# HSEF v5.3 review: Shingles Vaccination and Dementia Prevention

> Sources: 53 primary reports; claims: 58; verified: 58; unverified: 0. Verification is bounded attribution. Access: 3 partial-body and 50 bounded abstract/status examinations. Literature cutoff: 2026-10-13; source examination UTC date: 2026-10-13. No complete-paper audit or blinded calibration claimed.

> **Adversarial review:** Generated by Sol 6.1; adversarially reviewed by Claude Opus 5.5 (Anthropic) on 2026-10-14. 5 corrections applied (see ADVERSARIAL REVIEW AND APPROVAL). Status: **Approved as revised.**

## AT A GLANCE

**Shingles vaccination has promising local causal evidence for preventing or delaying dementia diagnoses; product-wide efficacy, causal magnitude and the mediator remain unsettled.**

| Level | Confidence | Causal role | Momentum | Stability | Divergence |
|---|---|---|---|---|---|
| Δ — First field-valid causal perturbation crossed | Moderate | reinforcing / stabilizing factor | Mechanistic only | Stable | Contested |

**Deciding experiment:** Combine an ethically designed randomized timing/product or vaccine-encouragement comparison within clinically permitted practice with adjudicated incident dementia, active comparators and competing-death analyses. Prespecify product, age and sex, assess baseline cognition and healthcare contacts, and separately reproduce eligibility estimates using individual uptake and counterfactual complier risks. A mechanistic substudy must distinguish viral reactivation from adjuvant immunity without replacing clinical endpoints.

**Belief versus evidence:** The preregistered primary-subclaim rule returns Δ from SC2: the strongest evidence is a positive local age-eligibility natural experiment for live-zoster vaccination and recorded dementia, with international eligibility corroboration. Multiple related discontinuity analyses do not automatically supply independently failing causal modalities. Receipt-effect assumptions, a quantitative 2026 critique, diagnosis ascertainment, product transfer and exploratory sex boundaries prevent a higher grade. Weakest-link returns Β; primary and scope-matched return Δ. This is stronger than an ordinary association but remains a bounded causal-inference judgment rather than completed randomized dementia-prevention proof.

**Best-supported rival:** Health selection and ascertainment (rank 2) — the nominal, a local vaccine effect on dementia diagnosis, itself ranks first, so it is not its own rival. The Welsh eligibility comparison PMID40175543 is the strongest separator between a local vaccine effect and ordinary healthy-user selection because nearly identical birth dates sharply change uptake. It does not eliminate every cutoff-specific intervention or determine receipt effect without instrumental-variable assumptions. The quantitative critique PMID42754179 supplies an essential independent analytical challenge.

## COMPETING MECHANISTIC MODELS

**Rank 1: Local vaccine effect on dementia diagnosis (Δ).** Live-zoster vaccination can prevent or delay recorded dementia in the local eligibility setting. Welsh exogenous birth-date uptake jump and positive clinical outcome, with international eligibility corroboration. Quantitative critique PMID42754179 and product/age/diagnosis limits require prospective resolution.

**Rank 2: Health selection and ascertainment (Β).** Healthy-vaccinee and prodromal-dementia selection generate some ordinary cohort associations. Welsh ordinary cohorts show broad disease/mortality associations PMID35434253; RZV nursing-emulation negative controls PMID42296498 retain residual confounding; uptake disparity PMID41435736 supplies a concrete selection path. Baseline balance and exact-date placebos make simple smooth health selection a weaker explanation of the local Welsh discontinuity.

**Rank 3: Infection suppression and immune modulation (Γ).** VZV suppression and nonspecific immune effects may both mediate a genuine vaccine benefit. Waning/dose infection correlations PMID41053450, neural culture PMID35754275 and AS01 vaccine comparisons PMID40562756 are compatible. Conflicting overt-HZ cohorts and absent mediator dissociation prevent necessity or exclusive adjuvant proof.

The tournament does not alter the nominal level. Compatible pathways and flexible alternatives are not automatically proved by explaining observations.

## FULL REVIEW

### Registered claim and scope

Shingles vaccination causally reduces incident dementia in older adults.

The precise lead clinical proposition is SC2. Mechanism, competing-cause, generalization and safety companions remain separately reported; they do not replace the primary clinical endpoint.

Selection: position 17 in the registered editorial traffic-priority list; registration timestamp 2026-10-13 T 16:06:25.540258+00:00. Ranked editorial traffic priority rather than measured global popularity or a preset verdict. Registered claim assessed against original sources; no independent blinded calibration is claimed.

### Evidence and causal interpretation

Rubric fit: Frozen01-neurodegeneration focuses on disease-mechanism perturbation. This registered preventive clinical claim uses a declared field-equivalent: a validated eligibility natural experiment with a clinically relevant incident-dementia endpoint and checked assignment assumptions. It is not literal laboratory removal of a protein. All frozen criterion text is retained; assay biomarkers and cell AD-like phenotypes cannot replace clinical dementia, and observational associations remain Β. Secondary immunology routing is considered for mediator questions, not used to borrow vaccine-class efficacy.

The Welsh eligibility experiment makes this question stronger than a routine vaccinated-versus-unvaccinated cohort. The original Methods specify 282541 dementia-free participants in the broad analysis frame, with local bandwidths near age 80, a seven-year main horizon and a fuzzy instrumental-variable receipt estimate. Its dementia outcome combines primary/specialist/hospital diagnoses and death certificates, and therefore tests recorded dementia rather than the first biological lesion. The primary analysis does not adjust directly for competing death, while examining alternate time and healthcare-pathway controls.

Australian and Canadian eligibility effects are real published positive diagnosis contrasts, but they are not interchangeable with the Welsh receipt estimate. In Canada, uptake is measured in a separate survey: the primary first stage is 27.4 points with 95% CI 0.0–54.7 and secondary 13.9 points with CI −5.8–33.6. The clinical difference is about 2 points. Women have a significant within-group effect and men do not, but interaction P=.218 does not establish different causal effects. A 2026 new quantitative analysis PMID42754179 reports implausibly large implied vaccine effectiveness outside Wales. Its aggregate-risk assumptions differ from local complier estimation; this disagreement is retained as an unresolved test of causal magnitude, not silently discarded.

Consensus discovered a further contrary English eligibility analysis. Its original repository identifies it as an unreviewed preprint: [Hamilton et al. 2026](https://www.medrxiv.org/content/10.64898/2026.07.20.26358345v1), doi 10.64898/2026.07.20.26358345. The independently read original abstract reports no detectable hospital-dementia effect despite a shingles reduction and states that linked Welsh comparisons do not support hospital-only ascertainment as the explanation. This is a serious non-counted challenge; full original Methods and peer review were not completed. It does not determine a higher causal boundary grade or erase the peer-reviewed local Welsh positive by assertion.

Recombinant-vaccine cohorts and the 2024 product transition are encouraging, yet ordinary treatment selection and calendar-time differences remain. Selected original Optum Methods assign person-time by changing vaccine status, preventing the crude interpretation that all follow-up belongs to the final dose category. Adjusted one/two-dose HRs 0.89/0.68 remain associations. The product-transition result 17% more diagnosis-free time and 164 days among affected participants is not 17% relative dementia-risk reduction. The nursing-facility emulation explicitly finds residual negative-control confounding.

Vaccine benefit does not establish a necessary VZV mechanism. UK and Danish clinical-zoster cohorts are null or inverse, while several Asian/Swedish cohorts are positive. Neural cultures show VZV can reactivate quiescent HSV1 and AD-like markers, but cell markers cannot establish human dementia mediation. The Welsh body treats both reactivation reduction and nonspecific immune effects as plausible exploratory explanations. AS01-adjuvanted RSV and zoster associations support a hypothesis without isolating the adjuvant.

The companion repurposing literature has genuine agent-specific positives and nulls. A randomized hypertension-care programme PMID40258956 lowers incident all-cause dementia; SystEur reports a small borderline positive. SPRINT lowers MCI and a composite but has a nonsignificant probable-dementia result; HYVET and ASPREE aspirin are also null for their dementia comparisons. GRADE has no cognitive differences between randomized diabetes agents; sildenafil improves some perfusion markers without dementia testing. Statin and metformin associations do not become randomized effects because they occur within parent trials. These results neither prove nor erase zoster prevention.

### Additional original-source checks outside the primary-report count

[Eligibility for shingles vaccination and hospital-coded dementia in England and Wales: a regression discontinuity analysis in England](https://www.medrxiv.org/content/10.64898/2026.07.20.26358345v1). Status: Unreviewed original preprint, posted 21 July 2026. Access: Original repository abstract and explicit status actually examined; no Methods/body audit. English eligibility analysis reports a clear shingles reduction and no detectable hospital-coded dementia difference; an independent HES extract and Welsh comparisons are reported. No proof of zero effect, all-product refutation or completed peer-reviewed causal boundary; uptake/complier and original full methods not independently audited. This source is not included in the primary-report target or treated as a complete-paper audit.

## HYPOTHESIS STATE EVALUATION FRAMEWORK (HSEF v5.3)

### Subclaim matrix and aggregation

| Subclaim | Scope | Level | Decision |
|---|---|---|---|
| SC1 Shingles vaccination precedes a lower incidence of dementia. | Baseline age, vaccine product, eligibility and dementia-free state | Β | Vaccination precedes lower recorded dementia incidence in multiple ordinary cohorts. These descriptive associations retain Β; larger samples and extensive adjustment alone do not create causal assignment. |
| SC2 Shingles vaccination causally reduces dementia incidence. | Baseline dementia-free older adults; age eligibility and follow-up fixed | Δ | The Welsh fuzzy eligibility discontinuity provides a positive local clinical causal-inference crossing, with strong first-stage uptake and baseline/placebo checks. Australian and Canadian eligibility analyses corroborate diagnosis differences but differ in receipt-effect identification and uptake precision. A 2026 quantitative challenge reports implausible implied effects outside Wales and deserves direct resolution. Δ is awarded for a restricted field-valid natural experiment, with Moderate confidence and no assertion of product-wide randomized confirmation. |
| SC3 Herpes-zoster prevention is the necessary mediator of the dementia effect. | Vaccine effect dissociated from infection and immune mechanisms | Γ | Clinical and subclinical VZV suppression, heterologous immune effects and healthcare pathways remain plausible. Infection cohorts conflict; neural-culture reactivation and vaccine-waning correlations do not independently isolate the necessary mediator. Same-uptake sex differences and observational AS01 comparisons do not prove an exclusive pathway. |
| SC4 Repurposed drugs and other vaccinations reproduce the same preventive dementia effect. | Exact agent, baseline eligibility and clinically diagnosed outcome | Γ | Other vaccines and repurposed drugs have agent-specific evidence, not a uniform reproduction of the zoster effect. The randomized Chinese hypertension-care programme and borderline SystEur positive retain clinical benefit in their own scopes; SPRINT probable dementia, HYVET, aspirin and GRADE cognitive outcomes include nulls. The conjunctive class transfer remains Γ, without downgrading those separate successes. |

Preregistered aggregation: primary-subclaim. The preregistered primary-subclaim rule returns Δ from SC2: the strongest evidence is a positive local age-eligibility natural experiment for live-zoster vaccination and recorded dementia, with international eligibility corroboration. Multiple related discontinuity analyses do not automatically supply independently failing causal modalities. Receipt-effect assumptions, a quantitative 2026 critique, diagnosis ascertainment, product transfer and exploratory sex boundaries prevent a higher grade. Weakest-link returns Β; primary and scope-matched return Δ. This is stronger than an ordinary association but remains a bounded causal-inference judgment rather than completed randomized dementia-prevention proof. Sensitivities: weakest-link Β; primary-subclaim Δ; scope-matched Δ.

