Flu vaccination linked to 40% reduced risk of Alzheimer’s disease

Epidemiologist in training here... There are quite a few comments in this thread already jumping on the 'correlation != causation' train. While that is true, I'd like to clarify a couple things:

1. The journal article didn't suggest it was causal. But such a correlation with such a large population warrants publication and further research into causation.

2. literally the first thing that any epidemiologist would consider is potential confounders. There is a big list of covariates they included into their model here: https://content.iospress.com/articles/journal-of-alzheimers-...

There are quite a few things that can be done to alleviate potential false correlations: DAGs, prior literature, removing confounders, and including covariates are all things at disposal.

3. Such a large sample size + previously reported findings + an inclusion of enough covariates still doesn't == causation, BUT it's important to publish and shout about so we can then look into the potential biological underpinnings that may cause this. Which by the way, those experiments may still use data science techniques.

4. If you are actually interested, there is a whole topic of this called 'causal inference' with one famous criteria list called the 'Bradford Hill Criteria': https://en.wikipedia.org/wiki/Bradford_Hill_criteria. This list is often argued about.

5. If all of this information was new to you, please stop spouting 'correlation != causation'. You probably don't know as much as you think

>There are quite a few comments in this thread already jumping on the 'correlation != causation' train.

Complaining about correlation not being causation is the perfect theme for the midwit meme. It's one of my pet peeves.

I do not remember where I read it (probably xkcd) but it was that correlation is not casuation, though the numbers are doing big winks to you (or something like that)

As a physicist who had to endure helping biologists with statistics and cooling down their enthusiasm: it make sense to have a deeper thought about the experiment.

As parent wrote, there may be various reasons for the correlations, sometimes you have random stuff, sometimes indirect stuff and in others extraordinary stuff. Many discoveries (especially older ones) fall into the last category.

The scientists said it was causation.

    "We found that flu vaccination in older adults reduces the risk of 
     developing Alzheimer’s disease for several years. The strength of 
     this protective effect increased with the number of years that a 
     person received an annual flu vaccine – in other words, the rate 
     of developing Alzheimer’s was lowest among those who consistently 
     received the flu vaccine every year"

Yes, that's in an interview for the article. That's the reason (1) the general public misunderstands (2) people scream about it. Scientists get points for "high-impact research," and there is strong incentive to be dishonest.

(As a footnote, personally, I do believe it is causation; I believe that as with COVID and EB, we've dramatically underestimated the long-term impact of many viral infections. But that's just a personal belief.)

>The scientists said it was causation.

Interesting. I don't read that as claiming causation, I read that as describing the correlation.

You're right though, it's confusing messaging, regardless.

I've always read "protective effect" to be making a predictive statement and not a causal one.

In the interview he seems to be more loose with his words, but that also doesn't say causation. In the paper the conclusion is "This study demonstrates that influenza vaccination is associated with reduced AD risk in a nationwide sample of US adults aged 65 and older."

That is distinctly not causal.

If they only intend to claim correlation, avoiding the word "linked" in the announcement would probably help with general public interpretation.

I feel like in English, "linked" usually implies some sort of potential causation, not just a general relationship. For example: "boyfriend linked to murder case".

Sitting down and talking it's often hard to be as precise as you would like to be. Especially when you don't have time to reflect on the implications of what you've just said.

I’d say they should show evidence that they know what they’re talking about, but then I’d be accused of credentialism.

I like the phrasing, "Correlation correlates with causation because causation causes correlations".

> where I read it (probably xkcd)

Yep, it's in the alt text for https://xkcd.com/552/

"Correlation doesn't imply causation, but it does waggle its eyebrows suggestively and gesture furtively while mouthing 'look over there'."

This reminds me of the intentionally-ambiguous "officer-involved shooting" turn of phrase.

https://www.mcsweeneys.net/articles/an-interactive-guide-to-...

https://www.cjr.org/analysis/officer-involved-shooting.php

With this sort of thing, lots of social factors weigh in too. Access the flu shots implies regular access to healthcare and perhaps other interventions.

