Systematic Review vs Meta-Analysis: What Is the Difference? (2026)
Dr. Rajni Garg
Executive Lead, Research & Content Strategy
7 min read
A systematic review is a complete research method for finding, appraising and summarizing all the studies that answer one focused question. A meta-analysis is a statistical method, usually used within a systematic review, that pools the results of two or more studies into a single overall estimate.
So you never really choose between them. A meta-analysis is one optional step inside a review, and plenty of good reviews leave it out. Below you will find a comparison table, pooling in plain words, when not to pool and how qualitative studies get combined. New to the method itself? Read our explainer on what is a systematic review first.
TL;DR
A systematic review is the whole project: question, protocol, search, screening, appraisal, synthesis and report.
A meta-analysis is the statistics step that turns several studies’ results into one estimate with a confidence interval.
You can publish a systematic review with no meta-analysis. An estimated 32% of health intervention reviews skip it.
For qualitative studies, the counterpart of a meta-analysis is a meta-synthesis, reported with ENTREQ.
Systematic review vs meta-analysis: key differences
Each row is checked against the Cochrane Handbook, chapter 10 (version 6.5) and the PRISMA 2020 statement.
Feature | Systematic review | Meta-analysis |
|---|---|---|
What it is | A full research method with written, repeatable steps | A statistical technique for combining study results |
The question it answers | What does all the relevant evidence say? | What is the best single estimate of the effect, and how precise is it? |
Where it starts | A focused question and a protocol | Effect estimates and their variances from studies already selected |
What it can include | Trials, observational studies or qualitative studies | Comparable numbers on the same outcome from two or more studies |
What you get | A synthesis that can be narrative, tabular or statistical | A pooled estimate with a confidence interval, usually drawn as a forest plot |
Can it stand alone? | Yes, it needs no meta-analysis | Technically yes, but its answer then depends on how someone picked the studies |
Biggest risk | Missing studies or choosing them in a biased way | Averaging studies too different to belong together |
Reporting guidance | PRISMA 2020 (27 items), plus SWiM if you combine results without pooling | The synthesis items of PRISMA 2020 |
Qualitative version | A qualitative systematic review, such as JBI’s method | A meta-synthesis, reported with ENTREQ |
The row that matters most is “Can it stand alone?” A review that ends in a careful written synthesis is still systematic, because of how it found and judged the evidence.
What does a meta-analysis actually do?
Each study reports its result as an effect size, a number that says how big the difference between groups was. For a yes or no outcome, like death or infection, that is often a risk ratio. A risk ratio of 0.80 means the outcome was 20% less likely in the treated group. For scale outcomes like pain scores, it is usually a mean difference, or a standardized mean difference when studies used different scales.
The meta-analysis then takes a weighted average of those effect sizes. Studies with more precise results, usually the larger ones, count for more. Why bother? Many single trials are too small to give a clear answer alone, and the Handbook names better precision as a core aim of pooling.
You will usually see the result as a forest plot. Each study gets a horizontal line for its confidence interval and a square at its point estimate. The bigger the square, the more weight that study carries. The diamond at the bottom is the pooled result, and its width shows the confidence interval. A vertical line marks no effect, so you can see at a glance which side each study lands on.
Next comes heterogeneity: how much the true effects seem to vary between studies. The I² statistic gives the share of the variation in results that reflects real differences rather than chance. The Handbook’s rough guide: 0% to 40% might not be important, 30% to 60% may be moderate, 50% to 90% may be substantial, and 75% to 100% is considerable. It warns that thresholds like these can mislead, so read I² next to the size and direction of the effects.
Last, you pick a model. A fixed-effect model assumes every study estimates one true effect, so any spread is chance. A random-effects model assumes the true effects differ and estimates their average. It also gives small studies relatively more weight, which matters when small studies are the likeliest to be biased.
Is every meta-analysis a systematic review, and does every systematic review need one?
No on both counts. A meta-analysis is a calculation, so it can run on any set of studies, such as a lab’s own series of experiments or the trials one author happens to know. The math stays the same, but the answer is only as good as the studies fed in. A patchy search can give a pooled estimate that looks precise and is still wrong, and the Handbook says meta-analyses can mislead seriously when design, bias and variation are not thought through.
Going the other way, the Handbook states that a systematic review need not contain any meta-analyses. The SWiM authors estimate that 32% of health-related reviews of interventions do not include one.
When not to do a meta-analysis
Chapter 12 of the Cochrane Handbook lists the legitimate reasons to hold back:
There is too little evidence to pool.
The studies report outcomes or effect estimates incompletely, or use effect measures that do not line up.
The evidence itself is biased, so one tidy number would carry that bias forward.