### Decisive-perturbation inventory and causal tally

[1] Welsh birth-date eligibility discontinuity identifies a local live-vaccine effect on new dementia diagnosis: reported receipt effect 3.5 percentage points over 7 years. CACE requires continuity, exclusion and monotonicity; diagnoses are not biological onset, age-near 80/product-specific and exploratory mediation is not necessity.

[2] Ontario eligibility thresholds associate with 2.0 percentage-point fewer dementia diagnoses over 5.5 years, with interprovincial trend corroboration. Eligibility is not receipt; uptake comes from a separate cross-sectional survey, secondary first stage includes zero and sex interaction P=.218 does not establish sex-specific effect.

[3] Clone-censor-weight emulation in 509926 nursing-facility entrants associates RZV with lower 4 year dementia risk. Only 1.73% vaccinated, most after discharge; negative controls suggest residual confounding despite 57 covariates and targeted emulation.

[8] UK CPRD cohort finds no increased dementia hazard after HZ, with null dementia-or-death composite and no antiviral variation. This challenges universal clinical-shingles mediation, while subclinical reactivation and vaccine-specific immune effects are not excluded.

[14] New quantitative reanalysis of published natural experiments reports implausibly large receipt-effect implications outside Wales and no corresponding national incidence drop. Aggregate uptake and comparator-risk assumptions do not directly reanalyse individual local compliers; critique must be resolved, not treated as automatic refutation of every eligibility result.

[15] Australian age 80 eligibility discontinuity raises uptake 16.4 points and reduces new dementia diagnosis 1.8 points over 7.4 years. Abstract reports eligibility effect, not receipt relative risk; local cutoff population differs from overall sample mean age and underlying vaccine-effect critique remains.

[16] Rapid product-transition comparison associates RZV with 17% more dementia-diagnosis-free time and 164 days among subsequently affected. Calendar-era treatment transition, measured matching and vaccine comparators are not randomized allocation; relative diagnosis-free time is not 17% risk reduction.

[21] Matched cohorts associate both AS01-adjuvanted RSV and zoster vaccination with lower 18 month dementia risk, with no between-product difference. Nonsignificance is not equivalence; shared adjuvant, healthy uptake and selection cannot isolate AS01 necessity or direct adjuvant efficacy.

[30] Danish long-term matched HZ cohort finds no increased dementia overall; rare CNS subgroup estimate is imprecise. Challenges overt-shingles universal mediation, not all subclinical immune pathways; CNS-attributable fraction is a model, not vaccine effect.

[41] 326-village randomized BP-care programme reduces all-cause dementia over 48 months in 33995 hypertensive adults. A genuine clinical prevention positive for its assigned care package, not proof of each antihypertensive molecule or the same vaccine mediator.

[50] SPRINT intensive BP assignment lowers MCI and composite cognitive impairment but not significant probable dementia. Early stopping and fewer dementia events limit power; positive composite must not replace null registered dementia endpoint.

[52] HYVET original 3336-person comparison lowersBP without significant incident dementia benefit. Early stopping/short follow-up and confidence interval permit benefit; pooled meta-component is not counted as another primary experiment.

[53] ASPREE randomized aspirin in 19114 older adults shows no dementia/MCI/cognitive benefit over 4.7 years. No class-wide preventive drug conclusion follows; statin analyses share cohort but aspirin assignment is distinct.

Tally: 1 positive qualifying study units across scoped subclaims; 28 screened discriminating contrast units. These are not independent causal lines or proof of full-claim necessity/sufficiency. Null and negative contrasts are discriminators, not positive threshold crossings. Reports and replications are not independent causal lines. Only the verified Welsh receipt-identification design earns exact positive SC2 crossing credit; international eligibility effects are supportive and remain different estimands. Hypertension clinical successes retain their drug/clinical scope and do not increase the vaccination causal tally. The 53 originals include comparator, mediation, uptake and analytic-challenge studies, not 53 zoster-prevention trials.

### Cap audit, precedence and Β/Γ/Δ boundary

Do not apply the association cap to a valid assembled natural experiment as if it were an ordinary cohort. Conversely, cohort size, culture markers and international repetition of eligibility designs do not alone establish higher independent causal convergence. Preserve clinical hypertension-prevention successes without lending their randomization to vaccination.

Universal six-cap audit: All six universal caps were considered: cell-only inference, unperturbed association, ectopic expression, biomarker-only intervention, forced transfer/seeding, and inherited-to-sporadic scope transfer. Each binds its offered inference; complementary same-scope causal classes are assessed together rather than re-capped in isolation.

Gamma hard-rule audit: Cell amyloid/phosphotau changes are not human dementia. An adjusted observational association is not randomized vaccine or mediator allocation. Diagnoses, MCI, death with dementia and biological disease onset have different scopes. Receipt effect requires a valid first stage and local instrumental-variable assumptions.

SC1 cap: “Pure association evidence — however large, however replicated — cannot exceed Β.” Vaccination precedes lower recorded dementia incidence in multiple ordinary cohorts. These descriptive associations retain Β; larger samples and extensive adjustment alone do not create causal assignment.

SC1 field-valid definition: “
- Endogenous genetic deletion, suppression, or degron depletion of the proposed mediator in vivo
- Selective pathway inhibition that changes onset/progression while preserving broader viability
- Vagotomy-type natural experiment, only if confounding is rigorously controlled AND effect is disease-initiation relevant rather than merely associated” Positive full-scope result: no. Vaccination precedes lower recorded dementia incidence in multiple ordinary cohorts. These descriptive associations retain Β; larger samples and extensive adjustment alone do not create causal assignment.

SC2 field-valid definition: “
- Endogenous genetic deletion, suppression, or degron depletion of the proposed mediator in vivo
- Selective pathway inhibition that changes onset/progression while preserving broader viability
- Vagotomy-type natural experiment, only if confounding is rigorously controlled AND effect is disease-initiation relevant rather than merely associated” Positive full-scope result: yes. The Welsh fuzzy eligibility discontinuity provides a positive local clinical causal-inference crossing, with strong first-stage uptake and baseline/placebo checks. Australian and Canadian eligibility analyses corroborate diagnosis differences but differ in receipt-effect identification and uptake precision. A 2026 quantitative challenge reports implausible implied effects outside Wales and deserves direct resolution. Δ is awarded for a restricted field-valid natural experiment, with Moderate confidence and no assertion of product-wide randomized confirmation.

SC3 field-valid definition: “
- Endogenous genetic deletion, suppression, or degron depletion of the proposed mediator in vivo
- Selective pathway inhibition that changes onset/progression while preserving broader viability
- Vagotomy-type natural experiment, only if confounding is rigorously controlled AND effect is disease-initiation relevant rather than merely associated” Positive full-scope result: no. Clinical and subclinical VZV suppression, heterologous immune effects and healthcare pathways remain plausible. Infection cohorts conflict; neural-culture reactivation and vaccine-waning correlations do not independently isolate the necessary mediator. Same-uptake sex differences and observational AS01 comparisons do not prove an exclusive pathway.

SC4 field-valid definition: “
- Endogenous genetic deletion, suppression, or degron depletion of the proposed mediator in vivo
- Selective pathway inhibition that changes onset/progression while preserving broader viability
- Vagotomy-type natural experiment, only if confounding is rigorously controlled AND effect is disease-initiation relevant rather than merely associated” Positive full-scope result: no. Other vaccines and repurposed drugs have agent-specific evidence, not a uniform reproduction of the zoster effect. The randomized Chinese hypertension-care programme and borderline SystEur positive retain clinical benefit in their own scopes; SPRINT probable dementia, HYVET, aspirin and GRADE cognitive outcomes include nulls. The conjunctive class transfer remains Γ, without downgrading those separate successes.

SC2 applied rubric adaptation (distinct from the unchanged literal criterion): Age eligibility is an exogenous treatment-encouragement perturbation. Continuity/exclusion/monotonicity and uptake establish the local clinical causal equivalent; recorded diagnoses remain distinct from biological onset.

SC4 applied rubric adaptation (distinct from the unchanged literal criterion): Exact drug/vaccine, clinical baseline and diagnosed endpoint require agent-specific attribution. Genuine hypertension-package prevention is preserved separately from generic class equivalence.

SC1 literal Γ hard rule: “In vitro aggregation alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC1 literal Γ hard rule: “In vitro seeding alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC1 literal Γ hard rule: “Droplet aging in vitro” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC1 literal Γ hard rule: “Organoid toxicity without organismal follow-through” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC1 literal Γ hard rule: “Exogenous seeding models alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC2 literal Γ hard rule: “In vitro aggregation alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC2 literal Γ hard rule: “In vitro seeding alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC2 literal Γ hard rule: “Droplet aging in vitro” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC2 literal Γ hard rule: “Organoid toxicity without organismal follow-through” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC2 literal Γ hard rule: “Exogenous seeding models alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC3 literal Γ hard rule: “In vitro aggregation alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC3 literal Γ hard rule: “In vitro seeding alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC3 literal Γ hard rule: “Droplet aging in vitro” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC3 literal Γ hard rule: “Organoid toxicity without organismal follow-through” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC3 literal Γ hard rule: “Exogenous seeding models alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC4 literal Γ hard rule: “In vitro aggregation alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC4 literal Γ hard rule: “In vitro seeding alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC4 literal Γ hard rule: “Droplet aging in vitro” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC4 literal Γ hard rule: “Organoid toxicity without organismal follow-through” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

SC4 literal Γ hard rule: “Exogenous seeding models alone” Not triggered literally for this assembled scoped case. Any adapted clinical, proxy or system-inference restriction is stated separately in the subclaim rationale.

### Upward threshold ladder

SC1 Δ: “Requires at least one genuine causal perturbation in a disease-relevant in vivo system, or equivalent human natural experiment that materially supports the mechanism.” Satisfied: no. A valid exogenous assignment or controlled clinical prevention contrast, not more ordinary cohorts.

SC2 Δ: “Requires at least one genuine causal perturbation in a disease-relevant in vivo system, or equivalent human natural experiment that materially supports the mechanism.” Satisfied: yes. 

