Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for the opportunity to review the manuscript titled ‘Mallet-Like Distal Phalanx Base Injuries in Skeletally Immature Patients: A PRISMA-ScR Scoping Review and Phenotype-Based Decision Framework.’

The title accurately signals the scope and methodology, but the term ‘Decision Framework’ risks overstating what is delivered. The authors describe it as ‘author-derived’ and ‘unvalidated’ (Figure 2 legend, Discussion 4.3). A subtitle qualifier, e.g., ‘proposed, unvalidated framework,’ or deleting it (‘and Phenotype-Based Decision Framework’) would better calibrate reader expectations.

The abstract is adequately structured.

The introduction is reasonably framed, distinguishing eponym from mechanism. However, the rationale for a new framework versus incremental extension of Kiely et al. (2022), already a systematic review/meta-analysis on this exact topic, is underdeveloped. The added value over existing syntheses needs a sharper, explicit statement in the final paragraph rather than being left implicit.

I have major concerns about the manuscript's methodology. Regarding database coverage, only PubMed and OpenAlex were searched; Embase, Scopus, and Web of Science were excluded for lack of institutional access. This limitation is disclosed, but for a scoping review claiming to ‘map the diagnostic and management evidence,’ single-database-plus-aggregator coverage is a significant threat to completeness that should have been mitigated before submission, not acknowledged after the fact. Regarding charting design, one reviewer charted, and the second ‘verified every 'field'—this is not independent dual extraction, and inter-rater metrics (agreement, kappa) are absent. The claim of methodological rigor is therefore weaker than presented. The authors state that ‘reviewer-level votes, timestamps, and disagreement counts were neither retained nor reconstructed' and that no pilot log was kept. For a PRISMA-ScR–compliant review, this reproducibility gap should be flagged prominently, not folded into a supplement footnote. Also, OpenAI Codex/GPT-assisted tools performed ‘metadata normalization, duplicate detection, language editing, and document formatting.’ The precise boundary between AI-assisted deduplication and human deduplication decisions is not fully transparent. Did the AI propose duplicate matches that were then accepted, or merely reformat records? This area needs explicit methodological detail, given that deduplication errors directly affect the reported n=161. The OpenAlex full-text search for ‘Seymour fracture’ (114 hits, described as ‘intentionally sensitive’) introduces substantial screening burden and potential inconsistency versus the more specific PubMed strings; the trade-off is asserted rather than justified quantitatively.

Tables 2 and 3 are useful, but several ‘signals’ (e.g., associations with antibiotics within 24 hours) are presented using retrospective, confounded, non-randomized data. The effect sizes (RR 0.21–0.30) are then translated into an action-threshold table (Table 3) using directive clinical language (‘prompt specialist assessment,’ ‘antibiotics under the local pathway’). The gap between an observational association and a prescriptive threshold is not adequately buffered by hedging language at the point of translation, even though it is acknowledged elsewhere. This is the manuscript's central internal tension and should be resolved by either softening Table 3's imperative phrasing or adding explicit strength-of-evidence tags per row.

In the discussion, the ‘Clinical concordance’ section (4.2) usefully triangulates with BSSH/RCH guidance, but concordance with existing non-systematic guidance is not independent validation and risks circularity; much of that external guidance likely derives from the same primary literature. This point should be stated explicitly.

The conclusions are appropriately cautious in tone (‘structures judgment; it does not replace... assessment’), but this caution is somewhat undercut by the confident, algorithmic tone of Figure 2 and Table 3, which most readers and clinicians will actually use. Internal consistency between the humble conclusion and the assertive figures/tables needs tightening.

References are extensive (60 primary and guideline documents) and relevant; reference 5 (Ross et al. 2026) appears to substantially overlap in scope with the present review, and its unique contribution relative to that source could be clarified.

Overall, the manuscript is a well-organized, PRISMA-ScR–compliant scoping review addressing a genuine clinical gap, but incomplete database coverage, non-independent charting, a lost audit trail, and a disconnect between cautious prose and directive tables/figures are substantive issues that should be addressed before further consideration.

Comments on the Quality of English Language

The English is fine and does not require any improvement at this stage. Any residual edits are cosmetic/typographical rather than linguistic.

