The numbers in a PRISMA flow diagram trace every record from the moment it is found in a database to the final set of studies included in your systematic review. Each box holds a specific count: records identified, duplicates removed, records screened and excluded, reports sought and retrieved, reports assessed for eligibility and excluded with reasons, and studies included. Getting these counts right, and making them reconcile from top to bottom, is the single most scrutinised part of the figure. You can enter them directly into our free PRISMA 2020 flow diagram generator and watch the diagram build as you type, but understanding what each box represents is what prevents reviewer queries.
The reason this trips up so many researchers is terminology. PRISMA 2020 deliberately separates records, reports, and studies, and the arithmetic only works when you count the right unit at each stage.
Records, Reports, and Studies Are Not the Same Thing
Before any number makes sense, fix these three definitions in your mind:
- A record is a single citation or entry returned by a search, such as one row in a PubMed or Scopus results export. The same article found in three databases counts as three records until deduplication.
- A report is a full document (a journal article, preprint, or thesis) that you retrieve and read in full.
- A study is the underlying piece of research. One study can be described across several reports, for example a trial with a primary paper and a secondary analysis.
The top of the diagram counts records. The middle counts reports. The bottom counts studies (and the reports of those studies). Mixing these units is the root cause of most reconciliation errors.
Identification: Counting What You Found
The identification phase records the raw yield of your search before any filtering.
- Records identified from databases: enter the count from each database separately (PubMed, Embase, Scopus, and so on). The generator lets you add one line per database so the totals are transparent.
- Records identified from registers: trial registries such as ClinicalTrials.gov are counted here.
- Records identified from other methods: if your template includes other sources, count citation searching, website searches, organisational contact, and grey literature in their own column.
This phase should reflect the search exactly as you ran it. If you searched four databases and two returned overlapping results, every record still counts at this stage. Deduplication happens next, not here.
Records Removed Before Screening
This box captures everything taken out before a human reads a single title:
- Duplicate records removed: the same article found in multiple databases, merged into one.
- Records marked ineligible by automation tools: if you used machine learning or a rule to flag clearly irrelevant records.
- Records removed for other reasons: for example, non-English records you decided in advance to exclude by an automated filter.
The number entering the screening stage equals total records identified minus everything removed here. Our guide on systematic review screening best practices explains how to document this deduplication defensibly.
Screening: Title and Abstract Review
In the screening phase you record two numbers:
- Records screened: the count that reached title and abstract review after deduplication.
- Records excluded: those rejected at title and abstract stage.
The records excluded here do not need itemised reasons in PRISMA 2020. Detailed reasons are required only at the next stage, when you have read the full text.
Retrieval: Sought and Not Retrieved
Two boxes track access to full documents:
- Reports sought for retrieval: the records that passed screening and for which you tried to obtain the full text.
- Reports not retrieved: those you could not access despite reasonable effort, such as unavailable conference abstracts.
Reports sought minus reports not retrieved gives the number that move into eligibility assessment.
Eligibility: Assessed and Excluded With Reasons
The eligibility phase is where exclusion reasons become mandatory:
- Reports assessed for eligibility: the full texts you read against your criteria.
- Reports excluded: each grouped under a specific reason with its own count, for example "Wrong population (n = 31)", "Wrong intervention (n = 22)", "Wrong outcome (n = 14)", and "Wrong study design (n = 9)".
The sum of all exclusion-reason counts must equal the total reports excluded at this stage. A mismatch here is one of the most common PRISMA diagram mistakes and a frequent reviewer flag.
Included: Studies and Reports
The final boxes report the result of the whole process:
- Studies included in review: the number of distinct studies.
- Reports of included studies: the number of documents describing them, which may be higher than the number of studies.
If your review is an update, the template adds boxes for studies and reports carried over from the previous version, so the totals combine old and new evidence.
Occasionally the chain runs all the way down to an included count of zero, when no study survives full-text assessment. That is still a valid, reportable result, and the figure must be drawn and the reconciliation shown exactly as for any other review. Our guide to what to report when a search yields zero included studies explains how to present an empty review so the absence of evidence is itself documented transparently.
A Worked Example
Suppose a search returns 1,240 records: 600 from PubMed, 400 from Embase, and 240 from Scopus. Deduplication removes 340 duplicates, leaving 900 records screened. Title and abstract screening excludes 760, so 140 reports are sought. You cannot retrieve 8, leaving 132 reports assessed for eligibility.
At full text you exclude 110 reports: 40 wrong population, 35 wrong intervention, 20 wrong outcome, and 15 wrong study design. That leaves 22 reports, which describe 18 studies because four studies each have two publications.
Reading the diagram top to bottom, every number reconciles: 1,240 minus 340 is 900; 900 minus 760 is 140; 140 minus 8 is 132; 132 minus 110 is 22; and 22 reports map to 18 studies. This is exactly the chain a peer reviewer checks.
Do the Numbers Always Have to Add Up?
Yes, with one nuance. The vertical flow must reconcile: what leaves one box must be accounted for in the next. The only place counts can legitimately diverge is the studies-versus-reports distinction at the bottom, because one study can generate multiple reports. Everywhere else, a number that does not subtract cleanly signals a counting error. Picking the correct starting layout helps, which our guide to the four PRISMA 2020 flow diagram templates explains.
Frequently Asked Questions
What is the difference between records, reports, and studies in PRISMA?
A record is a single search result (one citation), a report is a full document you retrieve and read, and a study is the underlying research project. One study can appear in several reports, which is why the included counts can differ.
Do the numbers in a PRISMA flow diagram have to add up?
Yes. Each stage must reconcile with the next: records screened minus records excluded equals reports sought, and so on. The only intentional difference is that the number of reports of included studies can exceed the number of included studies.
Where do I put duplicates in a PRISMA 2020 diagram?
Duplicates go in the "records removed before screening" box, separate from automation removals and other pre-screening removals. They are subtracted from total records identified before title and abstract screening begins.
How many exclusion reasons should I list?
List enough specific categories to account for every excluded full-text report, commonly three to seven. The counts under your reasons must sum exactly to the total reports excluded at the eligibility stage.
What number do I report if one study has several papers?
Report the number of studies and the number of reports separately. The "studies included" box counts distinct studies, while "reports of included studies" counts every document describing them.