(A much shorter version of this article has been published in The Huffington Post)
You’ve probably heard by now the WHO’s announcement that human consumption of processed meat is well associated with some kinds of cancer, most significantly colorectal cancer. I’ve seen a lot of reaction to this announcement, almost none of it moderate, balanced or thoughtful. At one end of the continuum, vegans are basking in their smug “Told ya so!” stance. Whereas, at the other end, carnivores (and, presumably, pork industry lobbyists) cling to any perceived holes in the science, and decry the usurpation of the WHO by PETA-style activism.
As an Epidemiologist and someone who styles himself as some sort of public educator, I felt I needed to weigh in.
First, my full disclosure: I am a carnivore. More accurately, I’m an omnivore who quite prefers green vegetables, but who consumes a veritable mammalian menagerie of critters every week. In fact, back in my 20s, I wanted to “eat one of everything that has ever lived”, and embarked on a globe-spanning tour of various cultures and cuisines, bravely swallowing all manner of avian, quadruped, and the occasional multiped. (That particular, ill-advised quest ended after I became predictably ill after eating a “free range” rat kebab in China.)
As I write this, I’m sitting in a park in Potsdam, Germany, having overdosed for several days on all manner of questionable (yet delicious) street meat. The closest thing to a vegetable that’s been near my mouth in the past week is my mint-flavoured toothpaste.
In my middle ages, I’ve become something of a soft-hearted animal rights crusader. I still eat non-human animals (hereafter in this article simply referred to as “animals”), but I do so guiltily and regretfully, and always after trying my best to avoid meats produced via factory farming or otherwise cruel domestication practices. If it were not for my moral weakness and well evolved hypocrisy, I would likely be one of the smug vegetarians who are currently basking in the warm Twitter glow of the WHO’s announcement.
But despite my personal political bias that celebrates anything that moves human civilization toward greater appreciation of the rights and liberties of animals, I approached the WHO announcement from the perspective of an Epidemiologist, that problematic profession that is so much fun at parties. To paraphrase a quote from my favourite play, Friedrich Durrenmatt’s “The Physicists“, let’s break it down to vulgar fractions…
How Was The Study Done?
The report in question is something of a systematic review, meaning that its authors did not collect any new primary data or perform any experiments. Rather, they reviewed existing studies, assessed them for strengths and biases, then synthesized their results to garner a final, informed estimate of the size of the effects in question. The classical hierarchy of evidence places randomized controlled trials (RCTs, also called clinical trials) near the top of the pyramid of quality study designs. We call it th “gold standard” of clinical evidence. However, a systematic review comprised of good RCTs is considered by many to be even better than a single, good RCT. This is what has led to the explosion of organizations and repositories committed to the production of good quality systematic reviews. The Cochrane Collaboration is the best known of such movements.
According to the article, “the greatest weight was given to prospective cohort studies done in the general population. High quality population-based case-control studies provided additional evidence.” Cohort and case-control designs (also collectively called “observational” designs) are not RCTs, and are therefore considered to be lower quality study designs. The main reason is that observational designs have poor “internal validity”, as compared to RCTs. This means that while strong and meaningful associations can be measured using observational designs, causality is very difficult to assess.
An example of a cohort observational design in this context would be if researchers identified some people who eat processed meat, and some other people from the same community who do not eat processed meat, and waited to see which group would manifest a greater proportional incidence of cancer.
An example of a case-control observational design would be if researchers identified some people with colorectal cancer, and found some very similar people without that type of cancer, then investigated to see how much processed meat both groups had consumed in the past.
Both are strong designs. However, I think it’s clear that in both scenarios, a researcher typically cannot control for other potentially causal factors, such as a family predisposition toward cancer, or other comorbidities. There is a concept in epidemiology called “confounding” wherein a variable or factor either masks a real relationship or creates the impression of an illusory relationsip. Consider, for example, the possibility that people who consume a lot of processed meat are probably also people who engage in other unhealthy activities, like smoking, alcohol abuse, or sedentary behaviour. Smoking, drinking and exercise would then be confounders, and would create a false impression of an association betweem meat consumption and cancer.
I like to cite the oft quoted statistic that married men live longer than single men, a (probably false) observation that allows many pro-relationship people to smugly conclude that bachelors should not cling to their single status for too long. However, it’s quite possible that unhealthy men tend not to get married, and that healthier men are more attractive marriage partners, thus creating the illusion that there is a causal relationship between marital status and longevity. The confounder in that case is men’s health status.
It’s possible to stratify analyses by such confounding factors, but in general observational designs cannot absolutely determine whether an exposure –like meat eating– causes an outcome –like cancer. The best it can do is to determine whether a behaviour is associated with that outcome.
For example, low income in the USA is associated with obesity. But low income is not the cause of obesity. Rather, it’s what we call a risk factor. In this example, the risk factor manifests its influence through preventing people from being able to afford healthier foods, or from not having the education to recognize healthier behaviours, or by causing them to associate with other people with unhealthy beheaviours, or to live in areas with poor opportunities for physical exercise. You get the picture.
