PRISMA 2020 Checklist Explained: What to Report and Where

Interpreting reporting adherence without false precision

A 27-item checklist does not create 27 simple yes-or-no decisions

PRISMA 2020 looks compact enough to invite a simple workflow: read an item, find a sentence in the manuscript, enter a page number, and move to the next row. That appearance is misleading. Many PRISMA items contain several reporting elements, some apply only under particular conditions, some information may legitimately sit outside the main manuscript, and different researchers can reach different adherence judgments unless they use explicit decision rules.

That distinction explains why two papers can both claim to have assessed “PRISMA 2020 adherence” while measuring meaningfully different things. One team may evaluate the 27 headline items as binary yes/no judgments. Another may decompose the same checklist into many elements, award partial credit, remove non-applicable content from the denominator, and inspect supplementary files. Both studies can produce a percentage called adherence, but the percentages do not necessarily represent the same construct.

This is no longer a minor methodological concern. Hamilton and colleagues systematically examined 24 tools used to assess PRISMA 2020 adherence across 2,766 reviews. No two research groups used the same assessment method, only five reported psychometric testing, and only one tool was described sufficiently for an independent researcher to implement it without contacting the developers.³ The problem is therefore not merely that systematic reviews omit PRISMA items. Researchers themselves have not yet converged on one standardized method for deciding exactly what “adherence” means.

The expanded checklist changes the unit of attention

The standard checklist is organized around 27 numbered items, with several divided into lettered sub-items. The expanded checklist goes further by identifying reporting elements within those items and by distinguishing different levels of reporting importance. Its introductory note is methodologically important: non-italicized elements are considered essential and should be reported in the main report or supplementary material for systematic reviews, except for elements beginning with “If…”, which apply only where the stated condition is met. Italicized elements are additional. They are not essential, although they may enhance completeness and usability.²

This means that an item number is often a container rather than a single observation. Consider study selection. Item 8 does not merely ask authors to say that two reviewers screened records. The expanded guidance addresses how many reviewers screened records and reports, whether they worked independently, how disagreements were resolved, processes used to confirm information, translation where relevant, and detailed reporting when automation, crowdsourcing or previously screened datasets were incorporated.² A manuscript can therefore contain enough information for a superficial “yes” while still leaving important parts of the process impossible to reconstruct.

The same pattern occurs in synthesis reporting. Item 13d covers methods used to synthesize results. When meta-analysis is performed, its expanded guidance reaches the model, statistical method, heterogeneity assessment, software, between-study variance estimator where relevant, confidence-interval method for random-effects models, approaches to statistical dependency, and reasons why planned syntheses were not considered possible or appropriate.² Treating this as one binary variable necessarily compresses a large amount of reporting information.

“Complete” is an operational judgment, not an official PRISMA grade

The official PRISMA documents tell authors what information should be reported. They do not provide a universal psychometric rule for translating the manuscript into categories such as complete, partial or missing. That translation is performed by whoever is auditing the report. This difference is easy to overlook because a reporting checklist naturally looks like an assessment instrument.

For an author preparing a manuscript, a useful operational definition of complete reporting is that all applicable core information represented by the item can be explicitly located. That is a working rule, not an official PRISMA definition. It has two advantages. First, it forces the auditor to identify actual evidence rather than infer completeness from a familiar phrase. Second, it separates the reporting judgment from an opinion about whether the method itself was appropriate.

Partial reporting becomes most meaningful for multi-part items. If a review names the risk-of-bias tool but does not report the assessors, independence or judgment process, the reporting is not entirely absent, yet a binary yes/no response conceals the missing information. A partial category can therefore be useful. Its weakness is reliability. Unless the audit rules specify what counts as material missing content, different reviewers may use “partial” in different ways. Hamilton et al. found that response options and operational guidance varied considerably among existing PRISMA 2020 assessment tools.³

This is why a practical audit should avoid arbitrary thresholds such as “more than half the elements = complete.” Such thresholds are easy to calculate but difficult to defend. An omitted element may be more consequential for transparency than several minor pieces of additional detail. A location-based audit is therefore better understood as a structured review of reporting coverage than a mathematical grade.

The denominator problem: why one adherence percentage may not equal another

A percentage requires a numerator and a denominator. In PRISMA adherence research, both can change. One study may treat each of the 27 headline items as one unit. Another may separate lettered sub-items. Another may assess many expanded-checklist elements. Some studies award partial credit. Some exclude non-applicable content. Others include it in the denominator. Some assess only selected domains. These choices alter the resulting percentage before any manuscript has changed.

Ivaldi et al. provide a useful example. The investigators assessed 222 systematic reviews of interventions published from 2022 onward and identified 67 that reported using PRISMA 2020. Using the expanded checklist, they found no completely adherent review in that subgroup and an average adherence of 42.64%.⁴ The finding is informative about those reviews under the investigators’ expanded-checklist assessment method. It should not be transformed into a claim that “systematic reviews are only 42.64% PRISMA compliant” in general.

