PRISMA flow diagram: a practical guide

Examiners and journal reviewers read the PRISMA flow diagram before they read your prose. It is a single figure that summarises how many records your searches found, how many you screened, how many full texts you assessed, and how many studies made it into the review. It is also the easiest part of a systematic review to audit: anyone with a pencil can check whether the numbers reconcile from box to box. If they do not, the reader stops trusting the review before reaching your synthesis, and the questions in your viva start from a defensive position.

This guide explains what each box in the PRISMA 2020 flow diagram means, walks through a worked example with numbers that add up, and lists the mistakes that get flagged in vivas and peer review. It assumes you are running a systematic review, or a systematic-style literature review for a thesis, and that you want the figure to be transcription of a well-kept log rather than end-of-project archaeology. If you want the full review workflow rather than just the figure, see the companion guide on running a PRISMA systematic review in CiteDash.

What is a PRISMA flow diagram?

PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses. It is a reporting guideline, not a method: it tells you what to report about your review, not how to search, screen, or appraise. The current version is PRISMA 2020, published by Page and colleagues in 2021, and it has two core artefacts: a 27-item checklist covering everything from the rationale to the funding statement, and the flow diagram.

The flow diagram depicts the flow of information through the phases of the review: how many records were identified, how many were screened and excluded, how many reports were retrieved and assessed for eligibility, and how many studies were finally included. It exists because prose descriptions of screening are unauditable; a figure with explicit counts forces the arithmetic into the open. Cite the PRISMA 2020 statement (Page et al., 2021) in your methods chapter and say explicitly that you used the PRISMA 2020 template, because the 2009 and 2020 diagrams have different boxes and an examiner who knows the difference will notice a mismatch.

The four stages of a PRISMA 2020 flow diagram

The standard template for a new review that searched databases and registers has four horizontal bands, read top to bottom:

• Identification: records found by your database and register searches, listed per source, plus the number of duplicate records removed before screening.
• Screening: records screened at title and abstract level, and the number excluded at that stage.
• Retrieval and eligibility: reports you sought in full text, reports you could not retrieve, reports assessed against your eligibility criteria, and reports excluded with a count for each stated reason.
• Included: the number of studies included in the review, and the number of reports those studies came from.

The optional second column: other search methods

PRISMA 2020 also provides an optional parallel column for records identified by methods other than database searching: citation searching through reference lists, hand-searching key journals, websites and organisational reports, and contact with authors. These sources matter in fields where the database coverage is thin, and reviewers increasingly expect citation searching for any serious review.

The structural point people miss is that reports found this way join the main flow at the retrieval stage, not at title-and-abstract screening. A paper found by chasing a reference list usually arrives as a document rather than a database record, so it enters the diagram as a report sought for retrieval, with its own retrieval and eligibility counts in the second column. Folding these finds silently into your database totals misstates both columns and is one of the specific things the 2020 revision was designed to prevent.

Records, reports, and studies: three words that must not blur

The 2020 template is strict about vocabulary because earlier diagrams kept conflating three different things:

• A record is a title-and-abstract entry returned by a search. One paper indexed in three databases produces three records, which is why deduplication exists.
• A report is a document: a journal article, a preprint, a conference paper, a thesis chapter, a trial registration.
• A study is the investigation itself. One study can be described in several reports, and one report can describe several studies.

A worked example with numbers that reconcile

The top half of the diagram counts records; the bottom half counts reports and studies. Suppose you are reviewing mindfulness interventions for postgraduate student anxiety. Your searches ran on 12 March 2026 across three databases and one trials register. Here is a full set of counts that reconcile:

• Records identified: 1,214 from databases (Scopus 412, PubMed 414, Web of Science 388) and 62 from the trials register, 1,276 in total.
• Duplicate records removed before screening: 453, leaving 823 records.
• Records screened at title and abstract: 823; records excluded: 704.
• Reports sought for retrieval: 119; reports not retrieved: 7.
• Reports assessed for eligibility: 112; reports excluded: 84 (wrong population 31, wrong intervention 26, wrong study design 19, conference abstract only 8).
• Studies included in the review: 28, described in 31 reports.

