PRISMA 2020 Flow Diagram Explained, With a Worked Example

A PRISMA flow diagram is an audit trail in one figure. A reader should be able to see how many records your searches produced, what happened to each of them, and why the final set of studies is the final set. Item 16a of the 27-item PRISMA 2020 checklist asks authors to describe the results of the search and selection process, "ideally using a flow diagram". It is also easy for a reader to check, because the numbers either add up or they do not.

This guide explains each box, shows a complete example and lists the checks to run before you submit.

Records, reports and studies are different things

PRISMA 2020 introduced three terms and uses them precisely. Most flow diagram errors come from mixing them up.

The diagram starts by counting records, switches to reports when you begin retrieving full texts, and ends by counting both studies and reports. If a trial was published as a protocol, a main paper and a follow-up paper, that is three reports of one study.

Which template to use

PRISMA provides four templates as Word files, and links to a Shiny app that generates the diagram:

The updated-review templates add a box on the left for studies and reports carried over from the previous version. If you checked reference lists or searched citing articles, which most reviews do, you need a template with the "other methods" column. The PRISMA 2020 statement says the grey boxes "should only be completed if applicable; otherwise they should be removed from the flow diagram."

The boxes, in order

Identification

Records identified from databases and registers. Give the total and, preferably, the number from each source: "Databases (n = 2,912): PubMed (n = 812), Scopus (n = 1,104) ..." The template's footnote asks you to "consider, if feasible to do so, reporting the number of records identified from each database or register searched (rather than the total number across all databases/registers)". A database is a bibliographic index such as PubMed or ERIC. A register is a study registry such as ClinicalTrials.gov or the WHO ICTRP.

Records removed before screening. Three lines:

• Duplicate records removed.
• Records marked as ineligible by automation tools. Use this only if a classifier or similar tool removed records without a human looking at them. Most student reviews report zero or delete the line.
• Records removed for other reasons. State the reason, for example retracted publications.

Records identified from other methods (templates 2 and 4). Counted separately for websites, organizations, citation searching and any other route.

Screening

Records screened. Records identified minus records removed before screening. This is title and abstract screening.

Records excluded. The number rejected at title and abstract stage. The template does not ask for reasons at this stage. Its own footnote says: "If automation tools were used, indicate how many records were excluded by a human and how many were excluded by automation tools."

Reports sought for retrieval. The records that passed screening, now counted as reports because you are going after the full text.

Reports not retrieved. Full texts you could not obtain, for example after trying interlibrary loan and emailing authors.

Reports assessed for eligibility. Reports sought minus reports not retrieved.

Reports excluded, with reasons. List each reason with its count. Assign each report a single primary reason. Decide the order in which you check criteria in your protocol (for example population, then intervention, then comparator, then outcome, then design) and record the first one failed.

Included

Studies included in review and reports of included studies. Give both numbers. They differ whenever a study has more than one report. For an updated review, the template also asks for the totals across the previous and current versions.

A worked example

This example is hypothetical. No such review exists, and every number below was made up so that the arithmetic can be followed. None of it is a finding about the literature. For a real example of reporting item 16a, see the one quoted in the PRISMA 2020 explanation and elaboration paper (Page, Moher, et al., 2021). The imagined review question is whether school-based mindfulness programmes reduce anxiety symptoms in adolescents aged 11 to 18, compared with usual curriculum.

IDENTIFICATION: databases and registers
  Records identified (n = 2,950)
    Databases (n = 2,912)
      PubMed (n = 812)
      Scopus (n = 1,104)
      Web of Science (n = 640)
      ERIC (n = 356)
    Registers (n = 38)
      ClinicalTrials.gov (n = 38)

  Records removed before screening (n = 1,032)
    Duplicate records removed (n = 1,020)
    Records marked as ineligible by automation tools (n = 0)
    Records removed for other reasons: retracted (n = 12)

SCREENING
  Records screened (n = 1,918)
  Records excluded (n = 1,764)

  Reports sought for retrieval (n = 154)
  Reports not retrieved (n = 6)

  Reports assessed for eligibility (n = 148)
  Reports excluded (n = 119)
    Wrong population, e.g. university students (n = 41)
    Intervention not mindfulness-based (n = 33)
    No usual-curriculum comparator (n = 22)
    No anxiety outcome reported (n = 15)
    Not primary research (n = 8)

IDENTIFICATION: other methods
  Records identified (n = 21)
    Websites (n = 5)
    Organisations (n = 2)
    Citation searching (n = 14)
  Reports sought for retrieval (n = 21)
  Reports not retrieved (n = 2)
  Reports assessed for eligibility (n = 19)
  Reports excluded (n = 15)
    Wrong population (n = 6)
    No anxiety outcome reported (n = 5)
    Already identified through databases (n = 4)

INCLUDED
  Studies included in review (n = 30)
  Reports of included studies (n = 33)

The other-methods column of the template has no screening box: it goes straight from records identified to reports sought for retrieval. The example therefore counts as "identified" only the 21 items that looked relevant enough to retrieve. If you screened a larger set of titles found this way, say so in the text.

