← All articles

Building a PRISMA Flow Diagram That Survives Peer Review

Building a PRISMA Flow Diagram That Survives Peer Review

Decorative PRISMA flow diagram title card illustration

A PRISMA flow diagram is the official PRISMA 2020 figure that traces how records were identified, screened, and selected for a systematic or scoping review. It shows every count moving through three phases: Identification, Screening, and Included. Reviewers use it to audit whether your selection process actually adds up.

You have four official templates to pick from, and grabbing the wrong one is the single most common formatting error editors send back. Download the correct PRISMA 2020 template directly from the PRISMA statement site before you touch a single number.

Picking the right one comes down to two questions:

  • Is this a new review or an update of a previous one?
  • Did you search only databases and registers, or did you also pull records from citation chasing, websites, or organizations?

Answer those two, and the template choice is automatic.

Key Takeaways

A PRISMA flow diagram only holds up in peer review when its box labels match the official PRISMA 2020 wording exactly and every count reconciles against the underlying screening log.

Point Details
Pick the right template first Match new vs. updated review and single vs. two column source layout before filling in a single box.
Know your terminology Records, reports, and studies are distinct concepts, and confusing them is the top cause of failed reconciliation.
Delete unused boxes Never zero out a box that doesn’t apply. Remove it so the diagram doesn’t misrepresent your process.
Reconcile before exporting Run arithmetic checks (identification minus duplicates equals screened, and so on) before rendering the final figure.
Export in the right format Use vector SVG or PDF for most journals, and confirm resolution requirements before submission.

Table of Contents

What Is a PRISMA Flow Diagram and Which Template Do You Need?

The PRISMA statement publishes four templates, and each one maps to a specific combination of review type and search sources, according to the PRISMA 2020 flow diagram guidance. Grabbing a generic version off a forum or an old paper’s supplementary file almost always means fixing box labels later.

  1. New review, databases and registers only. Use this if every record came from database or register searches and this is your first version of the review.
  2. New review, databases/registers plus other sources. Add this two-column layout if you also ran citation chasing, hand searched websites, or contacted organizations for records.
  3. Updated review, databases and registers only. Same single-column logic as template one, but with an added row tracking records from the previous version.
  4. Updated review, databases/registers plus other sources. The two-column version for updates, tracking both new and previous search results across all source types.

The decision flow is simple: if you did anything beyond running database queries, citation chasing, hand-searching journals, checking reference lists, or emailing authors, you need the “other sources” template with its second column. Skip it, and a reviewer will ask where those extra records came from.

On format, the official files ship as editable Word documents, which is fine for drafts but clunky for final figures. Most journals now prefer a vector file (SVG or high-resolution PDF) over a screenshot of a Word box diagram, since text stays sharp at any zoom level during typesetting.

Pro Tip: Save the Word template under a name that includes your review type and template number (“SR_new_dbonly_v1”) so you’re not guessing six months later which version you started from.

What Goes in Each Box: Records, Reports, and Studies

The diagram’s left spine runs Identification, then Screening, then Included, and everything flows downward with excluded counts branching off to the right at each stage, per the PRISMA statement’s official layout. Get this vertical logic wrong and the whole figure reads backward to a reviewer skimming it in thirty seconds.

The terminology trips up more researchers than any other part of the process. A PMC article clarifying the distinction notes that mixing these three terms up is the single most common reason counts fail reconciliation checks during peer review.

  • A record is one entry in a database or register, before you’ve looked at the actual paper. Two records can point to the same underlying study if it was indexed twice.
  • A report is a document describing a study: a journal article, a conference abstract, a preprint, a trial registry entry. One study can have multiple reports.
  • A study is the actual research project. One study might generate a journal paper, a protocol, and a dataset publication, all of which are separate reports of that single study.

Working top to bottom, the standard boxes read: records identified from databases/registers, records removed before screening (duplicates, automation tools, other reasons), records screened, records excluded, reports sought for retrieval, reports not retrieved, reports assessed for eligibility, reports excluded (with reasons and counts for each), and finally studies included with their total reports of included studies.

If your review has zero records removed by automation, delete that box entirely rather than writing “0” in it. A zeroed box implies you checked and found nothing; a deleted box correctly signals that step didn’t apply.

How Do You Fill in the Diagram Step by Step?

Start by finalizing your search log before you touch the diagram. Export raw record counts from every database and register separately, and note your deduplication method (EndNote, Rayyan, manual check) since reviewers increasingly expect that detail in the text.

  1. Total identification counts. Add up records from all databases and registers, and list other sources (citations, websites, organizations) separately if you used the two-column template.
  2. Remove duplicates and ineligible records. Document how many were removed by automation tools versus manually, and how many were excluded for other reasons before screening began.
  3. Screen titles and abstracts. Record how many made it through and how many were excluded, ideally with a top-line reason if your protocol tracked that.
  4. Seek full reports. Track how many reports you tried to retrieve and how many you couldn’t get, whether due to paywalls, unresponsive authors, or missing full texts.
  5. Assess eligibility. Count reports assessed against your criteria, then list every exclusion reason with its own count, not a single lumped “other” bucket.
  6. Tally final numbers. Count studies included and reports of included studies separately. Since one study often has multiple reports, these two numbers rarely match, and that’s expected.