Thank you for posting this. I doubt people can make it to adulthood--or even manage speaking in complete sentences--without having some rudimentary appreciation for "correlation != causation", yet it is so frequently posted--both here & the r-place.

And its vileness is far beyond midwittery. It is a lazy & incredibly weasely way to stifle unwelcome lines of inquiry: by pre-emptively accusing anyone who would continue along those lines as a mental defective.

That being said, modern research is mostly a profit/status driven false-positive factory.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1182327/

Being able to demonstrate technical understanding of a problem isn't credentialism. Conversely, having an undergraduate degree in X isn't evidence of expertise.

Causation frequently removes correlations that would otherwise exist.

This is a mischaracterization of even the words you've quoted.

The team did indeed find that flu vaccines reduced the risk of Alzheimers. But he specifically doesn't say they know why the flu vaccine lowers the risk. Ie. They can't claim causation if they don't understand the underlying reason for the change. That's not how science works.

In fact in the article in question he says the opposite: That the immune system is vastly complex, and all they can say is the immune system reacts to flu vaccines (all flu, not just a specific one) by diminishing an Alzheimers response. ¯\_(ツ)_/¯ More study needed plz.

Correlations do imply a link, though. They don't imply causation, but in the absence of selection bias and with enough of a sample size, it is almost certain there is a link somewhere.

The word for when there is no link is, in turn, 'coincidence'.

Causation, correlation, coincidence.

And then there's quantum entanglement.

I don't think that's right? In theory you could have A and B that were naturally correlated and add a causal relationship between the two that exactly canceled that correlation but that would be infinity unlikely to happen by chance. In practice the only way that A and B can be correlated is that either A causes B, B causes A, or some C causes both A and B. Interestingly if both A and B cause C the C and A as well as C and B will be correlated but A and B won't be, allowing you to work out causation from a strictly correlational graph.

Well let's test this theory out then. Would you say I was being ridiculous if I told you that the number of people who drown in a year is linked to how many movies nicolas cage appears in?

If so, then you might not actually believe linked is simply a synonym for correlation. I know I certainly would think that was a ridiculous way to phrase it.

What's wrong with implying potential causation? Correlation is potential causation. The whole reason one says the boyfriend is linked to the murder case is because one isn't saying he's responsible -- and yet he might be.

> the number of people who drown in a year is linked to how many movies nicolas cage appears in

For those who don't know the reference: http://www.tylervigen.com/spurious-correlations

That's just being facetious though. We've all seen that website with funny correlations. This line of argument while an important lesson to look deeper into systems before claiming a correlation doesn't really add much here.

Loads of people get the flu vaccine. Enough that I'd imagine the sample size is representative of the population at large. A large number of people also don't get the flu vaccine. So this is a surprising result. Especially since we are injecting a liquid into the people so there could reasonably be some effect we never thought of. Which is the whole point. These people aren't correlating things in completely unrelated fields and claiming a link.

The brainlet and sage text is “maybe x causes y.”

It's not by chance. If you have a heater controlled by a thermostat, it destroys the correlation that would otherwise exist between outside temperature and inside temperature. That particular example sounds silly because it is so obvious. But in economics you can find examples where you would be seeing a relationship, but it is masked by an active response to the cause, but it's not so obvious this is happening.

Probably just because it isn't a very high bar. "People who eat McDonald's linked to fewer car accidents." I mean some data might just show that, but causality is a lot further away there.

Hence, while a correlation might be interesting, when presented we need to also understand the methodology and go into the detail a little bit to see how extra factors have narrowed down that finding.

E.g, without looking into any of the data, 'more Flu shots reduces risk of Alzheimer's' might be because people predisposed to Alzheimer's have a poor diet and the flu shot just counters some other terrible health effect. So first bit of detail to look at is whether the selection factors for diet, comorbidities, environment, general health, history, etc.

Not saying anything about this article, but context is always important, hence those claiming 'correlation !== causation' are only correct as far as not looking any deeper. Same for 'well it could be' is useless without knowing how close to 'could be' it is.

