Meta-analysisSystematic review
A study of studies: pooling results statistically to get a more precise answer than any single trial gives.
A meta-analysis statistically combines the results of multiple studies on the same question to produce a single pooled estimate. It usually sits inside a systematic review, which is the transparent process of finding every relevant study rather than the ones that suit the argument.
Its advantage is statistical power. Individually underpowered trials, each too small to detect a modest effect, can together reveal one clearly, and pooling also shows how consistent the findings are across different populations and designs.
Its limitation is inherited quality. Combining flawed studies produces a precise estimate of a flawed answer, which is why the phrase garbage in, garbage out attaches to this design more than most. A meta-analysis of five poor trials is weaker evidence than one good large trial.
Two things are worth checking before believing one. Heterogeneity, usually reported as I squared, says how much the included studies disagreed; very high heterogeneity means the pooled number is an average of things that may not belong together. And publication bias means positive results are more likely to have been published in the first place, which inflates pooled effects.
Not all meta-analyses are equal in other ways either. Ones that pool individual participant data are stronger than ones pooling published summaries, and network meta-analyses, which compare treatments that were never tested head to head, add useful reach and extra assumptions.
On the evidence ladder they generally sit at or near the top, above single randomised trials, which is why our evidence tiers treat a well conducted meta-analysis of trials as the strongest support a claim can have here.
In practice, when a headline cites one, ask how many studies, how consistent, of what design, and how large the effect was in absolute terms.