Correlation and causation are topics that have become a part of the anti-vaxxer claims regarding the links between vaccines and adverse effects. I hear both terms thrown out so frequently, it’s hard to determine what is what.
Correlation is a statistical technique that tells us how strongly the pair of variables are linearly related and change together. It does not tell us why and how behind the relationship but it just says a relationship may exist.
Causation takes a step further, statistically and scientifically, beyond correlation. It is any change in the value of one variable that will cause a change in the value of another variable. It is often referred to as cause and effect.
For example, biological plausibility is one of the requirements to establish that correlation means causation. It is almost an essential requirement for one to claim a causal association. But biological plausibility must be consistent with our existing knowledge of biology, chemistry, physics, and medicine.
How many times has an anti-vaccine zealot tried to convince us that “mercury in vaccines causes autism” but ignores the basic scientific tenets of numerous fields of biomedicine like biochemistry, cell biology, toxicology, immunology, neurology – well, just about every field?
Or someone who claims that acupuncture treats a bunch of diseases, yet we cannot find any reasonable biological plausibility between sticking a needle in the arm to treating some medical condition like pain. They tend to ignore the need for biological plausibility by using their own personal anecdote as “proof.”
That’s why science is much harder than pseudoscience. Establishing correlation and causation requires a strong knowledge of a scientific or medical specialty to make the case. It’s much more than simply stating that plausibility does exist, you have to use actual real science, published in real scientific journals, to make the case.
So let’s talk a little bit about correlation and causality.
Correlation and causation – Bradford Hill
I’ve written about correlation, causality, and plausibility before, but I’ve never felt that I made the case appropriately. So I started to investigate more about how we determine when a correlation is equivalent to causation, and I saw that some researchers use something called the Bradford Hill criteria.
English statistician Sir Austin Bradford Hill was interested in developing a set of objective criteria that could be used to provide epidemiological evidence of whether correlation equals causation. It serves as a sort of checklist for scientists who can take data that establishes correlation and then logically determine if that supports causality.
He used his criteria to establish that smoking was linked to lung cancer (and other diseases). He essentially went through each point of his criteria to show how smoking and cancer were linked.
The Bradford Hill criteria include the following aspects:
- Strength (effect size)– this is one of the important parts of this criteria – the larger the effect from the cause, the higher the probability of a causal link. This doesn’t mean small effects aren’t important, it’s just that fields like science-based medicine favor larger effects. For example, if we say drug A cures the common cold, but the course of the disease is only reduced by ½ day, then it’s hard to tell if it’s a result of randomness in data, bias in results, or actual clinical effect.
- Consistency (reproducibility) – proposed causality needs to be observed in more than just one location. Consistent data published by different researchers in different locations with different population samples strengthen the possibility that there is a link between a cause and effect.
- Specificity – causation requires a very specific population with a very specific disease with no other possible explanations of that causation. Again, the more specific an association is between cause and effect, the larger the possibility of a causal link.
- Temporality – the proposed effect must occur after the cause, and within a likely time period for which a link between cause and effect.
- Biological gradient – there must be some sort of dose-response effect, that is, the higher the exposure to some cause should generally lead to a higher incidence of the effect. (There are cases where a lower exposure leads to a higher incidence, so we should observe the inverse effect.)
- Biological plausibility – as we will discuss next, there must be a biologically plausible mechanism between cause and effect. Of course, it is possible that we lack knowledge of all possible mechanisms, but inventing one out of thin air is not going to help the “cause.” Even then, the potential new mechanism must fall within the basic principles of biology, chemistry, and physics. Biological plausibility is probably the most important factor in this list.
- Coherence – does the proposed cause and effect fit with what we know about the possible adverse effect? If you want to claim that the HPV vaccine causes autonomic dysfunction, yet our knowledge of what causes it has nothing to do with the HPV vaccine, then it’s going to take a stack of evidence to establish why this might be.
- Experiment – does a group that lacks exposure to the effect exhibit a different outcome? For example, large case-control studies of vaccines examine the risk of a particular adverse event compared to a group that is unvaccinated. In this case with vaccines, it’s hard to establish correlation let alone causation.
Bradford Hill developed this checklist over 50 years ago, so you could assume that there has been some evolution to the list. Some people have added one or two items to the list, like examining confounding factors and experimental bias. Those are usually evaluated in the original epidemiological research that establishes correlation, but if not, they become an unofficial part of the checklist.
These criteria should be used as a checklist. The more points that you can check off, the closer you can come to support a claim that there is a causal link between cause and observed effect.
