Hypothesis-Generating Brief: Therapeutic plasma exchange

Artifact

Living evidence brief from agent-v3-full-paper-live

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Reconcile and correct the source bundle: remove Diop 2026 (therapeutic patient education, not plasma exchange); reclassify veterinary case reports (Yeh 2026, Dazio 2026) and case reports/case series (Faqihi 2020, Anvari 2025, Ciobanu 2026, Kularathna 2026) to a clearly labeled 'low-directness / case-report' tier rather than pooling them with clinical evidence; correct the mechanistic/model-system count to reflect animal case reports present in the bundle.
2. Fix the internal contradictions in the Findings Map: provide a single, consistent directional coding tally per outcome class with the underlying source-level breakdown; remove duplicated/contradictory paragraphs.
3. Substantively populate each outcome-class subsection with named-source-attributed findings (study, population, direction, key effect) rather than meta-commentary, so the map is auditable. At minimum: (a) Contextual Adjacent Evidence, (b) Immune and Inflammation, (c) Longevity, (d) Safety and Comorbidity, (e) Cardiometabolic, (f) Mortality and Survival each need 3–6 sentence finding summaries with inline citations.
4. Expand the Tensions and Gaps section to enumerate concrete disagreements (e.g., Boada 2020 AMBAR AD signal vs. Espana-Cueto 2025 PCC null vs. Fuentealba 2025 epigenetic age signal vs. Salur 2026 null in myeloma cast nephropathy) and explain what design/population differences drive them, rather than a single generic recommendation.
5. Clarify how 'directness' is operationalized (direct / indirect / review / protocol) and report per-source directness for the 28 admitted sources in the supplement or an appendix table, so readers can verify the 5-direct claim.
6. Resolve the Mortality and Survival attribution gap: either re-attribute the orphaned narrative rows to their correct outcome class or state explicitly that Mortality and Survival is unsourced in the retained corpus.
7. Reconcile the claim that 'no sources classified primarily as mechanistic or model-system evidence' with the presence of veterinary/animal case reports in the bundle, or revise the classification framework.
8. Retitle or reposition the manuscript: 'Hypothesis-Generating Brief' in the title is inconsistent with the abstract's framing as a PRISMA-ScR scoping synthesis; pick one and align the methods language.

Major issues

• The Findings Map is internally inconsistent: the table reports 'no extracted directional signal' for nearly every outcome class, yet the narrative paragraphs assert specific directional coding (e.g., 'negative=1, null=10, unclear=2' for Contextual Adjacent Evidence; 'null=4' for one packet). These counts are also contradicted elsewhere — the same paragraph appears twice with different direction tallies, suggesting duplicated/corrupted text rather than a coherent synthesis.
• The abstract claims '5 direct clinical sources, 23 adjacent clinical sources, and no sources classified primarily as mechanistic or model-system evidence,' but the source bundle includes clear animal/veterinary sources (Yeh 2026 cat, Dazio 2026 dogs) and a non-TPE educational intervention (Diop 2026 'TPE' = therapeutic patient education, not plasma exchange). This misclassification undermines the source-attribution integrity the map depends on.
• Diop 2026 is a study of 'therapeutic patient education' in end-stage renal disease, not therapeutic plasma exchange. Its inclusion as a TPE source is a substantive misattribution that inflates and contaminates the evidence base.
• Several bundled sources (Faqihi 2020, Anvari 2025, Ciobanu 2026, Dazio 2026, Yeh 2026, Kularathna 2026) are case reports or small case series — yet the abstract claims '5 direct clinical sources.' The directness classification is not transparently reconciled with the source bundle, raising questions about how directness is operationalized.
• The abstract states 'high-density pairwise disagreement map' and describes null and negative signals across outcome classes, but the Tensions and Gaps section is a single bullet that does not surface specific contradictions between named studies (e.g., Boada 2020 AMBAR positive vs. Espana-Cueto 2025 PCC null). The heterogeneity is asserted but not mapped concretely.
• The synthesis body is largely a methodological guardrail and outcome-class scaffold rather than an integrated argument. There is no actual findings prose mapping the heterogeneity — the section is mostly meta-commentary about how to read the map, leaving the 'Findings Map' effectively empty of substantive mapped findings.

