If you have ever submitted a systematic review or meta-analysis to a journal, you have almost certainly been asked for a PRISMA flow diagram. It is the single most recognizable figure in evidence-synthesis research — four clearly labelled stages, a column of counts, and arrows showing exactly how you went from thousands of database hits down to the handful of studies you actually analysed.
Despite being almost universal, many researchers find the diagram confusing on first encounter. Which counts go in which box? What is the difference between "records identified" and "reports sought"? When do you show a second column for "other methods"? This guide answers every one of those questions and walks you through building a complete, publication-ready PRISMA 2020 flow diagram from scratch.
What you will learn:
- The four stages of a PRISMA 2020 diagram and what each one means
- Exactly which numbers to put in every box
- How to handle database searches vs. other identification methods
- Common mistakes that cause peer reviewers to request revisions
- How to produce the finished figure quickly with SciDraw AI's PRISMA flow diagram generator
Why PRISMA Exists and Why the Diagram Matters
PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses. The 2020 update replaced the 2009 version and introduced clearer language about where records come from (databases vs. registers vs. citations vs. other methods) and added an explicit step for seeking and retrieving full-text reports.
The flow diagram is mandatory or strongly recommended by hundreds of journals including The BMJ, The Lancet, JAMA, Cochrane Database of Systematic Reviews, and most journals in medicine, psychology, education, and social sciences. Reviewers check it immediately — an inconsistent count or a missing box is one of the fastest ways to earn a "major revision" decision.
Understanding the Four PRISMA 2020 Stages
PRISMA 2020 organises the literature-search process into four sequential stages. Think of them as a funnel: wide at the top, narrow at the bottom.
| Stage | Plain-English meaning |
|---|---|
| Identification | Every record you found before applying any filter |
| Screening | Records you checked against your inclusion/exclusion criteria at title/abstract level |
| Eligibility | Full-text reports you retrieved and assessed in detail |
| Included | Studies that passed all criteria and entered your synthesis |
Each stage has one or more boxes showing how many records entered and how many were removed (with reasons).
The PRISMA flow works like a funnel, narrowing from all records found to the studies finally included.
Stage 1 — Identification
The two-column structure
PRISMA 2020 divides identification into two columns:
- Left column — Databases and registers: records found by searching bibliographic databases (PubMed, Embase, Web of Science, Cochrane CENTRAL, PsycINFO, Scopus, etc.) and trial registers (ClinicalTrials.gov, WHO ICTRP, etc.).
- Right column — Other methods: records found via citation searching (backward and forward), contacting authors, reviewing grey literature, or any other non-database source.
If you only searched databases, you can omit the right column entirely. If you used both, show both.
What to put in each identification box
Left column, box 1:
Records identified from databases (n = X) and registers (n = Y)
Combine all database hits into a single number before deduplication. If PubMed returned 1,243 records and Embase returned 2,187 records, the total is 3,430 — even though some are duplicates.
Left column, box 2 (removal):
Records removed before screening: Duplicate records removed (n = ?) Records marked as ineligible by automation tools (n = ?) Records not retrieved (n = ?)
List each reason separately. Automation tools means any machine-learning-assisted screening software (Rayyan, Covidence, etc.) used before human screening.
Right column (if applicable):
Records identified from: Citation searching (n = ?) / Websites (n = ?) / Organisations (n = ?) / Hand searching (n = ?) / Other methods (n = ?)
Break down by method, or combine into a single "Other methods" total if that suits your reporting.
Identification uses two columns -- databases and registers on one side, other methods on the other.
Stage 2 — Screening
After removing duplicates and automated exclusions, the remaining records enter screening.
Box: Records screened
This is the total number of records a human reviewer actually looked at (title and abstract). It equals: records identified minus records removed before screening.
Records screened (n = ?)
Box: Records excluded
Report the total excluded at title/abstract level. PRISMA 2020 does not require you to list reasons for title/abstract exclusion (unlike full-text exclusion) — you just need the total.
Records excluded (n = ?)
Stage 3 — Eligibility (Full-Text Assessment)
Records that survived title/abstract screening move to full-text retrieval and assessment.
Box: Reports sought for retrieval
This is the count of records for which you tried to obtain the full text. It is usually identical to "records not excluded" from screening, but occasionally a record has no retrievable full text.
Reports sought for retrieval (n = ?)
Box: Reports not retrieved
Any full texts you could not obtain (paywalled and no author response, conference abstract only, etc.).
Reports not retrieved (n = ?)
Box: Reports assessed for eligibility
The full texts you actually read and assessed against all criteria.
Reports assessed for eligibility (n = ?)
Box: Reports excluded with reasons
This is the most important exclusion box. You must list each exclusion reason and its count. Common reasons include:
- Population does not meet criteria
- Intervention/exposure outside scope
- Comparator not specified
- Outcome not reported
- Wrong study design
- Duplicate publication (same data as another included study)
- Conference abstract only / no full data
Reports excluded: Reason A (n = ?) Reason B (n = ?) …
Stage 4 — Included
Box: New studies included in review
Studies that passed full-text assessment and entered your review. If you are performing a meta-analysis, these are the studies that contribute data to at least one analysis.
Studies included in review (n = ?)
Box: Reports of new studies included
A single "study" can have multiple reports (a trial may have a primary paper, a secondary outcomes paper, a protocol paper). Track both the number of unique studies and the total number of reports.
Reports of new studies included (n = ?)