SC2 Ε: “Requires causal support in a defined context, but with clear restriction by subtype, disease stage, model, or tissue.

Use Ε when there is a real causal contribution, but generalization to the entire disease class fails.” Satisfied: no. Resolve the 2026 effect-size challenge using harmonized individual-level local uptake and counterfactual complier risks; then confirm live/recombinant clinical endpoints by independent assignment with prespecified age/product/sex tests.

SC3 Δ: “Requires at least one genuine causal perturbation in a disease-relevant in vivo system, or equivalent human natural experiment that materially supports the mechanism.” Satisfied: no. Separate viral suppression from adjuvant effects in a credible randomized product/component or mediator-specific design, including measured reactivation and clinical dementia.

SC4 Δ: “Requires at least one genuine causal perturbation in a disease-relevant in vivo system, or equivalent human natural experiment that materially supports the mechanism.” Satisfied: no. Agent-specific randomized or valid natural preventive clinical tests; a drug-class analogy or biomarker response cannot substitute.

Upper ladder: Natural experiments were assessed directly; their clinical phenotype adaptation is explicit. Same-modality international replication, retrospective product comparisons and cell-marker experiments do not automatically satisfy independently failing causal lines. No universal association cap is reimposed on the valid Welsh natural experiment.

Independence: No nominal/subclaim/tournament grade reaches Ζ. Shared Welsh endpoints, overlapping national datasets and repeated eligibility designs are conservatively grouped; the inference does not count papers as independent causal lines.

### Falsification and pitfalls

Association treated as causal mediation. VZV infection/dementia correlations, dose response and AS01 active comparison are not necessary-path perturbations.

Causal estimand conflation. Eligibility effect, receipt CACE, risk reduction and diagnosis-free time are different quantities.

Replication confused with independent causal lines. Welsh stage endpoints and repeated registry analyses share source platforms; preprints not counted separately.

Endpoint substitution. MCI, cerebral perfusion, memory correlations and biological AD onset do not replace incident dementia.

### Adversarial steelman: above, then below

The strongest Ε case uses independent-country replications and female-predominant effects as a mapped causal boundary. However, age/product transfers remain largely observational, the Canadian sex interaction is nonsignificant, and the 2026 implied-effect critique has not been resolved at individual-complier level. These observations do not yet map a reproducible causal failure context.

The strongest Γ case says unresolved instrumental-variable assumptions, diagnosis effects and implausible international implied receipt effects leave mechanistic plausibility alone. This would underweight the strong Welsh first stage, cutoff balance and positive diagnosed endpoint. The critique remains serious but does not by itself demonstrate failure of the Welsh local design.

Δ preserves the local positive causal-inference crossing while withholding mapped generalization and independent convergence. Neither universal protection nor definitive refutation follows. Confidence is Moderate because primary-body checks expose estimand and first-stage limits and an unresolved quantitative challenge.

### Final score, causal direction and M/S/D

The preregistered primary-subclaim rule returns Δ from SC2: the strongest evidence is a positive local age-eligibility natural experiment for live-zoster vaccination and recorded dementia, with international eligibility corroboration. Multiple related discontinuity analyses do not automatically supply independently failing causal modalities. Receipt-effect assumptions, a quantitative 2026 critique, diagnosis ascertainment, product transfer and exploratory sex boundaries prevent a higher grade. Weakest-link returns Β; primary and scope-matched return Δ. This is stronger than an ordinary association but remains a bounded causal-inference judgment rather than completed randomized dementia-prevention proof.

Vaccination precedes a potential reduction in diagnosed disease risk, so it is not merely downstream consequence. Neither necessity for all dementia nor an initiating/rate-limiting VZV dependency is established. The protective influence is classified as reinforcing/stabilizing maintenance of neuroimmune health, conditional on the local causal evidence.

Momentum: Mechanistic only. Recent natural experiments and product comparisons refine assignment and mediation questions. No verified named pending dementia-prevention result is used to predict a specific next grade.

Stability: Stable. No verified pending result was identified that demonstrably determines the next grade. The published 2026 critique is current evidence, not a forecasted result.

Divergence: Contested. No single field-implied level is defensible from conflicting framings, so no implied level and no rung arithmetic are stated: the Welsh, Australian and Canadian eligibility analyses are presented in the field as near-settled prevention, while the 2026 quantitative critique (PMID42754179) and the unreviewed English null preprint present the same evidence as not yet credible. No manufactured field consensus or arithmetic gap. These dimensions do not change level or confidence.

## PROPOSED DECISIVE EXPERIMENTS

Combine an ethically designed randomized timing/product or vaccine-encouragement comparison within clinically permitted practice with adjudicated incident dementia, active comparators and competing-death analyses. Prespecify product, age and sex, assess baseline cognition and healthcare contacts, and separately reproduce eligibility estimates using individual uptake and counterfactual complier risks. A mechanistic substudy must distinguish viral reactivation from adjuvant immunity without replacing clinical endpoints.

Readout: Adjudicated incident dementia and diagnosis-free survival over years, with mortality competing-risk sensitivity and prespecified product/age/sex interaction tests. Controls: Comparable healthcare contacts, baseline cognition, calendar time, vaccination uptake, placebo cutoffs and an active vaccine comparator; blinded dementia adjudication. Supporting: Replicated positive clinical effect under credible assignment, with coherent eligibility-to-receipt estimates and independent measurement. Falsifying: No clinically meaningful difference despite adequate uptake separation/power, or a diagnosis difference eliminated by ascertainment/cutoff correction.

## KNOWLEDGE GAPS

Resolve implausible international implied receipt effects and weak separate-survey first stages with individual-level local data.

Direct preventive efficacy of recombinant vaccination rather than ordinary vaccinee cohorts.

Biological onset versus delayed clinical diagnosis and competing-death effects.

Prespecified age/product/sex boundaries with formal interaction tests.

Separate infection suppression, heterologous immune effects and healthcare pathways.

Drug-specific clinical prevention rather than class transfer from surrogate or observational successes.

## REVIEWER NOTES

All 53 selected originals were individually presented and appraised. Selected Methods/Results from 3 papers were actually examined, including local-CACE definition, clinical ascertainment, separate Canadian uptake, interaction testing and recombinant person-time. Preprints are not extra reports, and 37291109 narrative review was removed after examination. No full-paper audit, independent blind calibration or scientific peer review is claimed. An additional original unreviewed English-null preprint was examined at repository abstract/status depth and kept separately from 53 peer-reviewed report credit. The nominal Δ crossing relies on the peer-reviewed Welsh clinical natural experiment; no higher grade or decisive causal boundary is borrowed from this preprint.

Search leads, reviews, protocols and correction notices do not pad the primary count. Downloaded XML does not earn examination credit. Linked notices remain attached; unexamined correction details limit inference. This is a coordinator source/claim consistency assessment, not independent blinded calibration.

## Consensus and original-source reconciliation

The original Welsh body distinguishes eligibility from receipt:3.5 points is the fuzzy CACE receipt estimate. It uses a local complier counterfactual-risk construction and not a simple division of the overall control risk. Diagnosis sources and lack of primary competing-death adjustment are explicit. [Original PMID40175543](https://pubmed.ncbi.nlm.nih.gov/40175543/).

Selected original Methods/Results confirm the eligibility estimand, uptake from a separate CHSS survey, primary 27.4 point first stage with lower CI 0.0 and secondary 13.9 point interval crossing zero. Women-significant/men-nonsignificant has interaction P=.218; this cannot establish a verified sex boundary. [Original PMID41579903](https://pubmed.ncbi.nlm.nih.gov/41579903/).

The original reports 17% increase in diagnosis-free time, with 164 days among participants subsequently diagnosed, not 17% relative risk reduction in all recipients. Calendar transition and active vaccine matching remain observational limitations. [Original PMID39053634](https://pubmed.ncbi.nlm.nih.gov/39053634/).

The original new quantitative secondary analysis reports 18% Wales, 177% Australia and 84%/111% Canada under its published-estimate scaling. These are analytical estimates rather than new vaccine trials. Their denominator/complier-risk assumptions require reconciliation with the Welsh local CACE method; provider strength labels do not adjudicate the critique. [Original PMID42754179](https://pubmed.ncbi.nlm.nih.gov/42754179/).

The original reports an association in a clone-censor-weight emulation, with negative controls suggesting residual confounding. Only 1.73% received vaccine and 87% of these did so after discharge; wording is qualified to association rather than proven lowering. [Original PMID42296498](https://pubmed.ncbi.nlm.nih.gov/42296498/).

The original is a matched observational comparison with no significant product difference. Nonsignificance is not equivalence, and sharing an adjuvant is not a randomized adjuvant-versus-virus dissociation; the mechanism remains a hypothesis. [Original PMID40562756](https://pubmed.ncbi.nlm.nih.gov/40562756/).

The original UK cohort reports no increased dementia after overt HZ, no dementia-or-death association, and no antiviral prescription variation. This important challenge is retained alongside positive registry studies and does not itself exclude subclinical viral effects. [Original PMID35170873](https://pubmed.ncbi.nlm.nih.gov/35170873/).

The original 326-village randomized care programme significantly lowers all-cause dementia over 48 months. This genuine clinical prevention positive is preserved for its exact package, alongside SPRINT dementia-null/MCI-positive and aspirin/HYVET nulls; it neither validates nor refutes vaccination. [Original PMID40258956](https://pubmed.ncbi.nlm.nih.gov/40258956/).

The original medRxiv repository abstract was independently read: Hamilton et al. 2026 doi 10.64898/2026.07.20.26358345v1 is explicitly unreviewed. It reports hospital-dementia eligibility RD −0.06 points,CI −0.40 to 0.27, P=.72, including Welsh ascertainment comparisons. It is retained as a separate non-counted preprint challenge, with no invented original Methods reading or inclusion among 53 peer-reviewed reports. [Original PMID40175543](https://pubmed.ncbi.nlm.nih.gov/40175543/).

Consensus was used to identify literature and cross-check specific claims about design, endpoints, scope and research gaps. Those aspects were checked against original primary abstracts/status records and selected original Methods/Results passages. Consensus output was not treated as original full text. Provider search omissions are not evidence that a study does not exist. Source access and examination receipts are retained. Scite use or monthly unavailability is recorded per case; provider citation counts do not change causal grades.

## Research timeline and gaps: Shingles Vaccination and Dementia Prevention

Selected publication milestones, not an exhaustive history or discovery priority claim.

**1998 — 1998 SystEur reports a small randomized dementia-prevention signal for a BP regimen.** [Original PMID9802273](https://pubmed.ncbi.nlm.nih.gov/9802273/).

**2008 — 2008 HYVET lowersBP but not significant dementia over its early-terminated horizon.** [Original PMID18614402](https://pubmed.ncbi.nlm.nih.gov/18614402/).

