422 / METHOD NOTE
S3 rules and preliminary assessment records
An experimental appendix, separate from the official rating: rules, uncertainties, source locations and update conditions.
S3 assessment: an experimental methodological appendix
Rule CG-S3-DECISION-0.3 · Review cutoff: September 29, 2026 · Not adopted as an official scoring standard
This appendix discloses a preliminary assessment of a specific blood outcome and experimental scoring rules. It does not issue a new final score for verdict 422. Its numbers are not effect sizes, probabilities of benefit or product rankings. Read the reader-facing explanation for the publication correction and historical score limitations.
Candidate rule table: CSV · Rules, inputs and calculation: JSON · Preliminary assessment: JSON
1. Three distinct numerical records
| Record | Scope and status |
|---|---|
| Published D/26 | Historical rating for the broader liposomal NAD+ cellular-energy and mitochondrial-vitality claim. Original item-level calculations and any preprint deduction remain unverified. |
| Earlier pilot S3=25; mean=12.5 | Preserved development records, not adopted public ratings and not recalculated under this version. |
| Current candidates {25, 50} | A conditional logical set for the separate Day 6 whole-blood NAD result, not a single final score, confidence interval or score for the broader claim. |
No new S3 or mean is established for the original claim. The new S3 cannot be averaged with old S1, S2 and S4 values: all four would require the same claim and rule set.
2. Scope and inputs
The profile is healthy adults aged 45–75, oral LNAD/PEG powder, five days, placebo comparison, Day 6 total whole-blood NAD, and the primary N50 analysis. The paper reports a 53.0% increase versus placebo, with a 95% CI of 41.2–65.7%. This is not a direct vitality or tissue-function measurement. See the paper, Methods; Results—Primary endpoints; Fig. 1.
S3 classifies support across a core analysis and prespecified reasonable sensitivity analyses of the same outcome. Sensitivity analyses test alternative assumptions; they are not independent studies. Here, an estimate above the defined target counts as support. The directional target is zero, not a threshold for clinically meaningful benefit.
The module covers individually randomized parallel trials. Other designs require appropriate modules; inapplicability is not zero. This is an application to acquired evidence, not a new comprehensive literature search.
3. Experimental five-level decision table
Use this table only when support and outcome-specific risk-of-bias judgments are known for the planned analyses. Unknown is not automatically nonsupport or high risk.
| Score | Conditions | Boundary |
|---|---|---|
| 0 | No analysis supports the target. | Unreported effects cannot be classified as nonsupport. |
| 25 | Support exists, but every supporting analysis is high risk. | Not awarded merely because evidence is indirect. |
| 50 | Low-risk or some-concerns support exists, alongside a high-risk or nonsupporting analysis. | Identify which analyses retain support. |
| 75 | All support; none high risk; some concerns remain. | Unassessed is not some concerns. |
| 100 | All support; all low risk. | Not complete certainty, replication or 100% efficacy. |
The 25-point spacing is an unvalidated policy choice, not evidence of equal scientific distances. High-risk analyses are not silently removed, nor is the most favorable result selected. Changing the scoring policy requires a new rule version.
4. Preliminary risk of bias for the N50 result
We used RoB 2 for individually randomized parallel trials, August 22, 2019, addressing assignment to intervention. This is an AI-assisted preliminary assessment, not external independent expert review. PN means probably no; NI means no information. NI follows question-specific pathways.
| Domain | Judgment | Reason and location |
|---|---|---|
| D1 Randomization | Some concerns | Stratified variable-block generation reported; concealment details and arm-level baseline data for all 60 not fully verified. Methods—Study design; Results—Sample demographics. |
| D2 Deviations/analysis | High | Post-unblinding exclusion of one placebo participant with observed data; numerical impact not independently checked against the comparison table. Methods—Statistical analyses; Discussion—Limitations. |
| D3 Missing outcome data | High | Arm-level flow and Day 6 observations for all 60, and alternative missingness assumptions, remain insufficiently documented in the acquired material. Methods—Study design and Statistical analyses; Fig. 1. |
| D4 Measurement | Low | Total-NAD assay purpose, repeatability, common sampling procedures and reported masking. Methods—Blood collection and Primary endpoints; registry Masking. |
| D5 Selection of reported result | Some concerns | A prior plan is reported, but the original plan, amendment timing and result comparison were not directly verified. Methods—Statistical analyses; Discussion—Limitations. |
D2: questions 2.6 and 2.7 are PN and NI. The authors report materially unchanged N51/N50 results. A separate AI assessment inferred PN for 2.7 from that report; the retained assessment uses NI because the numerical impact of an outcome-related exclusion remains unverified. This disagreement is preserved. The applicable pathway is in official Table 6, pages 28–32.
