PRISMA 2020 Flow Diagram Generator

A plausible diagram can still conceal an incoherent selection record

The PRISMA 2020 flow diagram compresses thousands of identification and selection events into a single visual summary. While this structure helps readers trace how search hits became included studies, it also removes the underlying case-level evidence. Completing a diagram from memory or inconsistent software exports can yield internally plausible numbers that actually represent mismatched units or conflicting review stages.

As a reporting guideline, PRISMA 2020 specifies what information must be made visible: records identified, removals prior to screening, records screened, reports sought and assessed, full-text exclusions with primary reasons, and included studies and reports.^1,2 It does not evaluate whether the search was comprehensive, whether criteria were methodologically justified, or whether inclusion decisions were correct. Flow records should always be generated directly from the study-selection ledger rather than created as a substitute for it.

The central reconciliation challenge stems from shifts in analysis unit. Initial screening counts database records; retrieval and full-text appraisal evaluate individual reports; final inclusion applies to underlying studies, which may span multiple reports. Cochrane explicitly designates the study as the primary unit of interest and requires grouping companion reports prior to synthesis.^4,5 Generators that simply subtract raw numbers across stages risk creating false equations if they mix report-level events with study-level outcomes without a linkage record.

Choose the official family before entering a count

Official PRISMA guidance provides four flow-diagram templates: new or updated reviews, each available as a databases-only version or a version that incorporates other identification routes.³ Selecting the correct layout changes the audit trail structure. Reviews using citation searching, web searches, or organizational contacts should not force those hits into a databases-only template. Updated reviews must separately represent both retained baseline studies and newly identified records.

Confirming template family early ensures all required transitions are logged during screening. Switching layout families after completing screening often reveals missing source data, particularly if secondary searches were recorded only as narrative notes rather than countable events.

Records, reports, and studies are not interchangeable

A record is the title/abstract entry evaluated during preliminary screening. A report is a document (journal paper, preprint, or trial register) supplying data about an investigation. A study is the underlying scientific trial itself. One study may yield several reports, while a single paper may describe multiple studies. Maintaining distinct counts for included studies and included reports prevents reporting errors.

Duplicates refer to repeated database records of the same publication, not separate reports describing the same trial. A trial registry entry, conference abstract, and full journal article represent distinct reports of one study. Removing secondary reports as “duplicates” risks discarding valuable outcome details; treating them as separate included studies artificially inflates trial numbers.

Use arithmetic only where units and pathways align

For standard database searches, expected screening totals equal identified records minus documented removals. Reports sought equal screened records minus title/abstract exclusions. Reports assessed equal reports sought minus unretrieved full texts. When primary exclusion reasons are mutually exclusive, their sum must equal total full-text exclusions. Discrepancies signal ledger tracking errors rather than reasons to alter displayed totals.

Investigate a mismatch before editing the number

An arithmetic mismatch often indicates a transcription mistake, an incomplete eligibility decision, an unlinked secondary report, or an unresolved conflict. Review the screening ledger to locate the specific record or report. Avoid altering displayed numbers purely to force arithmetic agreement.

Retrieved full texts that could not be obtained belong in the “reports not retrieved” section. Because they were never evaluated against full criteria, they must not receive a methodological exclusion reason. In updated reviews, baseline studies and newly identified records must remain separate before totals are calculated.

Trace every warning back to a decision event

Even a single discrepancy can alter review outcomes. It may signify a record screened twice, a late-retrieved paper, an unresolved full-text disagreement, or an incorrectly linked companion report. Locate the root cause in the primary database before regenerating counts.

Audit the final selection ledger in both directions. Forward auditing follows a sample of search hits through deduplication, abstract screening, retrieval, and full-text decisions. Backward auditing starts with included studies and excluded full texts to verify their original search sources. Conducting this verification across all identification routes ensures robust reporting.

A reconciled flow record supports transparency, not certification

This generator helps select the correct template family, maintain record-report-study distinctions, run arithmetic checks, and export preview records. It cannot certify whether search strategies were comprehensive or whether screening criteria were correctly applied. Methodological validity depends on adherence to your protocol and conduct standards.

Official PRISMA 2020 templates remain the reporting standard and are published under CC BY 4.0.³ Verify exported values against official templates prior to manuscript submission. Specialized review formats—such as scoping reviews or diagnostic test accuracy syntheses—may require modified flow structures.^7,8