SciSpace 详解 PRISMA 2020 流程图：每个框怎么填（附真实案例）

What's new?

Now you can research with Unlimited Agent Usage

5 Nov 2026·Deepak Raghava·4 min read

PRISMA Flow Diagram Explained: How to Fill Every Box (With a Real Example)

What each box in the PRISMA 2020 flow diagram means, how to count records, reports and studies, and a worked example from a real systematic review.

A PRISMA flow diagram shows how a systematic review moved from every record found in the search to the studies finally included. Journals expect it because it lets readers check that nothing was dropped without a reason.

In short: the PRISMA 2020 flow diagram has three stages. Identification counts the records found and removed before screening (mainly duplicates). Screening counts what was excluded at title and abstract, which full-text reports could not be retrieved, and which were excluded at full text with reasons. Included gives the final number of studies and reports. Every number must add up and match your screening log. You can draw it with the PRISMA flow diagram template or generate it with the PRISMA chart maker.

What the PRISMA flow diagram is

PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses. The current version, PRISMA 2020 (Page et al., BMJ, 2021), replaced the 2009 statement and comes with updated flow diagram templates. There are versions for new reviews that searched only databases and registers, for new reviews that also used other sources, and for updated reviews.

Use the 2020 template. Reviewers notice the older 2009 layout, and it does not separate records, reports and studies.

Records, reports and studies: the distinction that trips people up

PRISMA 2020 uses three different units, and the diagram changes unit as you move down it:

Unit

What it means

Where it appears

---|---|---

Record

A title and abstract indexed in a database or register

Identification and title/abstract screening

Report

A document you retrieve: article, preprint, thesis, conference abstract

Full-text retrieval and eligibility

Study

The investigation itself, which may be described in several reports

Included box

If one trial is published as two articles, you count two reports but one study.

How to fill every box

Identification

Records identified from databases and registers. Give the total and, ideally, the number from each source (for example PubMed n = 1,210; Scopus n = 1,894).
Records removed before screening. Split this into duplicates removed, records marked ineligible by automation tools, and records removed for other reasons.

Screening

Records screened. Records left after removals, screened on title and abstract.
Records excluded. Those removed at title and abstract screening. You do not need reasons here.
Reports sought for retrieval. The full texts you tried to get.
Reports not retrieved. Full texts you could not obtain. Report these rather than hiding them.
Reports assessed for eligibility. Full texts read against your criteria.
Reports excluded, with reasons. Give a count per reason, such as wrong population (n = 31), wrong outcome (n = 22) or ineligible study design (n = 18).

Included

Studies included in review. The final number of studies.
Reports of included studies. Add this when some studies have more than one report.

If you also found studies through citation searching, websites or experts, use the template with a second column for identification via other methods, and run those records through the same retrieval and eligibility boxes.

Worked example: from 4,634 records to 74 studies

In his SciSpace webinar, Dr. Faheem Ullah (University of Adelaide) walked through the screening pathway of his own systematic review. His team screened titles before removing duplicates, which is less common but allowed, as long as the diagram shows the order actually used.

Stage

Count

Excluded at this stage

---|---|---

Records identified from six databases

4,634

Retained after title screening

748

3,886

Retained after duplicate removal

516

232 duplicates

Retained after abstract screening

168

348

Retained after full-text screening

99 (with reasons)

Added through snowballing (other methods)

Studies included

Check the arithmetic the way an editor would: 748 − 516 = 232 duplicates, 516 − 168 = 348 excluded at abstract, 168 − 69 = 99 excluded at full text, and 69 + 5 = 74 included. If any of these subtractions fails in your own diagram, fix the screening log before you submit.

Key moments: 12:53 designing the search string and choosing databases · 14:51 PRISMA screening and snowballing · 19:30 how many papers a review should include.

Five mistakes reviewers spot in PRISMA diagrams

Numbers that do not add up. Every box should equal the box above it minus the exclusions beside it.
Mixing units. Counting reports in the Included box when you mean studies, or the reverse.
No reasons for full-text exclusions. A single "excluded (n = 99)" is not enough at full-text stage.
Missing other sources. Studies found through snowballing or experts appear in the final count but not in the diagram.
The diagram and the methods disagree. Databases, dates and screening order in the diagram must match the Methods section and your search log.

How to make a PRISMA flow diagram

Template: download the official PRISMA 2020 template from the PRISMA website and type in your counts, or use the SciSpace PRISMA flow diagram template.
R and Shiny app: the PRISMA2020 package and app (Haddaway et al., 2022) draw a compliant diagram from a table of counts.
AI-assisted: a systematic review agent can keep the counts as it screens and draw the diagram at the end. In her SciSpace session, Marija Đukić (University of Belgrade) shows the generated diagram and report at 15:39. Always reconcile the generated numbers with your own screening records.

Watch: automating PRISMA screening and data extraction

This SciSpace webinar shows how AI can speed up title, abstract and full-text screening while keeping the counts you need for every box of the PRISMA diagram.

注意转载授权 · 阅读原文