The truth about the shingles vaccine and dementia

Ground breaking intervention or senseless hype?

You may have seen headlines about a single, cheap, now withdrawn vaccine that appears to cut your risk of dementia by around a fifth; the live-attenuated shingles vaccine (now replaced by a recombinant vaccine). You likely saw it stated as plain fact from a source you trust. It is one of the most hopeful ideas in recent medicine, and as a vaccinologist, I would love it to be true.

This is why it is due more scepticism than it has received, and the last week has brought some important events.

Let’s look at the story and recent developments.

Why would anyone think this?

The underlying idea is not completely implausible. Previous lines of research suggested neurotropic herpesviruses such as Varicella zoster, the shingles virus, may play some role in dementia. If a dormant virus reactivating in the nervous system contributes to neurodegeneration, a vaccine that keeps it dormant could theoretically help.

The trouble started with studies that simply compared dementia rates in the vaccinated and unvaccinated. These reliably found lower rates in the vaccinated but are close to uninterpretable, because people who turn up for vaccines are healthier, wealthier and more health-engaged to begin with. A giveaway was that in several datasets the flu vaccine looks almost as "protective" as the shingles one. When every vaccine prevents dementia, you are not measuring vaccines. You are measuring the kind of person who gets them.

What changed things was a 2025 Nature paper from Wales. Eligibility for the live vaccine there turned on an arbitrary date-of-birth cut-off: born just before, ineligible for life; just after, eligible. People either side of that line are, on average, extremely similar except for their access to the vaccine. The design behaves somewhat like a natural experiment.

The study reported a jaw dropping 20% relative reduction in dementia over seven years. The situation presented a clever opportunity, and this was meant to be the study that finally escaped confounding. The same team then applied the same trick to rollouts in Australia and Ontario, and each time the result pointed the same way. To most observers the same finding across three countries looked like gold-standard confirmation, and was heralded as such by some.

The cracks

There were many reasons to have remained highly sceptical of these findings at the time.

The effect was enormous. A one-fifth reduction from a single jab, against a disease which, until now, we’ve struggled to even understand the pathophysiology of, let alone make any therapeutic progress, is the kind of effect size that should raise an eyebrow instead of a cheer.

It was also mostly in women, with little or no effect in men. A real biological effect confined to one sex is possible, but seems highly unlikely in a disease where the pathology is thought to be common, and seemingly scrambled explanations were put together after the result was known.

As Vinay Prasad has noted, where survival curves have been presented (showing differences in how many cases occur over time) the effect seems to occur implausibly early following vaccination. For a disease which takes years to develop and manifest, an almost immediate effect from vaccination is difficult to explain.

On a more technical note, the outcome was never dementia itself, but a dementia diagnosis in an administrative record. Whether a case is captured depends on who comes through the doctor’s door and how a clinician codes them. The eligibility moment is also a healthcare-contact moment, so anything that nudges how often the just-eligible are seen and worked up (e.g., by getting shingles) can shift recorded dementia without changing the real rate. It is also why the same design can give different answers in different health systems, which is precisely what happened next.

What a new analysis shows

My friend, the venerable George Davey Smith and colleagues did the next obvious thing and ran the identical design in a far larger population, which was presented at a guest lecture on June 17th. England used the same vaccine, the same date, the same cut-off, so an almost identical natural experiment exists, across 6.3 million people (roughly twenty times the Welsh sample).

As one would expect from this group, the study is persuasive. First they confirmed that eligibility produced a clear drop in recorded shingles. This is proof the design works (and the vaccine, which is excellent, does its actual job). Then they looked at dementia, and found….nothing.

A complete null result, stable across sensitivity analyses, with no bias they could find to explain the gap with Wales.

This does not quite close the question completely, as no single study ever does. England codes dementia from hospital records, which capture it less completely than Welsh primary care, so a defender could argue the effect is simply being missed. This criticism is weak however, because the same design plainly can detect the vaccine's effect on shingles in these very people. In addition, restricting the Welsh analysis only to hospital data (like the new English analysis) doesn’t alter their findings, showing it should be robust whichever of these types of data you use.

Speaking to the differences in these findings, a point made in the lecture was that the type of method deployed to these study (regression discontinuity design, for those who are interested), enables the researcher a huge degree of freedom in making analytical decisions. Each of these can potentially drastically alter the ultimate result.