Author Response

Author's Reply to the Review Report (Reviewer 1)

Manuscript ID: children-4487609  |  Journal: Children

We thank the reviewer for the careful and constructive assessment. Each comment is addressed point by point below. Revised wording is shown in red, and stable section/table/figure locations are provided because line numbering is not preserved reliably across Word renderers.

Comment 1

Qualify the title/framework as proposed and unvalidated, or remove ‘Decision Framework’; make the added value over Kiely et al. (2022) explicit.

Response 1

Accepted with evidence-calibrated wording. We removed ‘Decision Framework’, made the proposed/unvalidated status visible, and changed the population label to ‘Children and Adolescents’, which matches the mapped corpus more accurately than a universal claim of skeletal immaturity. No retrospective corpus restriction was performed. The Introduction also distinguishes the mapping contribution from Kiely et al.’s meta-analysis.

Change made in the revised manuscript: Mallet-Like Distal Phalanx Base Injuries in Children and Adolescents: A PRISMA-ScR Scoping Review and a Proposed, Unvalidated Phenotype-Based Assessment Framework

Location in the revised manuscript: Revised manuscript p. 1 (title) and pp. 1–2 (Introduction); Methods §2.2, p. 2; Limitations §4.4, p. 10; Supplement S2 §S2.8.

Comment 2

Methodological completeness: explain the PubMed + OpenAlex-only coverage and the absence of Embase, Scopus and Web of Science; mitigate where possible and state residual retrieval bias.

Response 2

Accepted with evidence-calibrated wording. We explain the complementary but incomplete PubMed/OpenAlex coverage and explicitly retain residual retrieval bias. Institutional-access checks on 25 August 2026 did not provide searchable document access to Embase, Scopus, or Web of Science Core Collection; no search string, result count, or export was therefore produced from those databases. The manuscript does not present OpenAlex as equivalent to them.

Change made in the revised manuscript: Coverage limitation retained; no undocumented access claim or database search was created.

Location in the revised manuscript: Revised manuscript Methods §2.3, p. 2; Limitations §4.4, p. 10; Supplement S1 §S1.9, p. 3.

Comment 3

Data charting was performed by one reviewer and verified by another, not independent dual extraction. State this precisely; address absent inter-rater metrics, pilot log and reviewer-level audit trail rather than implying stronger rigor.

Response 3

Accepted. We now distinguish independent source selection from data charting: two reviewers screened records and reports with consensus resolution, whereas one reviewer charted data and a second verified every displayed field and citation. No pilot log or agreement statistic was retained, and none was reconstructed. These limitations are explicit.

Change made in the revised manuscript: Reconciled screening outcomes and human roles are documented; no reviewer-level vote log, timestamp series, or formal agreement statistic was reconstructed.

Location in the revised manuscript: Revised manuscript Abstract, p. 1; Methods §§2.4–2.5, p. 3; Limitations §4.4, p. 10; Supplement S1 §§S1.8 and S1.12, pp. 2–4.

Comment 4

Explain what OpenAI Codex/GPT-assisted tools actually did in metadata normalization, duplicate detection, language editing and formatting; distinguish machine suggestions from human eligibility/deduplication decisions and explain the relevance to the n=161 denominator.

Response 4

Accepted. Deterministic matching and AI-assisted duplicate-candidate identification are separated from final human decisions. OpenAI Codex and GPT-assisted tools supported metadata normalization, duplicate-candidate identification, language editing, and formatting; they did not make final duplicate, eligibility, extraction, interpretive, or clinical decisions.

Change made in the revised manuscript: Methods, Supplement S1, and the formal disclosure now use one consistent description of tool functions and final human responsibility.

Location in the revised manuscript: Revised manuscript Methods §§2.4 and 2.7, p. 3; AI disclosure, p. 11; Supplement S1 §S1.8, pp. 2–3.

Comment 5

Justify the sensitive OpenAlex full-text search for ‘Seymour fracture’ (114 hits) and its screening trade-off.

Response 5

Accepted. We identify the OpenAlex field semantics, preserve the exact July API logic, and explain the sensitivity/specificity trade-off. The unchanged 20 August rerun returned three new records; FL and NL independently excluded all three, consensus was recorded on 25 August 2026, and the historical 60-source flow is unchanged.

Change made in the revised manuscript: “An OpenAlex Works API query using fulltext.search…can match indexed titles, abstracts, and full-text content where available and is not a full-text-only search.”