**2016 — 2016 influenza/CKD observational protection extends beyond infection season.** [Original PMID26945371](https://pubmed.ncbi.nlm.nih.gov/26945371/).

**2018 — 2018 Taiwanese HZ/antiviral associations motivate prevention hypotheses.** [Original PMID29244265](https://pubmed.ncbi.nlm.nih.gov/29244265/).

**2019 — 2019 SPRINT separates positive MCI/composite outcomes from nonsignificant probable dementia.** [Original PMID30688979](https://pubmed.ncbi.nlm.nih.gov/30688979/).

**2020 — 2020 ASPREE aspirin finds no cognitive/dementia benefit.** [Original PMID32213642](https://pubmed.ncbi.nlm.nih.gov/32213642/).

**2022 — 2022 Welsh ordinary cohorts explicitly retain healthy-life-expectancy selection.** [Original PMID35434253](https://pubmed.ncbi.nlm.nih.gov/35434253/).

**2022 — 2022 UK overt-zoster/dementia association is null or inverse.** [Original PMID35170873](https://pubmed.ncbi.nlm.nih.gov/35170873/).

**2022 — 2022 neural cultures demonstrate VZV–HSV1 reactivation with AD-like markers.** [Original PMID35754275](https://pubmed.ncbi.nlm.nih.gov/35754275/).

**2024 — 2024 product-transition analysis reports more dementia-diagnosis-free time after RZV.** [Original PMID39053634](https://pubmed.ncbi.nlm.nih.gov/39053634/).

**2025 — 2025 Welsh birth-date natural experiment identifies a local live-vaccine causal signal.** [Original PMID40175543](https://pubmed.ncbi.nlm.nih.gov/40175543/).

**2025 — 2025 Australian eligibility design corroborates a positive diagnosis contrast.** [Original PMID40267506](https://pubmed.ncbi.nlm.nih.gov/40267506/).

**2025 — 2025 AS01-adjuvanted RSV/zoster cohorts suggest a possible immune pathway.** [Original PMID40562756](https://pubmed.ncbi.nlm.nih.gov/40562756/).

**2025 — 2025 randomized hypertension care lowers incident all-cause dementia.** [Original PMID40258956](https://pubmed.ncbi.nlm.nih.gov/40258956/).

**2026 — 2026 Canadian eligibility contrasts extend geography while separate uptake and interaction limits remain.** [Original PMID41579903](https://pubmed.ncbi.nlm.nih.gov/41579903/).

**2026 — 2026 nursing-facility RZV emulation retains residual negative-control confounding.** [Original PMID42296498](https://pubmed.ncbi.nlm.nih.gov/42296498/).

**2026 — 2026 quantitative reanalysis challenges implied causal magnitudes outside Wales.** [Original PMID42754179](https://pubmed.ncbi.nlm.nih.gov/42754179/).

## Research gaps

Resolve implausible international implied receipt effects and weak separate-survey first stages with individual-level local data.

Direct preventive efficacy of recombinant vaccination rather than ordinary vaccinee cohorts.

Biological onset versus delayed clinical diagnosis and competing-death effects.

Prespecified age/product/sex boundaries with formal interaction tests.

Separate infection suppression, heterologous immune effects and healthcare pathways.

Drug-specific clinical prevention rather than class transfer from surrogate or observational successes.

## Boundary experiment

Combine an ethically designed randomized timing/product or vaccine-encouragement comparison within clinically permitted practice with adjudicated incident dementia, active comparators and competing-death analyses. Prespecify product, age and sex, assess baseline cognition and healthcare contacts, and separately reproduce eligibility estimates using individual uptake and counterfactual complier risks. A mechanistic substudy must distinguish viral reactivation from adjuvant immunity without replacing clinical endpoints.


## ADVERSARIAL REVIEW AND APPROVAL

**Authoring model:** Sol 6.1. **Adversarial reviewer:** Claude Opus 5.5 (Anthropic). **Date:** 2026-10-14.

**Scope of checks.** Reference PMIDs verified against PubMed metadata: 53/53 resolve to the stated first author, title, journal and year. Retraction and expression-of-concern screen: none found. Load-bearing claims spot-checked against the primary records: 12, with priority on every effect size, because this review's grade turns on magnitudes. Scoring logic re-derived against the framework reference and SKILL.md, including the upper-ladder clause that names natural experiments as a field-valid equivalent where deliberate human perturbation is not available. Internal consistency checked between AT A GLANCE, the subclaim matrix, the final-score block and the score card (53 sources, 58 claims, Delta with Moderate confidence, canonical descriptor).

**Numeric audit (absolute versus relative, and population).** Every figure checked matched its source exactly, and the review's separation of estimands is the strongest thing about it. The Welsh receipt effect is 3.5 percentage points over seven years, which the source also expresses as a 20.0% relative reduction; the review reports the absolute figure and never substitutes the relative one (PMID 40175543). The Australian discontinuity raised uptake by 16.4 percentage points and reduced new dementia diagnoses by 1.8 percentage points over 7.4 years, and the review correctly labels this an eligibility effect rather than a receipt effect (PMID 40267506). The Canadian clinical effect of about 2 percentage points over 5.5 years is right, and the review flags that uptake came from a separate survey (PMID 41579903). The recombinant-vaccine transition result is "17% more diagnosis-free time and 164 additional days in those subsequently affected", and the review states in two places that this is not a 17% relative risk reduction (PMID 39053634) — that is the single most commonly misquoted figure in this literature and the review gets it right. The Optum hazard ratios are 0.89 for one dose and 0.68 for two (PMID 39733478); the skilled-nursing emulation enrolled 509,926 people of whom 1.73% received a dose, 87.0% of those after discharge, with 57 covariates and negative controls indicating residual confounding (PMID 42296498); the AS01 comparison covers an 18-month risk window with no between-product difference, and the review refuses the source's own adjuvant-necessity inference on the ground that non-significance is not equivalence (PMID 40562756); the cluster-randomized blood-pressure programme covered 326 villages and 33,995 people with a dementia risk ratio of 0.85 over 48 months (PMID 40258956); and the 2026 quantitative critique's implied effectiveness figures (18% in Wales against 177%, 84% and 111% elsewhere) are reported as stated (PMID 42754179). No citation inversion, no population substitution, and no absolute-for-relative conflation was found.

| ID | Severity | Location | Original to corrected | Evidence |
|---|---|---|---|---|
| AR-1 | Major | Throughout the prose review and the score card | 100 fused tokens repaired in the review and 228 in the score card where the builder dropped a space next to a numeral, including numbers the grade depends on: "effect3.5percentage" to "effect 3.5 percentage", "diagnosis1.8points" to "diagnosis 1.8 points", "with95%CI0.0–54.7" to "with 95% CI 0.0–54.7", "despite57covariates" to "despite 57 covariates", "not17%risk" to "not 17% risk", "age80" to "age 80", "uptake16.4points" to "uptake 16.4 points". Leading zeros were restored where the renderer had dropped them ("HRs.89/.68" to "HRs 0.89/0.68", "CI−.40to.27" to "CI −0.40 to 0.27", "RD−.06" to "RD −0.06"), matching the cited sources. Thirty-eight bare seven- and eight-digit reference numbers fused to the preceding word were given their "PMID" prefix, matching the review's own convention elsewhere | The known "randomize3808" rendering-defect class. Here it degraded statistics rather than only prose, which is the reason for the Major rating. Identifier tokens (PMID, PMC, NCT, DOI, study_id, receipt handles, version strings, hashes, URLs, AS01/HSV/HbA1c) were guarded and verified byte-identical afterwards; the whitespace-insensitive comparison confirms that no wording or number was otherwise altered. |
| AR-2 | Minor | "Final score, causal direction and M/S/D", Divergence line; score card divergence justification | Divergence given the artifact the framework requires, naming the two conflicting framings that make a single implied level indefensible | Divergence requires a checkable artifact. "Contested" legitimately has no implied level and no arithmetic, but both the prose and the score card's justification asserted only the absence of one. Value unchanged at Contested. |
| AR-3 | Editorial | Same Divergence fields | "No manufactured field consensus or arithmetic gap" retained after the added artifact | Kept so the original disclaimer is not lost. |
| AR-4 | Major | AT A GLANCE, "Best-supported rival" line | "Local vaccine effect on dementia diagnosis" (the nominal, rank 1) corrected to "Health selection and ascertainment (rank 2)", with a clause saying the nominal ranks first and so is not its own rival | Score-card renderer defect: the "Best-supported rival" line is generated from `competing_models.models_evaluated[0]`, and in this case that entry carries `is_nominal: true` with tournament rank 1, so the prose named the nominal as its own rival. The score card has no separate best-supported-rival field, which is why the defect is prose-only. The body of the paragraph already discussed the genuine rival, so only the label was wrong; the tournament ranking, the level and Rule 9 are untouched. The paragraph's body already argued that the Welsh design is the strongest separator against ordinary healthy-user selection, which is precisely the rank-2 rival, so the corrected label makes the paragraph read as intended. |
| AR-5 | Minor | Score card `final_score.confidence_rationale` | "density, heterogeneous controls and incomplete access set the confidence, not correctness" corrected to "evidence density, independent replication and heterogeneity of controls set the confidence; access depth and belief in correctness do not" | The framework states that confidence "reflects evidence density and reproducibility" and lists what it does not reflect; the score card's boilerplate rationale named "incomplete access" as one of the things that "set the confidence". Access depth is a statement about the reviewer, not about the evidence, so it cannot be a determinant of the band. The rationale now rests on density, independent replication and heterogeneity of controls. The band itself was re-tested against both the Moderate and the High text and is unchanged. Moderate remains correct: the causal core is a small family of eligibility natural experiments with an unresolved published challenge, which is not "extensive independent replication ... reproducibility well-established". |

**Rejected findings (checked, no change).**

1. **The level.** A reduction to Beta was argued seriously. The case: the headline rests on one regression-discontinuity design, a published 2026 analysis concludes the implied vaccine effectiveness behind the non-Welsh natural experiments is internally inconsistent and in two cases above 100%, no population-level dementia decline followed the vaccination programmes, and an unreviewed English analysis finds no hospital-coded effect. Rejected. The framework states that where deliberate human perturbation is unethical or impossible the field-valid equivalents include natural experiments, and Delta requires exactly one field-valid perturbation with a positive result in a phenotype-relevant setting. The Welsh design has a first stage running from 0.01% to 47.2% uptake across one week of birth date, balanced covariates, placebo cutoffs and a positive clinical endpoint; the critique attacks the magnitude and the cross-country consistency rather than demonstrating failure of that local design, and the English analysis is an unreviewed preprint the review deliberately excludes from its source count. That combination is what Delta with Moderate confidence and a Contested divergence describes. Raising to Epsilon was also rejected: the sex boundary rests on a non-significant interaction and the product and age transfers are observational, so no causal failure context is mapped. Flagged for the lead as the review's most contestable judgement.
2. **Aggregation.** Primary-subclaim was pre-registered with SC2 named, and both alternatives are reported (weakest-link Beta, scope-matched Delta). No Epsilon subclaim is involved and the arithmetic checks out.
3. **Collective author.** Reference 50 has a consortium first author and renders correctly as "SPRINT MIND Investigators for the SPRINT Research Group, Williamson JD, Pajewski NM et al." — no leading comma and no dropped author, so the recurring builder defect did not fire here.
4. **Citation years.** References [4], [12], [33] and [49] print a year one later than the PubMed e-pub year. Rejected per the print-versus-e-pub convention.
5. **Momentum.** "Mechanistic only" is correct and its artifact names a trial type ("recent natural experiments and product comparisons") rather than a single study. Thin but compliant, and at a non-modal value; no change.
6. **Empty DOI fields, retrieval handles, "PMID not supplied" strings.** Screened for; none present.

