Full-Text Exclusion Reasons Log for Systematic Reviews

The flow diagram is a summary, not the source record

Full-text selection fundamentally alters the evidence base. Every exclusion must therefore be recoverable as a methodological decision: which report was assessed, which versioned criterion failed, what information supported that judgment, who resolved it, and when the outcome became final. PRISMA 2020 requires authors to report reasons for exclusions and to cite borderline studies that might appear eligible but were ultimately excluded. It does not, however, supply a universal taxonomy for every review question.^1,2

Weak exclusion logs damage a review in three ways. First, vague labels conceal whether you applied the protocol consistently. Second, counting every failed criterion mathematically inflates the number of excluded reports and breaks the flow diagram. Third, treating retrieval failures or unresolved records as eligibility exclusions invents a judgment you never actually made. A defensible system separates evidence, decision, reporting category, and workflow status while firmly preserving their relationships.^3,4

Design the reason system before you need it

Build your taxonomy directly from the operational eligibility criteria in your approved protocol. For each criterion, define the reportable label, the observable text evidence required, boundaries with adjacent reasons, and the protocol action for missing information. Generic labels like “wrong study” or “not relevant” conceal the scientific boundary and cannot be applied reproducibly. A robust reason is concise enough to aggregate visually but specific enough to point back directly to a scientific rule.

Because a single report can fail several criteria, the interactive log below assigns one reportable primary reason while retaining all other failed criteria as secondary detail. This constitutes a transparent design choice rather than a PRISMA mandate. The review team should pre-specify a priority order, pilot multi-failure cases, and document any subsequent changes to the hierarchy. A different prioritization system is entirely acceptable, provided it preserves the original decisions and clearly explains how category totals map back to the true number of excluded reports.

Build a hierarchy derived from the protocol

The hierarchy should directly match the scientific question. A population-first order makes sense when the core population strictly defines the review boundary; conversely, a design-first order proves more efficient when the vast majority of retrieved reports fail a clearly operationalized design rule. The hierarchy influences the reported distribution whenever reports fail multiple criteria, meaning you must not select the order retroactively after viewing the desired totals. Pilot the order on difficult cases, compare reviewer reasoning, and lock the agreed version into the screening documentation.

Categories must never silently combine different decisions. “Outcome” might mean the study did not measure an eligible outcome, the report simply omitted the outcome data, or the data cannot be extracted reliably. Those conditions are not methodologically equivalent. Eligibility follows the protocol; extractability and synthesis limitations belong in downstream data management stages unless the protocol explicitly designated them as selection conditions. Similarly, risk of bias is generally assessed after inclusion and should never be retrofitted as a full-text exclusion reason.

Pilot the log against real boundary cases

Before launching the main full-text assessment, instruct reviewers to apply the draft reason system independently to a highly varied sample: clear inclusions, single-reason exclusions, reports failing several criteria simultaneously, companion publications, incomplete reports, and cases sitting precisely on the eligibility boundary. Compare the selected primary reason, secondary reasons, text evidence notes, and criterion clauses. Disagreements reveal missing definitions or priority rules operating improperly. Revise and date the taxonomy before scale-up.^4,7

High agreement statistics alone are insufficient. Reviewers may agree on a vague label while applying entirely different reasoning, or disagree because the report genuinely lacks crucial information. Calibration must examine the cited evidence and underlying interpretation, not just the final exclusion code. If criterion wording changes, determine whether already assessed reports require re-evaluation and keep the old decisions preserved. A frozen dropdown menu without criterion versioning creates superficial consistency while concealing structural methodological changes.

Interpret every field as part of an audit trail

The report identifier anchors the decision directly to the document assessed; the study identifier connects companion publications without erasing their individual histories. Outcome or workflow status clearly states what occurred. The primary reason supplies the mutually exclusive reporting category only when the status is firmly “excluded.” The criterion clause identifies the exact governing rule, while the evidence note records the text, table, registry entry, or author clarification dictating the judgment.