Location in the revised manuscript: Revised manuscript Methods §2.3, p. 2; Results §3.1 and Figure 1, p. 4; Supplement S1 §§S1.4 and S1.10–S1.11, pp. 1 and 3–4.

Comment 6

Tables 2–3 translate retrospective/confounded associations, including antibiotic timing estimates, into directive clinical ‘action thresholds’. Soften imperative wording and/or add explicit strength-of-evidence tags; do not imply validated causal or safe-delay thresholds.

Response 6

Accepted with evidence-calibrated wording. “Action thresholds” was removed. Table 3 is a matrix of coexisting dimensions with directness tags. The visible framework is proposed/unvalidated, contains no rule-out branch or safe-delay cutoff, and explicitly states that immobilization duration was not comprehensively synthesized.

Change made in the revised manuscript: “Proposed evidence-informed responses by coexisting clinical dimension (unvalidated)” with directness and limitation in every row.

Location in the revised manuscript: Revised manuscript Results §§3.4–3.7, pp. 5–6; Figure 2, p. 7; Tables 2–3, pp. 7–8; Conclusion, p. 10.

Comment 7

Say explicitly that concordance with BSSH/RCH guidance is not independent validation and may be circular because guidance may draw on the same primary literature.

Response 7

Accepted. We explicitly state likely non-independence/circularity and that guidelines provide no performance estimate for the framework.

Change made in the revised manuscript: “Their concordance is not independent validation: guidance may draw on the same primary literature and provides no sensitivity, specificity, safety, or calibration estimate.”

Location in the revised manuscript: Revised manuscript Discussion §4.2, p. 9; Supplement S2 §§S2.4 and S2.7.

Comment 8

Reconcile the cautious prose/conclusions with the assertive algorithmic tone of Figure 2 and Tables 2–3; ensure the most visible clinical elements carry the same uncertainty language.

Response 8

Accepted. The visible elements now carry the uncertainty directly. Figure 2 has a prominent “PROPOSED—UNVALIDATED” banner, three communication states, no binary NO-as-safety branch, and no treatment thresholds. Table 3 uses conditional language.

Change made in the revised manuscript: Suspected, uncertain, and no-communication-identified states; “may be considered,” “supports evaluation,” and evidence tags.

Location in the revised manuscript: Revised manuscript Abstract, p. 1; Figure 2, p. 7; Table 3, pp. 7–8; Conclusion, p. 10.

Comment 9

Clarify the unique contribution relative to Ross et al. (2026), which may overlap in scope.

Response 9

Accepted. The Introduction now explicitly contrasts Ross’s narrative clinical review with the present cross-phenotype scoping map and author-derived unvalidated scaffold.

Change made in the revised manuscript: “Ross et al. provided a recent clinical narrative review…The present scoping review has a different purpose: it maps both open and closed…phenotypes across five interacting dimensions…”

Location in the revised manuscript: Revised manuscript Introduction, paragraph 3, pp. 1–2.

Reviewer 2 Report

Comments and Suggestions for Authors

This is an interesting and clinically relevant scoping review addressing the diagnostic overlap between pediatric mallet-like injuries and Seymour/physeal distal phalanx fractures. The manuscript is generally well written, the clinical message is useful, and the authors appropriately acknowledge the predominantly retrospective and heterogeneous nature of the available evidence. The phenotype-based framework is potentially valuable for clinical practice, particularly because it emphasizes nail-unit integrity, fracture morphology, joint congruity and stability rather than relying exclusively on diagnostic labels. I believe the manuscript is suitable for publication after minor revision.

Specific comments:

