Does the shingles vaccine prevent dementia — or only delay it?

Over the last few years, a strange and potentially important signal has appeared in the epidemiology of dementia: people vaccinated against herpes zoster, the virus that causes shingles, seem to have lower rates of dementia.

At first, this came mostly from ordinary observational studies. For example, analyses of Veterans Health Administration and MarketScan data found lower incident dementia among people who had received a herpes zoster vaccine [1]. Other vaccine studies have reported similar associations for shingles, influenza, pneumococcal, and other adult vaccines.

The problem is obvious: vaccinated people are not interchangeable with unvaccinated people.

They may be healthier, wealthier, more educated, more adherent to medical advice, more engaged with preventive care, less frail, and less likely to already have subtle cognitive impairment. This is the classic “healthy user” or “healthy vaccinee” bias. There are also more technical problems: confounding by healthcare access, confounding by contraindication, differential survival, differential dementia ascertainment, and, in some designs, immortal time bias. Even if an observational study adjusts for many measured variables, it usually cannot adjust for the fact that the type of person who gets vaccinated is different from the type of person who does not.

So I would not treat ordinary vaccinated-vs-unvaccinated comparisons as reliable causal evidence.

The story became much more interesting when natural experiments appeared.

The key idea is simple. Some countries introduced shingles vaccination with eligibility determined by date of birth. People born just before the cutoff were ineligible; people born just after the cutoff were eligible. These people are nearly the same age and should be very similar in health, behavior, socioeconomic status, and dementia risk. But one group suddenly has much higher vaccine access. This creates something close to a randomized experiment.

The Wales study exploited exactly this rule. Eligibility for the live zoster vaccine depended sharply on whether someone was born before or after 2 September 1933. Vaccination uptake jumped at the cutoff, while background characteristics were smooth. Over seven years, eligibility reduced new dementia diagnoses by about 1.3 percentage points, and the estimated effect of actually receiving the vaccine was about 3.5 percentage points, or roughly a 20% relative reduction [2].

Then Australia replicated the logic. Its rollout created a similar age cutoff around eligibility for free Zostavax. In primary-care records, eligibility reduced new dementia diagnoses by about 1.8 percentage points over 7.4 years [3].

Then Canada added another replication. Ontario’s publicly funded zoster vaccine program produced eligibility thresholds that could be analyzed using regression discontinuity and related quasi-experimental methods. The Canadian study found about a 2.0 percentage-point reduction in new dementia diagnoses over 5.5 years, with similar results at two different eligibility thresholds [4].

This agreement is remarkable. Wales, Australia, and Canada used different health systems, different datasets, and different policy cutoffs, but all pointed in the same direction: live zoster vaccine eligibility seemed to reduce, or at least delay, recorded dementia diagnoses by around 1-2 percentage points at the eligibility level.

At this point, the causal story looked unusually strong for an epidemiological claim.

Then came the plot twist: England.

A recent Oxford/NDPCHS presentation reported a regression discontinuity analysis of more than 6.3 million people in England, again using the 2013 UK shingles vaccine rollout. The study confirmed that eligibility reduced hospital-coded herpes zoster, so the policy instrument worked. But it found essentially no effect on hospital-coded dementia [5].

The slide I saw showed estimates extremely close to zero: roughly −0.06 percentage points with a confidence interval around −0.40 to +0.27 using rdrobust, and 0.00 percentage points with a confidence interval around −0.26 to +0.26 using RDHonest. If England were measuring the same outcome as Wales, Australia, and Canada, this would be a devastating contradiction.

But it is not measuring exactly the same outcome.

Wales used broader linked electronic health records: dementia diagnoses in primary or secondary care, plus dementia on death certificates. Australia and Canada used primary-care EHR dementia diagnoses. England, at least as described in the abstract, used hospital-coded dementia in Hospital Episode Statistics.

That difference matters. Primary-care dementia diagnosis is closer to the moment when dementia is first clinically recognized. Hospital-coded dementia is a later, narrower, and more contact-dependent endpoint. A person may have dementia recorded by their GP long before dementia appears in hospital data, if it ever appears there. So England’s null does not necessarily refute the claim that vaccination delays early dementia diagnosis. It does, however, make a large durable effect on severe or hospital-visible dementia much less likely.

One skeptical explanation deserves special attention: maybe shingles itself simply brings people into contact with doctors, and that extra contact leads to more dementia diagnoses. In that case, a shingles vaccine would not prevent dementia. It would merely prevent shingles-related visits, thereby preventing some dementia diagnoses from being discovered.