Their study also illustrates why deeper assessment can produce lower apparent adherence. A broad item marked yes can contain several unreported elements when opened. That does not prove that one particular element-level scoring algorithm is the uniquely correct measurement system. It shows that the granularity of the assessment changes what becomes visible.

Not applicable is a methodological decision, not an empty cell

Conditional reporting is one of the easiest places to distort an adherence audit. The official expanded checklist handles conditionality through wording such as “If…” rather than through a universal N/A scoring system.² Whether an element applies therefore depends on the design and methods of the specific review.

Some distinctions are straightforward. If sensitivity analyses were conducted, Item 13f asks for the methods and Item 20d asks for the results. If no sensitivity analysis existed, those reporting components have no corresponding analysis to describe. Likewise, reporting of investigations of heterogeneity is conditional on those investigations having been carried out.

Other items demonstrate why “we did not do it” and “not applicable” are not the same idea. Item 24a does not disappear when a review was unregistered. It explicitly allows the authors to state that the review was not registered. Item 24b similarly permits authors to state that a protocol was not prepared.² In these cases, absence of the underlying research object is itself reportable information.

This matters because an audit denominator can be manipulated unintentionally by removing difficult rows. If N/A is treated as a convenient escape for an unperformed activity, the audit can look more complete without the manuscript becoming more transparent. A robust audit therefore requires the reviewer to answer two separate questions: Does this reporting request apply to the review? and If it applies, is the relevant information present?

Where an item is reported is part of the audit

The conventional PRISMA checklist includes a column asking authors to identify the location where an item is reported. That seemingly administrative field contains an important idea: an adherence claim should be traceable. A reviewer should not have to search the entire manuscript to decide whether the authors’ claim is plausible.

A location may be a section of the manuscript, a table, a figure or a supplementary file. Some reporting requests extend naturally to external research objects. Registration information points to a registry. Protocol accessibility may be represented by a DOI, citation or link. Item 27 asks authors to report which data, code and other materials are publicly available and where they can be found.²

Search reporting provides an especially clear example. PRISMA 2020 Item 7 asks authors to present full search strategies, while PRISMA-S adds more detailed reporting guidance for literature searches.⁶ A lengthy multi-database search strategy may appropriately live in supplementary material. The audit question is therefore not “Is every search line in the Methods section?” It is “Can the reader identify the executed search strategy and find it from the report?”

This leads to a useful distinction between an evidence location and a cross-reference location. The evidence location is where the detailed information resides. The cross-reference is where the manuscript tells the reader how to get there. A repository may contain a perfect analytic dataset, but if the availability statement points to the wrong repository or does not identify it, the reporting problem remains.

Page numbers are useful for journal submission, but they are poor long-term identifiers because a revised Word file, generated submission PDF and final typeset article can each paginate differently. Section names, table numbers, figure identifiers, supplement identifiers and persistent external URLs can provide a more durable semantic location. This is an implementation recommendation rather than an official PRISMA rule.

Declared adherence is not the same as verified reporting

Authors commonly write that a review was conducted or reported “according to PRISMA.” That statement communicates intention, not proof that every applicable item has been fully reported. Meta-research repeatedly finds incomplete reporting among reviews that cite or declare use of PRISMA. Ivaldi et al.’s sample is especially relevant because the researchers specifically identified reviews that reported using PRISMA 2020 and then assessed them with the expanded checklist.⁴

This should not be interpreted as evidence of dishonesty. Reporting guidelines are complex, manuscript space is limited, journal processes vary, and authors may sincerely believe an item is complete when an assessor using a more granular rule disagrees. The methodological lesson is more restrained: a declaration of adherence and an independently verified item-level audit are different types of evidence.

The same caution applies to journal endorsement. Wang, Wolfram and Gilbert examined instructions for authors across 559 biomedical journals from major publishers and found inconsistent endorsement and implementation practices. Their study also distinguishes endorsement, author adherence and journal implementation as different concepts.⁷ A journal can recommend PRISMA without actively checking every item. A submitted checklist can be treated as an administrative file rather than an audited representation of the manuscript.

Agha et al.’s before-and-after study provides evidence that implementation can matter. After one surgical journal mandated submission of completed reporting-guideline checklists, measured PRISMA compliance increased from 48% to 76% in the study period.⁵ The result is useful but context-specific. It does not establish that a mandatory checklist will produce the same effect in every journal, nor does it show that an editor verified every item. It supports the narrower conclusion that reporting-guideline implementation can influence reporting behavior.

Why PRISMA adherence cannot stand in for methodological quality

A reporting guideline makes methodological decisions visible. It does not determine whether those decisions were defensible. A review can clearly report a flawed eligibility rule, an inadequate risk-of-bias process or an inappropriate statistical model. From a reporting perspective, transparency may be high. From a methodological perspective, the review may still warrant serious criticism.