1. Lines 41–46: The sentence “An open physis denotes skeletal maturity” appears incorrect. An open physis indicates skeletal immaturity or remaining growth potential. Please correct this sentence, as the distinction is particularly important in the context of this manuscript.
2. Lines 52–57: Please clarify how “otherwise skeletally immature” was operationally defined for study eligibility, particularly for patients older than 18 years or mixed-age cohorts. It would be helpful to specify whether skeletal immaturity had to be explicitly reported by the original authors or could be inferred from radiographs/clinical information.
3. Lines 60–63: The authors state that eligibility criteria, search logic and synthesis axes were fixed before the final reconciled screening. Please indicate whether a formal review protocol was prospectively registered or publicly available. If no registration was performed, this should simply be stated.
4. Lines 113–120 and 138–145: Since no formal critical appraisal or certainty grading was performed, the subsection title “Evidence architecture and certainty” may be slightly misleading. Consider changing this to “Evidence architecture and limitations” or “Evidence architecture and strength of evidence.”
5. Lines 149–156: The recommendation that nail-fold abnormalities “should be treated as possible communication” is clinically reasonable, but the manuscript simultaneously acknowledges the absence of diagnostic-accuracy studies. Please slightly temper the wording to make it explicit that this is an evidence-informed precautionary approach rather than a validated diagnostic rule.
6. Lines 158–167: The distinction between observational timing thresholds (<24 h for antibiotics and <48 h for debridement) and true biological/safe-delay thresholds is excellent and should be retained. I suggest emphasizing this point also in Table 2, as these numbers could otherwise easily be misinterpreted by readers as recommended maximum treatment delays.
7. Lines 174–180: The discussion of the NINJA trial is appropriate, but its indirect applicability to Seymour fractures could be made even clearer. Please explicitly state that the trial population differs from Seymour fractures in terms of fracture involvement and potential infection/instability, and therefore cannot directly determine nail-plate management in this specific injury pattern.
8. Lines 188–205: The authors recommend continuous DIP-extension splinting for convincingly closed and stable mallet injuries. For greater clinical applicability, please consider adding the usual duration of splinting reported in the included pediatric literature, if sufficiently consistent. If duration was too heterogeneous to support a recommendation, this should be explicitly stated.
9. Lines 206–220: Similarly, in delayed/infected presentations, it may be useful to clarify that the available literature does not support a universal antibiotic agent or treatment duration. This would complement the authors’ appropriate caution regarding the retrospective clindamycin versus cephalexin comparison.
10. Figure 2: The proposed phenotype-based framework is one of the major strengths of the manuscript. However, please improve the readability of the figure, particularly the font size within the decision boxes and the YES/NO pathways. The figure should remain easily readable when displayed at standard journal page width.
11. Table 2, lines 225 onward: Only a limited number of “high-yield” studies from the 60 included sources are presented. Please briefly explain in the text or table legend how these particular studies were selected for inclusion in Table 2. This would make the selection process more transparent.
12. Table 3, line 226: The term “action thresholds” may suggest that validated thresholds have been identified, whereas the table itself correctly states that these are not validated rules. Consider renaming the table “Evidence-informed management considerations” or “Proposed phenotype-based management framework.”
13. Lines 243–250: The research priorities section is useful. I suggest adding the need for prospective evaluation of interobserver agreement in identifying nail-unit communication and radiographic morphology, since reproducibility of phenotype assignment will be essential before the proposed framework can be clinically validated.
14. Lines 251–260: The limitations section is appropriate, but it would be useful to explicitly mention the absence of formal critical appraisal as an additional limitation when interpreting the proposed clinical framework, even if such appraisal is not mandatory for a scoping review.
15. Lines 262–269: The conclusions are clear but somewhat directive considering the low level of evidence and lack of prospective validation. Consider slightly moderating phrases such as “should enter” or “warrants” by reiterating that these recommendations represent an evidence-informed framework and not a validated decision rule.
16. Lines 287–291: Please ensure consistency regarding AI-assisted tools. The Methods mention both “OpenAI Codex and GPT-assisted tools,” whereas the formal AI disclosure mentions only OpenAI Codex. The tools and their roles should be reported consistently in both sections.

Overall, this is a thoughtful and useful review that addresses an important diagnostic and therapeutic problem in pediatric hand trauma. The proposed phenotype-based approach is clinically intuitive and the authors appropriately highlight the limitations of the existing evidence. The comments above mainly concern clarification, terminology, methodological transparency and moderation of a few clinical statements. I therefore recommend Minor Revision.

Author Response

Author's Reply to the Review Report (Reviewer 2)

Manuscript ID: children-4487609  |  Journal: Children

We thank the reviewer for the careful and constructive assessment. Each comment is addressed point by point below. Revised wording is shown in red, and stable section/table/figure locations are provided because line numbering is not preserved reliably across Word renderers.

Comment 1

Correct ‘An open physis denotes skeletal maturity’ to skeletal immaturity/remaining growth potential.

Response 1

Accepted. The factual inversion was corrected globally and the revised Table 1 repeats the distinction from an open fracture.

Change made in the revised manuscript: “An open physis denotes skeletal immaturity and remaining growth potential; it is not synonymous with an open fracture.”

