Controversial, I know, but that’s the theme of the recent blockbuster Eyting et al., “A natural experiment on the effect of herpes zoster vaccination on dementia”.
Apparently in Wales in 1933, there was a single cutoff date: those born immediately before that date generally did not receive a herpes zoster vaccine and for those born after that date there was a precipitous increase in herpes zoster vaccination rates. The authors borrowed an idea from econometrics called a regression discontinuity experiment. Since the people born right before and right after the arbitrary vaccination cutoff date are expected to be statistically similar, you can infer causal effects of the vaccine itself on subsequent lifetime health outcomes. This experimental design reduces confounding that corrupts traditional correlational studies.
What did they find?
Using these comparison groups in a regression discontinuity design, we show that receiving the zoster vaccine reduced the probability of a new dementia diagnosis over a follow-up period of 7 years by 3.5 percentage points (95% confidence interval (CI) = 0.6–7.1, P = 0.019), corresponding to a 20.0% (95% CI = 6.5–33.4) relative reduction.
I know Nature has been taking a lot of shit from my parts of the internet, but this is a carefully done paper. The authors go through many credible robustness checks. They’re even smart enough to realize that getting shingles (more likely without the herpes zoster vaccine) gives a person more exposure to the health system and might make them more likely to end up with a dementia diagnosis, an effect that would incorrectly inflate the effect size.
My main complaints are that “dementia” is obviously not a single thing and ideally they’d study something more constrained. Also, I did a little math on the reported effect size, and it seems a bit implausibly high, but overall this paper deserves serious attention and has some very interesting implications.
Infectious causation
Let’s go back in time to the year 2000—before the human genome was sequenced, before calamities like 9/11 and The Apprentice. Greg Cochran, Paul Ewald, and Kyle Cochran published “Infectious Causation of Disease: A Evolutionary Perspective”, henceforth CEC2000. This paper gives a brief history of the how medical science gradually accepted when certain diseases were caused by infectious agents, and the circumstances under which that happened. They go on to argue that given this history, many diseases—particularly those common ones which severely damage evolutionary fitness—are very likely caused by infectious agents even if how they’re caused is currently ambiguous.
Infectious diseases have simply not been in vogue in health science, probably for decades, and this has contributed to this view.
The CEC2000 argument goes something like this: genetic diseases that affect fitness are unlikely to be common because in equilibrium, they’d be eliminated by selection. Given the retrospectively embarrassing history of medical scientists missing infectious causation, if these diseases are not primarily genetically mediated, they’re quite likely caused by infections.
Rheumatic fever illustrates how these crypticity factors played out during this period. The lag between the early symptoms of acute streptococcal infection (e.g., sore throats) and rheumatic fever typically range from a few weeks to a few months [19]. Toward the end of the 19th century, epidemiological links were noted between rheumatic fever and both sore throats and scarlet fever; infectious causation was proposed in 1895 by Arthur Newsholme, who argued against the generally accepted view that it was a hereditary disease [19]. During the first two decades of the 20th century, researchers actively investigated infectious causation of rheumatic fever, but could not find bacteria at the sites of damage. From the 1930s through the mid-1950s, research first associated and then causally linked rheumatic fever with Streptococcus pyogenes. The experimental control of rheumatic fever through the use of antibiotics from 1939 through 1955 provided the last batch of evidence necessary for general acceptance of infectious causation [19]. The process from suspicion of infectious causation to general acceptance therefore spanned a half century of scientific effort.
So, causation is hard to rigorously determine when the disease course is long and in multiple clinical stages.
They then do a little math creating an index called “fitness load” that quantifies how likely a disease is to be caused by infectious agents; it’s a combination of the disease’s prevalence and its deleterious effect on fitness, kind of analogous to genetic load in evolutionary biology.
I remember reading this paper years ago and being quite impressed by this argument. It’s first principles thinking backed up by math that a smart sixth grader can understand. They make predictions about what future research will reveal about disease causation after stating
The key problem is how to facilitate recognition of infectious causation among these diseases. One step toward resolution of this problem involves increased awareness of the sources of crypticity that we are likely to encounter in ascribing infectious causation. One source of crypticity is the increasing difficulty in obtaining suitable animal models. Few mammals live as long as humans. It is therefore difficult to find experimental animals that can be infected by an organism thought to cause long-delayed chronic disease and that then survive long enough to demonstrate the same chronic disease found in humans. Even if possible, these procedures may be prohibitively expensive.