D3: responses 3.1/3.2/3.3/3.4 are PN/PN/NI/NI, mapping to high risk in official Table 10, page 47. The repeated-measures model relies on a particular missingness assumption; that alone does not demonstrate robustness to alternatives. The N51/N50 comparison addresses a different, observed-data exclusion. We have not classified all nine outside the N51 set as confirmed missing outcomes.
Overall risk is provisionally high. The alternative D2 interpretation does not change this while D3 remains high. This is not evidence establishing actual bias, no effect or misconduct. Bias direction is not determined, and this judgment is not automatically copied to the N51 analysis.
5. Why {25, 50}?
Holding the primary analysis's reported positive estimate, preliminary high-risk judgment and eligibility fixed, two branches remain:
- 25: all supporting analyses are high risk, with no low-risk or some-concerns support.
- 50: another eligible low-risk or some-concerns analysis supports the target, while the high-risk primary analysis remains included.
The retained positive result excludes zero; the retained high-risk result excludes 75 and 100. We have not verified that the additional supporting analysis actually exists. The incomplete analysis inventory and unresolved dependencies make this a logical outer set, not a probability distribution, confidence interval or continuous range.
New evidence changing the primary estimate, judgment or eligibility can move the result outside {25, 50}. These are conditional outputs, not bounds on future findings.
6. Transfer and update conditions
Formulation equivalence and a validated connection from whole-blood NAD to tissue energy or mitochondrial function remain unverified. The candidates therefore do not transfer to the broader claim. This is not proof that such a connection is impossible, and not a zero score. Indirectness is not counted again as measurement bias.
Arm-level participant flow and missing-data sensitivity results inform D3; N51/N50 tables inform D2 and support; dated analysis plans and amendments inform D5. Formulation or direct functional evidence informs transfer. Publication status alone carries no automatic adjustment in this candidate policy; that does not prove the historical preprint deduction was zero.
7. Reproducibility is not clinical validation
Checks covered 1,554 complete-input combinations, 12 unit tests and 31 synthetic boundary cases prepared by a separate AI without reading the calculator. Counterfactual input changes test the rules' behavior, not clinical causal effects under alternative patient histories.
These checks do not establish human inter-rater agreement, scale validity, reader comprehension or decades-long reliability. Verdict 422 was a known development case, not an independent external validation case.
Sources: published paper; trial registry; official RoB 2 tool. See the assessment record for answers and disagreements, and the rules for local policy and uncertainty handling.
Complete signalling-question record
Question labels are brief paraphrases, not the official wording. See the linked RoB 2 tool for original questions and branching. Y=yes; PY=probably yes; N=no; PN=probably no; NI=no information; NA=not applicable under the branch. These are preliminary assessment records, not responses supplied by the study authors.
D1 · Randomization · some
| Item / answer | Question summary | Rationale | Source location |
|---|---|---|---|
| 1.1 · Y | Random element in allocation order? | The report describes sex-stratified variable blocks and random sequence generation using blockrand. | Methods / Study design, randomization, and intervention Sex-stratified blocked allocation, replacement of pre-screening losses, and participant flow from 60 randomized to 51 dosed with follow-up sampling. RoB 2 (22 August 2019), Table 4; signalling question 1.1; guidance pp. 17–20 Source 1 · Source 2 |
| 1.2 · NI | Upcoming allocation concealed until enrolment and assignment? | Pre-enrolment access to the allocation list and details of concealment were not established. Opaque containers alone were not taken as evidence of sequential numbering or irreversible assignment. | Methods / Study design, randomization, and intervention Sex-stratified blocked allocation, replacement of pre-screening losses, and participant flow from 60 randomized to 51 dosed with follow-up sampling. Opaque masked bottles and the LNAD/PEG and placebo formulations. RoB 2 (22 August 2019), Table 4; signalling question 1.2; guidance pp. 17–20 Source 1 · Source 2 |
| 1.3 · NI | Baseline imbalances suggesting flawed randomization? | The authors report no baseline differences in the analysed sample. Comparisons covering all 60 randomized participants and the intended allocation ratio were not available. | Methods / Study design, randomization, and intervention; Results / Sample demographics; Table 1 Sex-stratified blocked allocation, replacement of pre-screening losses, and participant flow from 60 randomized to 51 dosed with follow-up sampling. Reported baseline comparisons for the analysed sample. RoB 2 (22 August 2019), Table 4; signalling question 1.3; guidance pp. 17–20 Source 1 · Source 2 |
Update trigger: Allocation concealment and access records, plus baseline data by group for all 60 randomized participants.
