Systematic review or meta-analysis: what is the difference?

A systematic review gathers and assesses research using planned methods to answer a particular question. A meta-analysis statistically combines results from separate studies. Not every review includes one. When an autism research headline uses these terms, it still matters who was studied, what was measured and how uncertain the conclusions are.

A review is more than a reading list

Cochrane's introduction to systematic reviews explains their role in bringing evidence together around a question. How authors searched for studies and decided what to include matters. A long list of references alone does not establish a systematic method.

In a summary, look for the participants, comparisons and outcomes. Research involving one age group may not directly answer a question about another. You are checking what the text means, rather than carrying out an assessment at home.

When are findings combined into a single number?

A review may use a meta-analysis when combining the findings is appropriate. The Cochrane methods handbook also describes other approaches to synthesis. A review without a pooled number is therefore not automatically a poor review.

Its meta-analysis chapter discusses the decisions that come before the calculation. Adding more studies does not make all their limitations disappear. When a headline says that a review of many studies has “confirmed” something, open the sections on uncertainty and limitations too.

Four questions to keep beside the link

What question did the authors ask? Who was included? What outcome did they measure? What limitations do they report? You can save those questions with a paper to return to when discussing it with a qualified professional.

Separate the method of combining evidence from publication status. A paper presented as a preprint can itself describe a review. Our preprint guide explains that label. The guide to 1 in 31 and 1 in 127 offers another example of why the population behind a number matters.

Follow one fictional question from search to summary

Imagine the question, “How do families describe sharing picture books?” This is an illustration of a research question, not a report of an actual finding. Review authors would first need to decide which families, reading arrangements and study designs belong in the review. They would then search for and select studies using a method they explain to readers.

Interviews, family diaries and questionnaires may not provide numbers that can sensibly be combined into one calculation. Authors can still give a systematic account of what the studies examined and how they differ. If suitable comparable numerical results are available, they might also plan a meta-analysis. A systematic review does not have to end with a single percentage to be useful.

A six-line reading note you can use

You do not need to understand every equation to notice when a headline goes beyond the original paper. A short note beside the publication can help you ask more precise questions. It is a reading aid, not a way to carry out your own clinical assessment.

1. Question: What exactly did the authors set out to investigate? Write one sentence, rather than the broad label “autism”.
2. People and setting: Who took part in the included studies? Findings about adults, children, classrooms and homes are not automatically interchangeable.
3. Outcome: What did the studies actually measure? A questionnaire score, a family's account and a long-term change provide different kinds of information.
4. Time: When did the search for studies end, and how long were participants followed? The publication date is not necessarily the date of the most recent included evidence.
5. Limitations: What problems do the authors identify? Read the discussion as well as the last sentence of the abstract.
6. Claim: Does the statement in the headline appear in the original paper, or was it added by a news article, social post or seller?

If you cannot access the full text, mark the unanswered questions. Missing information is a reason to describe uncertainty. It is not a reason to assume either that every methodological decision was sound or that the paper must be unreliable.

Why an average is not every participant's story

A pooled result describes the data included in that analysis. It does not tell you what happened to every individual. Look at how the studies differ and whether the result addresses the situation you are interested in. Research involving adults in one setting does not automatically answer a question about all children at home or in school.

When a graph includes a point estimate and a confidence interval, copying only the point leaves information behind. The interval helps describe uncertainty around an estimate; it is not a guarantee about an individual person's outcome. If a graph is unclear, note its label and your question instead of selecting whichever number looks most striking.

Certainty of evidence and risk of bias

Cochrane's chapter on certainty of evidence distinguishes high, moderate, low and very low certainty for a particular outcome. These labels express confidence in an estimate based on the available evidence. They are not a judgement about a family and do not automatically tell someone what decision to make.

The conduct of individual studies matters too. The chapter on assessing risk of bias explains why the way a study is carried out and assessed can affect its findings. A long reference list or a polished chart does not answer those questions on its own. If a popular summary omits this discussion, look for it in the original review.

When two reviews appear to disagree

Start by comparing their questions. Do they include the same ages, types of study and outcomes? Next, compare the search dates and the reasons studies were included or excluded. Two papers can have similar topic labels while answering different questions. Their conclusions may therefore need more context than a simple vote for one or the other.

For a discussion with a professional, bring both links and a specific note: “This review describes this group and this outcome. Have I understood how it differs from the other paper?” That is a more useful starting point than choosing whichever headline feels more hopeful. This guide explains how to read research language. It does not assess a particular treatment or instruct readers to change care.

Frequently asked questions

Is a meta-analysis the same as a systematic review?

No. They describe different things. Statistical pooling can be one part of a review.

Does a review automatically create a new recommendation?

No. A research paper and advice for a particular person are different things. This guide does not recommend changing a child's care.

Do I need to understand every calculation?

For an initial reading, concentrate on the question, participants, outcome and limitations. Decisions about care also need appropriate professional context.

Room for information and for stories

This is an evergreen explanation based on sources checked on 16 October 2026, rather than a report of a new discovery. For shared time away from research headlines, explore our own TAMITOS book and current edition information. A story does not replace an assessment or an expert evaluation of research.

A related guide: Association is not causation: reading a study.