This part of the paper aged the worst. In fact, we didn’t need expensive animal models to successfully identify long term infectious causation. As Eyting et al. demonstrate, we needed enormous modern datasets that were barely imaginable in 2000. Regression discontinuity design was invented in the 1950s but the data that supported it was just completely over the horizon.
What are CEC2000’s predictions?
Multiple Sclerosis
They state
Epidemic waves of MS also implicate infectious causation; for example, a surge of MS cases occurred in the Faeroe Islands from the early 1940s to the early 1970s [78]. Although various pathogens have been suggested, including a newly identified retrovirus, the matter remains unresolved [79–81]. Virtually all MS patients are infected with EpsteinBarr virus, raising the possibility that MS may result from coinfection of Epstein-Barr virus with one or more other pathogens.
In 2022, Bjornevik et al. was published providing very strong evidence that Epstein-Barr causes MS by way of meticulous analysis of a gigantic dataset, and just earlier this month, in the year of our lord 2025, the authors won the breakthrough prize for this work.
Pregnancy Stuff
They state
Infections, however, are known to contribute to chromosomal aberrations; hepatitis B, polyoma viruses, and papillomaviruses, for example, are associated with chromosomal damage [96–102], and human papillomaviruses occurred in the majority of material from first-trimester spontaneous abortions, being three times more common than in material from elective abortions [103]. These considerations suggest that spontaneous abortions and associated chromosomal damage, so prevalent in humans, may be caused by infection.
In 2020, Been et al. showed preterm birth was significantly reduced after COVID lockdown measures were undertaken in the Netherlands, and most suspiciously
Reductions in the incidence of preterm births after March 9 were consistent across gestational age strata and robust in sensitivity analyses. They appeared confined to neighbourhoods of high socioeconomic status, but effect modification was not statistically significant.
In other words, the rich people who were able to lock down and avoid disease were less likely to have a preterm birth.
Magnus et al., 2022 is a weaker result, but addresses CEC2000’s more specific point about chromosomal damage in fetuses caused by infection. They found a null result for miscarriage except for in the early months of lockdowns in Scandinavian countries.
We observed a slightly decreased risk of miscarriage during the first 4 months, with an HR of 0.94 (95% CI 0.90, 0.99) after lockdown.
Like the natural experiment of vaccination cutoff date in Wales, the lockdowns were a natural experiment that limited the spread of common communicable diseases suggesting that prenatal infections, whatever they might be, generally make a pregnancy less likely to succeed.
Implications
Back to Eyting et al. They’re not sure precisely how the Herpes zoster vaccine is causing decreased dementia prevalence, but they’re open-minded and explore several possibilities. This paper is timely given that current investments in the amyloid hypothesis for Alzheimers are unwinding due to poor scientific inference, misfeasant funding, and fraud. Whatever theory of Alzheimer’s causation replaces the amyloid hypothesis, it needs to consider a model of latent infectious causation. After 25 years, CEC2000 has absolutely won the day.
A few other ideas:
We should be more cautious regarding propositions like “COVID doesn’t effect children” or “COVID vaccines in children aren’t necessary”. We don’t know that yet, as the many examples above illustrate.
Maybe women should be taking more time off or more time at home during pregnancies, especially in third terms.
This one’s a bit self-serving as a data guy, but health policymakers should make efforts to improve health data quality and improve access to it for a larger group of researchers. There are clever ways data could be robustly anonymized and made more open. This has been done successfully in the Netherlands and Scandinavia. Contra what CEC2000 state, these recent breakthroughs in causal determination were made by sophisticated analysis of longitudinal, high resolution datasets, not wet lab work on animal models.
More resources could be put into vaccination attempts to inoculate against infectious insults that are merely regarded as nuisances now. We don’t know what they could be causing years down the road.