This is a valid concern in principle. But quantitatively it looks too small to explain the whole Wales result.

In Wales, receiving the vaccine reduced recorded shingles diagnoses by about 2.3 percentage points and reduced new dementia diagnoses by about 3.5 percentage points [2]. A naive logical upper bound would say: even if every prevented recorded shingles case would otherwise have generated a dementia diagnosis, this pathway could explain only 2.3 / 3.5, or about 66%, of the dementia effect.

But that is too generous. Most people who get shingles are not about to receive a new dementia diagnosis. If the baseline seven-year dementia diagnosis risk is around 12.5%, then the expected overlap is roughly:

2.3 percentage points × 12.5% = 0.29 percentage points.

That is only about 8% of the 3.5 percentage-point dementia effect. If shingles cases are enriched two- or threefold for dementia risk, perhaps this rises to 16-25%. But it still does not naturally explain the whole result.

More importantly, the Wales authors directly tested this pathway. They examined healthcare use around shingles episodes, adjusted for healthcare utilization, analyzed frequent healthcare users, controlled for shingles episodes, and asked whether shingles diagnoses were followed by increased dementia diagnosis. They found only short-term increases in healthcare use after shingles and no evidence that shingles diagnoses triggered new dementia diagnoses in the following months [2].

So I do not think “more shingles -> more doctor visits -> more dementia coding” is a sufficient explanation.

Here is my current reconciliation of all the evidence:

the vaccine may create a short protective window against dementia diagnosis, and that protection may be partly independent of preventing clinical shingles.

In this model, the vaccine does not permanently prevent the underlying neurodegenerative process. Instead, it temporarily reduces one or more accelerants that push vulnerable people across the clinical dementia threshold.

Those accelerants could include VZV reactivation, subclinical herpesvirus activity, systemic inflammation, sleep disruption, pain, delirium, vascular injury, or immune dysregulation. Alternatively, the important factor might not be herpes zoster protection per se, but the immune stimulation caused by vaccination.

Imagine someone already close to the dementia diagnosis threshold. The brain pathology is there, but the person is still functioning just well enough not to be diagnosed. A transient improvement in antiviral surveillance, innate immune tone, or microglial function could delay the crossing of that threshold by months or years. But if the underlying neurodegeneration continues, those delayed cases eventually catch up.

This would produce exactly the pattern we see in some data:

early divergence in dementia diagnoses -> then narrowing of the gap -> little or no durable effect on later hospital-coded dementia.

The Shingrix-vs-Zostavax study is important here [6]. In the United States, there was a rapid transition around 2017 from the live vaccine Zostavax to the recombinant adjuvanted vaccine Shingrix. Taquet et al. used this transition as a natural experiment, comparing people vaccinated just before vs just after the switch. The post-2017 group, mostly Shingrix recipients, had more dementia-diagnosis-free time than the pre-2017 group, mostly Zostavax recipients: RMTL ratio 0.83, corresponding to 164 additional diagnosis-free days among people later affected [6].

But the shape of the curves is just as interesting as the average effect. The dementia curves separate early, and then begin to converge. The time-varying hazard ratio falls below 1 early, then moves back toward 1, and in some robustness analyses exceeds 1 late in follow-up. The authors explicitly note that this could mean the vaccine delays, rather than prevents, dementia onset [6].

This is not proof of accelerated dementia progression. The outcome is first dementia diagnosis, not time from diagnosis to severe dementia or death. But it is consistent with delayed diagnosis followed by catch-up.

The same idea is strengthened by the AS01-adjuvant study [7]. Shingrix contains the AS01 adjuvant. Taquet, Todd, and Harrison later compared AS01-adjuvanted vaccines against shingles and RSV. Both were associated with reduced 18-month dementia risk, and there was no significant difference between the two AS01-adjuvanted vaccines. That is important because the RSV vaccine should not protect against herpes zoster. If the signal is similar for AS01-shingles and AS01-RSV vaccines, then part of the effect may come from the adjuvant or immune activation itself, not only from preventing shingles [7].

This is where sargramostim becomes relevant.