Bradford Hill criteria – vaccines and autism
Let’s take a look at the hypothesized effect between vaccines and autism spectrum disorder (ASD), which has been thoroughly debunked by scientific research. How do the observations of vaccines and autism fit within the Bradford Hill criteria?
It’s hard to do this exercise since we know the outcome – there is no correlation or causation. However, it will be a useful exercise to examine this central tenet of the anti-vaccine movement:
- Strength – there is simply little robust, unbiased published evidence that establishes a higher incidence of autism in vaccinated children than in unvaccinated children. So the strength of the association is nearly non-existent.
- Consistency – actually, the only consistent evidence is supporting the null hypothesis, that is, that vaccines are not linked to vaccines. There are over 150 studies of links between vaccines and autism, and there is no evidence of correlation or causation.
- Specificity – the claimed link between vaccines and autism may appear to meet the criteria of specificity, but remember, autism spectrum disorder is a wide range of neurodevelopmental changes that can be quite dissimilar. Furthermore, there are a large number of other, more scientifically-supported links to autism, including genetics.
- Temporal – some observations meet the temporal criteria, but it’s difficult to establish at what point a cause and effect are within some logical time frame. The people pushing the narrative about vaccines and ASD seem to think that the appearance of the symptoms of ASD at any point post-vaccine should be included. That’s not good science.
- Biological gradient – there just isn’t any robust published evidence of some sort of dose-response relationship between vaccines and ASD. If the supporters of the claim had strong evidence that the more vaccines a child gets, the higher the risk of ASD, we could check this criterion off for the link. But we can’t because there is no data supporting a biological gradient.
- Biological plausibility – this becomes one of the central issues of trying to determine if there is a link between vaccines and ASD. We don’t know a lot about the pathophysiology of ASD, but there is overwhelming evidence that it has a very strong genetic component. There just isn’t a biologically plausible mechanism that can lead from vaccines to ASD. The anti-vaccine squad loves to propose mechanisms that not only aren’t plausible but also require significant changes in our understanding of human physiology. For example, we understand that aluminum burden in children is 99.999999% from inhaled and consumed aluminum – yet the anti-vaccine zealots want us to believe that intramuscular injection of a sub-biological amount of aluminum somehow is more important than all other sources of the element, and only the aluminum from vaccines leads to autism. That is the definition of biologically implausible.
- Coherence – as we stated in biological plausibility, our understanding of ASD leads us to one basic idea – the initial symptoms occur at the same time as vaccines. And ASD has a large genetic component that is not induced by vaccines, which has only one effect – on the immune system.
- Experiment – once again, there is just no robust published evidence that shows that the non-vaccinated population has some lower risk of ASD. In fact, affirmative evidence is that they are almost the same.
Thus, at best, the claim that there is a link between vaccines and autism barely meets one of the Bradford Hill criteria – temporal. The problem with this one point is that irrespective of vaccination status, the initial diagnosis of ASD is often at the same age as a lot of recommended childhood vaccinations.
To me, the largest issue I have with the anti-vaccine claim about autism is biological plausibility. They keep moving the goalposts to try to twist logic into a pretzel to convince us of biological plausibility. They first claimed it thiomersal. Then aluminum. Then some random ingredients of the vaccine.
The only biologically plausible link to ASD is genetics. And it meets almost all of the Bradford Hill criteria. Imagine that.
No matter what you want to believe about any medical quackery, no matter how hard you want to convince yourself they are real, and no matter how much you want everyone to believe your anecdotes, finding a potential correlation then causation is very difficult. And it requires a logical process not a claim that it must be so because of anecdotes or belief.
There is a logical process that is required to get from correlation to causality. Those who attempt to shortcut that process to reach a pre-ordained conclusion means that they have neither established correlation nor causality.
- Fedak KM, Bernal A, Capshaw ZA, Gross S. Applying the Bradford Hill criteria in the 21st century: how data integration has changed causal inference in molecular epidemiology. Emerg Themes Epidemiol. 2015 Sep 30;12:14. doi: 10.1186/s12982-015-0037-4. eCollection 2015. PubMed PMID: 26425136; PubMed Central PMCID: PMC4589117.
- Mitkus RJ, King DB, Hess MA, Forshee RA, Walderhaug MO. Updated aluminum pharmacokinetics following infant exposures through diet and vaccination. Vaccine. 2011 Nov 28;29(51):9538-43. doi: 10.1016/j.vaccine.2011.09.124. Epub 2011 Oct 11. PubMed PMID: 22001122.