Checking the arithmetic

Run these checks on your own diagram.

In the results text of this hypothetical review, the matching sentence would read: "The searches identified 2,950 records. After removing 1,020 duplicates and 12 retracted publications, we screened 1,918 records and assessed 148 full-text reports, of which 29 met the inclusion criteria. Citation searching and website searches contributed a further 4 reports. In total, 33 reports of 30 studies were included (Figure 1)."

Keeping the counts as you go

Reconstructing a flow diagram months after screening is painful and error-prone. Keep a log from the first day.

1. On the day you search, record for each database the date, the exact search string and the number of results before any deduplication. PRISMA item 7 asks for the full strategy for every database, and the PRISMA-S extension (Rethlefsen et al., 2021) lists what to report.
2. Export everything to a reference manager or screening tool and note the combined total. It must equal the sum of the per-database counts.
3. Deduplicate once, and write down how many were removed. If you later find more duplicates during screening, add them to the duplicates line, not to "records excluded".
4. Screen with a tool that logs decisions. Dedicated screening tools such as Rayyan and Covidence record each decision. A spreadsheet works for small reviews if you add a column for stage and a column for reason.
5. Fix the order of exclusion reasons before full-text screening so that each report gets one reason and the reasons are comparable between reviewers.
6. Track reports and studies separately from the moment you notice two papers describing the same sample.

To draw the figure, use the Word templates from the PRISMA website or the free PRISMA2020 R package and Shiny app by Haddaway and colleagues (2022), which produces standardised diagrams in a variety of formats, including interactive web-based diagrams exported as HTML files.

Common errors

• Reporting post-deduplication numbers as "records identified". The first box is the raw total.
• Putting title and abstract exclusions in the "reports excluded" box. Reports are full texts.
• Leaving out "reports not retrieved". If you could not get six PDFs, say so. Hiding it makes the subtraction fail.
• Listing reasons that do not sum, usually because some reports were tagged with two reasons.
• Counting reports as studies. A review of "33 studies" that is really 30 studies inflates the evidence, and double-counts participants if it carries into a meta-analysis.
• Using the 2009 diagram. PRISMA 2020 replaced the 2009 statement, so use the 2020 templates and cite Page et al. (2021).
• Treating the diagram as the search documentation. It summarizes the flow. The full search strings still belong in an appendix. See the Boolean search strings guide for how to write and record them.

What the diagram does not show

The flow diagram says nothing about the quality of the included studies, and nothing about whether your eligibility criteria were sensible. Those are covered elsewhere in the checklist. Criteria are item 5, and the inclusion and exclusion criteria guide covers how to write them. A tidy diagram with poorly justified criteria is still a weak review.

References

• Haddaway, N. R., Page, M. J., Pritchard, C. C., & McGuinness, L. A. (2022). PRISMA2020: An R package and Shiny app for producing PRISMA 2020-compliant flow diagrams, with interactivity for optimised digital transparency and Open Synthesis. Campbell Systematic Reviews, 18(2), e1230. https://doi.org/10.1002/cl2.1230
• Page, M. J., McKenzie, J. E., Bossuyt, P. M., et al. (2021). The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ, 372, n71. https://doi.org/10.1136/bmj.n71
• Page, M. J., Moher, D., Bossuyt, P. M., et al. (2021). PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews. BMJ, 372, n160. https://doi.org/10.1136/bmj.n160
• PRISMA. PRISMA 2020 flow diagram templates. https://www.prisma-statement.org/prisma-2020-flow-diagram
• Rethlefsen, M. L., Kirtley, S., Waffenschmidt, S., et al. (2021). PRISMA-S: an extension to the PRISMA Statement for Reporting Literature Searches in Systematic Reviews. Systematic Reviews, 10, 39. https://doi.org/10.1186/s13643-020-01542-z
• Tricco, A. C., Lillie, E., Zarin, W., et al. (2018). PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and explanation. Annals of Internal Medicine, 169(7), 467–473. https://doi.org/10.7326/M18-0850