Meta-analysis (MA) uses numerical tools to synthesise effect measures from the data discovered in the literature search of randomised controlled trials (RCTs) for a systematic review (SR). An effect measure is a single number that contrasts the results of two different treatments. Commonly used effect measures for binary data are the risk ratio (RR) or the odds ratio (OR); the one or the other is calculated for each included study. Statistical methods for an MA are straightforward. A summary effect measure for data from all included studies is the desired output of the MA and is displayed in a forest plot. It is critically important to minimise bias in estimating this value. Techniques in MA to identify and/or minimise bias include the funnel plot, the exploration of clinical and statistical heterogeneity and the use of sensitivity analysis.

Introduction

Suppose that an anaesthesiologist reads an SR of most improbable research studies comparing cyclopropane versus diethyl ether general anaesthesia for aortic valve surgery. The conclusion offered is that the cyclopropane is favoured because the summary RR for mortality is 0.6 (95% confidence interval (CI) 0.45 to 0.8) with an I2 statistic of 80%. From whence did these numbers appear? What do they mean?

A fundamental distinction must be drawn between an SR and an MA.