D2 · Deviations and assignment-effect analysis · high
| Item / answer | Question summary | Rationale | Source location |
|---|---|---|---|
| 2.1 · PN | Participants knew their allocation? | Participant blinding and masked containers are reported. This does not establish that maintenance of blinding was independently verified. | Methods / Study design, randomization, and intervention Trial design, blinding, registration timing and reported pre-unblinding specification of endpoints. Opaque masked bottles and the LNAD/PEG and placebo formulations. ClinicalTrials.gov: protocolSection.designModule.designInfo.maskingInfo, including its masking description. RoB 2 (22 August 2019), Table 6; signalling question 2.1; guidance pp. 28–32 Source 1 · Source 2 · Source 3 |
| 2.2 · PN | Carers or intervention providers knew allocation? | Blinding of study personnel and intervention providers is reported. | Methods / Study design, randomization, and intervention Trial design, blinding, registration timing and reported pre-unblinding specification of endpoints. ClinicalTrials.gov: protocolSection.designModule.designInfo.maskingInfo, including its masking description. RoB 2 (22 August 2019), Table 6; signalling question 2.2; guidance pp. 28–32 Source 1 · Source 2 · Source 3 |
| 2.3 · NA | Deviations caused by the trial setting? | Both 2.1 and 2.2 are PN, so the official question flow does not enter this branch. | Official RoB 2 question flow and decision table for question 2.3. RoB 2 (22 August 2019), Table 6; signalling question 2.3; guidance pp. 28–32 Source 1 |
| 2.4 · NA | Deviations likely to affect the outcome? | Not applicable because the 2.3 branch was not entered. | Official RoB 2 question flow and decision table for question 2.4. RoB 2 (22 August 2019), Table 6; signalling question 2.4; guidance pp. 28–32 Source 1 |
| 2.5 · NA | Deviations balanced between groups? | Not applicable because the 2.3 branch was not entered. | Official RoB 2 question flow and decision table for question 2.5. RoB 2 (22 August 2019), Table 6; signalling question 2.5; guidance pp. 28–32 Source 1 |
| 2.6 · PN | Analysis suitable for the effect of assignment? | One placebo participant with observed outcome data was excluded after unblinding. Suitability for estimating assignment effects was not established, nor were the conditions permitting post-randomization exclusion of an ineligible participant. | Methods / Statistical analyses N=51 and N=50 analysis sets, exclusion after unblinding, and the authors' comparison referring to Supplementary Figure 2. RoB 2 (22 August 2019), Table 6; signalling question 2.6; guidance pp. 28–32 Source 1 · Source 2 |
| 2.7 · NI | Potentially important impact of departures from randomized-group analysis? | Similarity between the N=51 and N=50 analyses is reported by the authors; exact estimates were not available. A small exclusion proportion was not considered sufficient to establish a small impact when exclusion was related to observed outcomes. | Methods / Statistical analyses; Discussion / Limitations N=51 and N=50 analysis sets, exclusion after unblinding, and the authors' comparison referring to Supplementary Figure 2. Authors' claim of robustness across analysis sets and methods. RoB 2 (22 August 2019), Table 6; signalling question 2.7; guidance pp. 28–32 Source 1 · Source 2 |
Update trigger: Original N=51 and N=50 result tables, prior eligibility of the excluded participant, and the numerical impact of exclusion.
