How to Make a PRISMA 2020 Flow Diagram: Every Box, Every Number, and the Four Templates

Most PRISMA flow diagrams that reach peer review have the right shape and the wrong numbers. The boxes are copied from someone else's paper, the labels drift one word away from the official template, and the counts do not reconcile — which a methods reviewer will notice, because reconciling them is the only thing the diagram is for.

Everything below comes from the official PRISMA 2020 templates rather than from a redrawn version of them. Where a label is quoted, it is quoted verbatim from the Word template that the PRISMA statement website distributes.

What the PRISMA 2020 flow diagram is for

The diagram exists to answer one question a reader cannot otherwise answer: of everything your search returned, what happened to each item, and why.

PRISMA 2020 checklist item 16a asks authors to "describe the results of the search and selection process, from the number of records identified in the search to the number of studies included in the review, ideally using a flow diagram". Item 16b adds the part people forget — cite the studies that looked like they met the inclusion criteria but were excluded, and say why.

That pairing is the point. The diagram gives the counts; item 16b makes you name names for the near-misses. A diagram that cannot be traced back to a real screening log is not doing its job, however tidy it looks.

Which of the four PRISMA 2020 templates do you need?

There are four official templates, not one. Picking the wrong one is the easiest mistake to make and the hardest to spot afterwards.

The templates are Word documents. The PRISMA statement website also offers a Shiny app for generating the diagram, which is worth knowing about if you would rather produce it from your screening numbers than lay it out by hand.

"Other sources" is broader than people assume. If you did any citation chasing — following reference lists, checking who cited an included study — you searched by another method, and you need the two-column version. The same applies to hand-searching websites or contacting organisations.

Every box in the PRISMA 2020 flow diagram, in order

This is the v1 template, for a new review that searched databases and registers only. Labels are verbatim; (n = ) marks a count you must supply.

Phase 1 — Identification

• Records identified from*: Databases (n = ), Registers (n = )
• Records removed before screening: Duplicate records removed (n = ), Records marked as ineligible by automation tools (n = ), Records removed for other reasons (n = )

Phase 2 — Screening

• Records screened (n = ) → Records excluded** (n = )
• Reports sought for retrieval (n = ) → Reports not retrieved (n = )
• Reports assessed for eligibility (n = ) → Reports excluded: Reason 1 (n = ), Reason 2 (n = ), Reason 3 (n = ), etc.

Phase 3 — Included

• Studies included in review (n = )
• Reports of included studies (n = )

The three phase labels — Identification, Screening, Included — run down the left edge as a spine. The left-hand column is the main flow; each box on the right is an exit, recording what left the process at that step.

Note what the exclusion boxes ask for. Records excluded at title-and-abstract screening takes a single number: nobody expects a reason for every one of several thousand. Reports excluded at full-text stage takes a number per reason. This is the box reviewers read most carefully, and "Reason 1, Reason 2, Reason 3" is a placeholder — replace it with your actual criteria ("Wrong population", "No control arm", "Conference abstract only").

Records, reports and studies are three different counts

This is the concept the diagram is built around, and the one most often flattened into "papers".

The switch happens at retrieval. Up to "Records screened" you are counting search hits; from "Reports sought for retrieval" you are counting documents you went and found.

And the last two boxes are deliberately two boxes. Studies included in review and Reports of included studies are usually different numbers — a trial with a primary paper and a two-year follow-up is one study and two reports. If those two boxes carry the same number in your diagram, that is possible, but check it rather than assuming it: it is the single most common sign that the distinction was not applied.

The second column: identification via other methods

If you used the v2 template, a second column appears on the right, headed "Identification of studies via other methods". Its first box is:

• Records identified from: Websites (n = ), Organisations (n = ), Citation searching (n = ), etc.

That column then runs its own retrieval and eligibility path — Reports sought for retrieval → Reports not retrieved, then Reports assessed for eligibility → Reports excluded — before merging into the same shared Studies included in review box at the bottom.

Two things follow from the structure that catch people out:

1. The other-methods column has no screening box. Records found by citation chasing go straight to retrieval, because you are not screening thousands of titles — you already have a specific reference in hand.
2. The two columns merge, they do not sum in parallel. The final box counts studies from both routes together, so a study found by both a database search and citation chasing is counted once, in the route where you actually found it first.

The two footnotes almost everyone deletes

The template carries two footnotes marked on the box labels themselves. They are part of the template, and dropping them removes information a reviewer may expect.

*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).

**If automation tools were used, indicate how many records were excluded by a human and how many were excluded by automation tools.

The first is a recommendation, and following it is a mark of a carefully reported review — "Records identified from: MEDLINE (n = 1,204), Embase (n = 987), CENTRAL (n = 311)" tells a reader far more than a single total, and lets someone reproduce your search database by database.

