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. Remove or relabel the Diop 2026 'e-TPE' study from the cardiometabolic row. It is a web-based therapeutic patient education trial in ESRD, not therapeutic plasma exchange. Replacing it with a true TPE cardiometabolic source or deleting the row is required.
2. Reconcile directional coding with the source bundle. Boada 2020 (AMBAR primary) and Boada 2021 should be coded 'mixed' or 'positive on select endpoints, null on others', not 'null'; Espana-Cueto 2025 should be coded 'negative' or 'mixed' for efficacy, not 'null'; Ipe 2021 should be coded 'positive for TPE vs IVIG response rate'. Either correct the bundle or correct the prose and Findings Map so they match.
3. Reclassify the Longevity outcome class. Maier 2025 (COVID-19 hyperviscosity RCT), Krzych 2021 (COVID-19 systematic review), and Thomas 2026 (pediatric autoimmune DAH) do not test longevity. Either move them to a more appropriate outcome class or replace them with sources that actually address aging-relevant endpoints; only Fuentealba 2025 plausibly belongs here.
4. Resolve the scope framing. Either (a) retitle and reframe the evidence map as 'Clinical applications of therapeutic plasma exchange across heterogeneous indications' and drop the anti-aging framing, or (b) restrict the map to aging-relevant evidence and rebuild the corpus around it. The current mixed framing is internally inconsistent.
5. Make the directional tallies auditable. Provide, in the supplement or inline, the per-source direction/directness/tier table so the 'null=10/13', 'null=3/3', etc. counts in the prose can be verified against the 28 retained sources.
6. Fix the funnel arithmetic in Search Summary so the admission buckets reconcile to 28 admitted sources, and remove the duplicated 'Evidence-honesty note' / 'Source-bundle reconciliation note' blocks that appear in abstract, scope, and limitations.

Major issues

• Scope mismatch: article is filed as evidence_map on 'Therapeutic plasma exchange' with a longevity/anti-aging framing, but the retained corpus is dominated by TPE for autoimmune neurological disease, TTP, sepsis, COVID-19, myeloma cast nephropathy, pediatric neuroimmune disease, and replacement-fluid hemostasis studies. Only Fuentealba 2025 plausibly maps to an anti-aging/biological-age question, and the corpus has no long-term mortality or hard-aging-endpoint RCTs. The abstract and limitations both acknowledge this, but the evidence map never reconciles why these heterogeneous clinical TPE contexts are pooled under a 'longevity' evidence-map label rather than treated as a clinical TPE evidence map.
• Outcome-class coding is inconsistent and partially fabricated: the manuscript repeatedly cites directional-coding counts (e.g., 'Directional coding within this packet is negative=1, null=10, unclear=2'; 'mixed=1, null=4, unclear=1'; 'null=3' for longevity) that do not sum to the stated source counts in a transparent way, and several paragraphs appear duplicated or pasted with stale totals. The mapping from the 28 retained sources to these directional tallies is not auditable from the bundle.
• Longevity outcome class is under-supported and partly misclassified: Maier 2025 (COPLEX) is a COVID-19 hyperviscosity RCT, Krzych 2021 is a COVID-19 TPE systematic review, and Thomas 2026 is pediatric autoimmune DAH. None of these directly tests longevity. Their grouping under 'Longevity Outcomes' with a null=3 tally is not source-attributable in the bundle; this is a faithfulness defect for an evidence map.
• Findings Map and prose disagree with the source bundle on key directional coding: the bundle shows Ipe 2021 coded 'negative' but the manuscript narrative treats it as null/indirect; Espana-Cueto 2025 is coded 'null' with clear negative efficacy conclusions ('did not lead to any discernible improvement'), which the bundle direction field understates; Boada 2020/2021 are coded 'null' despite the AMBAR primary results describing statistically significant co-primary endpoint differences. The evidence map's directional tallies therefore do not faithfully represent the cited sources.
• The Cardiometabolic row maps Diop 2026 — a cost-utility analysis of a web-based patient-education platform for ESRD — to TPE as the intervention. This is a clear source-misattribution: the study acronym 'e-TPE' refers to 'electronic Therapeutic Patient Education', not therapeutic plasma exchange. The evidence map is therefore citing a non-TPE source as a direct clinical TPE efficacy source for cardiometabolic outcomes, which is a material integrity defect.