More critical imo, is the use of '40% reduction.' If this is relative risk (this study is not, it's absolute), the actual risk could be 0.0037%, where a reduction of 40% is practically nothing. So next check is: is this relative or absolute risk? If relative, relative to what baseline? So many studies are distorted by using relative risks against a meaningless baseline.

Congratulations, you've discovered p-hacking.

I said "almost certain". What you did is taken a database of someone who made millions of comparisons, such that coincidences that are very unlikely to occur happen via pure coincidence.

Additionally, I said "enough of a sample size". Given that the analysis you're talking about was over 11 pairs of points, that doesn't even qualify as very unlikely.

I disagree that's a mischaracterization.

> The team did indeed find that flu vaccines reduced the risk of Alzheimers. But he specifically doesn't say they know why the flu vaccine lowers the risk

I don't think that's an accurate statement. They didn't find it reduced risk. They found that there seemed to be a pattern. That's different then saying the flu vaccine reduced risk. Its possible that the vaccine is just a coincidence.

But the pattern looks pretty strong so its interesting news nonetheless.

Access to, and willingness to take, flu vaccinations is the first thing that comes to mind for me. It does narrow the set somewhat to people with certain habits.

"Reduce" is also an action verb that means causation.

“The team did indeed find that suntan lotion reduced the risk of rain. But he specifically doesn't say they know why the suntan lotion lowers the risk.” - spandrew, probably

<--- Best comment ever.

Indeed there is a link but we don't even know the direction of causation. I find it quite plausible that people with as-yet undiagnosed early-onset Alzheimer's forget to get flu shots.

Yet, you also find it plausible that people who spent years on this project would never have thought of that.

Your misplaced trust in your own plausibility filter causes you to make foolish posts on HN.

This comment (and often the "correlation != causation" discussion more generally) seems to equate "causation" with "proof of causation" (in a mathematical sense), which is a flawed criteria for science IMO. The causation part (not the mere association) is the substance, so it's appropriate to use causal language.

Of course any such explanation is provisional. I might think your comment is just to complain the provisional aspect isn't mentioned in the quote (I think it'd be tiresome to say that in every statement, and the public's understanding of science being provisional is underestimated), but you seem to think it's a problem that causal language is used at all?

While quote may have been offhand, look at what the actual paper says [1]: "[the study] design prevent[s] strong conclusions regarding causation." Note that it doesn't say it prevents any conclusions regarding causation.

By the way, it seems weird that you yourself have come to believe it's causal, while seemingly denying that this paper provides any basis. How did you come to that conclusion then?

[1] In the last paragraph: https://content.iospress.com/download/journal-of-alzheimers-...

> In theory you could have A and B that were naturally correlated and add a causal relationship between the two that exactly canceled that correlation but that would be infinit[el]y unlikely to happen by chance.

That hardly seems relevant, since we're talking about causal relationships and chance is the opposite of causation.

> Yet, you also find it plausible that people who spent years on this project would never have thought of that.

I find it plausible that the few people who spent hours reviewing this paper didn't think of that, and that the people publishing the paper ignored it since ignoring it could help them gain more funding.

> Your misplaced trust in your own plausibility filter causes you to make foolish posts on HN.

It is foolish to believe that scientists does things properly. As an outsider you only see the outlier results, meaning that most of what reaches HN could be faked or statistical trickery or have simple explanations due to poor science since poor science is more likely to produce headlines, while most scientists could still do the right thing.

Scientists are very frequently wrong. Any scientist who doesn't admit that is none. One example is adjacent to the topic here, continued adherence to the amyloid beta hypothesis.

But their mistakes are generally subtle, and not what somebody who knows nothing about the subject invents in the first seconds after first hearing about the topic.

> They can't claim causation if they don't understand the underlying reason for the change. That's not how science works.

Sorry but I had to nitpick here. This is exactly how science works. We first observe things that we cannot explain, and we can definitely infer causation without a complete mechanism or even a proper theory for it.

