PRISMA Flow Diagram

The PRISMA flow diagram is the standardized figure PRISMA 2020 uses to document, box by box, how many records moved through each stage of a systematic review or meta-analysis — how many were identified, screened, excluded, and ultimately included, with a stated reason for every full-text exclusion. It is one of PRISMA 2020’s four core deliverables alongside the 27-item checklist, the expanded checklist, and the explanation-and-elaboration paper (see CASRAI’s guide to PRISMA and systematic review methodology for how the diagram fits into the reporting standard as a whole). This entry covers the diagram specifically: its structure, its two template variants, why it exists, and how to build one.

References

• Page MJ, McKenzie JE, Bossuyt PM, et al. “The PRISMA 2020 statement: an updated guideline for reporting systematic reviews.” BMJ. 2021;372:n71.
• Page MJ, Moher D, Bossuyt PM, et al. “PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews.” BMJ. 2021;372:n160.
• Haddaway NR, Page MJ, Pritchard CC, McGuinness LA. “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. 2022;18(2):e1230.
• PRISMA statement, official flow diagram page and downloadable templates: prisma-statement.org/prisma-2020-flow-diagram.

Structure: three labeled phases, not four

A common point of confusion is the box count. The original 2009 PRISMA diagram was organized under four phase headings — Identification, Screening, Eligibility, and Included. PRISMA 2020 restructured this to three labeled phases — Identification, Screening, and Included — folding the old standalone Eligibility phase into Screening as a set of boxes within it, rather than dropping the eligibility step itself. The boxes that appear on the standard PRISMA 2020 template, in order, are:

• Identification — records identified from each database and register searched, plus records identified from other sources (e.g. citation searching, organisational websites) if that variant of the template is used; records removed before screening (duplicates, records marked ineligible by automation tools, records removed for other reasons).
• Screening — records screened and records excluded at title/abstract; reports sought for retrieval and reports not retrieved; reports assessed for eligibility and reports excluded, with reasons and a count given for each reason.
• Included — studies included in the review, and the number of reports those studies are associated with (a single study is often reported across multiple publications).

The distinction PRISMA 2020 tightened relative to 2009 is terminological as much as structural: a record is a database entry (e.g. a search-result hit, which may be a duplicate), a report is a document describing a study (e.g. a journal article, preprint, or conference abstract), and a study is the actual piece of research, which may be described across several reports. Collapsing these three into one undifferentiated count — reporting only how many “papers” went in and came out — is exactly the kind of imprecision the 2020 revision was written to close.

Two template variants

The PRISMA statement publishes four downloadable Word templates (CC BY 4.0), which reduce to two structural variants depending on what the review needs to show:

• New systematic reviews — the standard flow described above, for a review with no prior version. Offered in a “databases and registers only” form and a “databases, registers, and other methods” form (the latter adds a parallel Identification column for records found via citation searching, hand-searching, or other non-database methods).
• Updated systematic reviews — adds a second Identification arm specifically for studies included in a previous version of the review, which are carried forward into the new totals alongside newly identified records. This variant is the template a living systematic review or any other periodically updated review uses, so a reader can see at a glance how many included studies are newly added in this update versus retained from before.

Why it exists

PRISMA 2020 checklist item 16a asks authors to describe the search-and-selection results, from records identified through to studies included, “ideally using a flow diagram” — and in practice the diagram is treated as close to mandatory by journals that require PRISMA compliance, because it makes the attrition path auditable at a glance in a way a paragraph of prose does not. A reader — or a future reviewer trying to update or replicate the search — can see immediately how large the initial search was relative to the final included set, where the largest drops happened, and whether exclusion reasons were actually documented at the full-text stage or only asserted in aggregate. This is the same underlying transparency goal that drives the rest of PRISMA 2020: a systematic review’s conclusions are only as trustworthy as a reader’s ability to verify how the evidence base was assembled, and the flow diagram is the single figure that makes the selection process itself checkable rather than taken on faith.

Creating one

Three practical paths, in increasing order of automation:

• Manual template completion. Download one of the four official Word templates from prisma-statement.org and fill in the box counts by hand once screening is complete. Simplest option for a small review, but error-prone for larger ones and easy to let drift out of sync if counts change late in the process.
• The PRISMA2020 Shiny app. A free, browser-based generator built by the authors of the PRISMA2020 R package (Haddaway, Page, Pritchard & McGuinness, Campbell Systematic Reviews 2022), available at estech.shinyapps.io/prisma_flowdiagram. A reviewer enters the box counts and exclusion reasons into a form and the app renders a correctly formatted diagram, exportable as an image or as an interactive HTML version whose boxes can link out to supplementary detail — useful for open-synthesis reviews that want the diagram itself to be explorable rather than static.
• The PRISMA2020 R package. The same tool as an R package (on CRAN) for reviewers who want the diagram generated programmatically from a data file as part of a reproducible analysis pipeline, rather than through the web form.

Whichever route is used, the numbers in the diagram must reconcile exactly with the numbers reported in the text and in any PRISMA checklist submitted alongside the manuscript — a diagram that disagrees with the methods section is a common desk-review flag at journals that require PRISMA compliance.

The next major revision, PRISMA 2026, adds a structured JSON representation of the diagram’s box counts, from which the visual diagram itself can be regenerated — the first PRISMA reporting item to get a machine-readable companion format.

Common mistakes

• Reporting a single combined count instead of the record/report/study breakdown described above, which obscures where records were actually lost.
• Omitting reasons for full-text exclusions, or giving reasons without matching counts per reason.
• Using the “new review” template for an updated review (or vice versa), which misrepresents whether included studies are newly found or carried forward from a prior version.
• Diagram totals that don’t match the numbers stated in the methods or results text — a reconciliation error rather than a reporting-format error, but one the diagram makes immediately visible to a careful reader.