D3 · Missing outcome data · high
| Item / answer | Question summary | Rationale | Source location |
|---|---|---|---|
| 3.1 · PN | Outcome availability across the randomized population? | Of 60 randomized participants, 51 met the dosing and at-least-one follow-up blood-draw criteria. Availability of Day 6 outcomes for nearly everyone was not established. All remaining nine participants were not assumed to have confirmed missing outcomes. | Methods / Study design, randomization, and intervention; Methods / Statistical analyses; Results / Primary endpoints: intracellular and circulating NAD+; Figure 1 caption Sex-stratified blocked allocation, replacement of pre-screening losses, and participant flow from 60 randomized to 51 dosed with follow-up sampling. N=51 and N=50 analysis sets, exclusion after unblinding, and the authors' comparison referring to Supplementary Figure 2. Displayed primary-analysis sample sizes and baseline, Day 4 and Day 6 measurements. RoB 2 (22 August 2019), Table 10; signalling question 3.1; guidance pp. 40–47 Source 1 · Source 2 |
| 3.2 · PN | Evidence against distortion from missing outcomes? | MMRM under a missing-at-random assumption depends on that assumption. Robustness to other plausible missingness assumptions for the reduction from 60 to 51 participants was not established. The N=51 versus N=50 comparison addresses a separate one-person exclusion. | Methods / Statistical analyses; Methods / Changes to trial protocol-statistical analysis plan MMRM implementation and the missing-at-random assumption. N=51 and N=50 analysis sets, exclusion after unblinding, and the authors' comparison referring to Supplementary Figure 2. Authors' claim of robustness across analysis frameworks. RoB 2 (22 August 2019), Table 10; signalling question 3.2; guidance pp. 40–47 Source 1 · Source 2 |
| 3.3 · NI | Possible link between outcome values and missingness? | Group-specific reasons for the reduction from 60 to 51 and Day 6 observation status were unavailable, leaving insufficient grounds to infer that missingness was unrelated to actual NAD values. | Methods / Study design, randomization, and intervention Sex-stratified blocked allocation, replacement of pre-screening losses, and participant flow from 60 randomized to 51 dosed with follow-up sampling. RoB 2 (22 August 2019), Table 10; signalling question 3.3; guidance pp. 40–47 Source 1 · Source 2 |
| 3.4 · NI | Likely link between outcome values and missingness? | Group-specific missingness proportions, reasons, and evidence about outcome dependence were unavailable. The assessment does not claim that such a relationship was observed. | Methods / Study design, randomization, and intervention Sex-stratified blocked allocation, replacement of pre-screening losses, and participant flow from 60 randomized to 51 dosed with follow-up sampling. RoB 2 (22 August 2019), Table 10; signalling question 3.4; guidance pp. 40–47 Source 1 · Source 2 |
Update trigger: Day 6 observation status and dropout reasons by group for all 60, with sensitivity analyses of missingness assumptions. Exclusion despite observed data belongs in D2.
D4 · Outcome measurement · low
| Item / answer | Question summary | Rationale | Source location |
|---|---|---|---|
| 4.1 · PN | Measurement unsuitable for the specified outcome? | The report describes an assay for whole-blood total NAD and duplicate-measurement precision. Whether it is a valid surrogate for mitochondrial function is a separate transferability question. | Methods / Primary endpoints: intracellular and circulating NAD+ Laboratory and whole-blood versus plasma assay descriptions. Assay procedure, total-NAD measurement and duplicate-measurement precision. RoB 2 (22 August 2019), Table 12; signalling question 4.1; guidance pp. 54–57 Source 1 · Source 2 |
| 4.2 · PN | Potential differences in outcome measurement between groups? | Common blood collection, transport, and assay procedures are reported. Evidence of different testing procedures between groups was not identified. | Methods / Blood collection; Methods / Primary endpoints: intracellular and circulating NAD+ Collection, cold-chain transport and scheduled blood-sampling visits. Heading introducing NAD measurement methods. Laboratory and whole-blood versus plasma assay descriptions. Assay procedure, total-NAD measurement and duplicate-measurement precision. RoB 2 (22 August 2019), Table 12; signalling question 4.2; guidance pp. 54–57 Source 1 · Source 2 |
| 4.3 · PN | Assessors knew which intervention participants received? | This is a provisional inference from the registry's report of blinding for assessors and the molecular testing facility. | ClinicalTrials.gov: protocolSection.designModule.designInfo.maskingInfo, including its masking description. RoB 2 (22 August 2019), Table 12; signalling question 4.3; guidance pp. 54–57 Source 1 · Source 2 |
| 4.4 · NA | Could that knowledge affect assessment? | The official question flow does not enter this branch because 4.3 is PN. | Official RoB 2 question flow and decision table for question 4.4. RoB 2 (22 August 2019), Table 12; signalling question 4.4; guidance pp. 54–57 Source 1 |
| 4.5 · NA | Was such influence on assessment likely? | Not applicable because the 4.4 branch was not entered. | Official RoB 2 question flow and decision table for question 4.5. RoB 2 (22 August 2019), Table 12; signalling question 4.5; guidance pp. 54–57 Source 1 |
Update trigger: Evidence about group-specific assay procedures, blinding, or measurement-validity defects. Uncertain surrogacy for vitality alone does not change D4 to high.