It's a little more nuanced. Causation can be inferred in an experimental design. If the researchers can manipulate the independent variable (the vaccine) using an experimental and a placebo group, and then they can measure a statistically (and clinically) significant difference, then we can assume that this is not just correlation.

They were very precise about causation in the interview.

They just can't include it in the paper because reviewers will make sure to remove it

The quote isn't offhand, it's directly and explicitly making claims of causality, e.g. talking about the "effect" and saying the vaccination "reduces the risk".

Yes the paper says differently. So what? We're used to this from epidemiology by now. The claims they make to the government/media/public about disease frequently don't match their actual data. To discover this you have to not only read the paper but often dig through the most obscure parts of it. The fact that their claims are wrong will only emerge in, like, table 3 of Appendix 2 which is by the way only available on GitHub if at all. Here it emerges in literally the last paragraph. Then someone blogs about this and they get kicked off Twitter for spreading "misinformation".

The public's understanding of science being provisional is actually excellent and far better than the supposedly elite decision making classes. That's why the public increasingly doesn't trust claims made by scientists, and correctly so. Scientists will make bold claims of causation whilst actually having a sketchy P=0.049 regression at best and a fictional model built on circular logic at worst.

We need to hold scientists to higher standards, and especially epidemiologists. The amount of damage their sloppy "offhand" approach has caused is astronomical. Or, quicker, just accept that they aren't going to improve, have learned nothing from COVID and cut them out of society and the public conversation entirely.

That's the thing though. A heater controlled by a thermostat will considerably reduce the correlation between outside and inside temperature but because it's imperfect it wont' eliminate it entirely.

I have no idea what you were saying by "Causation frequently removes correlations that would otherwise exist" then.

You missed my point. I don't actually think the quote was offhand either (another comment suggested that possibility), but the lines from the paper I cited also uses causal language. My point is that using causal language is fine (understanding that it's provisional as all science is), and I think the scientist quote is fine.

You think the scientist quote is dishonest? It seems you too conflate causality with something like proven mathematically or 100% confidence.

If they're going to claim causality for vaccines->less Alzheimers then yes, they need pretty close to 100% confidence for that because this is the sort of thing that gets translated into mandatory government policies. What they have here is literally nothing, it's just a correlation. They don't have any evidence of a causal relation, and they don't have any suggested biological pathway either. It's malpractice to assert causality given such a total absence of evidence.

Moreover, as a killed comment elsewhere in this thread points out, it's very likely that they made a mistake somewhere. This claim is absurd on its face. A 40% effect size is enormous. Alzheimer's is tracked very closely and there has been no change in incidence over time:

https://ourworldindata.org/grapher/prevalence-of-dementias

Flu vaccines on the other hand have become far more prevalent over the last 20 years especially amongst the age groups most at risk for Alzheimers. So, where's the impact? There isn't any. If flu vaccines really reduced the risk that much then we'd see it in the actual numbers, as a 40% reduction is hard to hide.

It's not facetious. It's quite the stretch to say that if you assume literally any injection could cure any disease, it's OK to describe any found correlation as a "link".

This sort of intellectual looseness is not free. People are learning to treat claims by scientists as untrue, and it's partly because of this sort of press release/paper hacking.

Finding a correlation between two random medical data sets does not mean there is a "link" in any English that normal people would use it. It definitely does not mean there's an "effect" or that one thing "reduces" the other. It might mean there's something worth a followup investigation there, though given the prevalence of non-reproducible p-hacked results in science, also maybe not.

Regardless, before doing press releases and going to the public with such a claim there is a large amount of work needing to be done to actually prove causality. Moreover you'd then want to ask why does such a thing happen when there is no prior reason to suspect such an impact.

We're talking about epidemiology though. Their papers routinely contain major, grievous errors that anyone can spot in five minutes. There might well be subtle errors too but it doesn't matter when the field is overrun with papers that are quite obviously invalid on their face to any outsider, yet none of the insiders care.

