The PRISMA flow diagram is a four-phase chart that traces every record through a systematic review, from the total identified by the search, through the numbers screened and excluded, to the studies finally included. It is the single figure that lets a reader audit your study selection at a glance, and reporting it is a core requirement of the PRISMA 2020 guideline.
Why every number on the diagram has to reconcile
The flow diagram is more than decoration. It is an arithmetic statement: the records that enter each phase, minus those excluded, must equal the number passing to the next phase. When the figures do not add up, an editor or peer reviewer immediately doubts the rest of the review, because it suggests records were lost or counted inconsistently. Building the diagram from counts captured at the moment each decision is made, rather than reconstructing them at write-up, is the only reliable way to keep the arithmetic honest.
The four phases, in order
1. Identification
The top box reports how many records the search found, split by source such as databases and registers, and counts the duplicates removed before screening. These figures come straight from de-duplicating your search results, so the number of unique records entering screening is fixed here.
2. Screening
The screening phase reports the records assessed at title and abstract screening and the number excluded. At this level exclusions are a single count, not an itemised list, because an abstract rarely supports a precise reason.
3. Eligibility
The eligibility phase covers full-text screening. Here the diagram needs the reports sought, those not retrieved, the reports assessed, and the number excluded broken down by reason. This is the only phase where exclusions are itemised, so each excluded full text contributes to one reason category.
4. Included
The final box reports the number of studies included in the review and, where relevant, the number of reports of those studies. Because one study can have several reports, this is also where companion papers are reconciled so a single study is counted once even when it appears in a protocol paper and a results paper. This is also the count that feeds every downstream figure: the number of studies in the included box should match the rows in your characteristics table, the entries in your appraisal table, and the data points on any forest plot. A mismatch between the diagram and those tables is one of the first things a careful reviewer cross-checks.
Common mistakes that get a diagram sent back
The recurring problems are predictable. Numbers that fail to reconcile are the most common. Mixing the older PRISMA 2009 layout with the 2020 version is another, since the 2020 diagram separates databases from other methods such as citation chasing. A third is omitting the not-retrieved count, which quietly hides reports you never assessed. Each of these undermines the auditability the diagram exists to provide.
Two subtler errors trip up experienced teams. The first is confusing records, reports, and studies, which the 2020 diagram treats as distinct: a record is a database hit, a report is a retrievable document, and a study is the underlying piece of research, which may span several reports. The included box counts studies, not reports, so a single trial published as a protocol paper and a results paper is one study but two reports. The second is itemising exclusion reasons at the wrong phase. Reasons belong only at the eligibility, full-text level; attaching them to title and abstract exclusions implies a precision an abstract cannot support, and reviewers notice. When in doubt, resolve any ambiguous count back to the screening records that also drive your agreement statistics, because those records are the single source of truth for every figure.
A worked example of reconciling the counts
The arithmetic is easiest to trust when you can follow a single set of numbers through. Suppose two databases return 3,200 and 1,900 records, and a citation search adds 60, for 5,160 in total. After removing 1,160 duplicate records, 4,000 unique records enter screening. Title and abstract screening excludes 3,650, leaving 350 reports sought for full text. Of those, 12 are not retrieved, so 338 full texts are assessed. Full-text screening excludes 310, itemised by reason (for example, 140 wrong population, 90 wrong study design, 80 no usable outcome), leaving 28 studies. Every box now reconciles: 4,000 minus 3,650 equals 350; 350 minus 12 equals 338; 338 minus 310 equals 28. If any subtraction failed, a reviewer would stop reading the results and start questioning the search.
Database records versus records found by other methods
The redesigned PRISMA 2020 layout splits the top of the diagram into two parallel columns: records identified through databases and registers, and records identified through other methods such as citation chasing and snowballing, contact with experts, or grey literature sources. Each column runs through its own screening and eligibility counts before the two streams merge at the included box. Reviewers mix this up by funnelling everything into the database column, which hides how studies found by hand searching were assessed. Keeping the two streams separate is not cosmetic: it lets a reader see how much of the evidence the formal search strategy actually captured.
Where the diagram fits in reporting
The flow diagram is one element of full PRISMA reporting, sitting alongside the full PRISMA 2020 checklist, the full search strategy, and the risk of bias assessment. It is generated from the same screening records that drive conflict resolution, which is why doing your screening inside a structured screening tool pays off: the counts you need for the diagram are produced as a by-product of the work rather than as a separate, error-prone exercise at the end. The same records also tell you how many reviewers screened each phase, which PRISMA asks you to report alongside the figure.