Location in the revised manuscript: Revised manuscript Introduction, paragraph 2, p. 2; Table 1, p. 5.

Comment 2

Define ‘otherwise skeletally immature’, including mixed-age cohorts and whether immaturity must be reported or may be inferred.

Response 2

Accepted. Age, cohort description, direct physeal evidence, and the role of mixed-age evidence are now distinguished. Because skeletal immaturity is not reported across the entire mapped corpus, the title now uses ‘Children and Adolescents’. Two legacy mixed-age sources with inseparable age strata are identified as contextual only and are not used for pediatric-specific quantitative inference. No source was silently removed, and physeal status is not inferred from age alone.

Change made in the revised manuscript: Population/maturity categories are explicit; physeal status is not inferred from age alone, the title uses ‘Children and Adolescents’, and source-level classifications and the contextual-only role of two mixed-age sources are retained in Supplement S2.

Location in the revised manuscript: Revised manuscript title, p. 1; Methods §2.2, p. 2; Table 1, p. 5; Supplement S2 §S2.8.

Comment 3

State whether a protocol was prospectively registered/publicly available; if not, state that plainly.

Response 3

Accepted. The revised Methods now state plainly that no prospective protocol registration or publicly accessible protocol was documented for this review. The absence of prospective registration is also acknowledged in the Limitations and checklist.

Change made in the revised manuscript: No prospective protocol registration or publicly accessible protocol was documented; this is stated in Methods and Limitations without reconstructing a retrospective chronology.

Location in the revised manuscript: Revised manuscript Methods §2.1, p. 2; Limitations §4.4, p. 10; PRISMA-ScR checklist item 5.

Comment 4

Rename ‘Evidence architecture and certainty’ to a label consistent with the absence of formal appraisal/certainty grading.

Response 4

Accepted. The heading no longer implies a certainty assessment.

Change made in the revised manuscript: “Characteristics and structure of the mapped evidence”

Location in the revised manuscript: Revised manuscript Results §3.2, p. 5.

Comment 5

Temper nail-fold abnormality language as an evidence-informed precaution, not a validated diagnostic rule.

Response 5

Accepted with evidence-calibrated wording. We preserve a safety-oriented urgent-evaluation route while explicitly stating that individual signs are not validated diagnostic tests.

Change made in the revised manuscript: Signs “raise concern for possible communication” and “support urgent specialist evaluation”; diagnostic accuracy is unknown.

Location in the revised manuscript: Revised manuscript Results §3.3 and Table 1, p. 5; Figure 2, p. 7; Table 3, pp. 7–8.

Comment 6

Repeat in Table 2 that <24 h antibiotic and <48 h debridement observations are not biological/safe-delay limits.

Response 6

Clarified and retained. This qualification was already present in the submitted prose and Table 2. We retained it, made the non-causal language more prominent, and propagated it to every visible element rather than implying a new result.

Change made in the revised manuscript: The same qualification remains, now also stating “not safe-delay windows, biological cutoffs, or universal regimens.”

Location in the revised manuscript: Revised manuscript Results §3.4, p. 5; Figure 2, p. 7; Table 2, p. 8.

Comment 7

Clarify why the NINJA trial cannot directly determine Seymour fracture nail-plate management.

Response 7

Clarified and retained. NINJA is now explicitly labelled “external contextual randomized evidence,” outside the 60-source denominator, and indirect for Seymour injury because infected injuries and fractures requiring fixation were excluded. S2 topology and the claim ledger carry the same designation.

Change made in the revised manuscript: External contextual randomized evidence; outside the 60-source map; indirect for Seymour injuries.

Location in the revised manuscript: Revised manuscript Results §3.5, p. 6; Table 2, p. 8; Supplement S2 §S2.7.

Comment 8

Add splint duration if sufficiently consistent; otherwise state heterogeneity.

Response 8

Accepted. Because source-by-source immobilization-duration extraction is incomplete, we do not convert incomplete synthesis into an evidence-of-absence claim. The same bounded sentence now appears in the abstract, Results, Figure 2, Table 2, Table 3, Conclusion, response, S2, and claim ledger.

Change made in the revised manuscript: Immobilization duration was not comprehensively synthesized in this review; therefore, no universal duration is proposed.

