KONSİLWhat is a meta-analysis?
A meta-analysis pools the numerical results of two or more separate studies statistically, producing one summary estimate with a confidence interval. It can be one step within a systematic review, and Cochrane says the variation among the pooled studies (heterogeneity) must be considered.
The Cochrane Handbook defines meta-analysis as “the statistical combination of results from two or more separate studies.” It describes two stages: a summary statistic such as a risk ratio or mean difference is calculated for each study in the same way, then the results are combined as a weighted average. In the inverse-variance method, more precise studies receive more weight. Cochrane adds that combining results can improve precision but can also mislead seriously if study designs, within-study biases, variation across studies and reporting biases are not carefully considered. Results are usually shown in a forest plot.
A fixed-effect model assumes every study estimates the same true effect; a random-effects model assumes the effects differ but follow a distribution. The Handbook says the two give identical results when there is no heterogeneity. The I² statistic describes the percentage of variability in effect estimates that is due to heterogeneity rather than sampling error. The Handbook advises considering a meta-analysis only when studies are similar enough in participants, interventions and outcomes to give a meaningful summary.
A meta-analysis is not the same as a systematic review. The Handbook treats the decision whether to combine results as one step within a systematic review and warns that a meta-analysis can mislead seriously unless the question, eligibility criteria and selection of studies have been handled with care; on the reading of that warning, statistical pooling does not replace systematic selection of studies. PRISMA 2020 covers systematic reviews with and without synthesis, so a review need not include a meta-analysis.
Example
In a hypothetical meta-analysis, a risk ratio is calculated for each of several randomized trials comparing the same drug with a placebo; the ratios are weighted by each trial’s precision, combined, and displayed in a forest plot as a single summary estimate.
A point to watch
The Handbook says the choice between a fixed-effect and a random-effects meta-analysis should never be made on the basis of a statistical test for heterogeneity. It also says a random-effects model does not make heterogeneity disappear and that exploring its possible causes is always preferable.