**Residual limitations.** Two corrections in this pass (AR-4, AR-5) came from systematic artifacts reported by the lead mid-pass and were applied after the first three. Several load-bearing figures — the Welsh analysis frame of 282,541 dementia-free participants, the Canadian first-stage estimates of 27.4 and 13.9 percentage points, and the sex-interaction p value of .218 — sit in Methods and Results text rather than abstracts, so this pass could confirm them only against the review's own partial-body receipts, not independently. Access depth is unchanged at 3 partial-body and 50 bounded abstract or status examinations. The shipped validator is version 2.4 and enforces the v5.4 lay-summary requirement; this is a v5.3 package, so it returns one FAIL, the missing lay summary, both before and after this pass, and no lay summary was fabricated to clear it.

**Approval:** Following adversarial review and the corrections listed above, the revised document is approved. — Claude Opus 5.5 (Anthropic), 2026-10-14.

## REFERENCES

1. Eyting M, Xie M, Michalik F et al. A natural experiment on the effect of herpes zoster vaccination on dementia. Nature. 2025. [PMID 40175543](https://pubmed.ncbi.nlm.nih.gov/40175543/). DOI: 10.1038/s41586-025-08800-x. Access: full_text_partial. Study: Welsh_eligibility_SAIL.

2. Pomirchy M, Chung S, Bommer C et al. Herpes zoster vaccination and incident dementia in Canada: an analysis of natural experiments. The Lancet. Neurology. 2026. [PMID 41579903](https://pubmed.ncbi.nlm.nih.gov/41579903/). DOI: 10.1016/S1474-4422(25)00455-7. Access: full_text_partial. Study: Canadian_CPCSSN_CHSS.

3. Hayes KN, Harris DA, McConeghy KW et al. Dementia Risk After Recombinant Herpes Zoster Vaccination in Older Adults With a Recent Skilled-Nursing Facility Stay : A Target Trial Emulation. Annals of internal medicine. 2026. [PMID 42296498](https://pubmed.ncbi.nlm.nih.gov/42296498/). DOI: 10.7326/ANNALS-25-04689. Access: abstract_metadata. Study: PMID42296498.

4. Tang E, Ray I, Arnold BF et al. Recombinant zoster vaccine and the risk of dementia. Vaccine. 2025. [PMID 39733478](https://pubmed.ncbi.nlm.nih.gov/39733478/). DOI: 10.1016/j.vaccine.2024.126673. Access: full_text_partial. Study: Optum_OLDW_possible_overlap.

5. Scherrer J, Salas J, Jacobs C et al. Lower dementia risk in patients vaccinated against herpes zoster. Annals of family medicine. 2022. [PMID 36696249](https://pubmed.ncbi.nlm.nih.gov/36696249/). DOI: 10.1370/afm.20.s1.2680. Access: abstract_metadata. Study: VHA_and_MarketScan.

6. Zheng B, Cao W, Zhang L et al. Association between herpes zoster and cognitive impairment among Chinese community-dwelling elderly: a cross-sectional study. BMC public health. 2026. [PMID 42443821](https://pubmed.ncbi.nlm.nih.gov/42443821/). DOI: 10.1186/s12889-026-28485-9. Access: abstract_metadata. Study: PMID42443821.

7. Chen VC, Wu SI, Huang KY et al. Herpes Zoster and Dementia: A Nationwide Population-Based Cohort Study. The Journal of clinical psychiatry. 2018. [PMID 29244265](https://pubmed.ncbi.nlm.nih.gov/29244265/). DOI: 10.4088/JCP.16m11312. Access: abstract_metadata. Study: PMID29244265.

8. Warren-Gash C, Williamson E, Shiekh SI et al. No evidence that herpes zoster is associated with increased risk of dementia diagnosis. Annals of clinical and translational neurology. 2022. [PMID 35170873](https://pubmed.ncbi.nlm.nih.gov/35170873/). DOI: 10.1002/acn3.51525. Access: abstract_metadata. Study: PMID35170873.

9. Lopatko Lindman K, Hemmingsson ES, Weidung B et al. Herpesvirus infections, antiviral treatment, and the risk of dementia-a registry-based cohort study in Sweden. Alzheimer's & dementia (New York, N. Y.). 2021. [PMID 33614892](https://pubmed.ncbi.nlm.nih.gov/33614892/). DOI: 10.1002/trc2.12119. Access: abstract_metadata. Study: PMID33614892.

10. Israel A, Weizman A, Israel S et al. Atovaquone/proguanil use and zoster vaccination are associated with reduced Alzheimer's disease risk in two cohorts: implications for a latent Toxoplasma gondii mechanism. Brain, behavior, and immunity. 2026. [PMID 41619985](https://pubmed.ncbi.nlm.nih.gov/41619985/). DOI: 10.1016/j.bbi.2026.106473. Access: abstract_metadata. Study: Israel_and_TriNetX_possible_overlap.

11. Weinmann S, Rawlings A, Koppolu P et al. Herpes zoster diagnosis and treatment in relation to incident dementia: A population-based retrospective matched cohort study. PloS one. 2024. [PMID 38271405](https://pubmed.ncbi.nlm.nih.gov/38271405/). DOI: 10.1371/journal.pone.0296957. Access: abstract_metadata. Study: PMID38271405.

12. Bae S, Yun SC, Kim MC et al. Association of herpes zoster with dementia and effect of antiviral therapy on dementia: a population-based cohort study. European archives of psychiatry and clinical neuroscience. 2021. [PMID 32613564](https://pubmed.ncbi.nlm.nih.gov/32613564/). DOI: 10.1007/s00406-020-01157-4. Access: abstract_metadata. Study: PMID32613564.

13. Xie M, Eyting M, Bommer C et al. The effect of shingles vaccination at different stages of the dementia disease course. Cell. 2025. [PMID 41338191](https://pubmed.ncbi.nlm.nih.gov/41338191/). DOI: 10.1016/j.cell.2025.11.007. Access: abstract_metadata. Study: Welsh_eligibility_SAIL.

14. Kivimäki M, Frank P, Livingston G et al. Natural experiments of herpes zoster vaccination: Quantifying the implied vaccine effectiveness against dementia. Brain, behavior, and immunity. 2026. [PMID 42754179](https://pubmed.ncbi.nlm.nih.gov/42754179/). DOI: 10.1016/j.bbi.2026.107021. Access: abstract_metadata. Study: PMID42754179.

15. Pomirchy M, Bommer C, Pradella F et al. Herpes Zoster Vaccination and Dementia Occurrence. JAMA. 2025. [PMID 40267506](https://pubmed.ncbi.nlm.nih.gov/40267506/). DOI: 10.1001/jama.2025.5013. Access: abstract_metadata. Study: Australian_PrimaryCare_eligibility.

16. Taquet M, Dercon Q, Todd JA et al. The recombinant shingles vaccine is associated with lower risk of dementia. Nature medicine. 2024. [PMID 39053634](https://pubmed.ncbi.nlm.nih.gov/39053634/). DOI: 10.1038/s41591-024-03201-5. Access: abstract_metadata. Study: PMID39053634.

17. Levine KS, Leonard HL, Blauwendraat C et al. Virus exposure and neurodegenerative disease risk across national biobanks. Neuron. 2023. [PMID 36669485](https://pubmed.ncbi.nlm.nih.gov/36669485/). DOI: 10.1016/j.neuron.2022.12.029. Access: abstract_metadata. Study: PMID36669485.

18. Polisky V, Littmann M, Triastcyn A et al. Varicella-zoster virus reactivation and the risk of dementia. Nature medicine. 2025. [PMID 41053450](https://pubmed.ncbi.nlm.nih.gov/41053450/). DOI: 10.1038/s41591-025-03972-5. Access: abstract_metadata. Study: US_EHR_TriNetX_possible_overlap.

19. Bukhbinder AS, Ling Y, Hasan O et al. Risk of Alzheimer's Disease Following Influenza Vaccination: A Claims-Based Cohort Study Using Propensity Score Matching. Journal of Alzheimer's disease : JAD. 2022. [PMID 35723106](https://pubmed.ncbi.nlm.nih.gov/35723106/). DOI: 10.3233/JAD-220361. Access: abstract_metadata. Study: PMID35723106.

20. Roh JH, Jung I, Suh Y et al. A potential association between COVID-19 vaccination and development of Alzheimer's disease. QJM : monthly journal of the Association of Physicians. 2024. [PMID 38806183](https://pubmed.ncbi.nlm.nih.gov/38806183/). DOI: 10.1093/qjmed/hcae103. Access: abstract_metadata. Study: PMID38806183.

21. Taquet M, Todd JA, Harrison PJ. Lower risk of dementia with AS01-adjuvanted vaccination against shingles and respiratory syncytial virus infections. NPJ vaccines. 2025. [PMID 40562756](https://pubmed.ncbi.nlm.nih.gov/40562756/). DOI: 10.1038/s41541-025-01172-3. Access: abstract_metadata. Study: PMID40562756.

22. Cairns DM, Itzhaki RF, Kaplan DL. Potential Involvement of Varicella Zoster Virus in Alzheimer's Disease via Reactivation of Quiescent Herpes Simplex Virus Type 1. Journal of Alzheimer's disease : JAD. 2022. [PMID 35754275](https://pubmed.ncbi.nlm.nih.gov/35754275/). DOI: 10.3233/JAD-220287. Access: abstract_metadata. Study: PMID35754275.

23. Rayens E, Sy LS, Qian L et al. Recombinant zoster vaccine is associated with a reduced risk of dementia. Nature communications. 2026. [PMID 41663414](https://pubmed.ncbi.nlm.nih.gov/41663414/). DOI: 10.1038/s41467-026-69289-0. Access: abstract_metadata. Study: PMID41663414.

