PRISMA flow diagram example
Updated 2026-10-27
The easiest way to understand a PRISMA flow diagram is to follow the numbers through one. This page works through four examples, from a simple database-only review to one that uses trial registers and automation tools. It then gives you a diagram with three mistakes to find.
The numbers are made up for teaching. They are not taken from a published review, but each one is realistic and every subtraction is shown, so you can check your own diagram the same way.
Example 1: databases only
A review of a drug treatment searched three databases. Every included study had one report, so the number of studies and the number of reports are the same.
| Stage | Number | Check |
|---|---|---|
| PubMed | 412 | |
| Embase | 520 | |
| CENTRAL | 168 | 412 + 520 + 168 = 1,100 records identified |
| Duplicates removed | 260 | 1,100 − 260 = 840 screened |
| Records excluded at title and abstract | 700 | 840 − 700 = 140 reports sought |
| Reports not retrieved | 10 | 140 − 10 = 130 assessed |
| Reports excluded with reasons | 90 | 130 − 90 = 40 studies included |
The exclusion reasons add up to 90: wrong population (35), wrong outcome (30) and wrong study design (25).
Here is the diagram this example produces with the new-review, databases-and-registers-only template:
How to describe Example 1 in your results
Journals expect the text to match the figure. A results paragraph for this example could read:
The database searches identified 1,100 records (PubMed n = 412; Embase n = 520; CENTRAL n = 168). After 260 duplicates were removed, we screened 840 titles and abstracts and excluded 700. We sought 140 reports for retrieval and could not obtain 10. Of the 130 full-text reports assessed for eligibility, 90 were excluded (wrong population n = 35; wrong outcome n = 30; wrong study design n = 25). Forty studies met the inclusion criteria (Figure 1).
Every number in that paragraph appears in a box, and nothing in a box is missing from the paragraph.
Example 2: adding other methods
Now suppose the same review also searched websites and organisations and checked the reference lists of included studies. These go in the right-hand column of the template for databases, registers and other sources.
| Stage (other methods) | Number | Check |
|---|---|---|
| Websites | 22 | |
| Organisations | 8 | |
| Citation searching | 30 | 22 + 8 + 30 = 60 records identified |
| Reports sought for retrieval | 48 | 12 records were clearly irrelevant on sight |
| Reports not retrieved | 6 | 48 − 6 = 42 assessed |
| Reports excluded with reasons | 30 | 42 − 30 = 12 studies included |
The 30 exclusions break down as not a primary study (14), wrong outcome (9) and wrong population (7).
Two things often confuse people here. First, the other-methods column has no records-screened box. The official template goes straight from records identified to reports sought, so the drop from 60 to 48 isn't drawn. If the gap is large, explain it in your methods. Second, the two columns merge at the bottom, so the review reports 40 + 12 = 52 studies included, not two separate totals.
Example 3: more reports than studies
PRISMA 2020 counts reports and studies separately because one study can be published several times: a main paper, a conference abstract, a long-term follow-up, a trial register entry with results. Suppose the 130 full-text reports in Example 1 had led to 40 included reports, but these described 34 studies:
- 29 studies had a single report (29 reports)
- 4 studies had two reports each (8 reports)
- 1 study had three reports (3 reports)
29 + 8 + 3 = 40 reports, from 29 + 4 + 1 = 34 studies. The final box then reads Studies included in review (n = 34) and Reports of included studies (n = 40). The arithmetic above that box still uses reports, because you assessed documents, not studies. If you write "40 studies" here, every meta-analysis count further on in the paper will be off.
Example 4: trial registers and automation tools
A larger review searched three databases and two trial registers. Before screening, it used a deduplication step and a machine-learning classifier that flagged records as ineligible.
| Stage | Number | Check |
|---|---|---|
| MEDLINE | 1,254 | |
| Embase | 1,873 | |
| CINAHL | 402 | Databases: 3,529 |
| ClinicalTrials.gov | 96 | |
| WHO ICTRP | 71 | Registers: 167. Total: 3,529 + 167 = 3,696 |
| Duplicate records removed | 1,108 | |
| Records marked as ineligible by automation tools | 214 | |
| Records removed for other reasons | 0 | 1,108 + 214 + 0 = 1,322 removed; 3,696 − 1,322 = 2,374 screened |
| Records excluded | 2,201 | 2,374 − 2,201 = 173 sought |
| Reports not retrieved | 7 | 173 − 7 = 166 assessed |
| Reports excluded with reasons | 139 | 166 − 139 = 27 reports included |
| Studies included | 21 | 27 reports describe 21 studies |
The 139 exclusions break down as wrong population (48), wrong comparator (31), wrong study design (37) and register entry with no results available (23), which adds up to 139.
Lessons from this example:
- Keep boxes that are zero. "Records removed for other reasons (n = 0)" shows that the step was considered. Removing the box makes readers wonder what happened.
- Name the automation tool in your methods. The 214 records are only meaningful if readers know what the classifier was, how it was trained or validated, and whether a person checked its decisions.
- Give register counts separately from databases. Trial registers are a separate source in PRISMA 2020, so they get their own line instead of being merged into the database total.
Spot the errors
This diagram, written out as a table, has three mistakes of the kind peer reviewers often find. Try to find them before opening the answers.
| Stage | Number |
|---|---|
| Records identified: PubMed 400, Embase 350, Scopus 250 | 950 |
| Duplicates removed | 200 |
| Records screened | 800 |
| Records excluded | 690 |
| Reports sought for retrieval | 110 |
| Reports not retrieved | 5 |
| Reports assessed for eligibility | 110 |
| Reports excluded: wrong population 40, wrong study design 30 | 80 |
| Studies included | 25 |
Show the answers
1. The total doesn't match the sources. 400 + 350 + 250 = 1,000, not 950. Records screened (800) only works with 1,000 identified (1,000 − 200), so the 950 is a typo in the top box.
2. Reports not retrieved weren't subtracted. 110 sought minus 5 not retrieved is 105 assessed, not 110. Studies included (25) only works with 105 (105 − 80).
3. The exclusion reasons don't add up. 40 + 30 = 70, but the box says 80. Ten excluded reports have no reason. Find them in your screening log, or add the missing category, for example "wrong outcome (n = 10)".
The checks every diagram must pass
- Sources add up to records identified, separately for databases and for registers.
- Records identified minus every "removed before screening" box equals records screened.
- Records screened minus records excluded equals reports sought for retrieval.
- Reports sought minus reports not retrieved equals reports assessed.
- Reports assessed minus reports excluded equals included reports, and the exclusion reasons add up to the excluded total.
- Included studies are no more than included reports, and the figure matches the text and abstract.
Try it with your own numbers
Type any of these examples into the free PRISMA flow diagram generator to see it drawn, then replace the numbers with your own. Example 1 and Example 4 use the databases and registers only template. Example 2 needs the version with other methods. For a review that adds to an earlier one, see the worked numbers in the updated review guide. New to the layout? Read what a PRISMA flow diagram is or how to make one.
Quick answers
Do the numbers in the examples matter?
No. They are illustrative. What matters is that your own boxes add up and that every exclusion has a reason.
Why is there no screening box for other methods?
The official PRISMA 2020 template goes straight from records identified to reports sought for retrieval in the other-methods column. If many records were set aside before retrieval, say so in your methods section.
Can included reports be more than included studies?
Yes, and in many reviews they are. One study can have several reports, such as a main paper, a follow-up and a conference abstract. Report both numbers in the final box.
Source of the flow diagram wording: Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi:10.1136/bmj.n71. Licensed CC BY 4.0. This site is independent and not affiliated with the PRISMA or EQUATOR Network groups.