Previous studies (if updating a prior review)
If your systematic review updates an earlier one, PRISMA 2020 adds boxes for studies and reports carried over from the previous version. These are shown in a shaded area at the bottom of the Included stage.
Complete Box-and-Count Reference Table
| Box label | What the number represents | Calculation check |
|---|---|---|
| Records from databases and registers | All database hits, pre-deduplication | Sum of all individual database search yields |
| Records from other methods | All non-database records | Sum by method type |
| Records removed before screening | Duplicates + automation exclusions + not retrieved | Subtracted from identification total |
| Records screened | Records seen by a human reviewer | Identification total minus removed-before-screening |
| Records excluded (title/abstract) | Failed title/abstract screen | Screened minus advancing to full-text |
| Reports sought for retrieval | Full texts you tried to get | = Records advancing from screening |
| Reports not retrieved | Full texts you could not access | Sought minus actually retrieved |
| Reports assessed for eligibility | Full texts you read | = Reports retrieved |
| Reports excluded (with reasons) | Failed full-text assessment | Assessed minus included; reasons must sum to total |
| Studies included | Unique studies in your synthesis | Final analytical sample |
| Reports of included studies | Total publications for included studies | ≥ number of studies |
The Arithmetic Must Balance
A surprisingly common revision request is an inconsistency in the numbers. Use these checks before you submit:
- Identification → Screening: (Records from databases) + (Records from other methods) − (Removed before screening) = Records screened
- Screening → Eligibility: Records screened − Records excluded (title/abstract) = Reports sought for retrieval
- Eligibility → Included: Reports assessed − Reports excluded (with reasons) = Studies included
- Reports ≥ Studies: Reports of included studies must be ≥ number of included studies
Run through these four checks every time you update any count.
At every stage the counts must balance: records in, minus those removed, equals records advancing.
Common Mistakes to Avoid
Reporting post-deduplication instead of pre-deduplication in the first box. The first identification box should show the raw total from each source before you removed any duplicates. The deduplication step is shown explicitly in the removal box.
Omitting reasons for full-text exclusion. Title/abstract exclusion reasons are optional; full-text exclusion reasons are required. List every reason and its count.
Mixing studies and reports. The included stage distinguishes between unique studies and the publications reporting them. Keep these separate.
Not showing the right column when you did other searching. If you hand-searched journals, screened reference lists, or contacted experts, those records belong in the right column — omitting them is a reportable bias.
Using a PRISMA 2009 template for a 2020 submission. The two versions look similar but differ in terminology and structure. Most journals now require the 2020 version. Check the PRISMA 2020 statement (Page et al., 2021, BMJ) if you are unsure.
Formatting the Diagram
PRISMA does not dictate fonts or colours, but convention and readability point toward a few best practices:
| Element | Recommendation |
|---|---|
| Font | Sans-serif (Arial, Helvetica, or Calibri); 8–10 pt for box text |
| Box style | Rounded rectangle for stage boxes; plain rectangle for count/removal boxes |
| Arrows | Single-headed, solid; no decorative arrowheads |
| Colour | Monochrome is safe for all journals; light grey fill for removal boxes aids readability |
| Stage labels | Bold, all-caps or title case; placed outside or above the first box in each stage |
| File format | Export as TIFF (≥300 DPI) or PDF for journal submission; PNG for preprints |
Building Your PRISMA Diagram with SciDraw AI
Drawing the diagram manually in PowerPoint, Word, or Illustrator is tedious — every time a count changes you must re-edit boxes and recheck alignment. A faster path is to use a dedicated tool.
SciDraw AI's PRISMA flow diagram generator lets you enter your counts directly into a structured form and generates a properly formatted, publication-ready diagram automatically. You can adjust stage labels, add or remove the "other methods" column, edit exclusion reasons, and export to the format your target journal requires.
If your work extends beyond systematic reviews, the same platform's workflow diagram generator handles general process flows, experimental pipelines, and multi-step methodology diagrams — useful for methods sections in primary research papers as well.
Frequently Asked Questions
Do I need to use the official PRISMA 2020 template? PRISMA provides a fillable Word template on the PRISMA website, but journals do not require it. Any diagram that contains all required elements and accurate counts is acceptable. Using a purpose-built tool or drawing the figure in vector software is fine as long as the content matches the checklist.
What if a record appears in multiple databases? Count it once in each database's yield (contributing to the pre-deduplication total in the first box), then remove all copies except one in the "Duplicate records removed" step. The key is that the numbers in the first box reflect raw search results so readers can assess database overlap.
Do I need separate PRISMA diagrams for different outcomes? Generally no — one diagram covers the whole review. If you are running separate searches for distinct review questions, separate diagrams may be appropriate, but this is unusual. Discuss with your co-authors and check the journal's author guidelines.
Can I use PRISMA for a scoping review? The PRISMA-ScR (PRISMA extension for Scoping Reviews) provides a modified checklist. The flow diagram structure is essentially identical to PRISMA 2020, so the same box logic applies.
What is "automation tools" in the removal step? Any software that uses machine learning or rule-based algorithms to flag records as likely irrelevant before human screening. Common examples are Rayyan's AI prioritisation, Covidence's deduplication engine, and ASReview. If you did not use such tools, omit that line from the removal box.
My journal asks for a "study flow diagram" — is that the same thing? Yes, in most cases. "Study flow diagram," "PRISMA flow chart," and "literature search flowchart" all refer to the same figure. Check whether the journal specifies PRISMA 2020 compliance explicitly; if so, follow this guide.
