Article Text

Download PDFPDF

Understanding and interpreting systematic reviews and meta-analyses. Part 2: meta-analyses
  1. Nick Freemantle, MA1,
  2. John Geddes, MD2
  1. 1Medicines Evaluation Group, Centre for Health Economics, University of York, UK
  2. 2Editor, Evidence-Based Mental Health

Statistics from

In part 1 (August 1998 issue) we introduced the rationale for the systematic review and described the first part of how to appraise critically such articles before using them clinically. The user would want to know: did the review focus on a specific question? Was a comprehensive and clearly described search strategy used? Were the appropriate studies selected? And did the raters agree about which articles should be included?

In part 2, we focus on the statistical combination of the results of a series of studies (meta-analysis). Our objective is to suggest questions that a reader should ask of an article describing a meta-analysis. We will also outline the main methods used in meta-analyses.

Were the results of the individual studies combined and was this appropriate?

Meta-analyses seek to provide the best estimates of treatment effect based upon all the available valid evidence. The simplest type of meta-analysis involves simply counting up the number of statistically significant studies (vote counting). Although vote counting is straightforward and superficially easy to interpret, it leads to various potential biases and other problems. Many treatment effects that are of potential clinical importance are only moderately sized. When they have only been investigated in small trials with insufficient statistical power, a simple vote count may fail to identify a true treatment effect that may be important clinically. Of course, such a situation is why many meta-analyses are conducted, and this is therefore a major deficiency of the vote count. Similarly, vote counts treat every study the same, when larger and more powerful studies may appropriately be attributed greater weight than small lower powered ones. A further and important problem is that vote counting does not provide a useful estimate of the magnitude of an effect across a group of studies. For these reasons, more sophisticated methods are needed to synthesise the results of several experimental studies.


A …

View Full Text

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.