Could a shingles vaccine protect your brain? Here’s what the science actually shows

A series of studies — including a Welsh natural experiment now published in Nature — suggests these vaccines may help reduce dementia risk — and they’re already sitting on the shelf. The evidence isn’t perfect. But it’s considerably stronger than most people realise, and it isn’t reaching the people who need it.

In 2021, an estimated 57 million people were living with dementia worldwide — a number projected to nearly triple by 2050. Dementia is the seventh leading cause of death globally. Women are disproportionately affected in both outcomes and unpaid care burden: they account for the majority of people living with the condition and provide around 70% of dementia care hours worldwide. The search for prevention strategies has been running for decades — and while lifestyle factors and vascular risk management get considerable attention, one signal has been building quietly in the epidemiological literature: routinely available vaccines may reduce dementia risk.

This is not a fringe hypothesis. It is supported by large observational studies, at least one quasi-experimental natural experiment, and a plausible set of biological mechanisms. But the evidence has not translated into clinical guidance, public health messaging, or informed patient conversations. The gap between what the science shows and what most clinicians and patients know is striking — and worth examining carefully.

This post deals with the evidence base. What do we actually know, and how confident should we be?

The shingles signal: where the evidence is strongest

The most compelling data concerns the herpes zoster (shingles) vaccine. Multiple large observational studies, conducted across different populations and healthcare systems, have found consistent associations between shingles vaccination and lower dementia incidence. The direction and approximate magnitude of the effect are coherent across studies. That alone is meaningful.

But the study that should command the most attention is a quasi-experimental natural experiment using administrative data from Wales, published in Nature in 2025 by Eyting and colleagues.¹ It takes a particular kind of national health service to accidentally run a quasi-randomised trial on 80,000 people. Wales managed it. New Zealand has also managed it.

People who received the shingles vaccine were approximately 20% less likely to develop dementia over a 7-year follow-up period, compared to equally eligible individuals born just after the date-of-birth eligibility cut-off — a finding consistent with a protective effect that observational data alone cannot definitively establish. Eyting et al., Nature, 2025

A note on vaccine type is important here: the Welsh programme used the live-attenuated vaccine (Zostavax), not Shingrix. This matters for interpreting the findings. Both the Nature paper and the December 2025 Cell paper by Xie and colleagues² used the same Welsh natural experiment design — Xie et al. extended the analysis to study outcomes at different stages of the dementia disease course. Among people already living with dementia at the start of the programme, vaccination was associated with a 29.5% reduction in dementia-related deaths — a finding the authors interpret as consistent with an effect on outcomes later in the disease course, though the exact mechanism remains unclear and the comparison is between vaccine-eligible and vaccine-ineligible groups rather than directly vaccinated and unvaccinated individuals.

The comparison between Shingrix and Zostavax comes from a different study: Taquet and colleagues in Nature Medicine (2024)³ used US electronic health records and a natural experiment created by the rapid transition from live to recombinant vaccine in 2017. They found that receiving Shingrix was associated with a 17% increase in time lived without a dementia diagnosis — equating to 164 additional dementia-free days among those subsequently affected. Both shingles vaccines also outperformed influenza and Tdap vaccines in this analysis.

How much should we trust this evidence?

Confidence in a body of evidence is not binary. The GRADE framework, which is used to rate evidence quality in clinical guidelines, rates certainty across four levels: High, Moderate, Low, and Very Low. Standard practice automatically downgrades observational studies — but this can be inappropriate when a study has design features that substantially reduce bias. Epidemiologists call this being methodologically rigorous. Critics of the field call it burying good evidence under procedural caution. Both have a point. GRADE explicitly allows for upgrading when effects are large and consistent, dose-response relationships exist, and residual confounding would likely bias results in the opposite direction.

A newer tool, ROBINS-E (2024), was specifically developed to assess observational studies of exposures against the framework of a hypothetical target trial. Under ROBINS-E analysis, the Welsh natural experiment has low risk of bias in the confounding domain — the date-of-birth cut-off achieves something close to quasi-randomisation, and negative control outcomes (diabetes, cancer) behave as expected, providing further assurance that the design is sound.

Applying these criteria to the shingles evidence, my appraisal is that this body of evidence merits Moderate certainty — potentially upgradeable to Moderate-High when the full profile is considered. This is substantially stronger than the reflexive assumption that all observational data defaults to Low certainty. It is not a formal guideline body rating; it is an expert appraisal using established methodology.