Big differences between studies are another commonly cited reason. The Handbook pushes back a little and asks you to consider meta-analysis methods even then. What you should avoid is quoting an average of results that scatter widely, especially in opposite directions.
If you decide not to pool, you still need a method. The Handbook accepts three: summarizing the effect estimates (for example, their range), combining P values, and vote counting based on the direction of effect. Vote counting based on statistical significance, where you tally “significant” and “not significant” studies, is ruled unacceptable. To report whichever method you use, follow SWiM (Synthesis Without Meta-analysis), a nine-item checklist built for this exact situation.
Meta-analysis vs meta-synthesis (for qualitative studies)
A meta-analysis needs numbers. Qualitative studies, such as interviews and focus groups, give you themes and quotes instead. Combining those is called meta-synthesis, or qualitative evidence synthesis, and it ends in themes or statements about what people experienced, not a pooled estimate.
ENTREQ, the reporting guideline for this kind of synthesis, names meta-ethnography, thematic synthesis and framework synthesis among others. Interpretive methods such as meta-ethnography aim to build new explanations. JBI’s qualitative review method, meta-aggregation, takes a different line: it groups similar findings into categories, then into synthesized findings, and tries to present what the original authors found rather than reinterpret it.
Report a meta-synthesis with ENTREQ’s 21 items, which cover five areas: introduction, methods and methodology, literature search and selection, appraisal, and synthesis of findings. The quick test: numbers you can line up point to a meta-analysis, and themes point to a meta-synthesis.
How long a systematic review really takes
Longer than most people plan for. Borah et al. (2017) looked at 195 published reviews registered in PROSPERO. From the registered start date to publication, the mean was 67.3 weeks, well over a year (SD 31.0 weeks, range 6 to 186 weeks). The mean team size was 5 authors.
The screening load varied hugely. Searches turned up anywhere from 27 to 92,020 records, and on average just under 3% of what a search found made it into the review.
Two caveats. The sample was drawn from PROSPERO in July 2014, so newer screening tools may shorten the job. And the figure covers the whole review, so a meta-analysis is one stage inside those weeks, not an extra project.
Two published reviews: one with a meta-analysis, one without
1. A Cochrane review that pools the data
McGoldrick et al. (2020) updated the Cochrane review of antenatal corticosteroids for women at risk of preterm birth. They included 27 studies with 11,272 women and pooled the results: a single course of steroids reduced perinatal death (risk ratio 0.85, 95% CI 0.77 to 0.93; 14 studies, 9,833 infants; high-certainty evidence).
The topic has a place in meta-analysis history. Cochrane’s logo is based on a forest plot of what a meta-analysis of the neonatal death trials available in 1982 would have shown.
2. A review that uses SWiM instead
Shah et al. (2025), writing in BMC Nephrology, reviewed observational studies and randomized trials on high-flow arteriovenous fistulas in people on hemodialysis. They found the studies too heterogeneous to pool, so they used vote counting based on the direction of effect and followed SWiM. Higher fistula flow was consistently linked with more heart failure.
Which AnswerThis tools help?
If you are preparing numbers for a meta-analysis, our effect size calculator gives unlimited free calculations of Cohen’s d, Hedges’ g, paired effects, odds ratios and Pearson r, with no login. It also reports intervals for odds ratios and Pearson r.
Pooled or not, your review needs a flow diagram. The PRISMA flow diagram generator builds a PRISMA 2020 diagram, or a scoping-review adaptation, from your counts and exports PNG, SVG, editable PowerPoint or JSON, free and with no login.
FAQs
How many studies do you need for a meta-analysis?
Two, since a meta-analysis combines the results of two or more studies. With only a few, it is hard to judge how much the true effects vary, so read the pooled result with care.
What does I² tell you?
It estimates what percentage of the spread in study results comes from real differences rather than chance. A high value means the studies disagree more than chance would explain, so look for the reason before you trust the average.
Is a meta-analysis stronger evidence than a systematic review?
Not by itself. A meta-analysis is only as trustworthy as the search, selection and appraisal behind it. A well-run review with no pooling can beat a pooled figure built from studies that never belonged together.
Does PRISMA 2020 apply if you do not pool?
Yes. PRISMA 2020 covers reviews with any kind of synthesis, and even reviews with no synthesis at all. Add the SWiM checklist to explain how you combined results without a meta-analysis.
What does “systematic review and meta-analysis” mean in a paper title?
It tells you the authors ran a full systematic review and then pooled at least some of its results. If a title says only “systematic review,” check the methods section to see how the findings were combined.
When your screening counts are final, turn them into a figure with the PRISMA flow diagram generator, whether your review ends in a forest plot or not.
.png)
