the need for sound, compassionate clinical reasoning. Statistical heterogeneity should be anticipated and welcomed. It forces a consideration of clinical heterogeneity as well as variation in study protocol and quality. Statistical tests for homogeneity are insensitive and do not indicate sources of heterogeneity, making such consideration imperative. The most common and popular measures of efficacy for a meta-analysis are the standardized difference between two means, the relative risk, and the odds ratio. An additional measure, the number needed to treat, with its 95% confidence interval is the most clinically useful measure of the effects of an intervention and is useful for comparing the relative effectiveness of different interventions for the same condition. Important parts of meta-analysis and sensitivity and subgroup analyses are best considered a priori and should be used to explore heterogeneity and to test for publication bias and variation in study quality.