Before exporting anything, run the arithmetic checks a university library’s PRISMA guide recommends: identification total minus duplicates should equal records screened, and screened minus excluded should equal reports sought. If a box doesn’t reconcile, the error is almost always in the screening log, not the diagram itself.

Pro Tip: Keep exclusion reasons in a running spreadsheet as you screen, not reconstructed from memory afterward. Reviewers will ask for exact counts per reason, and rebuilding that list from PDF folders two months later wastes an entire afternoon.

Which Tools Can Generate a Publication-Ready Diagram?

You have three real options, and the right one depends on how many times your counts will change before submission.

  • Official Word templates work fine for a first draft or a diagram that won’t change again, but every count edit means manually repositioning text boxes and arrows.
  • The PRISMA Shiny app and similar generators let you paste in counts and instantly render a diagram with correct spacing, official wording, and no manual box alignment.
  • Third-party generators like the PRISMA Flow Diagram Generator or the Vizcept PRISMA tool accept plain-text counts and export directly to SVG or PNG, which saves real time when a co-author asks for “one more revision” the night before a deadline.

Most generators offer two modes: a precise, reproducible layout that matches the official PRISMA specification exactly, and a styled “illustrative” mode meant for slide decks rather than manuscripts. Use the precise mode for anything going to a journal.

Before you export, run through a short checklist: box wording matches the official template verbatim, every count reconciles against your screening log, fonts stay legible at column width, and the file format matches your target journal’s figure requirements (SVG or vector PDF for most, though a few still want PNG at 300 DPI). Getting the figure export resolution right the first time avoids a frustrating round of “please resubmit at higher resolution” from production editors.

Can Automation Speed Up Reconciling Your Counts?

The slowest part of building a PRISMA diagram usually isn’t the drawing. It’s manually recounting records after every database export, every deduplication pass, and every screening round, then double checking that the numbers still add up. That manual tallying is also where transcription errors creep in, since a missed row in a spreadsheet doesn’t announce itself until a reviewer’s math doesn’t match yours.

Importing your reference exports directly from Scopus or Web of Science into a platform like Papersynapse lets AI read titles and abstracts and populate structured screening tables automatically, cutting the counting bottleneck instead of doing it by hand row by row. Once your inclusion and exclusion labels are normalized, pulling the exact numbers for each PRISMA box becomes a matter of reading totals off a table rather than re-tallying a spreadsheet at 11 p.m.

  • Import raw RIS or CSV files straight from your search, no manual retyping of citation lists.
  • Let AI extraction flag likely exclusions and reasons as it reads each abstract.
  • Export reconciled counts as a table, then paste them straight into your chosen diagram template or generator.

Pro Tip: Always spot check a sample of AI-tagged exclusions against the original abstracts before finalizing counts. Automation speeds up the first pass, but the reconciliation check still belongs to you.

For deeper reading on where AI fits across the broader research workflow, the workflow-based guide to AI for researchers covers use cases well beyond screening.

Getting Published Without a Rejected Figure

Papersynapse handles the bottleneck that turns a straightforward diagram into a week of spreadsheet archaeology: reconciling how many records became how many studies. The platform imports your Scopus or Web of Science exports, uses AI to read abstracts and populate structured tables, and can process up to 200 papers in under two minutes according to the PaperSynapse product page, leaving you with clean, exportable counts ready to drop straight into the official template.

Hands navigating tablet for data extraction

That doesn’t replace the official PRISMA template or the arithmetic checks covered above. It just means you’re not the one manually recounting a spreadsheet at midnight before a submission deadline.

What the Research Actually Supports About Building This Diagram

Most advice on PRISMA diagrams focuses on the wrong problem. Guides obsess over box placement and arrow direction, but the templates already fix that for you. The real failure point is upstream: researchers don’t track their screening log carefully enough to produce numbers that reconcile, and they discover the gap only when a peer reviewer asks why 340 records screened became 312 excluded plus 30 sought instead of 28.

Conventional advice also underplays how easy it is to conflate records, reports, and studies. That distinction sounds pedantic until a reviewer catches your “studies included” count matching your “reports of included studies” count, which almost never happens honestly since most studies generate more than one report.

If you take one thing from this, prioritize your screening log over the diagram itself. Build the numbers first, check the arithmetic twice, and treat the actual figure as the last five minutes of the job rather than the main event. A generator or an automation tool speeds up the counting, but it can’t fix a screening log that was never kept straight in the first place.

— Ubada

Sources

Building a PRISMA Flow Diagram That Survives Peer Review | PaperSynapse