What is a systematic review? Definition, steps, pros and cons
By Mubarak MutesasiraUpdated 12 min read
Short answer:
A systematic review is a study of studies. It answers one clearly defined question by searching for every eligible study with a documented, reproducible method, screening them against criteria set in advance, appraising their quality and combining their results. The aim is a conclusion that is less biased than any single study or expert opinion. It is the method behind most clinical guidelines.
This guide is for students, researchers and analysts deciding whether they need a systematic review and what it involves. It is the pillar of our systematic review series, which also covers meta-analysis, scoping reviews, search strategies, PRISMA flow diagrams and Cohen's kappa for screening.
What is a systematic review?
A systematic review is a review that uses explicit, systematic methods to collate and synthesize the findings of studies that address a clearly formulated question. That is the definition in the PRISMA 2020 statement [3]. The Cochrane Handbook adds that a systematic review "attempts to collate all the empirical evidence that fits pre-specified eligibility criteria" and uses methods "selected with a view to minimizing bias" [1].
Four features make a review systematic:
- A question and plan fixed in advance. The protocol states the question, the eligibility criteria and the methods before you look at results.
- A search that tries to find everything. Several databases, plus trial registers, reference lists and grey literature, with every search string recorded.
- Selection and appraisal by more than one person. Two reviewers decide on eligibility independently, then each included study is assessed for risk of bias.
- A transparent synthesis. Results are combined in a table, a narrative synthesis or a meta-analysis, and reported so that someone else could repeat the review.
Is a systematic review primary or secondary research?
A systematic review is secondary research. It collects no new data from people. Its "participants" are published and unpublished studies. That doesn't make it a lesser form of research: a good review needs as much planning as a trial, and it sits at or near the top of most levels of evidence because it weighs all the trials, not one.
What are the steps of a systematic review?
There are eight steps, from question to report. The Cochrane Handbook and the JBI Manual for Evidence Synthesis describe them in detail [1][7]; this is the order you work in.
- Write the question. Use a framework: PICO (population, intervention, comparison, outcome) for effectiveness, SPIDER for qualitative questions. See how to write a research question.
- Write and register the protocol. State eligibility criteria, databases, screening process, extraction fields, risk-of-bias tool and synthesis plan. Health-related reviews are registered on PROSPERO, ideally before screening starts [6].
- Search. Build a search string per database with controlled vocabulary (MeSH in PubMed, Emtree in Embase) and free-text synonyms. Cochrane reviews must search CENTRAL, MEDLINE and Embase where available [2]. See systematic review search strategy.
- Remove duplicates. Merge the exports into one set and remove records that appear in more than one database.
- Screen titles and abstracts, then full texts. Exclude clearly irrelevant records first, then read the full text of the rest. Record a reason for every full-text exclusion.
- Extract data and assess risk of bias. Fill the same fields for every included study (design, sample, intervention, outcomes, results) and judge each study with a tool such as RoB 2 for trials.
- Synthesize. Pool the results in a meta-analysis if the studies are similar enough; otherwise group and compare them in a structured narrative synthesis. See systematic review vs meta-analysis.
- Report. Follow the 27-item PRISMA 2020 checklist and draw the flow diagram that shows how many records you found, screened, excluded and included [3]. See how to make a PRISMA flow diagram.
A worked example of the numbers
This is an invented example. A team asks: "In adults with chronic low back pain, does supervised exercise therapy reduce pain at 12 weeks compared with usual care?"
| Stage | Records |
|---|---|
| Found in PubMed / Embase / CENTRAL | 1,240 / 1,610 / 530 (3,380 total) |
| Duplicates removed | 1,120 |
| Titles and abstracts screened | 2,260 |
| Excluded at title and abstract | 2,140 |
| Full texts sought | 120 |
| Full texts not retrieved | 4 |
| Full texts assessed | 116 |
| Excluded at full text (with reasons) | 98 |
| Studies included | 18 |
Each line follows from the one above. A yield this low (18 of 3,380) is normal: in one analysis of PROSPERO reviews, the mean share of retrieved records that were included was 2.94% [4].
How is a systematic review different from a narrative review?
A narrative review lets the author choose what to cite; a systematic review commits in advance to a method that decides for them. A narrative review is a summary of the literature on a topic in which the author selects and interprets the sources without a pre-specified, reproducible method. Narrative reviews are useful for background, theory and expert overviews. They are not designed to answer "does X work?" without bias.
| Systematic review | Narrative (traditional) review | |
|---|---|---|
| Question | One focused question (often PICO) | Broad topic |
| Protocol | Written and often registered in advance | Usually none |
| Search | Several databases, documented strings, grey literature | Author's choice, rarely reported |
| Selection | Explicit criteria, two independent reviewers | Author's judgment |
| Quality appraisal | Required (risk of bias for each study) | Optional, informal |
| Synthesis | Meta-analysis or structured narrative synthesis | Narrative summary |
| Reporting standard | PRISMA 2020 | None required |
| Typical time | Months to over a year | Weeks |
| Reproducible | Yes, in principle | No |
Two related review types sit between these. A scoping review maps what research exists on a broad question, usually skips quality appraisal and is reported with PRISMA-ScR [5][9]; see scoping review vs systematic review. A rapid review follows the systematic method but simplifies some steps, such as fewer databases or one screener with a second checking a sample, to finish faster [8].
