Meta-Analysis vs Systematic Review: What Is the Difference?

What is a meta-analysis? ¶

Every good meta-analysis sits inside a systematic review. Not every systematic review contains a meta-analysis.

A meta-analysis is a statistical method that combines the numerical results of two or more independent studies addressing the same question to produce a single pooled estimate of effect. Each study is weighted, usually by the inverse of its variance, so that larger and more precise studies contribute more. The result is a more precise estimate than any single study gives.

A systematic review, by contrast, is the full research process: a focused question, a registered protocol, a comprehensive search, independent screening, data extraction, risk of bias assessment, synthesis and reporting. Our systematic review guide walks through each step.

So the short answer to "systematic review vs meta-analysis" is that they work at different levels. The systematic review is the container; the meta-analysis is one possible way of synthesizing what the container holds.

Difference between a meta-analysis and a systematic review ¶

A paper titled "a systematic review and meta-analysis" did both: it reviewed the literature systematically and then pooled at least one outcome. A scoping review is a third option that maps evidence and almost never includes a meta-analysis.

When is a meta-analysis appropriate? ¶

The Cochrane Handbook frames this as a question of whether combining studies makes sense clinically and methodologically, not only statistically. Before pooling, check that:

1. The studies answer the same question. Participants, interventions, comparators and outcomes are similar enough that an average effect is meaningful.
2. Outcomes can be expressed on a common scale. For example, a risk ratio for a binary outcome or a mean difference for a continuous one.
3. Designs are compatible. Randomized and non-randomized studies are usually analyzed separately.
4. Each study reports usable data, such as event counts, or means with standard deviations and sample sizes.
5. Risk of bias is understood. Pooling biased studies produces a precise but misleading answer, so plan sensitivity analyses that exclude high-risk studies.

Two studies are technically enough to pool, but estimates of heterogeneity are very imprecise with few studies. When pooling is not appropriate, use a structured narrative synthesis and report it with the SWiM guideline. Avoid simple vote counting of significant versus non-significant results.

How to conduct a systematic review and meta-analysis ¶

The meta-analysis comes near the end of the review, after the groundwork that makes it trustworthy.

1. Register a protocol (for example on PROSPERO) that states your question, outcomes, effect measures, model and planned subgroup and sensitivity analyses.
2. Search, screen and select studies systematically, and record the counts for your PRISMA flow diagram.
3. Extract outcome data in duplicate: event counts, or means, standard deviations and sample sizes for each group.
4. Assess risk of bias in each study with a tool such as RoB 2 or ROBINS-I.
5. Pool the data with the pre-specified model, then examine heterogeneity and run sensitivity analyses.
6. Check small-study effects with a funnel plot when you have enough studies.
7. Rate certainty of the evidence for each outcome, commonly with GRADE, and report everything against the PRISMA 2020 checklist.

Effect measures used in meta-analysis ¶

Ratio measures are analyzed on the log scale and transformed back for presentation.

Fixed-effect vs random-effects models ¶

The choice of model reflects what you assume about the true effects behind the studies.

Random-effects models are the usual choice in health and social sciences, where studies rarely share identical populations and settings. Choose the model on conceptual grounds in your protocol rather than switching after a heterogeneity test.

Heterogeneity: Q, I² and tau² ¶

Heterogeneity is variation in study results beyond what chance alone would produce. Three statistics describe it:

• Cochran's Q (chi-squared test) tests whether the variation exceeds chance. It has low power with few studies, so the Cochrane Handbook notes that a P value of 0.10 is sometimes used, and a non-significant result is not evidence of no heterogeneity.
• I² is the percentage of variability in effect estimates due to heterogeneity rather than chance. Cochrane's rough guide: 0% to 40% might not be important; 30% to 60% may be moderate; 50% to 90% may be substantial; 75% to 100% is considerable. The ranges overlap on purpose, because interpretation depends on the size and direction of effects.
• tau² is the estimated between-study variance in true effects, on the scale of the effect measure. Its square root, tau, is the standard deviation of true effects.

A prediction interval uses tau² to show the range in which the true effect of a new, similar study would likely fall. Explore heterogeneity with pre-specified subgroup analyses or meta-regression, not by dropping studies until I² falls.

How to read a forest plot ¶

A forest plot is the standard figure for a meta-analysis. Each row is a study: a square marks its estimate, with size proportional to its weight, and a horizontal line shows its 95% confidence interval. A vertical line marks no effect. The diamond at the bottom is the pooled estimate, its width the confidence interval.

With 10 or more studies, add a funnel plot to look for small-study effects such as publication bias.

Meta-analysis software ¶

If you are running the analysis yourself and want someone to check the model, code and interpretation, our systematic review and meta-analysis support works through it with you. For hands-on help with the software, see R help or Stata help.