The Vaccine–Brain Connection

The vaccine that may help protect your brain — and the system that can’t say so

A vaccine already in your GP’s fridge. A growing body of evidence. And a structural silence that keeps both apart.

There is a sentence buried in a GSK press release from July 2025, issued while the company was presenting real-world data at the world’s largest Alzheimer’s conference, that captures this entire story in sixteen words:

GSK was simultaneously presenting evidence of an association between their vaccine and reduced dementia risk, while being legally and regulatorily required to tell you that the vaccine is not approved for that purpose. Both statements are true. Neither one cancels the other out. And together, they define exactly the problem this post is about.

If you have been following this series, you have seen the evidence build across three posts: the natural experiments from Wales, Australia, and the United States;^1,2,3 the adjuvant hypothesis;⁴ the striking and under-reported signal in women.^2,5 The evidence is not perfect — no epidemiological evidence ever is — but it is substantial, convergent, and increasingly difficult to attribute entirely to residual confounding. And yet almost none of it has translated into a clinical conversation that routinely reaches older adults.

This final post asks two questions. First, why not? And second: now that you know, what should you do?

---

The regulatory paradox

Regulatory agencies approve medicines and vaccines for specific, rigorously defined indications. The FDA label for Shingrix covers two populations: adults aged 50 and older, and adults aged 18 and older who are at increased risk of shingles due to immunodeficiency or immunosuppression caused by known disease or therapy.⁶ Dementia prevention appears nowhere in that label. No neurological benefit. No brain health language.

This is not an oversight. It reflects how the regulatory system works. An indication requires evidence from prospective clinical trials designed to test that specific outcome. The randomised controlled trials that underpinned Shingrix’s approval were designed, as they should have been, to measure shingles prevention. Dementia was not a pre-specified endpoint. It could not have been: the shingles trials were designed years before the dementia signal was identified.

The result is a structural asymmetry. The pharmaceutical company, regulatory agencies, and health systems are constrained in how they communicate a potential benefit that independent researchers are actively publishing. That constraint is not dishonest — it is legally and procedurally required. But it leaves an information gap, and information gaps do not stay empty.

---

Who fills the gap — and the limits of each

When official channels are silent, other voices fill the space. In the case of the shingles-dementia evidence, those voices include academic researchers, science journalists, independent science communicators — and, on the other side, people with a commercial or ideological interest in undermining vaccine confidence.

The research community has done its job. The Welsh natural experiment was published in Nature.¹ The recombinant vaccine analysis was published in Nature Medicine.³ The full disease-course paper appeared in Cell.⁵ These are flagship journals with stringent peer review. The Gavi Vaccine Alliance published a summary in April 2026 noting eight vaccines now linked to lower dementia risk.⁸ Eric Topol devoted a detailed Substack analysis to it. The evidence is not hidden.

But research publication is not the same as clinical reach. A paper in Nature is read by researchers. It is rarely read by the 75-year-old woman in the waiting room who last had her shingles vaccine discussion three years ago and has since developed a mild cognitive complaint she hasn’t mentioned to anyone.

The counter-narrative problem is harder still. The same information gap that allows credible science communicators to share this evidence also allows bad actors to construct a mirror image. When regulatory agencies are silent on a vaccine benefit, the silence can be weaponised: if this were real, they would tell you. The absence of official endorsement becomes, in the wrong hands, evidence of suppression. The antidote is not more silence — it is more calibrated communication from credible sources.

---

How certain is certain enough?

This is the most important question in the post, and it deserves a science-based answer rather than a diplomatic one.

The evidence is observational. Causality has not been established by a randomised controlled trial. That is an important distinction. It is worth being precise about why, and equally precise about why that caveat does not mean the evidence can be dismissed.

The case for caution

Healthy vaccinee bias is the central threat to this literature. People who receive vaccines are not, on average, identical to people who do not. They tend to be more health-engaged, better-resourced, more likely to attend preventive care appointments, and on average healthier. Any or all of these factors could independently reduce dementia risk. If the association between shingles vaccination and dementia is partly or entirely explained by who gets vaccinated (rather than by the vaccination itself), then the apparent benefit is illusory.

This bias cannot be fully eliminated in observational data, no matter how sophisticated the statistical adjustment. It can be reduced; it cannot be zeroed out. This is not a minor caveat — it is why the authors of the most careful studies have been scrupulous about language, consistently writing “associated with” rather than “prevents.”

There is also the question of follow-up. Most studies have followed participants for five to seven years. Dementia is a disease with a decades-long preclinical trajectory. We do not yet know whether the apparent protection persists, attenuates, or disappears at ten or fifteen years of follow-up.

The honest case against waiting indefinitely for RCT certainty

The natural experiments that anchor this evidence base were specifically designed to address healthy vaccinee bias — not to adjust for it statistically, but to structurally eliminate it. The Welsh programme’s birth-date eligibility cutoff means that people born one week apart in age were either eligible or ineligible for the vaccine — not because of any choice, health behaviour, or characteristic of their own, but because of an arbitrary administrative date.¹ As Eyting and colleagues wrote, a confounding variable can bias their analysis only if it changes abruptly at the September 2, 1933 date-of-birth threshold — a near-impossible condition for any plausible confounder to meet.

The same logic was independently replicated in Australia, where a different age-based eligibility cutoff at programme launch in November 2016 produced a structurally identical natural experiment in a different health system.¹⁰ The Wales and Australia findings are concordant. The Geldsetzer group subsequently extended this approach to Ontario, Canada, with findings published in The Lancet Neurology in February 2026 — a third independent replication in a third health system.¹¹ The convergence across independent populations, health systems, and programme designs is not easy to explain by confounding.