What are the advantages and disadvantages of a systematic review?
The main advantage is a less biased, reproducible answer; the main disadvantage is the time and people it takes. The table sets out both.
| Advantages | Disadvantages |
|---|---|
| Less selection bias: the method, not the author, decides what is included | Slow: about 67 weeks on average from registration to publication [4] |
| Reproducible: others can rerun the search and check the decisions | Needs at least two reviewers and often an information specialist [2] |
| Can pool small studies to give a more precise estimate (with meta-analysis) [10] | Only as good as the included studies: "garbage in, garbage out" |
| Shows where the evidence is missing or weak | Can be out of date by publication; living reviews exist to address this |
| Accepted basis for guidelines, policy and grant justifications | Publication bias: unpublished null results are hard to find |
| Clear reporting standard (PRISMA 2020) that journals and examiners recognize | Combining different studies can mislead if they are too different to pool |
When is a systematic review the wrong choice?
Choose another review type when:
- Your question is broad or exploratory ("what is known about...?"). A scoping review fits better [5].
- A decision is needed in weeks. Consider a rapid review and say which steps you shortened [8].
- A recent, good-quality systematic review already answers the question. Search PROSPERO and the Cochrane Library first, and only update it if there is new evidence.
- You are working alone with no second reviewer. Without independent screening, the review will not meet Cochrane or most journal standards [2].
How long does a systematic review take?
Expect about a year. Borah and colleagues analyzed 195 completed reviews registered in PROSPERO and found a mean of 67.3 weeks from registration to publication, with a mean of five authors per review [4]. Searches in those reviews returned anywhere from 27 to over 92,000 records [4].
The time scales with two things: the number of records (screening is done twice) and the number of included studies (extraction and risk of bias are done twice). Full-text retrieval through paywalls and interlibrary loans adds weeks. Screening is the stage AI tools most often target; the judgments that follow it do not get faster with AI.
How many people do you need for a systematic review?
You need at least two people who can screen and extract independently. Cochrane's rules make two independent reviewers mandatory for deciding whether each study meets the eligibility criteria, with a disagreement process defined in advance [2]. Title and abstract screening by one person is acceptable to Cochrane, but final inclusion decisions need two [2]. Agreement between the two is often reported with Cohen's kappa.
A typical team adds a content expert, a librarian for the search, and a statistician if you plan a meta-analysis.
Who needs a systematic review?
You need one when your conclusion must stand up to scrutiny:
- Clinicians and guideline groups deciding what treatment to recommend.
- PhD and master's students whose thesis chapter or whole thesis is a review, or who must justify a research gap.
- Grant applicants showing that a new trial is needed because existing evidence is weak.
- Policy and public health teams deciding whether an intervention is worth funding.
For background in an essay or introduction, a narrative review is enough.
How to run a systematic review in AI for Research
AI for Research has a Systematic Review workspace that holds the whole process in one place: protocol, search, screening, extraction and the PRISMA diagram. You and your co-reviewers make every inclusion decision. The AI only suggests.
- Pick a review type. Eight presets: systematic, scoping, rapid, meta-analysis, mapping, qualitative evidence synthesis, realist and living.
- Write the protocol. Enter the question with PICO, PCC or SPIDER and your eligibility criteria. An AI draft in PROSPERO's sections costs 3 credits; you still register it on PROSPERO yourself.
- Search. Builders for PubMed, Europe PMC and OpenAlex show a live record count, and each search is logged for the report. Scopus and Web of Science aren't searched from inside the product: run those searches through your library and upload the export (RIS, BibTeX, CSV or PubMed .nbib). Duplicates are removed by DOI, PMID and title.
- Screen. Title and abstract, then full text: include, exclude with a reason, or maybe. AI suggestions with a reason cost 1 credit per 25 records. Add a project editor as the second reviewer, or use the AI as the second reviewer. Conflicts are listed with Cohen's kappa and percent agreement.
- Extract. A table with 9 standard columns, custom columns and risk-of-bias fields. The AI fills cells from abstracts and shows the quoted words (3 credits per 10 studies). Check each cell against the full paper.
- Report. The PRISMA 2020 diagram is drawn from the live counts (SVG and PNG, 0 credits). Export CSV or RIS, and get an AI draft of the methods section (3 credits).