24. Harris K, Ling Y, Bukhbinder AS et al. The Impact of Routine Vaccinations on Alzheimer's Disease Risk in Persons 65 Years and Older: A Claims-Based Cohort Study using Propensity Score Matching. Journal of Alzheimer's disease : JAD. 2023. [PMID 37574727](https://pubmed.ncbi.nlm.nih.gov/37574727/). DOI: 10.3233/JAD-221231. Access: abstract_metadata. Study: Optum_Clinformatics_possible_overlap.

25. Oh J, Lee K, Yeo D et al. Herpes zoster vaccination and cognitive disorders in older adults. Alzheimer's & dementia : the journal of the Alzheimer's Association. 2026. [PMID 41988837](https://pubmed.ncbi.nlm.nih.gov/41988837/). DOI: 10.1002/alz.71410. Access: abstract_metadata. Study: PMID41988837.

26. Dos Reis S, Tran P, Mohanty K et al. Reduced risk of dementia with recombinant zoster vaccine in US adults age 65 or older. Alzheimer's & dementia : the journal of the Alzheimer's Association. 2026. [PMID 42050365](https://pubmed.ncbi.nlm.nih.gov/42050365/). DOI: 10.1002/alz.71407. Access: abstract_metadata. Study: PMID42050365.

27. Wiemken TL, Salas J, Hoft DF et al. Dementia risk following influenza vaccination in a large veteran cohort. Vaccine. 2021. [PMID 34420785](https://pubmed.ncbi.nlm.nih.gov/34420785/). DOI: 10.1016/j.vaccine.2021.08.046. Access: abstract_metadata. Study: VHA_possible_overlap.

28. Lehrer S, Rheinstein PH. Herpes Zoster Vaccination Reduces Risk of Dementia. In vivo (Athens, Greece). 2021. [PMID 34697158](https://pubmed.ncbi.nlm.nih.gov/34697158/). DOI: 10.21873/invivo.12622. Access: abstract_metadata. Study: PMID34697158.

29. Schnier C, Janbek J, Lathe R et al. Reduced dementia incidence after varicella zoster vaccination in Wales 2013-2020. Alzheimer's & dementia (New York, N. Y.). 2022. [PMID 35434253](https://pubmed.ncbi.nlm.nih.gov/35434253/). DOI: 10.1002/trc2.12293. Access: abstract_metadata. Study: Welsh_population_overlap_possible.

30. Schmidt SAJ, Veres K, Sørensen HT et al. Incident Herpes Zoster and Risk of Dementia: A Population-Based Danish Cohort Study. Neurology. 2022. [PMID 35676090](https://pubmed.ncbi.nlm.nih.gov/35676090/). DOI: 10.1212/WNL.0000000000200709. Access: abstract_metadata. Study: PMID35676090.

31. Ukraintseva S, Yashkin AP, Akushevich I et al. Associations of infections and vaccines with Alzheimer's disease point to a role of compromised immunity rather than specific pathogen in AD. Experimental gerontology. 2024. [PMID 38548241](https://pubmed.ncbi.nlm.nih.gov/38548241/). DOI: 10.1016/j.exger.2024.112411. Access: abstract_metadata. Study: PMID38548241.

32. Luo CS, Chi CC, Fang YA et al. Influenza vaccination reduces dementia in patients with chronic obstructive pulmonary disease: a nationwide cohort study. Journal of investigative medicine : the official publication of the American Federation for Clinical Research. 2020. [PMID 31941664](https://pubmed.ncbi.nlm.nih.gov/31941664/). DOI: 10.1136/jim-2019-001155. Access: abstract_metadata. Study: PMID31941664.

33. Barry EVH, Suffel AM, Walker J et al. Disparities in uptake of Shingrix® vaccine in immunosuppressed individuals in England: a population-based cohort study. Vaccine. 2026. [PMID 41435736](https://pubmed.ncbi.nlm.nih.gov/41435736/). DOI: 10.1016/j.vaccine.2025.128078. Access: abstract_metadata. Study: PMID41435736.

34. Choi HG, Park BJ, Lim JS et al. Herpes Zoster Does Not Increase the Risk of Neurodegenerative Dementia: A Case-Control Study. American journal of Alzheimer's disease and other dementias. 2021. [PMID 33882722](https://pubmed.ncbi.nlm.nih.gov/33882722/). DOI: 10.1177/15333175211006504. Access: abstract_metadata. Study: PMID33882722.

35. Liu JC, Hsu YP, Kao PF et al. Influenza Vaccination Reduces Dementia Risk in Chronic Kidney Disease Patients: A Population-Based Cohort Study. Medicine. 2016. [PMID 26945371](https://pubmed.ncbi.nlm.nih.gov/26945371/). DOI: 10.1097/MD.0000000000002868. Access: abstract_metadata. Study: PMID26945371.

36. Shim Y, Park M, Kim J. Increased incidence of dementia following herpesvirus infection in the Korean population. Medicine. 2022. [PMID 36254002](https://pubmed.ncbi.nlm.nih.gov/36254002/). DOI: 10.1097/MD.0000000000031116. Access: abstract_metadata. Study: PMID36254002.

37. Walker VM, Kehoe PG, Martin RM et al. Repurposing antihypertensive drugs for the prevention of Alzheimer's disease: a Mendelian randomization study. International journal of epidemiology. 2020. [PMID 31335937](https://pubmed.ncbi.nlm.nih.gov/31335937/). DOI: 10.1093/ije/dyz155. Access: abstract_metadata. Study: PMID31335937.

38. Taylor-Bateman V, Bothongo P, Walker V et al. Repurposing drugs for the prevention of vascular dementia using evidence from drug target Mendelian randomization. Nature aging. 2026. [PMID 42009889](https://pubmed.ncbi.nlm.nih.gov/42009889/). DOI: 10.1038/s43587-026-01106-1. Access: abstract_metadata. Study: PMID42009889.

39. Berger S, Moseholm KF, Hegelund ER et al. Association of Tumor Necrosis Factor-α Inhibitors with Incident Dementia: Analysis Based on Population-Based Cohort Studies. Drugs & aging. 2024. [PMID 38609734](https://pubmed.ncbi.nlm.nih.gov/38609734/). DOI: 10.1007/s40266-024-01112-1. Access: abstract_metadata. Study: PMID38609734.

40. Rosell-Díaz M, Petit-Gay A, Molas-Prat C et al. Metformin-induced changes in the gut microbiome and plasma metabolome are associated with cognition in men. Metabolism: clinical and experimental. 2024. [PMID 38871078](https://pubmed.ncbi.nlm.nih.gov/38871078/). DOI: 10.1016/j.metabol.2024.155941. Access: abstract_metadata. Study: PMID38871078.

41. He J, Zhao C, Zhong S et al. Blood pressure reduction and all-cause dementia in people with uncontrolled hypertension: an open-label, blinded-endpoint, cluster-randomized trial. Nature medicine. 2025. [PMID 40258956](https://pubmed.ncbi.nlm.nih.gov/40258956/). DOI: 10.1038/s41591-025-03616-8. Access: abstract_metadata. Study: PMID40258956.

42. Zhou Z, Ryan J, Ernst ME et al. Effect of Statin Therapy on Cognitive Decline and Incident Dementia in Older Adults. Journal of the American College of Cardiology. 2021. [PMID 34167639](https://pubmed.ncbi.nlm.nih.gov/34167639/). DOI: 10.1016/j.jacc.2021.04.075. Access: abstract_metadata. Study: ASPREE.

43. Jing B, Liu X, Graham LA et al. Deprescribing of Antihypertensive Medications and Cognitive Function in Nursing Home Residents. JAMA internal medicine. 2024. [PMID 39312220](https://pubmed.ncbi.nlm.nih.gov/39312220/). DOI: 10.1001/jamainternmed.2024.4851. Access: abstract_metadata. Study: PMID39312220.

44. Luchsinger JA, Rosin SP, Kazemi EJ et al. Glucose-Lowering Medications, Glycemia, and Cognitive Outcomes: The GRADE Randomized Clinical Trial. JAMA internal medicine. 2025. [PMID 40388190](https://pubmed.ncbi.nlm.nih.gov/40388190/). DOI: 10.1001/jamainternmed.2025.1189. Access: abstract_metadata. Study: PMID40388190.

45. Webb AJS, Birks JS, Feakins KA et al. Cerebrovascular Effects of Sildenafil in Small Vessel Disease: The OxHARP Trial. Circulation research. 2024. [PMID 38832504](https://pubmed.ncbi.nlm.nih.gov/38832504/). DOI: 10.1161/CIRCRESAHA.124.324327. Access: abstract_metadata. Study: PMID38832504.

46. Wang W, Davis PB, Qi X et al. Associations of semaglutide with Alzheimer's disease-related dementias in patients with type 2 diabetes: A real-world target trial emulation study. Journal of Alzheimer's disease : JAD. 2025. [PMID 40552638](https://pubmed.ncbi.nlm.nih.gov/40552638/). DOI: 10.1177/13872877251351329. Access: abstract_metadata. Study: PMID40552638.

47. Lai Y, Her QL, Zhang Y et al. The heterogeneous treatment effects of statins on dementia: a target trial emulation with causal machine learning using integrated genetic and real-world data. Alzheimer's & dementia : the journal of the Alzheimer's Association. 2026. [PMID 41748529](https://pubmed.ncbi.nlm.nih.gov/41748529/). DOI: 10.1002/alz.71178. Access: abstract_metadata. Study: PMID41748529.

48. Inoue K, Saliba D, Gotanda H et al. Glucagon-Like Peptide-1 Receptor Agonists and Incidence of Dementia Among Older Adults With Type 2 Diabetes : A Target Trial Emulation. Annals of internal medicine. 2025. [PMID 40690769](https://pubmed.ncbi.nlm.nih.gov/40690769/). DOI: 10.7326/ANNALS-24-02648. Access: abstract_metadata. Study: PMID40690769.

49. Schroevers JL, Eggink E, Hoevenaar-Blom MP et al. Antihypertensive medication classes and the risk of dementia over a decade of follow-up. Journal of hypertension. 2023. [PMID 36394298](https://pubmed.ncbi.nlm.nih.gov/36394298/). DOI: 10.1097/HJH.0000000000003324. Access: abstract_metadata. Study: PMID36394298.

50. SPRINT MIND Investigators for the SPRINT Research Group, Williamson JD, Pajewski NM et al. Effect of Intensive vs Standard Blood Pressure Control on Probable Dementia: A Randomized Clinical Trial. JAMA. 2019. [PMID 30688979](https://pubmed.ncbi.nlm.nih.gov/30688979/). DOI: 10.1001/jama.2018.21442. Access: abstract_metadata. Study: PMID30688979.

51. Forette F, Seux ML, Staessen JA et al. Prevention of dementia in randomised double-blind placebo-controlled Systolic Hypertension in Europe (Syst-Eur) trial. Lancet (London, England). 1998. [PMID 9802273](https://pubmed.ncbi.nlm.nih.gov/9802273/). DOI: 10.1016/s0140-6736(98)03086-4. Access: abstract_metadata. Study: PMID9802273.