Minor issues

• The abstract, scope, and limitations repeat near-identical 'Evidence-honesty note' and 'Source-bundle reconciliation note' blocks verbatim; this padding suggests the underlying manuscript may be a templated scaffold rather than a finished synthesis.
• The Search Summary reports 157 records in the receipt-candidate union but the funnel arithmetic is internally inconsistent: 37 classified source candidates minus 24 no-extractable-claims minus 11 none-only minus 61 mixed partial-or-none minus 19 partial-only minus 5 strict high-confidence does not reconcile to 28 admitted sources.
• Several named sources cited in the Findings Map (Kimber 2026, Eichinger 2025, Xu 2026, Yeh 2026) are only briefly mentioned and never have their directional coding or effect statistics integrated into the prose; they function as name-drops rather than attributed findings.
• Duplicate narrative paragraphs under 'Contextual Adjacent Evidence Outcomes' suggest editorially unpolished text with possibly copy-paste errors in directional counts.
• The reference to Studenski 2011, Cesari 2009, and Cruz-Jentoft 2019 cutoffs in Limitations is inserted without appearing in the source bundle; these are valid scholarly references but are not part of the admitted 28 sources, which slightly undermines the 'all claims traceable to admitted sources' claim of an evidence map.

Reviewer note

This evidence map attempts a faithful landscape of therapeutic plasma exchange evidence across 28 retained sources. The protocol disclosure, PRISMA-ScR framing, search-summary structure, and explicit 'no clinical efficacy conclusion' caveats are appropriate for the article type and match the calibration rules for evidence maps. The hypothesis-generating framing, attention to directness, and explicit 'Tensions and Gaps' section are genuine strengths. However, several material defects prevent acceptance. First, Diop 2026 is a cost-utility analysis of a web-based patient-education platform (e-TPE) in ESRD — not therapeutic plasma exchange — and is misattributed as a direct clinical TPE cardiometabolic source. Second, directional coding for several key sources (Boada 2020/2021 AMBAR, Espana-Cueto 2025 PCC trial, Ipe 2021 MG review) does not match the bundle's direction field or the published findings, so the 'null' tallies presented in the Findings Map are not faithful to the cited sources. Third, the Longevity outcome class is filled with COVID-19 and pediatric DAH studies that do not test aging endpoints, and only Fuentealba 2025 plausibly maps to a longevity/biological-age question. Fourth, the scope label of 'longevity' is inconsistent with a corpus dominated by autoimmune neurology, TTP, sepsis, COVID-19, and myeloma indications, and this mismatch is not resolved in the synthesis. The limitations section correctly notes the absence of long-term mortality RCTs, head-to-head comparisons against canonical geroscience comparators, and prespecified geriatric-syndrome endpoints, and the Tensions and Gaps section appropriately calls for adequately powered human studies. These are good features for an evidence map and should be preserved in revision. Recommendation: revise. The manuscript is salvageable with bounded edits — primarily correcting the Diop 2026 misattribution, reconciling directional coding with the bundle, reclassifying the Longevity outcome slice, and either retitling or rescoping to match the actual corpus. Source grounding is otherwise strong: 27/28 cited bundle entries appear genuinely citable and most are recent (within 5 years), and the abstract's evidence-honesty framing is consistent with the limitations and scope sections. Once the integrity issues above are fixed, this could plausibly become an accept-quality evidence map.

Panel metadata

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

Route: consensus

Prompt: reviewer-v11-research-synthesis