The influenza picture: real but messier

Influenza vaccination has also been associated with lower dementia risk across multiple studies, but the evidence base is less coherent and more methodologically vulnerable. Meta-analyses have reported varying estimates of protective association — some substantial — but studies applying stricter controls for healthy-vaccinee bias find substantially attenuated effects. A large Danish population study found no meaningful protection in the general population after controlling for this bias, though a dose-response signal emerged, with protection apparent only among those receiving multiple vaccinations over time.

The more consistent signal is in high-risk populations: people with chronic kidney disease, COPD, or pre-existing cardiovascular disease show more robust protective associations. This is biologically plausible — preventing acute inflammatory episodes that may accelerate vascular pathology could matter more in those with existing vulnerability.

What about the mechanisms?

Several non-exclusive biological pathways have been proposed, and understanding them matters for assessing whether the epidemiological signal is plausible rather than artifactual.

The most direct pathway involves VZV reactivation itself. Shingles and subclinical VZV reactivation have been linked to vascular damage and neuroinflammatory pathways, and have been hypothesised to contribute to Alzheimer-like pathological features — amyloid deposition and tau phosphorylation — based on animal and cell model data. The evidence is plausible rather than settled, but is sufficient to support a biological rationale for the epidemiological signal. Preventing reactivation may help prevent these downstream consequences.

One important clarification: the Welsh study used Zostavax, a live-attenuated vaccine that contains no AS01 adjuvant. The ~20% reduction in dementia risk observed in Wales must therefore be explained by the immune response to the VZV antigen itself — not by adjuvant-mediated effects. This is significant: it means the baseline neuroprotective signal exists independently of Shingrix’s adjuvant system. If AS01 does confer additional benefit, the recombinant vaccine may be even more protective — but the Welsh findings stand on their own regardless.

A second, more speculative pathway does concern the AS01 adjuvant system in Shingrix and in the RSV vaccine Arexvy. AS01 contains two immunostimulatory components — MPL (monophosphoryl lipid A) and QS-21. Animal model data suggest that TLR4 stimulation by MPL induces interferon-gamma production, which may in turn reduce amyloid deposition. A large propensity-matched study in humans found similar dementia risk reductions for both AS01-containing vaccines compared to influenza vaccine, which would be consistent with an adjuvant-mediated effect beyond VZV prevention alone — but a methodological critique has been published in npj Vaccines, and the debate remains open. Crucially, we cannot yet decouple the adjuvant from the antigen effect in human populations. This is frontier science — interesting, but not established.

A third pathway, relevant to influenza vaccination, is cardiovascular: preventing the acute inflammatory response to influenza infection may reduce the risk of the thrombotic and embolic events that drive vascular dementia.

The sex difference that isn’t being discussed

One of the most striking — and least publicised — findings in this evidence base is a stronger protective signal in women across multiple analyses. Xie et al.’s Cell analysis found an approximately 50% reduction in dementia-related mortality among vaccinated women, compared to a smaller, statistically non-significant effect in men. Taquet et al. in Nature Medicine similarly found the dementia-protective association was greater in women (22% more time lived diagnosis-free) than men (13%). These findings are consistent across two independent datasets and designs. A 50% reduction in dementia deaths in women, sitting quietly in a supplementary table. If this were a drug trial result, it would be on the front page of every newspaper. It’s a vaccine finding, so here we are. Given that women are disproportionately affected by both dementia outcomes and dementia care responsibilities, the sex-differential signal deserves far more attention than it is currently receiving.

The mechanism is unclear. It may involve sex differences in VZV reactivation patterns, differences in immune senescence trajectories, hormonal modulation of neuroinflammatory responses, or baseline differences in amyloid and tau pathology. What is clear is that the finding is being effectively buried in supplementary tables rather than communicated to the demographic most affected. That is a science communication failure with real consequences. It will be the subject of a separate post in this series.

What this means, and what it doesn’t

The evidence supports a provisional conclusion: shingles vaccination, particularly with the recombinant vaccine, is associated with a meaningful reduction in dementia risk. The certainty of that association — assessed across multiple studies rather than any single one — is sufficient to inform clinical conversations and public health messaging. This doesn’t require certainty about mechanism, and it doesn’t require RCT evidence — the JCVI’s 2023 advice on using the meningococcal B vaccine (Bexsero) for gonorrhoea prevention, based on observational data without a formal indication, demonstrates that advisory bodies are capable of acting on evidence of this structure when they judge it sufficient. That parallel will be examined in Post 4.

What it does justify is a frank conversation about why this evidence is not reaching clinicians or patients — and who benefits from that silence. That is the more uncomfortable question.