Sargramostim is recombinant GM-CSF, an innate immune modulator. In a small randomized trial in mild-to-moderate Alzheimer’s disease, a short course of sargramostim activated innate immune markers, improved MMSE scores at the end of treatment and follow-up, and shifted several plasma biomarkers in a favorable direction, including reductions in total tau and markers of neuronal injury [8]. This was a small trial and should not be overinterpreted. But it demonstrates an important biological point: stimulating innate immunity is not automatically bad for Alzheimer’s disease. Under some conditions, a short, patterned immune activation may be beneficial.

This matters because people often talk about “inflammation” as if it were one thing. It is not. Chronic unresolved inflammation, infection, and delirium are likely harmful. But transient innate immune activation, trained immunity, improved pathogen surveillance, or better myeloid cleanup of debris could plausibly be protective.

So the model I propose is:

a vaccine or adjuvant produces a temporary, organized immune perturbation -> this improves antiviral control and/or innate immune cleanup -> some near-threshold people cross into dementia later than they otherwise would -> the effect wanes -> diagnoses catch up -> and later endpoints such as hospital-coded dementia may show little or no durable difference.

This model can reconcile the four apparently conflicting facts:

Ordinary observational studies find lower dementia risk after shingles vaccination, but are vulnerable to healthy-user bias.
Natural experiments in Wales, Australia, and Canada find strikingly similar reductions in primary-care or broad-EHR incident dementia diagnoses.
England’s much larger hospital-coded analysis appears null.
Shingrix-vs-Zostavax curves suggest early protection followed by catch-up, while AS01 and sargramostim studies make an immune-mediated, herpes-zoster-independent mechanism biologically plausible.

The practical implication is that we should be careful with the slogan “the shingles vaccine prevents dementia.” A more precise version may be:

herpes zoster vaccination, or perhaps certain vaccine-induced immune states, may delay dementia diagnosis/onset in some older adults, but the durability of this effect and its impact on severe dementia remain uncertain.

That is still potentially very important. Even a one- or two-year delay in dementia onset would have enormous personal and public-health value. But it is different from permanent prevention.

The key next studies should not only ask whether vaccinated people get fewer dementia diagnoses. They should ask when the protection appears, how long it lasts, whether boosters restore it, whether the effect is seen for MCI and biomarkers, whether it affects hospitalizations and dementia deaths, and whether AS01 or other immune adjuvants reproduce the same signal without herpes zoster being the target.

If the short-window hypothesis is right, the most important finding may not be “shingles causes dementia.” It may be that the aging brain is surprisingly responsive to temporary immune modulation.

What practical implications could this have? If this hypothesis is true, then a one-time recombinant zoster vaccination series may not be enough to produce a lasting effect on dementia risk. We might need periodic immune re-stimulation — perhaps with recombinant herpes zoster vaccine boosters every 3–5 years, or with other interventions that mimic the same immune-modulating effect without necessarily targeting herpes zoster.

This is not a clinical recommendation yet. Existing shingles-vaccine schedules are designed to prevent herpes zoster, and Shingrix protection against shingles appears long-lasting. But if the dementia signal comes from a shorter protective window — for example, transient innate immune activation, improved antiviral surveillance, or AS01-like adjuvant effects — then booster timing, adjuvant choice, and alternative immune modulators become central research questions.

Author — Aleksandr Fedintsev

Literature list
[1] Scherrer et al. Impact of herpes zoster vaccination on incident dementia: A retrospective study in two patient cohorts. PLOS ONE, 2021.
[2] Eyting, Xie et al. A natural experiment on the effect of herpes zoster vaccination on dementia. Nature, 2025.
[3] Herpes Zoster Vaccination and Dementia Occurrence. JAMA, 2025. Natural experiment in Australia.
[4] Pomirchy et al. Herpes zoster vaccination and incident dementia in Canada: an analysis of natural experiments. The Lancet Neurology, 2026.
[5] Oxford/NDPCHS seminar abstract. Little evidence that live herpes zoster vaccine eligibility reduces hospital-coded dementia in England: a regression discontinuity analysis of 6.3 million individuals. 2026.
[6] Taquet et al. The recombinant shingles vaccine is associated with lower risk of dementia. Nature Medicine, 2024.
[7] Taquet et al. Lower risk of dementia with AS01-adjuvanted vaccination against shingles and respiratory syncytial virus infections. npj Vaccines, 2025.
[8] Potter et al. Safety and efficacy of sargramostim (GM-CSF) in the treatment of Alzheimer’s disease. Alzheimer’s & Dementia: Translational Research & Clinical Interventions, 2021.

Google Sites

Report abuse