How many studies do you need for a meta-analysis?

Technically a meta-analysis needs two studies. In practice the number of studies decides which analyses you can do and how far you can trust them. The thresholds below are conventions from methods guidance, not laws, but reviewers and editors will expect you to respect them.

The number also decides whether pooling is a good idea at all: a few studies that differ a lot are better described than combined.

Practical thresholds

Two studies: a pooled estimate can be calculated, but it says little about how much the studies differ.
Fewer than about five studies: the estimate of between-study variance is very imprecise. With a random-effects model, an adjustment such as the Hartung-Knapp-Sidik-Jonkman method gives more honest confidence intervals when studies are few. Always show the individual studies as well.
About ten studies or more: enough to look at funnel plot asymmetry for possible publication bias, using the tests designed for the outcome type. With fewer, the tests have low power and the plot is hard to read.
About ten studies per characteristic: a rough rule for subgroup analysis and meta-regression. With fewer, the analysis cannot separate real differences between subgroups from chance, and it should be labelled exploratory.
Network meta-analysis: the number of studies matters less than whether the network is connected and whether the trials are similar enough for indirect comparisons to be valid.

What matters more than the number

Comparability: ten studies of different populations and outcomes tell you less than five studies of the same thing.
Size and weight: when one large trial gets most of the weight, the pooled estimate mostly repeats it.
Risk of bias: a precise estimate from biased studies is still unreliable, which is why the certainty of the evidence is graded with GRADE.
Number of events: for rare outcomes, the total number of events matters more than the number of participants, and methods for rare events may be needed.

Let SynthGap run the first pass

Enter an area. SynthGap proposes candidate topics, checks each against existing systematic reviews, and lists the primary studies behind it. Start on the Free plan: no card needed, and every run shows every topic it finds.

What to do when there are too few

Report the studies in a table and describe the results narratively, following reporting guidance for synthesis without meta-analysis (SWiM).
Pool anyway if the question is clear, but present the result as provisional, show the prediction interval and the individual studies, and avoid claims of precision.
Broaden the question, if that can be done without mixing things that should not be mixed.
Change the question: some topics are valuable because they show that the evidence is thin and say what a future trial should do.

Questions

Can I do a meta-analysis with two studies?

You can calculate one, and some published reviews have. It is most defensible when the two studies are very similar, large and of good quality, and when you present the result as limited.

Does a small number of studies mean the review is not worth doing?

No. A review that shows that the evidence is sparse, and what is missing, is useful to researchers and funders. It just should not claim certainty it does not have.

How many studies for a prevalence meta-analysis?

The same logic applies, but heterogeneity is usually very high in prevalence meta-analyses, so the pooled estimate is less meaningful without subgroups. Report prediction intervals.

Keep reading

General methodological guidance, not medical advice. SynthGap's suggestions are AI-generated from the published literature, so verify them before registering a protocol.