Location in the revised manuscript: Revised manuscript Abstract, p. 1; Results §3.6, p. 6; Figure 2, p. 7; Tables 2–3, pp. 7–8; Conclusion, p. 10; Supplement S2 §S2.7.

Comment 9

State that no universal antibiotic agent or duration is supported for delayed/infected cases.

Response 9

Accepted. The revised manuscript makes this explicit for delayed/infected presentations and does not generalize the retrospective clindamycin/cephalexin comparison.

Change made in the revised manuscript: “No universal antibiotic agent, dose, route, or duration is supported…”

Location in the revised manuscript: Revised manuscript Results §3.7, p. 6; Figure 2, p. 7; Tables 2–3, pp. 7–8.

Comment 10

Improve Figure 2 font size and YES/NO pathway readability at standard journal width.

Response 10

Accepted. Figure 2 was redrawn as a deterministic, publication-scale vector schematic with larger type, three communication states, conditional pathway language, and a prominent proposed/unvalidated label. No generative image is used in the manuscript.

Change made in the revised manuscript: Landscape assessment scaffold with explicit uncertainty and no binary “NO = closed” inference.

Location in the revised manuscript: Revised manuscript Figure 2 and legend, p. 7; separate 600-DPI PNG and SVG figure files.

Comment 11

Explain how the high-yield studies were selected for Table 2.

Response 11

Accepted. The eight rows are labelled illustrative, non-ranked, and selected post hoc based on direct relevance, sample size where informative, uniqueness of the signal, and relevance to high-stakes diagnostic or management decisions.

Change made in the revised manuscript: Eight illustrative, non-ranked rows; selection was post hoc and the rationale is now stated in the caption.

Location in the revised manuscript: Revised manuscript Table 2 caption and table, p. 8.

Comment 12

Rename Table 3 to avoid implying validated ‘action thresholds’.

Response 12

Accepted. The table is now a four-column matrix of coexisting clinical dimensions, proposed responses, and evidence basis/directness. “Thresholds” and exclusive phenotype categories were removed.

Change made in the revised manuscript: “Proposed evidence-informed responses by coexisting clinical dimension (unvalidated).”

Location in the revised manuscript: Revised manuscript Table 3 and legend, pp. 7–8.

Comment 13

Add prospective interobserver agreement as a research priority.

Response 13

Clarified and retained. This research priority was already present. We retained it and expanded future validation to reliability, safety, discrimination, and calibration.

Change made in the revised manuscript: “requires prospective testing of interobserver reliability, safety, discrimination, and calibration…”

Location in the revised manuscript: Revised manuscript Discussion §4.3, p. 9.

Comment 14

Add absence of formal critical appraisal to limitations.

Response 14

Accepted. The absence of formal critical appraisal and certainty grading is now explicit in Methods, Results, S2, and Limitations.

Change made in the revised manuscript: “No formal critical appraisal or certainty grading was performed.”

Location in the revised manuscript: Revised manuscript Methods §2.6, p. 3; Results §3.2, p. 5; Limitations §4.4, p. 10; Supplement S2 §S2.2.

Comment 15

Moderate directive conclusion language, such as ‘should enter’ and ‘warrants’, given low evidence and lack of validation.

Response 15

Accepted with evidence-calibrated wording. The safety principle is retained, but treatment language is conditional. Published emergency-department pathways are described as observed multi-component pathways, not as a validated standard or universal eligibility checklist. Routine pinning is recalibrated to a lack of established benefit in the mapped evidence.

Change made in the revised manuscript: “The mapped evidence does not establish a benefit for routine pinning”; ED components are descriptive, not universal conditions.

Location in the revised manuscript: Revised manuscript Results §§3.4–3.6, pp. 5–6; Figure 2, p. 7; Table 3, pp. 7–8; Conclusion, p. 10.

Comment 16

Use one consistent description of AI-assisted tools in Methods and formal disclosure.

Response 16

Accepted. Methods, the formal disclosure, and Supplement S1 now use the same description: OpenAI Codex and GPT-assisted tools supported metadata normalization, duplicate-candidate identification, language editing, and formatting; all final duplicate, eligibility, extraction, interpretive, and clinical decisions remained human.

Change made in the revised manuscript: One consistent disclosure now appears in Methods, the formal AI statement, and Supplement S1.

Location in the revised manuscript: Revised manuscript Methods §2.7, p. 3; AI disclosure, p. 11; Supplement S1 §S1.8, pp. 2–3.