52. Peters R, Beckett N, Forette F et al. Incident dementia and blood pressure lowering in the Hypertension in the Very Elderly Trial cognitive function assessment (HYVET-COG): a double-blind, placebo controlled trial. The Lancet. Neurology. 2008. [PMID 18614402](https://pubmed.ncbi.nlm.nih.gov/18614402/). DOI: 10.1016/S1474-4422(08)70143-1. Access: abstract_metadata. Study: PMID18614402.

53. Ryan J, Storey E, Murray AM et al. Randomized placebo-controlled trial of the effects of aspirin on dementia and cognitive decline. Neurology. 2020. [PMID 32213642](https://pubmed.ncbi.nlm.nih.gov/32213642/). DOI: 10.1212/WNL.0000000000009277. Access: abstract_metadata. Study: ASPREE.

## Source-specific evidence cards

### [1] A natural experiment on the effect of herpes zoster vaccination on dementia.

**Shows:** Welsh birth-date eligibility discontinuity identifies a local live-vaccine effect on new dementia diagnosis: reported receipt effect 3.5 percentage points over 7 years.

**Does not show:** CACE requires continuity, exclusion and monotonicity; diagnoses are not biological onset, age-near 80/product-specific and exploratory mediation is not necessity.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. Selected body sections: 6:Explorations of the effect mechanism:paragraph4, 6:Explorations of the effect mechanism:paragraph8, 6:Explorations of the effect mechanism:paragraph10, 6:Explorations of the effect mechanism:paragraph12, 8:Methods:paragraph6, 8:Methods:paragraph7, 8:Methods:paragraph8, 8:Methods:paragraph12, 8:Methods:paragraph14, 8:Methods:paragraph15. No complete-paper or supplement audit claimed. UpdateOf: PMID37292746. Subclaims: SC1, SC2, SC3.

### [2] Herpes zoster vaccination and incident dementia in Canada: an analysis of natural experiments.

**Shows:** Ontario eligibility thresholds associate with 2.0 percentage-point fewer dementia diagnoses over 5.5 years, with interprovincial trend corroboration.

**Does not show:** Eligibility is not receipt; uptake comes from a separate cross-sectional survey, secondary first stage includes zero and sex interaction P=.218 does not establish sex-specific effect.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. Selected body sections: 1:Methods:paragraph2, 1:Methods:paragraph4, 1:Methods:paragraph5, 2:Results:paragraph1, 2:Results:paragraph3, 2:Results:paragraph5, 2:Results:paragraph7. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2, SC3.

### [3] Dementia Risk After Recombinant Herpes Zoster Vaccination in Older Adults With a Recent Skilled-Nursing Facility Stay : A Target Trial Emulation.

**Shows:** Clone-censor-weight emulation in 509926 nursing-facility entrants associates RZV with lower 4 year dementia risk.

**Does not show:** Only 1.73% vaccinated, most after discharge; negative controls suggest residual confounding despite 57 covariates and targeted emulation.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2.

### [4] Recombinant zoster vaccine and the risk of dementia.

**Shows:** Optum claims cohort associates time-updated one/two RZV doses with dementia HR 0.89/0.68 and antiviral use with lower hazards.

**Does not show:** Individuals contribute person-time by dose; covariate adjustment and dose response do not remove health-seeking confounding or establish infection mediation; reported NNT is observational.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. Selected body sections: 1:Methods:paragraph2, 1:Methods:paragraph4, 1:Methods:paragraph6, 2:Results:paragraph3, 2:Results:paragraph4. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2, SC3.

### [5] Lower dementia risk in patients vaccinated against herpes zoster.

**Shows:** VHA and MarketScan cohorts with preventive visits associate HZ vaccination with 31–35% lower dementia hazard.

**Does not show:** Well-visit restriction and E-values do not prove absence of unmeasured bias; possible database overlap with other reports and product-specificity remain.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2, SC3.

### [6] Association between herpes zoster and cognitive impairment among Chinese community-dwelling elderly: a cross-sectional study.

**Shows:** Chinese cross-sectional 1233-person study associates past HZ with education-adjusted cognitive impairment; antiviral estimate is nonsignificant.

**Does not show:** Questionnaire history and simultaneous cognitive testing cannot establish temporality, diagnosed dementia prevention or antiviral efficacy.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3.

### [7] Herpes Zoster and Dementia: A Nationwide Population-Based Cohort Study.

**Shows:** Taiwan matched cohort associates HZ with HR 1.11 and antiviral prescriptions with lower dementia hazard.

**Does not show:** Treatment selection and diagnosis codes leave confounding; antiherpetic association is not randomized prevention or VZV necessity.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3, SC4.

### [8] No evidence that herpes zoster is associated with increased risk of dementia diagnosis.

**Shows:** UK CPRD cohort finds no increased dementia hazard after HZ, with null dementia-or-death composite and no antiviral variation.

**Does not show:** This challenges universal clinical-shingles mediation, while subclinical reactivation and vaccine-specific immune effects are not excluded.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3.

### [9] Herpesvirus infections, antiviral treatment, and the risk of dementia-a registry-based cohort study in Sweden.

**Shows:** Swedish registry associates overt HSV/VZV without treatment with greater dementia and antivirals with HR.89.

**Does not show:** Mixed herpes exposure, treatment indication and registry confounding do not isolate a necessary VZV path.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3, SC4.

### [10] Atovaquone/proguanil use and zoster vaccination are associated with reduced Alzheimer's disease risk in two cohorts: implications for a latent Toxoplasma gondii mechanism.

**Shows:** Israeli drug-screen associations for atovaquone/proguanil and VZV vaccines recur in matched TriNetX cohorts.

**Does not show:** Medication-wide discovery, travel/health-user selection and exploratory interaction do not establish latent Toxoplasma mechanism or drug benefit.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC3, SC4.

### [11] Herpes zoster diagnosis and treatment in relation to incident dementia: A population-based retrospective matched cohort study.

**Shows:** Kaiser matched cohort finds HZ-dementia HR.99; antiviral association attenuates to.88 with no duration/cumulative-dose trend.

**Does not show:** Null HZ association and unstable subgroup treatment signals limit universal mediation; treatment is not randomized.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3, SC4.

### [12] Association of herpes zoster with dementia and effect of antiviral therapy on dementia: a population-based cohort study.

**Shows:** Korean cohort associates HZ with adjusted HR 1.12 and antiviral-treated HZ with lower dementia hazard after matching.

**Does not show:** Shared national data platform can overlap other infection papers; measured treatment adjustment does not randomize therapy.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3, SC4.

### [13] The effect of shingles vaccination at different stages of the dementia disease course.

**Shows:** Welsh eligibility analysis extends signals to MCI and dementia-related deaths among people already diagnosed.

**Does not show:** Same eligibility/cohort as original study, not independent replication; progression and MCI are different scopes from dementia-free incident prevention.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC2, SC3.

### [14] Natural experiments of herpes zoster vaccination: Quantifying the implied vaccine effectiveness against dementia.

**Shows:** New quantitative reanalysis of published natural experiments reports implausibly large receipt-effect implications outside Wales and no corresponding national incidence drop.

**Does not show:** Aggregate uptake and comparator-risk assumptions do not directly reanalyse individual local compliers; critique must be resolved, not treated as automatic refutation of every eligibility result.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC2.

### [15] Herpes Zoster Vaccination and Dementia Occurrence.

**Shows:** Australian age 80 eligibility discontinuity raises uptake 16.4 points and reduces new dementia diagnosis 1.8 points over 7.4 years.

**Does not show:** Abstract reports eligibility effect, not receipt relative risk; local cutoff population differs from overall sample mean age and underlying vaccine-effect critique remains.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. UpdateOf: PMID38978672. Subclaims: SC1, SC2.

### [16] The recombinant shingles vaccine is associated with lower risk of dementia.

**Shows:** Rapid product-transition comparison associates RZV with 17% more dementia-diagnosis-free time and 164 days among subsequently affected.

**Does not show:** Calendar-era treatment transition, measured matching and vaccine comparators are not randomized allocation; relative diagnosis-free time is not 17% risk reduction.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2, SC3.

### [17] Virus exposure and neurodegenerative disease risk across national biobanks.

**Shows:** FinnGen discovery and UKBiobank replication identify viral-exposure associations with later neurodegeneration.

**Does not show:** Hospital-coded infection severity, ascertainment and immune vulnerability remain; cross-disease association does not isolate VZV vaccination causality.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3.

### [18] Varicella-zoster virus reactivation and the risk of dementia.

**Shows:** Large EHR analysis links recurrent HZ, vaccine dose and waning patterns to dementia associations after extensive adjustment.

**Does not show:** Temporal correlation of protection against HZ and dementia is not randomized mediator dissociation; shared TriNetX networks constrain independent-line counts.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC3.

### [19] Risk of Alzheimer's Disease Following Influenza Vaccination: A Claims-Based Cohort Study Using Propensity Score Matching.

**Shows:** Propensity-matched US claims associate influenza vaccination with lower 4 year AD diagnoses.

**Does not show:** Matched risk differences and calculated NNT are observational, not established clinical prevention or replication of the same zoster mechanism.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [20] A potential association between COVID-19 vaccination and development of Alzheimer's disease.

**Shows:** Korean cohort associates COVID mRNA vaccination with more AD/MCI diagnoses within 3 months.

**Does not show:** Short latency and differential contact/diagnostic ascertainment cannot establish vaccine-caused neurodegeneration; a different product does not refute zoster effects.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [21] Lower risk of dementia with AS01-adjuvanted vaccination against shingles and respiratory syncytial virus infections.

**Shows:** Matched cohorts associate both AS01-adjuvanted RSV and zoster vaccination with lower 18 month dementia risk, with no between-product difference.

**Does not show:** Nonsignificance is not equivalence; shared adjuvant, healthy uptake and selection cannot isolate AS01 necessity or direct adjuvant efficacy.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3, SC4.

### [22] Potential Involvement of Varicella Zoster Virus in Alzheimer's Disease via Reactivation of Quiescent Herpes Simplex Virus Type 1.

**Shows:** VZV infection of quiescent HSV1 neural cultures reactivates HSV1 and AD-like amyloid/phosphotau changes; VZV alone lacks those main markers.

**Does not show:** Cell-marker phenotype is not human dementia or a necessary vaccine mediator; forced viral exposures retain system limits.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3.

### [23] Recombinant zoster vaccine is associated with a reduced risk of dementia.

**Shows:** Kaiser RZV two-dose cohort has lower dementia hazard, attenuated from.49 versus unvaccinated to.73 versus Tdap.

**Does not show:** Vaccine active comparator reduces but does not eliminate bias; excluding early post-dose dementia changes eligibility/estimand, not randomized prevention.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2.

