A PRISMA flow diagram traces a systematic review's search from the records identified, through screening and full-text assessment, to the studies included. For a new review, use the PRISMA 2020 template for databases and registers, or the version that adds other sources. Count records, reports and studies separately, give reasons for excluded reports, and check every subtraction.
PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) 2020 replaced the 2009 statement, which its authors say should no longer be used. Item 16a of the 27-item checklist asks for the results of the search and selection process "ideally using a flow diagram".
A randomised trial counts people, not records. For trials, see how to draw a CONSORT 2025 flow diagram; for a review that does not claim a systematic search, see how to write a mini review.
Pick one of the four PRISMA flow diagram templates
The PRISMA group publishes four official Word templates on its PRISMA 2020 flow diagram page. Choose by whether the review is new or an update, and whether you searched beyond databases and registers.
| Your review | Columns in the template |
|---|---|
| New; databases and registers only | Identification of studies via databases and registers |
| New; also websites, organisations or citation searching | Adds Identification of studies via other methods |
| Update; databases and registers only | Adds a Previous studies column |
| Update; also other sources | Previous studies, plus both search columns |
In the update templates, a previous-version box and a new-studies box both feed a total box, for studies and for reports. If an update drops a previously included study or finds a new report of one, previous plus new will not equal the total; say why. The figure in the PRISMA 2020 statement shades optional boxes grey and says to remove them when they do not apply. The templates are licensed under Creative Commons Attribution (CC BY 4.0), so you may adapt them if you cite the statement; keep the "Source:" line at the foot of each template.
Records, reports and studies are counted separately
- Record: the title or abstract of a report as indexed in a database or website. Records pointing to the same report are duplicates.
- Report: a document about a study, such as a journal article, preprint, conference abstract or register entry.
- Study: an investigation with defined participants, interventions and outcomes. One study can have several reports.
Records are counted until title and abstract screening ends. From full-text retrieval onwards the diagram counts reports, and only the final box counts studies. That box gives both studies included and reports of included studies. A trial published at three months and again at two years is one study with two reports. A paper on common flow diagram questions, co-written by the PRISMA 2020 lead author, says the split was added because some diagrams lost reports between "assessed" and "included".
Match reports to studies before you fill the last box. The Cochrane Handbook suggests comparing registration numbers, authors, setting, intervention details, participant numbers and dates, and writing to the authors when in doubt. Counting two reports of one study as two studies counts its participants twice, the same harm duplicate publication causes in reviews.
Fill the database and register column box by box
| Box in the template | What you count |
|---|---|
| Records identified from: Databases (n = ) Registers (n = ) | Every record each search returned; per-source counts are optional ("if feasible") |
| Records removed before screening | Duplicates; records marked as ineligible by automation tools; records removed for other reasons, explained in the Methods |
| Records screened | Titles or abstracts read |
| Records excluded | Records rejected at title or abstract; if tools were used, split human and automated exclusions |
| Reports sought for retrieval | Full texts you tried to obtain |
| Reports not retrieved | Reports you could not obtain, for any reason; these are not exclusions |
| Reports assessed for eligibility | Full texts checked against the criteria |
| Reports excluded: Reason 1, Reason 2, etc. | Reports failing the criteria, by primary reason |
| Studies included in review; Reports of included studies | Studies after matching, and the reports that describe them |
The template asks for reasons only at the report stage. The explanation and elaboration paper asks for the number of retrieved reports that failed the inclusion criteria and the primary reasons for exclusion, and Cochrane says numbers are sufficient for title and abstract exclusions. Give each excluded report one primary reason, so that the reason counts add up to the box total. Ongoing studies and studies awaiting classification have no box, so add one rather than losing them from a sum.
Fill the other-methods column for websites, organisations and citation searches
The second column, "Identification of studies via other methods", opens with "Records identified from: Websites (n = ) Organisations (n = ) Citation searching (n = ) etc."
- Trial registers stay in the first column with the databases.
- Count every record each source produced, not only those that qualified. Where a full count is not feasible, report what is.
- Citation searching may sit in either column; citation indexes are databases, so column 1 suits records that went through the main screening.
- The column has no "Records removed before screening", "Records screened" or "Records excluded" box. It runs from records identified to reports sought, not retrieved, assessed and excluded, then joins the shared included box. Describe how those reports were selected in the Methods rather than inventing a subtraction.
PRISMA flow diagram example: an invented two-column review with copy-ready box text
The example below is invented to illustrate the structure; its numbers are not real data. A new review searched three databases, two trial registers and three other sources, without automation tools. The registers were ClinicalTrials.gov and the World Health Organization's International Clinical Trials Registry Platform (ICTRP). Each line below, apart from the bold headings, is one box; swap in your own counts.