The December 2025 Cell paper added another dimension: HZ vaccination was associated not only with reduced dementia diagnoses but also with reduced mild cognitive impairment diagnoses and — among people already living with dementia — reduced dementia-related mortality.⁵ The authors are explicit that confirmatory natural experiments, not conventional observational studies, are the appropriate next step precisely because standard associational analyses remain vulnerable to healthy vaccinee bias. This is responsible science: the researchers themselves are calling for the highest-quality replication their methods can provide.

And then there is the RCT question itself. A simple placebo-controlled trial of Shingrix with dementia as the primary endpoint would be difficult to justify ethically in settings where it is standard recommended care, since participants randomised to placebo would be withheld a vaccine with proven shingles prevention benefit. Alternative designs would be required: active comparators, adaptive methods, or trials in populations where uptake remains genuinely low. Pascal Geldsetzer’s group at Stanford is seeking philanthropic funding for an RCT of the older live-attenuated vaccine (now off-patent), which avoids this problem — but even that trial, if funded and completed, would address a discontinued product rather than current clinical practice.¹²

This is not an argument for abandoning epistemic standards. It is an argument for applying them proportionately. Public health regularly acts on convergent observational evidence when the intervention carries an established safety profile, an existing indication, and a favourable benefit-risk ratio. Screening programmes, dietary recommendations, and occupational exposure limits are routinely set on observational evidence. The standard we apply should be consistent — and acknowledged for what it is.

---

The broader signal — vaccines beyond shingles

The shingles vaccine story would be noteworthy in isolation. But one of the most important developments in this space over the past two years is that it no longer stands alone.

The AS01 adjuvant hypothesis (explored in Post 2 of this series) connects the shingles and RSV vaccine signals via a shared biological mechanism. A propensity-matched cohort study of 436,788 individuals found that both the AS01-adjuvanted shingles vaccine and the AS01-adjuvanted RSV vaccine were individually associated with reduced 18-month dementia risk, with no significant difference between them — suggesting it is the adjuvant system, not the specific pathogen target, that carries the effect.⁴

Beyond AS01, the influenza vaccine association has accumulated across multiple independent datasets. A 2022 meta-analysis pooling 17 studies and over 1.8 million participants found influenza vaccination associated with a 31% reduction in dementia risk (RR 0.69, 95% CI 0.57–0.83).¹³ A larger 2025 meta-analysis in Age and Ageing, covering 104 million participants across 21 studies, found a more modest 13% reduction (RR 0.87, 95% CI 0.77–0.99) — consistent with a real but smaller effect when heterogeneity is more conservatively pooled.¹⁴ In April 2026, a retrospective cohort study in Neurology by Bukhbinder and colleagues compared high-dose versus standard-dose influenza vaccination in adults aged 65 and older, finding that high-dose influenza vaccination was associated with lower Alzheimer’s risk than standard-dose vaccination — with effect estimates larger than previously reported for standard-dose vaccination versus unvaccinated controls in earlier work by the same group. This is a direct comparison of two active formulations, not of vaccination versus no vaccination, and the finding requires independent replication.¹⁵

The same 2025 Age and Ageing meta-analysis found Tdap vaccination associated with a 33% reduction in any dementia risk (RR 0.67, 95% CI 0.54–0.83) and herpes zoster vaccination with a 24% reduction in any dementia and 47% reduction in Alzheimer’s disease specifically.¹⁴ A 2026 review in Aging Clinical and Experimental Research proposed a broader immunological model in which vaccines may exert protective effects on neuroinflammatory pathways through multiple mechanisms — AS01-type adjuvant activity, trained innate immunity effects, and general reduction of infectious burden — independently of their specific antigens.¹⁶

The consistency across vaccine types with different antigens, different manufacturers, and different study designs is the most important feature of this data landscape. It is difficult to construct a confounding story that explains why influenza vaccination, shingles vaccination, and tetanus-containing vaccination would all separately associate with lower dementia risk in unrelated populations — unless there is a shared biological mechanism at work. That mechanism may be the AS01 adjuvant for some vaccines; it may be a broader effect of trained innate immunity; it may be partly mediated by reduction of infectious burden. The evidence does not yet cleanly distinguish these explanations.

---

What should change — and what you can do now

The structural problem is clear: the evidence has outpaced the communication infrastructure. Regulatory approval lags the science. Clinical guidelines are slow to incorporate emerging signals. Health system messaging is constrained by what is officially endorsed. In the meantime, the people for whom this evidence is most directly relevant — older adults — remain largely unaware that it exists.

Here is what each part of the system can reasonably do, calibrated to the current level of evidence.

---

A final note on what this series has and hasn’t claimed

Across five posts, this series has not claimed that the shingles vaccine prevents dementia. It has not recommended that clinicians depart from their national guidance. It has not dismissed the limitations of observational evidence or the importance of randomised trial replication.

What it has claimed is this: there is a convergent, methodologically sophisticated, and increasingly replicated body of evidence that shingles vaccination is associated with meaningful reduction in dementia risk. That evidence deserves to be known, discussed, and acted on proportionately — which means neither dismissed nor overclaimed. The structural silence that currently surrounds it is not the result of the evidence being weak. It is the result of how regulatory and clinical communication systems are designed. Those systems have important purposes, and this post is not an argument against them. It is an argument for science communicators, clinicians, and engaged members of the public to understand the gap and to fill it responsibly.

The knowledge exists. The vaccine exists. The remaining gap is increasingly one of translation and communication — not simply absence of evidence.