Meta-Analysis vs Systematic Review: Which One Are You Actually Writing?

A surprising number of manuscripts arrive at medical journals with a title that says "meta-analysis" and a methods section that describes a narrative literature search, or the reverse. The authors usually mean well. They searched PubMed, read forty papers, and summarized what they found. Somewhere along the way the project picked up a label that promised more than the work delivered, and the editor noticed within a few minutes.

If you are asking about the difference between meta-analysis and systematic review, the short answer is that one describes a way of finding and selecting evidence, and the other describes a way of combining numbers. They overlap often. They are not the same thing, and getting the relationship right is the first decision in an evidence synthesis project.

Systematic Review vs Meta-Analysis: The Core Difference

The Cochrane Handbook describes a systematic review as an attempt to collect all the empirical evidence that fits prespecified eligibility criteria in order to answer a defined research question, using explicit and reproducible methods chosen to reduce bias. The defining feature is the process. You write down the question, the inclusion rules and the search strategy before you look at results, and you apply them the same way to every record.

A meta-analysis is narrower and more mechanical. It is the statistical combination of results from separate studies, usually weighting each study by its precision and producing a pooled estimate with a confidence interval. It says nothing on its own about whether the studies were found fairly, selected consistently or judged for risk of bias. You can run a meta-analysis on a handful of papers a colleague emailed you. That would be a meta-analysis in the narrow sense, and a poor piece of evidence.

So the answer to "is a systematic review a meta-analysis" is no, not necessarily. Many systematic reviews contain one. Many do not, because the included studies measure different outcomes, use incompatible designs, or are too few and too heterogeneous to pool sensibly. The Cochrane Handbook is explicit that a meta-analysis is not always appropriate, and a well-run review that concludes "we did not pool, and here is why" is a legitimate and publishable result.

Meta-Analysis vs Systematic Review: Which One Does Your Project Need?

The decision is easier if you stop asking which label sounds more impressive and start asking what your question needs. Begin with the question itself. If you want to know whether a drug lowers mortality in sepsis, the useful product is a pooled effect size, and that requires comparable outcome data across trials. If you want to know how diagnostic criteria for a condition have varied over the last two decades, no number can be pooled and the product is a structured description.

Then look at what the literature will actually support. Count the eligible studies. Check whether they report the same outcome on the same scale at comparable time points. Look at the variety of populations and comparators. A good test is to build the extraction sheet for five studies before you commit to the plan. If the columns do not line up, a pooled estimate will probably mislead, and a systematic review with a narrative or structured synthesis is the more honest design.

Finally, decide before you start. Whether to pool should be set out in the protocol, with the criteria you will use to judge that studies are similar enough. Deciding after you have seen the forest plot is a recognized source of bias, and reviewers who know the field will ask about it.

What Editors Mean by Each Term

Medical journals tend to handle the two labels differently at the front door. Most major titles list "systematic review" and "meta-analysis" as article types, often together, and then attach a methods checklist. Editors screen for a short list of things: a registered or published protocol, a search across more than one database with the full strategy available, two independent reviewers at screening and extraction, a risk-of-bias assessment using an accepted tool, and a clear account of heterogeneity if anything was pooled.

The volume of submissions explains the strictness. A JAMA Network Open editorial reported that in 2021 the journal received 655 meta-analyses and 175 systematic reviews, accepting about 9 percent of the former and 17 percent of the latter. I could not open the original text while preparing this post, so treat those figures as a pointer to the editorial rather than as a settled benchmark. The direction is plausible and consistent with other editors' commentary: evidence synthesis is one of the most heavily submitted article types, and a large share is written to a template.

A 2024 study of author instructions across 168 journals found that guidance on systematic reviews is uneven and often thin, and its authors recommended that editorial teams require reporting guidelines and protocol registration more consistently. If that recommendation spreads, authors who already work to those standards will find it a non-event. Authors who do not will find their manuscripts returned at the desk.

Reporting Guidelines for Systematic Reviews and Meta-Analyses

Both article types are reported with PRISMA, the Preferred Reporting Items for Systematic reviews and Meta-Analyses. The current widely used version is PRISMA 2020, published in the BMJ in March 2021 by Page and colleagues and replacing the 2009 statement. It keeps a 27-item checklist, adds a separate checklist for abstracts, and asks authors to report the certainty of evidence, competing interests, availability of data and code, and any automation tools used during screening. Its flow diagram has also been revised. Our PRISMA 2020 guide walks through the items one by one, and the PRISMA update posttracks what has changed since. Check the journal's own instructions for which version it expects, because I found no published statement that supersedes the 2020 version on the sources I could reach.