Identification of studies via databases and registers
Records identified from: Databases (n = 1,660): PubMed (n = 642); Embase (n = 815); Cochrane Central Register of Controlled Trials (n = 203). Registers (n = 120): ClinicalTrials.gov (n = 71); ICTRP (n = 49)
Records removed before screening: Duplicate records removed (n = 538); Records marked as ineligible by automation tools (n = 0); Records removed for other reasons (n = 0)
Records screened (n = 1,242)
Records excluded (n = 1,163)
Reports sought for retrieval (n = 79)
Reports not retrieved (n = 4)
Reports assessed for eligibility (n = 75)
Reports excluded (n = 63): ineligible study design (n = 29); ineligible population (n = 18); ineligible comparator (n = 9); ineligible intervention (n = 7)
Identification of studies via other methods
Records identified from: Websites (n = 18); Organisations (n = 5); Citation searching (n = 46)
Reports sought for retrieval (n = 11)
Reports not retrieved (n = 1)
Reports assessed for eligibility (n = 10)
Reports excluded (n = 6): ineligible study design (n = 4); ineligible population (n = 2)
Included
Studies included in review (n = 12); Reports of included studies (n = 16)
Column 1 yields 12 reports of 9 studies: seven with one report, one with three and one with two. Column 2 adds 4 reports: 3 new studies and a further report of a study already included.
Trace each count from records identified to studies included
PRISMA states no formulas; the rows follow from the order of the template's boxes and from one primary reason per excluded report. Example figures come from the invented review.
| Step | Template arithmetic | Example |
|---|---|---|
| Records identified, column 1 | Databases + registers | 1,660 + 120 = 1,780 |
| Records screened | Identified − removed before screening | 1,780 − 538 = 1,242 |
| Reports sought | Screened − records excluded | 1,242 − 1,163 = 79 |
| Reports assessed | Sought − not retrieved, in each column | 79 − 4 = 75; 11 − 1 = 10 |
| Reasons | Reason counts = reports excluded, if each report has one primary reason | 29 + 18 + 9 + 7 = 63; 4 + 2 = 6 |
| Reports included | Assessed − excluded, in each column | 75 − 63 = 12; 10 − 6 = 4 |
| Reports of included studies | Sum across columns | 12 + 4 = 16 |
| Studies included | Count after matching reports to studies | 9 + 3 = 12, fewer than 16 reports |
If one record led to two reports, or a missed duplicate meant two records led to one report, reports sought will not equal screened minus excluded; say so in the text.
Habits from the 2009 diagram that break a 2020 one
The 2009 diagram had four phases: identification, screening, eligibility and included. The 2020 templates label three stages: identification, screening and included.
- Counting after deduplication. "Records after duplicates removed" is gone: enter what each search returned, then the number removed. If you remove duplicates within each database before counting, both "Records identified" and "Duplicate records removed" come out too low.
- One lump for removals. Records marked ineligible by automation tools have their own line.
- Full-text articles. These are now reports, and unobtainable ones go in "Reports not retrieved".
- Synthesis boxes. The qualitative and quantitative synthesis boxes were replaced by studies and reports included.
- "Additional records". Other-source records now have their own column, counted in full.
A rerun search's new records can be added to each database's total; change the diagram, abstract and Results in one sitting, guided by a dated figure log across revisions.
Cross-check the diagram against the abstract, Results and near-miss list
- The abstract's total of included studies (PRISMA 2020 for Abstracts, item 7) equals "Studies included in review".
- The table of study characteristics (item 17) lists the same number of studies.
- Exclusion reasons quoted in the text match the diagram's counts.
- Item 16b, new in 2020, is met: studies that might appear to meet the inclusion criteria but were excluded (near misses) are cited, with reasons, in the text, a table or a supplement.
A matching abstract sentence for the invented review:
Database and register searches identified 1,780 records; after 538 duplicates were removed, 1,242 were screened and 75 reports were assessed in full. Other methods added 10 assessed reports. Twelve studies (16 reports), with 1,948 participants, were included.
Keep the diagram in the body of the paper, not a supplement; deciding what goes in supplementary files explains why.
At Directive Publications, the article types page says of systematic reviews and meta-analyses: "A completed PRISMA checklist and flow diagram are required, together with a Methods section describing the full search strategy and eligibility criteria." The same page says systematic reviews receive full external double-blind peer review by at least two qualified reviewers. Our author guidelines ask you to match the manuscript to the relevant checklist from the EQUATOR Network. We publish no rule on how a diagram's numbers are checked, so the arithmetic is yours to get right. Readers who find a published Directive review whose boxes do not reconcile can send us a problem report.