What it costs, for the example above: 2,260 records with AI screening suggestions is about 91 credits, extraction for 18 studies is 6 credits, and the protocol and methods drafts are 3 each: about 103 credits, which fits in the Plus plan's 750 a month. Searching, manual screening, the PRISMA diagram and exports cost nothing. The free plan includes one review of up to 50 records, enough to pilot your criteria. See pricing.
Limits to know: there is no meta-analysis or forest plot in the workspace yet, so pool effects in R, Stata or RevMan. Records from "other sources" are counted with database records in the PRISMA diagram, so adjust that box by hand if you used them. AI extraction reads abstracts, not full texts.
Before the review, use Find Papers (0 credits) to check whether a recent systematic review already answers your question.
FAQ
How long does a systematic review take?
Plan for about a year. An analysis of 195 completed reviews registered in PROSPERO found a mean of 67.3 weeks from registration to publication. Narrow questions with few studies go faster; broad questions, large result sets and meta-analysis take longer. Most of the time goes on screening, full-text retrieval and data extraction, not on the search itself.
Is a systematic review primary or secondary research?
A systematic review is secondary research. It collects no new data from participants. It finds, appraises and combines the results of primary studies that others have already done, such as trials, cohort studies or interviews. Its own method is still rigorous research: it has a protocol, a reproducible search, set eligibility criteria and a reported synthesis.
How many people do you need for a systematic review?
At least two. The Cochrane Handbook makes it mandatory that two people decide independently whether each study meets the eligibility criteria, with a set process for resolving disagreements. Title and abstract screening may be done by one person, but final decisions need two. Published reviews in PROSPERO had a mean of five authors.
What is the difference between a systematic review and a literature review?
A systematic review follows a protocol written in advance: a fixed question, a documented search of several databases, explicit inclusion criteria, two independent screeners, quality appraisal and a reported synthesis. A traditional narrative literature review has none of these requirements. The author chooses what to cite, which is faster and fine for background, but open to selection bias.
Can AI do a systematic review?
AI can speed up parts of one: suggesting search terms, ranking records for screening, and pulling data from abstracts into a table. It should not make the inclusion decisions or judge study quality on its own. Use AI as a suggestion or as one of two screeners, check its output against the papers, and say in your methods how you used it.
Next steps
- Decide whether you need pooled numbers: systematic review vs meta-analysis.
- Check whether your question is better suited to mapping: scoping review vs systematic review.
- Build the search: systematic review search strategy.
- Report your numbers: how to make a PRISMA flow diagram and Cohen's kappa for screening.
Sources
- Lasserson TJ, Thomas J, Higgins JPT. Chapter 1: Starting a review. In: Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA (eds). Cochrane Handbook for Systematic Reviews of Interventions, version 6.5. Cochrane, 2024. https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-01
- Lefebvre C, Glanville J, Briscoe S, et al. Chapter 4: Searching for and selecting studies. In: Cochrane Handbook for Systematic Reviews of Interventions, version 6.5. MECIR C24 and C39; section 4.6.4. https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-04
- Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. https://doi.org/10.1136/bmj.n71
- Borah R, Brown AW, Capers PL, Kaiser KA. Analysis of the time and workers needed to conduct systematic reviews of medical interventions using data from the PROSPERO registry. BMJ Open. 2017;7(2):e012545. https://doi.org/10.1136/bmjopen-2016-012545
- Munn Z, Peters MDJ, Stern C, Tufanaru C, McArthur A, Aromataris E. Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC Medical Research Methodology. 2018;18:143. https://doi.org/10.1186/s12874-018-0611-x
- PROSPERO: International prospective register of systematic reviews. Inclusion criteria. https://www.crd.york.ac.uk/prospero/AboutPROSPERO/InclusionCriteria.php
- Aromataris E, Lockwood C, Porritt K, Pilla B, Jordan Z (eds). JBI Manual for Evidence Synthesis. JBI, 2024. https://doi.org/10.46658/JBIMES-24-01
- Garritty C, Gartlehner G, Nussbaumer-Streit B, et al. Cochrane Rapid Reviews Methods Group offers evidence-informed guidance to conduct rapid reviews. Journal of Clinical Epidemiology. 2021;130:13-22. https://doi.org/10.1016/j.jclinepi.2020.10.007
- Tricco AC, Lillie E, Zarin W, et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): checklist and explanation. Annals of Internal Medicine. 2018;169(7):467-473. https://doi.org/10.7326/M18-0850
- Deeks JJ, Higgins JPT, Altman DG, McKenzie JE, Veroniki AA, Simmonds M (eds). Chapter 10: Analysing data and undertaking meta-analyses. In: Cochrane Handbook for Systematic Reviews of Interventions, version 6.5. https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-10