PRISMA has extensions for specific designs, including scoping reviews (PRISMA-ScR), diagnostic test accuracy reviews (PRISMA-DTA), individual participant data reviews (PRISMA-IPD) and network meta-analyses (PRISMA-NMA). If your review fits one, the journal may expect both the main checklist and the extension. Meta-analyses of observational studies have their own older guidance in MOOSE. Choose the guideline that matches the design you actually carried out, not the one you hoped to carry out.

Do not treat the checklist as a form to complete at the end. Reviewers read it as a map. If you cite item 13a (synthesis methods) and the manuscript contains no explanation of how heterogeneity was handled, the mismatch is visible at once.

Register the Protocol Before You Search

The cheapest step you can take to improve a synthesis manuscript is to register the protocol. PROSPERO, hosted by the Centre for Reviews and Dissemination at the University of York, accepts registrations for systematic reviews with a health-related outcome, and the Open Science Framework can host protocols for reviews that fall outside its scope. Registration must precede data extraction to be credible, and many journals now ask for the identifier in the abstract. Our PROSPERO registration guide covers eligibility and the common reasons a record is rejected.

Registration serves the question of whether to pool, too. A protocol that states the effect measure, the model (fixed or random effects), the planned subgroup analyses and the threshold for heterogeneity cannot easily be accused of choosing the analysis that gave the best p-value. When deviations occur, as they often do, report them plainly. Reviewers forgive a documented change much more readily than a silent one.

Writing It Up: What Changes Between the Two

If you are writing a systematic review without a meta-analysis, spend your words on the synthesis. Group studies by population, intervention or outcome in a way that is defensible, say how you decided on the groups, and present effect directions with the evidence behind them. The Synthesis Without Meta-analysis (SWiM) guideline exists for this purpose and gives editors something to point to when they ask how a narrative synthesis was structured. Do not fall back on vote counting by statistical significance, which discards information and can point the wrong way.

If you are writing a meta-analysis, the weight shifts to statistics and sensitivity. Say which model you used and why, report heterogeneity with I-squared and, ideally, a prediction interval, which tells readers what a future study might find rather than only where the average sits. Show a funnel plot only if you have enough studies for it to mean something (the Cochrane Handbook suggests at least ten), and be careful about reading small-study effects as proof of publication bias. Rate the certainty of evidence with GRADE or a similar framework so the pooled number is not read as stronger than it is.

In both cases, risk of bias deserves its own paragraph in the discussion. A pooled estimate built from studies with serious flaws is a precisely measured error. Name the tool you used (RoB 2 for randomized trials, ROBINS-I for non-randomized studies of interventions, or the Newcastle-Ottawa Scale if the journal accepts it) and say how the assessments affected your conclusions.

Mistakes That Trigger Desk Rejection

Editors report the same patterns repeatedly. The first is a question already answered by several recent, high-quality reviews, with no case for why another is needed. Check for existing reviews, including registered ones, before you begin, and state in the introduction what yours adds. The second is a search limited to one database, or a strategy that is not reproducible. The third is a mismatch between the title and the design, as described above. The fourth is mixing very different studies into one pooled estimate and calling the variation "moderate heterogeneity".

A newer problem is integrity. Evidence syntheses inherit the flaws of the papers they include, and fabricated or retracted trials have slipped into pooled analyses and guidelines. It is worth checking your included studies against a retraction database before submission, and our posts on checking retracted citations and paper mill contamination in systematic reviews explain how. AI tools are also now used for screening and extraction. If you used one, report it, as PRISMA 2020 asks, and verify a sample of its output by hand.

Frequently Asked Questions

What is the difference between meta-analysis and systematic review?

A systematic review is a structured, prespecified process for finding and appraising all relevant studies. A meta-analysis is a statistical method for combining their results. One concerns method of selection, the other concerns method of calculation.

Is a systematic review a meta-analysis?

Not by definition. A systematic review can end in a narrative or tabular synthesis if pooling would be inappropriate. A meta-analysis nested inside a systematic review is the common case, but it is a choice rather than a requirement.

Can I do a meta-analysis without a systematic review?

Technically yes, since the arithmetic does not care how studies were found. Practically, most journals will not publish it as evidence, because a pooled estimate from a convenience sample has no defensible meaning.

Which reporting guideline do I use?

PRISMA 2020 for both, plus the relevant extension if your design fits one. Confirm the exact requirement in the target journal's instructions.

What to Do Before You Start

Write one paragraph that states your question, your eligibility criteria and whether you plan to pool, and decide whether the literature can support it by extracting five studies by hand. Search for existing reviews. Register the protocol. Pick the reporting guideline that matches the design you will really use, and read the target journal's instructions for the article type before you finalize the title. If you cannot defend the word "systematic" or the word "meta-analysis" to a skeptical editor in a single sentence, use a plainer label and describe the work accurately.

The label is a promise about method. Keep it.

Further Reading