Research Synthesis: NAD+ Effects

Artifact

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

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Remove or explicitly reclassify sources where 'NAD' does not refer to nicotinamide adenine dinucleotide (Han 2022, Simonis 2025, Zhao 2024). At minimum, flag the acronym collision and explain why they are retained; preferably exclude them from the NAD+ evidence map since they are not NAD+ biology.
2. Resolve the Curran 2023 vs Curran 2025 attribution: either discuss the actual systematic review (Curran 2025, infection/sepsis animal models) in the safety section, or correct the citation. Do not present Curran 2023 as if it were the safety/meta-analysis evidence.
3. Integrate the retained human NAD+ RCTs that are currently absent from the prose (Liao 2021, Vreones 2022, Ren 2023, Pei 2024, Holmes 2026) or explain their exclusion. The map cannot claim 29 retained sources and then ignore 5 of them.
4. Correct the mischaracterization of Elhassan 2019 as 'preclinical/animal' — the bundle excerpt confirms it is a human RCT (n=12, crossover, NR vs placebo). Recalibrate the muscle function class accordingly.
5. Tie the long p-value lists (Xue 2022, Simon 2024, Elhassan 2019) to specific named endpoints rather than presenting them as undifferentiated clusters, so the 'mixed signal' claim is auditable.
6. Expand the 'Tensions and Gaps' section to enumerate concrete cross-domain tensions surfaced in the corpus (e.g., biomarker engagement vs functional null; direct vs indirect evidence asymmetry; precursor-specific vs class-level effects; acute vs chronic dosing) rather than a generic call for more trials.
7. Reconcile the abstract's '6/29 direct human evidence' claim with the Findings Map table, which implies different directness counts per class, and define the directness categories explicitly.
8. Verify the search funnel arithmetic: 167 receipt candidates → 47 source candidates → 29 admitted, with stated 'no post-screening exclusions' — these numbers are inconsistent and should be reconciled.

Major issues

• Several cited sources are mis-mapped to the NAD+ effects topic. 'NAD' in Han 2022 refers to 'non-accidental death' (air pollution epidemiology), not nicotinamide adenine dinucleotide; in Simonis 2025 'NAD' refers to 'non-acute decompensation' in cirrhosis; in Zhao 2024 'NAD' refers to 'non-anxious depression'. These are included in the longevity/related outcome classes and are not actual NAD+ evidence, which is a fundamental scope and attribution error for a synthesis paper.
• The synthesis treats acronym collisions (NAD = nicotinamide adenine dinucleotide vs NAD = non-accidental death/non-acute decompensation/non-anxious depression) as if they were relevant NAD+ biology. This is a major methodological error that inflates the apparent evidence base and contaminates outcome classes (especially longevity and safety/comorbidity).
• The 'muscle function' outcome class cites 'Yu 2025' as a cardiac-muscle endpoint, but its inclusion under muscle function is strained; Elhassan 2019 is correctly human skeletal muscle but Connell 2021 is presented in a way that obscures its small N (n=14) and null findings. The mechanistic-vs-functional gap is acknowledged but the framing around 'Elhassan 2019' as 'preclinical' is incorrect — the bundle excerpt describes a human RCT.
• The 'Contextual Adjacent Evidence' class bundles very heterogeneous sources (canine RCT, SSNHL trial, redox polymorphism study, RiaGev crossover, CNM-Au8 brain MRS, liposomal formulation in vitro) under a single category of 16 sources, which obscures what is actually direct NAD+ human evidence versus adjacent mechanistic work and makes the '4 direct; 11 indirect' accounting difficult to verify.
• The 'Safety and Comorbidity' section discusses Curran 2023 (a review/commentary) but the source bundle shows Curran 2025 is a separate systematic review/meta-analysis of niacin/NAD metabolites in infection models that is never discussed; the misattribution between Curran 2023 and Curran 2025 suggests a citation/curation error.
• Several p-value clusters are presented as long lists (e.g., Xue 2022, Simon 2024, Elhassan 2019) without tying them to specific endpoints, which makes the 'mixed signal' characterization hard to verify and risks looking like p-value dumping rather than synthesis.

