Systematic review · Reporting
How to build a PRISMA flow diagram that reconciles
A flow diagram is an arithmetic claim about your screening process. Reviewers check whether the numbers close, and the usual reason they do not is that the counts were reconstructed afterwards from memory.
In this article
The PRISMA flow diagram looks like a formatting task and behaves like an audit. Each box is a count, each arrow is a claim that one count produced another, and a reader can check the whole chain in about thirty seconds. This describes how to produce one whose numbers reconcile, using the official templates rather than a redrawn approximation.
It covers template choice, count capture, the handling of studies awaiting classification, and the final reconciliation. It does not decide your eligibility criteria, and it is not a substitute for your protocol.
1. Pick the right template — there are four
Most writers know of one PRISMA flow diagram. The PRISMA 2020 flow diagram page actually distributes four Word templates, and choosing the wrong one creates work you cannot undo cleanly later. They split along two axes: whether the review is new or an update, and whether it searched databases and registers only or also used other sources.
| Review | Sources searched | What the template adds |
|---|---|---|
| New | Databases and registers only | The baseline three-stage layout |
| New | Databases, registers and other sources | A parallel column for websites, citation searching, and contact with authors |
| Updated | Databases and registers only | Columns separating previously included studies from newly identified ones |
| Updated | Databases, registers and other sources | Both of the above |
The “other sources” column is the one most often dropped. If any study reached your included set through citation chaining, a preprint server, or an author who sent you an unpublished report, the databases-only template cannot represent how it arrived, and the diagram will imply a search path that did not happen.
2. Record the counts while screening, not afterwards
The Cochrane Handbook's chapter on searching for and selecting studies puts this plainly: during selection it is crucial to keep track of the number of references, and subsequently the number of studies, so that a flow diagram can be constructed. The order matters. The diagram is an output of the screening log, not a document you assemble from it later.
The same chapter is explicit that at least two people should work independently to decide whether each study meets the eligibility criteria, and that title-and-abstract screening should ideally be done in duplicate as well. That has a bookkeeping consequence: your log needs a column for each screener's decision and a column for the resolved decision, because the count that enters the diagram is the resolved one.
3. “Awaiting classification” is not “excluded”
This is the error that most often makes a diagram fail to reconcile, and it is a categorisation mistake rather than an arithmetic one. When you learn of a possibly relevant report you cannot yet assess — an unobtainable full text, an ambiguous conference abstract, a translation you are waiting on — Cochrane directs that it be added to studies awaiting classification rather than to excluded studies, and that this be reflected in the flow diagram.
Filing those records under excluded is convenient and wrong. It asserts that you applied your criteria and the study failed them, when in fact you never completed the assessment. It also destroys the audit trail: an update of the review has no way to find the records that were pending rather than rejected.
Note the last row. Twenty-three studies are reported in twenty-seven reports, because four studies each produced two publications. Studies and reports are different units, and collapsing them is how a review ends up double-counting evidence — the same distinction that governs a deduplication pass.
4. Make the arithmetic close
Before the diagram goes into the manuscript, test every stage as a subtraction and confirm each figure traces to a row set in your log rather than to a number you remember. Each of these should hold exactly:
- Records identified minus records removed before screening equals records screened.
- Records screened minus records excluded equals reports sought for retrieval.
- Reports sought minus reports not retrieved equals reports assessed for eligibility.
- Reports assessed minus the sum of all exclusion reasons minus studies awaiting classification equals included studies.
- The individual exclusion reasons sum to the stated exclusion total — this is the one that usually fails.
When a stage does not close, the cause is almost always a record moved between categories without the log being updated, or a duplicate removed at the wrong stage. Both are recoverable if the original exports were preserved unchanged, which is what a reproducible search log exists to guarantee.
5. Reporting, and the licence
The diagram is a summary, not the record. The review text still has to state the databases and platforms searched, the date of each search, and the eligibility criteria the numbers were produced by. Where a count reflects a judgement — what made a design ineligible, how companion reports were linked — the reasoning belongs in the text, because the box only shows the total.
One practical note: the PRISMA 2020 flow diagram templates are distributed under a Creative Commons Attribution 4.0 licence, which permits adaptation and commercial use provided the original work is properly cited. Adapting the layout is allowed; citing the source is the condition. The broader PRISMA 2020 statement covers the reporting items the diagram sits inside.
Once the counts close, carry the included set into your evidence matrix, where each retained study is joined to the passages that will support the synthesis. The flow diagram then documents how that set was reached, which is the question a reader asks first.