Auditing the arithmetic

Every transition in that example can be checked. 1,276 identified minus 453 duplicates is 823 screened. 823 screened minus 704 excluded is 119 sought for retrieval. 119 sought minus 7 not retrieved is 112 assessed. 112 assessed minus 84 excluded is 28 included studies, and the four exclusion reasons sum to 84. Three of the included studies each had a second linked report, a protocol or a follow-up paper, which is why 28 studies map to 31 reports.

Do this arithmetic on your own diagram before anyone else does. The commonest way it breaks in real theses is time: screening continued for another fortnight after the diagram was drafted, or a late hand-searched paper was added to the included set without touching the figure. Treat the diagram as generated from the log, never edited by hand in isolation, and regenerate it as the last step before submission.

Where each number comes from

You cannot reconstruct these counts at write-up time; they have to be captured as you go. The identification numbers come from your dated database export files, one per source. The duplicate count comes from your deduplication step, whether that is reference-manager matching or manual checking by DOI and title. The screening numbers come from a decision log in which every record carries an include or exclude verdict and a date. The retrieval gap is a list of reports you requested but never obtained, with the reason: paywalled with no interlibrary option, author unreachable, thesis embargoed. The eligibility exclusions come from full-text screening notes, where each excluded report carries exactly one primary reason drawn from a fixed list you defined in advance.

Note the asymmetry that trips people up: PRISMA 2020 does not require itemised reasons for title-and-abstract exclusions, but it does require reasons with counts at the full-text stage. If your full-text log only says excluded, you will be rebuilding your reasons from memory in the week before submission, and it will show.

What changed from PRISMA 2009 to PRISMA 2020

Many university guides and older theses still show the 2009 diagram, so it is worth knowing the differences rather than copying whichever template a previous student used. PRISMA 2020 separates database records from register records in the identification box, adds the explicit pair of reports sought for retrieval and reports not retrieved, lets you report records excluded by automation tools as their own count, and adds the optional second column for other search methods. The 2009 wording of eligibility as a labelled phase became the assessment of retrieved reports, and the ambiguous full-text articles language was replaced by the record, report, and study vocabulary above.

The practical rule: if your methods chapter cites PRISMA 2020, use the 2020 template. Citing the current statement while pasting the old figure is exactly the kind of internal inconsistency examiners circle in the margin, because it suggests the guideline was cited rather than read.

Common PRISMA flow diagram mistakes

These are the failures that come up again and again in vivas and in peer review:

• Numbers that do not reconcile between boxes, usually because screening continued after the diagram was drafted.
• Exclusion reasons reported at title-and-abstract stage but not at full-text stage, which is the reverse of what PRISMA 2020 asks for.
• Records counted where the template asks for reports, so the bottom half of the diagram cannot be audited.
• Register searches left out entirely, or citation-searching finds folded into the database totals.
• Deduplication done after screening, which double-counts screening work and makes the duplicate box meaningless.
• A diagram that disagrees with the results text or the abstract because one was updated and the other was not.

Keeping the counts without spreadsheet sprawl

None of those mistakes are fatal if caught before submission, and all of them are avoidable with a screening log kept from day one: one row per record, one dated decision per row, one reason per full-text exclusion. The practical difficulty is bookkeeping across months of screening and several tools, which is where most reviews fray.

Keeping search and screening in one workspace helps. In CiteDash, the Literature Finder searches an internal corpus plus live OpenAlex, PubMed, Semantic Scholar and arXiv, so your searches and their results live alongside your screening decisions rather than in a folder of exports. The Synthesis Lab supports PRISMA-friendly systematic review work with an evidence matrix across the papers you include, and the Library tracks reading statuses and shows retraction badges, with retracted papers blocked from citation, so a withdrawn trial cannot quietly survive into your included set.

Before you submit: a five-minute check

Reconcile the arithmetic line by line, top to bottom. Check the diagram against the results text and against the abstract, which usually quotes the included-studies number and is usually the last thing edited. Confirm the figure caption cites PRISMA 2020 and that the search date shown matches the methods chapter. Confirm the exclusion reasons in the figure match the fixed list your methods chapter promises. If you updated the search before submission, the diagram must reflect the combined totals, or you need a second diagram for the update, clearly labelled.

A PRISMA flow diagram that reconciles will not make a weak review strong. But a diagram that fails its own arithmetic makes a strong review look careless, and it is the one part of the thesis every examiner can check in five minutes. Spend the five minutes first.