The primary COVID model that triggered lockdowns was full of programming errors. It had never been peer reviewed, and its prediction of deaths varied by 80,000 depending on whether you engaged a data loading optimization or not. It gave totally different results depending on available CPU features! There were no tests and the results had never been validated against anything. Outsiders pointed out these problems, and the team didn't care, nor did anyone else in the field of epidemiology.

That's just one example of many. Epidemiology is kinda like the phrenology of our era (one of quite a few). It's not built on a firm scientific foundation.

The presence of protective effects reduces the expectation of future observations of something, and merely describes temporal correlation. I so far just see standard language used in literature.

> It's not facetious. It's quite the stretch to say that if you assume literally any injection could cure any disease, it's OK to describe any found correlation as a "link".

I don't see it that way. Nick Cage has nothing to do with Alzheimers. We're not finding correlations with Nick Cage movies and then saying "Nick Cage linked to reduction in Alzheimers".

They are suggesting that flu vaccination may have some 2nd order effect beyond protecting you from the flu. Which could be reasonable, science reporting and poor research notwithstanding.

Either way, the refrain "Correlation does not imply causation" is over used in my view. And I'd rather learn the specific ways the research is flawed.

Just that. Variables that are causally related frequently show no correlation, or a weak correlation with the opposite sign from the one you would expect. There's already an example sidethread from me - the correlation between the temperature outside a house and the temperature inside a house would ordinarily be close to 1. But a normal modern house tampers with the indoor temperature, reducing the correlation to something close to 0 instead. This is as dramatic of a change in correlation as it's possible to see, altering "basically the same thing" (the correct answer) to "basically unrelated" (correct in an observational sense, but wildly off in a causal sense).

It could be reasonable but like I said, it's a massive stretch. Flu vaccines are not designed to stop Alzheimer's. There is no argued or known biological pathway through which that might happen. Alzheimer's experts have not previously identified flu vaccines as doing anything that might help. And it probably isn't flu itself causing Alzheimer's because, as I've noted elsewhere in this thread, flu vaccines don't appear to have actually reduced overall flu mortality and some studies show negative effectiveness.

https://www.healio.com/news/primary-care/20200302/flu-vaccin...

I don't think there's a really killer argument here in the abstract: I personally find it unreasonable to imply causation from any medical intervention to any possible outcome based on just a correlation. Yes, it's more reasonable than Nick Cage being associated, but not reasonable enough. That's a judgement call however. I am guided in it by the massive costs and problems created when scientists claim vaccines are miracle cures without sufficiently robust data.

> Flu vaccines are not designed to stop Alzheimer's. There is no argued or known biological pathway through which that might happen.

You might want to read the paper before saying untrue things like that so confidently:

“Mounting evidence indicates that systemic immune responses can have lasting effects on the brain and can influence AD risk and/or progression. A diverse range of microorganisms and infectious diseases have been associated with an increased risk and/or rate of cognitive decline, particularly among older adults, including influenzal respiratory infections [5, 6], pneumonia [4, 7], herpes infections [7], chronic periodontitis [8], urinary tract infections [4], gastrointestinal infections [9], sepsis [4], and most recently COVID-19 [10]. Prevention or attenuation of microbe-related inflammation may therefore represent a rational strategy to delay or reduce the risk of neurodegenerative disease. Consistent with this hypothesis, studies have found a decreased risk of dementia associated with prior exposure to various adulthood vaccinations, including those for tetanus, diphtheria, pertussis (Tdap) [11–13]; poliomyelitis [11]; tuberculosis [14, 15]; herpes zoster (i.e., shingles) [6, 13, 16, 17]; and influenza [11, 18–21].”

Unlike your Nick Cage theory, this has a clear mechanism and is compatible with the understanding of similar effects.

That supports my point, no? The paragraph you cited boils down to "there is an association between being less sick and not getting Alzheimer's (i.e. being less sick" + "there seems to be an association between literally any vaccine regardless of mechanism or target and less Alzheimers". These are not specific biological mechanisms of action, they're just correlations supported by an ultra-vague causal hypothesis of the form "maybe microbes cause Alzheimers". That's not actually a causal biological explanation.