The reverse problem is also possible. A review team may have conducted a careful method but describe it inadequately. Readers then cannot fully assess what happened. Reporting and methodological quality can therefore be associated empirically without representing the same construct.

This distinction is particularly important when researchers create overall “PRISMA scores.” A high aggregate reporting value can look like a quality grade even when the scoring tool was never designed to measure methodological validity. Hamilton et al.’s findings show how much variation already exists in tools developed specifically to assess PRISMA 2020 adherence.³ The safest interpretation is item-specific: what reporting information is available, what is absent, and what cannot yet be verified?

PRISMA primarily helps answer the first question. Other methodological standards and appraisal instruments are needed for the second. This guide therefore does not teach AMSTAR 2, ROBIS, GRADE or statistical appraisal. Doing so here would blur both the user’s task and the Academy’s evidence architecture.

Extensions change what needs to be reported, but not in one uniform way

PRISMA 2020 is a central reporting framework, but systematic reviews differ in purpose and methods. Official extensions address particular review types and reporting domains. The relationship between the core statement and an extension should be checked from the relevant extension rather than assumed.

PRISMA-S is a useful example because it extends the detail of search reporting. PRISMA-LSR is explicitly an add-on for living systematic reviews. Other extensions such as PRISMA-ScR and PRISMA-NMA were developed for distinct review designs and predate the 2020 core update. Reporting-guideline updates and new extensions continue to develop, so authors should check the current official PRISMA extension register rather than maintain a permanent local list.

PRISMA-P deserves a separate clarification. It is guidance for systematic review and meta-analysis protocols. It is not a second checklist that should simply be appended to every completed systematic-review manuscript. Its value at the publication stage is different: it can help authors compare what they prespecified with what they ultimately report, especially when amendments must be explained.

Similarly, the detailed audit of the abstract belongs to PRISMA 2020 for Abstracts rather than being forced into the same full-report tracker. The abstract and full report must be consistent, but their reporting checklists solve different practical problems.

Automated adherence checking is promising, but the judgment problem remains

Large language models and other automated systems can search manuscripts for reporting information, but locating plausible text is not the same as making a valid adherence judgment. This distinction becomes especially important for items whose evidence is distributed across the manuscript and supplementary material, or where applicability has to be inferred from the review design.

Emerging benchmarking work suggests that structured checklist information can improve automated assessment performance compared with asking a model to infer every requirement from a manuscript alone. At the same time, reported specificity can remain inadequate in some configurations, which means a tool may incorrectly label missing reporting as present. These systems are therefore best treated as experimental triage or retrieval aids rather than substitutes for final human verification.

For a researcher completing a manuscript today, the practical implication is simple: structure the audit so that an automation tool could understand it in the future, but do not depend on automation to decide whether the report is complete. Explicit item identifiers, clear locations, defined statuses and concise audit notes create useful structured information regardless of whether a human or a machine later reads the record.

What a defensible PRISMA audit should preserve

The most rigorous audit is not necessarily the most complicated one. A system with hundreds of variables can become burdensome enough that authors abandon it. A system with twenty-seven green ticks can become too coarse to reveal important omissions. The useful middle ground is a transparent decision record.

For each item, the auditor should know which reporting content is being checked, whether conditionality matters, where the evidence appears, what remains incomplete, and whether the judgment has been independently verified when verification is needed. The audit should also identify the manuscript version because a checklist completed against one draft can become inaccurate after revision.

This is why the paired PRISMA 2020 checklist and compliance tracker separates reporting status from workflow status. “Partial” describes the reporting record. “Revision required” describes the next action. “Complete” and “Needs verification” can coexist conceptually because completeness is the first auditor’s reporting judgment while verification is a workflow state.

The tracker also avoids an overall percentage. Instead, it surfaces unresolved reporting gaps. That design is deliberately less impressive-looking than a score, but more useful. A single missing registration-amendment explanation, excluded-study list or synthesis-method detail can matter even when dozens of other rows are complete.

Use PRISMA as a reporting map, not a seal

The strongest interpretation of PRISMA 2020 is not that it certifies a review. Its value is that it gives authors, editors, reviewers and readers a common vocabulary for asking whether important features of a systematic review have been reported transparently.

That purpose is undermined when adherence is reduced to an unexplained percentage or when a checklist is completed by searching for keywords after the manuscript is finished. It is strengthened when the checklist becomes an audit trail: every applicable reporting request is connected to evidence, conditional decisions are explicit, omissions trigger revisions, and the final checklist refers to the version actually submitted.

The result is not proof that the review is methodologically strong. It is something more precise and useful at the reporting stage: another person can see what the authors say they did, where that information is located, and where transparency remains incomplete.

Move from interpretation to reporting work

Frequently asked questions