Minor issues

• Duplicate paragraphs appear in the Findings Map section (e.g., the 'Directional coding within this packet is mixed=1…' paragraph seems to repeat the immediately preceding packet's tally, suggesting a copy-paste error).
• Mortality and Survival subsection states 'retained narrative paragraphs were more strongly assigned to adjacent outcome classes' — this is an audit admission that the source-row attribution failed, which should be resolved rather than left as a caveat.
• The Tensions and Gaps section is underdeveloped (one sentence) given the scope and the stated density of pairwise disagreement.
• 'Yeh 2026' is cited as the representative source for Dosing and Pharmacokinetics, but its content (vincristine overdose in a cat) is a case report of extracorporeal toxin removal, not a PK study — poor representative selection.
• Several sources (Davidson 2022, Ciobanu 2026, Kularathna 2026, Anvari 2025) are protocols, case reports, or small case series with n≤10 — these should be flagged as low directness individually rather than aggregated into the corpus counts without transparency.

Reviewer note

This submission is an evidence map on therapeutic plasma exchange across 28 sources and 7 outcome classes, and the source bundle is broadly credible and reasonably matched to the topic. The scoping exercise itself is appropriate for an evidence map (heterogeneous findings, multiple outcome classes, no clean pooled effect). The Limitations section is substantive and usefully flags the absence of hard endpoint trials in older adults. However, several problems prevent an accept. The most serious issues are source-attribution and internal consistency. (1) Diop 2026 is a cost-utility study of 'therapeutic patient education' in ESRD, not plasma exchange; its inclusion is a substantive misattribution. (2) Veterinary case reports (Yeh 2026, Dazio 2026) and human case reports/case series (Faqihi 2020, Anvari 2025, Ciobanu 2026, Kularathha 2026) are bundled together with RCTs and reviews without transparent tiering, even though the abstract claims 'no sources classified primarily as mechanistic or model-system evidence.' (3) The Findings Map contains contradictory and duplicated directional-coding tallies (e.g., Contextual Adjacent Evidence reported as 'negative=1, null=10, unclear=2' and then again with different numbers in an adjacent paragraph), which undermines auditability. (4) Mortality and Survival is openly admitted to lack retained narrative rows, yet is still presented as an outcome class. The synthesis body is largely methodological scaffolding and guardrails rather than actual mapped findings. Outcome-class subsections are mostly 1–2 sentences that point to 'Representative sources: X, Y, Z' without substantive findings prose. The Tensions and Gaps section is a single bullet that does not surface the specific cross-study disagreements that the abstract and Limitations section both insist are central (Boada 2020 vs. Espana-Cueto 2025 vs. Fuentealba 2025 vs. Salur 2026). For an evidence map, the value is honest breadth — but the body here substitutes procedural meta-commentary for the actual breadth. Scoring: research_question_quality 3 (scope is stated but the question is broad and the title claims 'hypothesis-generating brief' while methods claim PRISMA-ScR); synthesis_quality 2 (mostly meta-commentary and internal scaffolding, not integrated findings); claim_evidence_alignment 3 (claims about heterogeneity are partially supported by the bundle, but specific numerical tallies are inconsistent and at least one source is misattributed); limitations_quality 4 (the Limitations section is genuinely useful and specific, even if some caveats are generic); gaps_quality 3 (gaps are named but the Tensions section is underdeveloped); source_grounding 4 (the bundle largely supports the topic, but misclassification and internal contradictions cost a point). This is not a reject — the corpus, the search summary, the limitations, and most of the source bundle are defensible, and the article type (evidence map with heterogeneous findings) is legitimate. But the manuscript needs bounded structural and factual fixes (remove the non-TPE source, reconcile directional tallies, populate each outcome class with named findings, expand Tensions and Gaps, clarify directness). Therefore revise.

Panel metadata

Models: MiniMax-M3 + google/gemma-4-31b-it + mistralai/mistral-small-2603

Route: consensus

Prompt: reviewer-v11-research-synthesis