The second has become sharply more relevant. If you used a screening tool to auto-exclude records, the diagram is where you disclose how much of the exclusion was machine and how much was human. The PRISMA 2020 paper also notes that grey boxes should only be completed if applicable and otherwise removed — so if you used no automation, delete that line rather than writing (n = 0) next to it.

How to make a PRISMA flow diagram: four routes compared

Whichever route you take, the sequence that avoids rework is the same:

1. Finalise your screening log first. The diagram is a rendering of counts you already have. Drawing it before screening is finished guarantees a redraw.
2. Pick the template by your actual search, not by the one you have seen most.
3. Paste the labels from the official template. Do not retype them. "Reports sought for retrieval" is not "Full texts sought" — the wording is the standard.
4. Delete the boxes you did not use, rather than zeroing them.
5. Reconcile before you style anything.

Why your numbers do not add up

Run these three checks. Almost every flow diagram that fails review fails one of them.

• Identification balance: records identified (databases + registers) − records removed before screening = records screened.
• Screening balance: records screened − records excluded = reports sought for retrieval. Then reports sought − reports not retrieved = reports assessed for eligibility.
• Eligibility balance: reports assessed − sum of all exclusion reasons = the reports that became your included studies.

The three most common breaks:

Duplicates removed in the wrong place. Deduplication belongs in "Records removed before screening", not silently between the identification box and the screening box. If your identified total minus your screened total is a number that appears nowhere, this is why.

Exclusion reasons that do not sum. Every full-text exclusion needs exactly one reason. A report excluded for two reasons still gets counted once, under the reason you actually applied first.

Studies and reports conflated at the bottom. Covered above — and worth re-checking after any late addition, because adding one follow-up paper changes the reports count without changing the studies count.

Can you reuse the PRISMA diagram? The licence, plainly

Yes, and this is unusually clean for a reporting standard.

The official templates carry a Creative Commons Attribution 4.0 (CC BY 4.0) licence, which permits redistribution, adaptation and commercial use provided attribution is given. That covers redrawing it in your own tool, restyling it to a journal's house style, and publishing it in a paper behind a paywall.

The attribution line is printed on the templates themselves:

Source: Page MJ, et al. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. This work is licensed under CC BY 4.0.

Keep that line, or cite the PRISMA 2020 statement in your reference list and point to it in the figure legend. The diagram itself is adapted from earlier flow diagrams proposed by Boers, Mayo-Wilson et al. and Stovold et al., which the PRISMA 2020 paper credits — you do not need to reproduce that chain, but it is the reason the citation goes to the 2020 statement rather than to "PRISMA".

Formatting it for submission

A flow diagram is a figure, and it goes through the same checks as every other figure in your manuscript. Two things matter more here than elsewhere:

It must be vector. This diagram is entirely text and lines — the exact content that turns to mush when rasterised. Most journals ask for line art as EPS, AI or PDF for precisely this case, and a screenshot of a Word diagram is the most common reason a flow diagram gets sent back.

It must survive reduction. The diagram is usually the widest figure in a systematic review and the one with the most text. If your target journal typesets at a single column, every one of those (n = ) labels has to stay readable at that width. Build it at final print size and check — do not design at slide scale and hope.

Our Scientific Reports figure requirements guide works through what those checks look like at one specific journal, including the minimum text size and line weight that a text-dense diagram like this one runs into first.

Frequently asked questions

What are the boxes in a PRISMA 2020 flow diagram? Records identified from Databases and Registers; Records removed before screening; Records screened; Records excluded; Reports sought for retrieval; Reports not retrieved; Reports assessed for eligibility; Reports excluded with reasons; Studies included in review; Reports of included studies — grouped under Identification, Screening and Included.

What is the difference between a record, a report and a study? A record is a search hit. A report is the retrieved document. A study is the research itself, which may span several reports.

How many PRISMA 2020 templates are there? Four — new and updated reviews, each in a databases-and-registers version and a version that adds other sources.

Do I need the two-column version? If you did any citation chasing, website searching or contacted organisations, yes.

Is the flow diagram mandatory? Item 16a says the search and selection process should be described, "ideally using a flow diagram". In practice, journals that require PRISMA compliance expect it.

Can I publish the PRISMA diagram in my paper? Yes, under CC BY 4.0, with attribution to Page MJ, et al. BMJ 2021;372:n71.

What should I do with boxes that do not apply? Delete them. The PRISMA 2020 paper says grey boxes should only be completed if applicable and otherwise removed.

Where to go next

If you are building the diagram now, download the template that matches your review from the PRISMA statement site, fill in the counts, and run the three balance checks above before you spend any time on styling.

If you are rebuilding it as a figure — because Word will not give you the vector output your journal wants — the figure tool comparison covers which options actually export publication-grade vector, and which only claim to.

And if the thing you are diagramming is a trial rather than a review, PRISMA is the wrong template. The CONSORT 2025 flow diagram is its counterpart for randomised trials, with its own boxes, its own counts and a set of labels that changed in 2025.