D5 · Selection of the reported result · some
| Item / answer | Question summary | Rationale | Source location |
|---|---|---|---|
| 5.1 · PN | Result analysed under a plan fixed before unblinding? | A pre-existing SAP is reported, but the N=50 result includes an exclusion after unblinding. The original SAP and timing of changes were unavailable, so finalization of the analysis before unblinding was not established. | Methods / Study design, randomization, and intervention; Methods / Statistical analyses; Methods / Changes to trial protocol-statistical analysis plan Trial design, blinding, registration timing and reported pre-unblinding specification of endpoints. N=51 and N=50 analysis sets, exclusion after unblinding, and the authors' comparison referring to Supplementary Figure 2. Reported January 2022 SAP and changes to blood sampling and visit windows. Change from mixed ANOVA/ANCOVA to MMRM. Change to two-sided tests and addition of bootstrap intervals. Authors' claim of robustness across analysis frameworks. RoB 2 (22 August 2019), Table 14; signalling question 5.1; guidance pp. 63–67 Source 1 · Source 2 |
| 5.2 · NI | Likely result-driven selection among eligible measurements or time points? | Selection among multiple time points and outcome measurements could not be determined because the original plan was unavailable. | Methods / Changes to trial protocol-statistical analysis plan Reported January 2022 SAP and changes to blood sampling and visit windows. Change from mixed ANOVA/ANCOVA to MMRM. Change to two-sided tests and addition of bootstrap intervals. Authors' claim of robustness across analysis frameworks. ClinicalTrials.gov: protocolSection.statusModule, including registration submission and first-posted dates. RoB 2 (22 August 2019), Table 14; signalling question 5.2; guidance pp. 63–67 Source 1 · Source 2 · Source 3 |
| 5.3 · NI | Likely result-driven selection among eligible analyses? | Multiple analysis methods are reported and the original SAP was unavailable. This does not demonstrate selection of a favourable result. | Methods / Statistical analyses; Methods / Changes to trial protocol-statistical analysis plan N=51 and N=50 analysis sets, exclusion after unblinding, and the authors' comparison referring to Supplementary Figure 2. Reported January 2022 SAP and changes to blood sampling and visit windows. Change from mixed ANOVA/ANCOVA to MMRM. Change to two-sided tests and addition of bootstrap intervals. Authors' claim of robustness across analysis frameworks. RoB 2 (22 August 2019), Table 14; signalling question 5.3; guidance pp. 63–67 Source 1 · Source 2 |
Update trigger: The original SAP, dated amendments, and comparison of planned outcomes and analyses with those reported.
Downloads and version checks
The CSV contains five candidate scoring conditions, not an actual score ledger. JSON files contain rule and preliminary assessment records. Hashes verify file identity, not medical validity.
- CSV · Candidate rule table ↓
SHA-256: eb5917436f9ca90d839f6d8ad1040b53881d941e7974d7da84d0200d3f216793 - JSON · Rules and applicability ↓
SHA-256: b789da8cb2981dbb6b195864e2d6e78358295e7a49aa52b81354bafa9ff322af - JSON · 22 signalling questions and preliminary assessment ↓
SHA-256: f96c9b1ea778ef2a7fed539f086b9f6dbc4d5f2fda455563303b2a7b0b6081d4
Paper attribution: Kornilov SA et al. (2026), GeroScience. The article is licensed CC BY 4.0. This page is a separate summary and commentary, not a document endorsed by the study authors. DOI: 10.1007/s11357-026-02399-1 · CC BY 4.0