Reviewer and date fields describe the final accountable outcome, but an institutional screening platform should additionally preserve all independent first decisions and conflict resolution dialogues. A free-text note is not a substitute for a controlled reason category, and a controlled category is not a substitute for cited evidence. Together, they allow the decision to be aggregated visually, interpreted accurately, and re-evaluated seamlessly during peer review.

Keep eligibility outcomes distinct from operational status

These states shift over time. Append a later decision when a report is obtained, a study completes, or author clarification resolves uncertainty; do not overwrite the earlier event. Event history is absolutely essential in living and updated reviews because it explains exactly why successive flow diagrams differ. It also prevents a current status from being mistaken for the only state the record ever held.

Reconcile from identifiers before transferring totals

First, confirm that every report assessed for eligibility holds exactly one terminal state for the reporting period. Next, confirm that every excluded report possesses one primary reason and that the sum of primary reasons perfectly equals the number of excluded reports. Keep non-retrieval, ongoing, and awaiting-classification records strictly outside that sum. Finally, connect included reports to study identifiers so the number of included reports and included studies can be reported separately. This order respects the underlying units and rapidly exposes missing or duplicate rows.^1,2

Do not repair a discrepancy by editing the aggregate total. Locate the affected report record, determine whether its status, reason, or linkage is flawed, and correct the source data with a documented explanation. If the taxonomy changed, preserve the old code and document its mapping to the new reporting category. Software validation detects structural contradictions, but human reviewers must determine whether the underlying criterion and evidential interpretation remain scientifically defensible. Diagram software improves presentation and basic arithmetic, but it cannot repair invalid source classifications.⁸

Prepare the reportable borderline list

PRISMA 2020 item 16b asks authors to cite studies that might appear to meet inclusion criteria but were ultimately excluded, and to explain why. The complete internal log can and should be larger. Select the candidate examples that readers are most likely to question, provide citations, and deliver concise, criterion-linked reasons.

Item 16b is not a demand to publish the entire internal log. Select studies whose apparent eligibility makes their absence likely to concern an informed reader, cite them, and state the decisive criterion-linked reason. Preserve the full record for audit even when journal space limits require a shorter publication table. The explanation should remain factual and specific; “wrong design” is significantly weaker than identifying the specific allocation method that failed the protocol’s eligible-design definition.

Preserve versions for amendments and review updates

If an eligibility rule changes during screening, record the former and revised wording, approval authority, effective date, and whether earlier reports underwent reassessment. Each individual decision must retain the criterion version applied to it. Overwriting the reason or rule makes it impossible to tell whether a later count shifted because of new evidence, conflict resolution, or a major methodological amendment. In an update, retain prior identifiers and append new events so the old and new evidence bases can be compared dynamically without reverse-engineering history from published totals.

Scope and Technical Limitations

The default categories provided here are starting points. Adapt them aggressively to your specific protocol before screening begins. Browser storage is convenient but is not a controlled multi-user audit system, and this resource does not algorithmically decide which exclusions merit citation under PRISMA item 16b. Do not enter confidential personal data.

Browser storage may be cleared inadvertently, is not designed for simultaneous multi-user work, and does not provide institutional access control or immutable audit history. Export records regularly and migrate them to your review’s approved controlled system. The default categories are prompts, not strict eligibility criteria; specialist review designs may require different taxonomies and specific PRISMA extensions.^5,6

Conclusion

A full-text reason becomes scientifically defensible when it remains permanently linked to the report, the study, the criterion, the text evidence, the reviewer, and the decision history. Utilizing a single reportable primary reason ensures flow totals remain mutually exclusive; tracking secondary criteria preserves the richer methodological record. Strongly separating scientific exclusion from non-retrieval and unresolved statuses prevents the review from claiming analytical judgments it did not actually make. Use the exported log as raw source data for reconciliation, not as a shortcut to bypass your approved protocol or accountable reviewer assessment.