### [24] The Impact of Routine Vaccinations on Alzheimer's Disease Risk in Persons 65 Years and Older: A Claims-Based Cohort Study using Propensity Score Matching.

**Shows:** Optum matched Tdap/HZ/pneumococcal cohorts show lower AD risks over 8 years.

**Does not show:** Multiple protective vaccine associations are compatible with healthy-user or immune effects; common source population is not independent causation.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC4.

### [25] Herpes zoster vaccination and cognitive disorders in older adults.

**Shows:** Korean weighted live-vaccine cohort shows lower memory-disorder and AD hazards with modest restricted-mean-time gains.

**Does not show:** Initial 2.5 million frame differs from 1041964 analytic sample; dose/time and behaviour associations are not randomized mediation.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1.

### [26] Reduced risk of dementia with recombinant zoster vaccine in US adults age 65 or older.

**Shows:** Medicare matched two-dose RZV cohorts show lower dementia/AD/vascular-dementia hazards across follow-up periods.

**Does not show:** Preventive-visit comparator and weights do not remove latent prodromal decline or healthcare-contact differences.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2.

### [27] Dementia risk following influenza vaccination in a large veteran cohort.

**Shows:** VHA influenza cohort associates repeated vaccination with lower dementia hazard, strongest for at least 6 vaccinations.

**Does not show:** Repeated adherence can reflect healthy-user selection and opportunity time; national VHA observations may overlap zoster reports.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [28] Herpes Zoster Vaccination Reduces Risk of Dementia.

**Shows:** BRFSS survey links HZ vaccination with less reported social restriction from disorientation or memory loss.

**Does not show:** Self-reported cross-sectional social symptoms are not incident diagnosed dementia despite causal title wording; small selected sample.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1.

### [29] Reduced dementia incidence after varicella zoster vaccination in Wales 2013-2020.

**Shows:** Welsh vaccinated-versus-unvaccinated cohort has lower dementia and several other disease/mortality hazards without HZ-diagnosis mediation.

**Does not show:** Authors explicitly retain healthy-life-expectancy selection bias; same national setting overlaps later eligibility analyses.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2, SC3.

### [30] Incident Herpes Zoster and Risk of Dementia: A Population-Based Danish Cohort Study.

**Shows:** Danish long-term matched HZ cohort finds no increased dementia overall; rare CNS subgroup estimate is imprecise.

**Does not show:** Challenges overt-shingles universal mediation, not all subclinical immune pathways; CNS-attributable fraction is a model, not vaccine effect.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3.

### [31] Associations of infections and vaccines with Alzheimer's disease point to a role of compromised immunity rather than specific pathogen in AD.

**Shows:** HRS sample associates multiple infections and adult vaccines with later AD more than other dementias.

**Does not show:** Pseudorandomized sampling is not randomized vaccination; common immune vulnerability and exposure ascertainment remain.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. UpdateOf: PMID38106098. Subclaims: SC1, SC3, SC4.

### [32] Influenza vaccination reduces dementia in patients with chronic obstructive pulmonary disease: a nationwide cohort study.

**Shows:** Taiwan COPD cohort associates influenza vaccination frequency with lower dementia hazard.

**Does not show:** Chronic-disease selection, adherence opportunity and residual bias preclude clinical causal or universal-vaccine inference.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [33] Disparities in uptake of Shingrix® vaccine in immunosuppressed individuals in England: a population-based cohort study.

**Shows:** English immunosuppressed-adult cohort finds substantially lower RZV uptake in dementia, care homes and deprivation.

**Does not show:** Different clinical population and descriptive uptake do not estimate prevention, but demonstrate a plausible selection path for ordinary cohorts.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC1, SC2.

### [34] Herpes Zoster Does Not Increase the Risk of Neurodegenerative Dementia: A Case-Control Study.

**Shows:** Korean case-control histories show no positive HZ-neurodegenerative-dementia association.

**Does not show:** Retrospective coding and shared national platform constrain independence; inverse association is not proof HZ protects.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3.

### [35] Influenza Vaccination Reduces Dementia Risk in Chronic Kidney Disease Patients: A Population-Based Cohort Study.

**Shows:** Taiwan CKD cohort associates influenza vaccination with lower dementia in both influenza and noninfluenza seasons.

**Does not show:** Stronger off-season association permits nonspecific effects or residual selection; recorded vaccines and dementia remain observational.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [36] Increased incidence of dementia following herpesvirus infection in the Korean population.

**Shows:** Korean HSV/VZV cohorts show modest adjusted increased dementia hazards, particularly AD.

**Does not show:** Shared registry and mixed virus subtypes do not establish mediation or explain conflicting UK/Danish results.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC3.

### [37] Repurposing antihypertensive drugs for the prevention of Alzheimer's disease: a Mendelian randomization study.

**Shows:** Antihypertensive-target MR finds limited evidence lowering systolic pressure affects AD risk.

**Does not show:** Lifelong instruments, exclusion assumptions and AD subtype differ from clinical BP-program all-dementia endpoints; null does not erase randomized prevention.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [38] Repurposing drugs for the prevention of vascular dementia using evidence from drug target Mendelian randomization.

**Shows:** Drug-target VaD MR is largely null after multiple-testing correction, with tentative beta1/ACE signals.

**Does not show:** Tentative target proxies and imaging surrogates are not clinical drug efficacy or established pharmacovigilance harm.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [39] Association of Tumor Necrosis Factor-α Inhibitors with Incident Dementia: Analysis Based on Population-Based Cohort Studies.

**Shows:** Danish/German rheumatology cohorts have nonsignificant lower dementia estimates for TNF inhibitors.

**Does not show:** Clinical-indication confounding and wide confidence intervals leave no demonstrated preventive benefit; no equivalence conclusion.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [40] Metformin-induced changes in the gut microbiome and plasma metabolome are associated with cognition in men.

**Shows:** Observational plus small randomized metformin cohorts show microbiome/metabolome changes and male-stratified memory correlations.

**Does not show:** Whole metformin-treated cohort ratio-cognition association is null; biomarker/correlation and sex stratification do not establish incident dementia prevention.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [41] Blood pressure reduction and all-cause dementia in people with uncontrolled hypertension: an open-label, blinded-endpoint, cluster-randomized trial.

**Shows:** 326-village randomized BP-care programme reduces all-cause dementia over 48 months in 33995 hypertensive adults.

**Does not show:** A genuine clinical prevention positive for its assigned care package, not proof of each antihypertensive molecule or the same vaccine mediator.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [42] Effect of Statin Therapy on Cognitive Decline and Incident Dementia in Older Adults.

**Shows:** Baseline statin use in ASPREE is not associated with dementia, MCI or cognitive decline.

**Does not show:** Statin exposure was observational even though parent aspirin trial was randomized; subgroup interaction does not create drug assignment.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [43] Deprescribing of Antihypertensive Medications and Cognitive Function in Nursing Home Residents.

**Shows:** VA target-trial emulation associates antihypertensive deprescribing with less cognitive-category decline in nursing-home residents.

**Does not show:** Frail baseline setting and many existing dementia cases differ from incident prevention; medication change selection remains.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [44] Glucose-Lowering Medications, Glycemia, and Cognitive Outcomes: The GRADE Randomized Clinical Trial.

**Shows:** GRADE randomized second-line diabetes agents show no cognitive-test differences over 4.1 years; glycemic exposure correlates with tests.

**Does not show:** No diagnosed dementia endpoint or placebo metformin contrast; association with HbA1c is not randomized glucose-mediator efficacy.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [45] Cerebrovascular Effects of Sildenafil in Small Vessel Disease: The OxHARP Trial.

**Shows:** OxHARP randomized crossover sildenafil has null primary cerebral pulsatility but better secondary perfusion/reactivity.

**Does not show:** Three-week surrogate benefit is not dementia prevention; crossover participants and secondary testing do not yield independent causal lines.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [46] Associations of semaglutide with Alzheimer's disease-related dementias in patients with type 2 diabetes: A real-world target trial emulation study.

**Shows:** US semaglutide target emulation associates use with lower ADRD than several diabetic comparators.

**Does not show:** Confounding by indication, metabolic improvement and diagnosis remain; FTD/Lewy-body nulls prevent a uniform effect across dementias.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [47] The heterogeneous treatment effects of statins on dementia: a target trial emulation with causal machine learning using integrated genetic and real-world data.

**Shows:** Statin emulation shows null overall 5 year dementia difference and exploratory genotype-defined subgroup signals.

**Does not show:** Data-derived treatment heterogeneity and all-dementia interval crossing zero do not validate a beneficiary group or randomized effect.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [48] Glucagon-Like Peptide-1 Receptor Agonists and Incidence of Dementia Among Older Adults With Type 2 Diabetes : A Target Trial Emulation.

**Shows:** Medicare GLP1-versus-DPP4 target emulation finds no clear overall dementia difference, with age-varying estimates.

**Does not show:** Wide compatible effects, unmeasured BMI/glycemia/duration and short follow-up limit efficacy or age-boundary inference.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [49] Antihypertensive medication classes and the risk of dementia over a decade of follow-up.

**Shows:** preDIVA antihypertensive-class associations attenuate by 10.4 years, with confidence intervals crossing null.

**Does not show:** Post-hoc medication exposures are not randomized drug classes; changing age/adherence and competing risks constrain attribution.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [50] Effect of Intensive vs Standard Blood Pressure Control on Probable Dementia: A Randomized Clinical Trial.

**Shows:** SPRINT intensive BP assignment lowers MCI and composite cognitive impairment but not significant probable dementia.

**Does not show:** Early stopping and fewer dementia events limit power; positive composite must not replace null registered dementia endpoint.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [51] Prevention of dementia in randomised double-blind placebo-controlled Systolic Hypertension in Europe (Syst-Eur) trial.

**Shows:** SystEur randomized BP regimen reports 21 versus 11 dementia cases over 2 years atP=.05.

**Does not show:** Small borderline event set and multi-agent regimen do not establish each drug, longer horizon or zoster-equivalent mediation.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [52] Incident dementia and blood pressure lowering in the Hypertension in the Very Elderly Trial cognitive function assessment (HYVET-COG): a double-blind, placebo controlled trial.

**Shows:** HYVET original 3336-person comparison lowersBP without significant incident dementia benefit.

**Does not show:** Early stopping/short follow-up and confidence interval permit benefit; pooled meta-component is not counted as another primary experiment.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.

### [53] Randomized placebo-controlled trial of the effects of aspirin on dementia and cognitive decline.

**Shows:** ASPREE randomized aspirin in 19114 older adults shows no dementia/MCI/cognitive benefit over 4.7 years.

**Does not show:** No class-wide preventive drug conclusion follows; statin analyses share cohort but aspirin assignment is distinct.

**Access/status:** Bounded original primary abstract Results/Findings/conclusion passages and linked-status metadata examined; no complete-abstract audit asserted. No complete-paper or supplement audit claimed. none_found. Subclaims: SC4.