Minor issues

• The search summary lists 12 databases and 10 queries but the actual retention is only 29 sources; the funnel numbers (167 → 47 → 29) do not add up cleanly and the exclusion reasons section says 'no records were excluded' which contradicts the funnel.
• Some p-values reported in the prose (e.g., 'P = 0.088, P = 0.115' attributed to Yu 2025) cannot be checked against the bundle excerpt provided, which only describes the design; per calibration rules this is acceptable for reference-only bundles but should still be flagged as not verifiable from the supplied source material.
• The Liao 2021 NMN + exercise in amateur runners source is in the bundle but never mentioned in the synthesis body despite being one of the few human RCTs of an NAD+ precursor with a functional endpoint.
• The Mevenkamp 2024 31P-MRS method paper is listed in the bundle and as 'additional corpus source' but is not discussed in any outcome class; it is highly relevant to non-invasive NAD+ measurement and should be integrated or explicitly excluded.
• Vreones 2022 (NR in neuronal extracellular vesicles, older adults RCT) is in the bundle but not discussed anywhere in the synthesis body.
• The Pei 2024 NAD+ in older HF patients and Holmes 2026 NRPT in menopause are in the bundle but absent from the body, suggesting incomplete coverage of the retained sources.
• Ren 2023 (CNM-Au8 brain NAD+/NADH) is in the bundle but absent from the body.
• The 'Tensions and Gaps' section is a single sentence that lacks the structure promised by the review checks; it should enumerate specific cross-domain tensions rather than a generic call for more trials.
• Inconsistent terminology: 'direct' vs 'review' vs 'indirect' directness categories are defined loosely; the count '6/29 direct human evidence' in the abstract is not reconciled with the Findings Map table.

Reviewer note

This evidence map attempts a tiered, audit-trail-style synthesis of 29 sources on NAD+ precursor effects across multiple outcome classes. The protocol-level disclosures (search scope, directness stratification, RoB instruments) are well articulated and the scope is auditable in principle. The abstract and Limitations sections appropriately avoid strong causal or clinical conclusions, and the evidence-honesty note about 23/29 sources being indirect is a useful framing. However, several issues materially weaken the synthesis. First, the acronym 'NAD' is treated as a single referent across the corpus, but Han 2022 uses NAD for 'non-accidental death' (air pollution mortality), Simonis 2025 uses it for 'non-acute decompensation' (cirrhosis), and Zhao 2024 uses it for 'non-anxious depression'. Including these as 'longevity' or 'contextual' NAD+ evidence is a fundamental scope error that an evidence map cannot afford. Second, several retained human NAD+ RCTs in the bundle (Liao 2021, Vreones 2022, Ren 2023, Pei 2024, Holmes 2026, Mevenkamp 2024) are never discussed in the body, so the map under-represents its own evidence base. Third, Elhassan 2019 is described as 'preclinical/animal' when the bundle confirms it is a human RCT. Fourth, the Curran 2023 vs Curran 2025 attribution appears confused. Fifth, the Tensions and Gaps section is a single sentence and does not honor the review-check requirement to surface contradictions explicitly. Synthesis quality is weak because the per-outcome-class narratives lean on long p-value lists rather than endpoint-level interpretation, and because the direct/indirect accounting is internally inconsistent (the abstract says 6 direct, the table implies different counts). Source grounding is partial: the core cardiometabolic, muscle, and dosing sources map well to the bundle, but the longevity and several contextual mappings do not. Limitations are well stated in the dedicated section. The map is salvageable but requires a scope/curation pass to remove non-NAD+ acronym collisions, integrate the missing retained RCTs, correct the Elhassan mischaracterization, and expand the Tensions and Gaps section. Recommendation: revise.

Panel metadata

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

Route: consensus

Prompt: reviewer-v11-research-synthesis