Ultimately, the failure to observe an effect size so large it should be visible from space, in the largest and most robust replication study to date, is devastating evidence against this hypothesis.

Suddenly, a trial is urgent?

Within days of the null replication, a Nature Medicine commentary argued that large randomised trials of the live vaccine for dementia are urgently needed, and that we have never had a more robust body of observational evidence for any dementia intervention. Notably, its author is the researcher behind the Welsh, Australian and Canadian observational data.

He is right that a trial would ultimately be the only way to provide a definitive answer, that observational evidence has gross limitations, and may be right that the live vaccine is the one to test. But the framing is exactly what this episode should teach us to distrust. "The most robust observational evidence we have ever had for a dementia intervention" is, to be blunt, an extremely low bar: that field is a graveyard of observational findings (HRT, vitamin E, ginkgo, B vitamins) that looked wonderful until someone ran the trial. Let’s also be aware that three of the four headline experiments come from one group running one design, while the single genuinely independent large-scale replication is the one that found nothing.

As is happens, such a clinical trial is already underway in Denmark; but with the new, recombinant vaccine, not the live-attenuated vaccine to which all this evidence refers. Recruitment is underway and over 100,000 people will be needed to assess if the vaccine reduces cardiovascular complications or dementia over a 3y period.

This trial is possible in Denmark as no shingles vaccine is currently offered on the national schedule there, unlike in the US or the UK.

This raises an important point - it’s not clear who the control group would be if a study of the old, live-attenuated vaccine was to be done in the US (as I understand the NIH are considering). If the control arm receives the current, recombinant vaccine (which they are entitled to), and this also prevents dementia (which hopefully we’ll discover from the Danish study), the trial would show no difference between the two even though they both prevent dementia.

This leaves me to ask, can we ethically randomise patients to not receive a vaccine which we already know is highly effective at the thing it is actually supposed to do: protecting against shingles? I am aggressively pro-clinical trials, and even I find this idea questionable. The commentary completely avoids referencing the control group at any point.

In my opinion, much more evidence from different streams of data (such as mechanistic work) are required to justify such an endeavour.

In aforementioned lecture, George Davey Smith speaks to the importance of triangulation of evidence; not relying on a single source or method of research, but collating multiple different types of observational and experimental evidence to try and establish causality. Even for associations as strong as smoking and lung cancer, such triangulation was necessary to eventually convince the scientific community of the relationship (this was a major source of disagreement between the two titans of medical statistics and epidemiology of the 20th century, Ronald Fisher and Austin Bradford Hill).

The part that worries me

The messaging about these studies from the people we trust to communicate science has completely missed the mark.

From very early on, the finding that shingles vaccines prevent dementia was being relayed to the public as established fact. One prominent, widely trusted, emphatically pro-science communicator told a national audience that the shingles vaccine reduces Alzheimer's and dementia "by at least 20 to 25 percent", not as a hypothesis but as something we had simply learned, folded into advice on how to age well.

The finding was exciting, and it flattered a “pro-vaccine” worldview, so scepticism got switched off. When people who are fervently anti-vaccine overstate a harm we rightly demand rigour and replication; when a result flatters the pro-vaccine side, the same people wave it through. It is the identical error, and when an oversold finding wobbles, the people who never trusted the messaging get to say, accurately, that scientists hyped a vaccine benefit that did not hold up. More of public health's credibility, which is hard won, is lost.

I am dismayed at how “Sci-Com” keeps losing its way, and how often the biggest names are guilty. The job is not simply to parrot a studies findings but to rigorously interrogate them: what was the design, how big was the effect, does it replicate, what would have to be true for it to be wrong? Reposting a conclusion with an admiring adjective is not communicating science; it is laundering a press release. At this point you’re playing the role of a publicist, not a scientist.

That appraisal, slow and unglamorous and sometimes deflating, is the whole value we are meant to add as science communicators.

Summary

The initial hype over an enormous risk reduction in dementia from receiving the shingles vaccine has not held up in the largest independent replication study to date. The hypothesis is not dead (although it is on life support), but it’s hard to see how a trial is currently the answer. The lesson is not that the vaccine is useless, nor that natural experiments are bad, but that an unbelievable result deserves the same hard scrutiny whether or not it tells us what we hoped to hear. Science Communicators must give due effort to appraising the research that they cite, not simply repeating the headlines. At present it is not good enough.

So the truth: does the shingles vaccine prevent dementia? Probably not - although I would love it to be proven so.