Public health researchers love to identify and measure risk factors. It often doesn’t matter that one cannot completely determine causality, since we can effect changes in the outcome simply by addressing the risk factor. For example, low income might not be the proximal cause of obesity, but programs directed toward improving income will eventually manifest as decreased obesity rates. Similarly, if meat consumption is a strong risk factor for cancer, then reducing one’s consumption of meat will quite possibly (though not always) result in a lowered cancer risk, regardless of the exact causal pathway.
The authors of the study in question, representing the International Agency for Research on Cancer (IARC), assessed 800 studies, and found that 18 cohort studies and 14 case-control studies were of sufficient usefuless and quality to be examined. They then placed their assessment of the risk posed by meat consumption in the appropriate risk category along with other potentially carcinogenic exposures.
They put processed meat into group 1, which is the group of exposures known to be carcinogenic to humans. As well, they put the red meat into group 2A, which is the group of exposures which are “probably” carcinogenic to humans. If you’re wondering how they defined processed and red meat, you can read the WHO’s FAQ on the issue.
Now, some journalists have gone a bit crazy over this study. The Guardian’s headline was, “Processed Meats Rank Alongside Smoking as Cancer Causes.” This is a misleading headline. The report simply put processed meat in the same category (group 1) as cigarette smoking (and arsenic and asbestos), but did not claim that it was as dangerous as smoking.
One convenient way to think about the way in which this grouping masks some wisdom is to consider this truism: There probably is a safe dose of processed meat. But there is no realistically safe dose of smoking, asbestos, or arsenic.
Statistical vs Clinical Significance
The WHO team also conducted something called a meta-analysis, meaning that they collected all of the individual risk estimates from each of the reviewed studies and computed a grand statistical estimate of the overall risk. This is the number that has been reported by many journalists. From the risk estimates extracted from 10 cohort studies, it was found that the overall increase in risk of colorectal cancer was 17% from red meat and 18% from processed meat.
These numbers were statistically significant. It’s important to understand what this means. Statistical significance is not necessarily clinical significance. In this case, the statistical test being applied was the computation of something called a relative risk. This means that statistical significance means only that if the study were repeated an infinite number of times, there is a vanishingly small probability that the measured difference in risk would be zero. In other words, statistical significance tells us that something is likely going on, but the magnitude of that “something” has some wiggle room.
The authors reported more than whether or not their findings were statistically significant, however. They reported something called a confidence interval. Specifcally, they computed that if the study were repeated an infinite number of times, 95% of the time the increased risk posed by red meat would be somewhere between 5% and 31%, with the most likely estimate being 17%. Similarly, the increased risk posed by processed meat is somewhere between 10% and 28%, with the most likely number being 18%.
The easiest way to interpret this is to consider that those who eat processed meat have an 18% higher likelihood of developing colorectal cancer than those who do not eat processed meat. (The dosage they used was risk per 50 g per day of meat).
Relative vs Absolute Risk
Probably the biggest misunderstading of this study concerns the nature of relative risk. Think about a city of one million people. If I tell you that the incidence of Malaria tripled in this city over the past year, you might be horrified. You might wonder why this public health crisis has manifested and what the government intends to do about it. But when you look more closely at the actual data, you see that the incidence of Malaria did in fact increase… from one case to three cases.
A change from 1 to 3 is indeed a tripling. It can also be expressed as a 200% increase in risk from one year to the next. But I think you will agree that it does not constitute real, meaningful risk. The number or proportion of new cases in a given group or time is called an absolute risk. The ratio of two absolute risks is a relative risk, or risk ratio.
So which one is more important, the relative risk or the absolute risk? That depends on the clinical or policy question you’re asking.
An 18% increase in risk from a non-meat eating state to a meat-eating state is a relative risk. Technically, we say it represents a risk ratio of 1.18. The question we need to ask is, how much baseline risk is there initially? It’s been argued that pretty much everything we are exposed to is in some way carcinogenic, meaning that there exists a baseline risk for colorectal cancer even for vegans. But if that baseline risk is vanishingly small, then an 18% increase really is not appreciable.
In fact, it’s already been reported that eating these meats increases one’s lifetime risk for colon cancer from a baseline of 5% to 6%. Is a one percent increase in lifetime risk a big deal? Again, that’s a debatable point.
According to the WHO, there are about 8 million cancer deaths globally per year. About 700,000 of those (less than 9%) are colorectal cancer deaths. In the USA and Canada, about 36% of new diagnoses of colorectal cancer will be fatal, compared to a global fatality rate of 51%.
To keep things in perspective, in the USA more people will be killed by unintentional poisoning or influenza. But cancer is a big deal. And among cancers, the colorectal type is usually the 10th or 11th most proportionately fatal.
What To Do?
Prevention is always the preferred strategy for avoiding disease. If we can modify our behaviours and minimize risk factors, we can maximize our chances for a long healthy life. While the avoidance of processed meat potentially offers a reduction in colorectal cancer incidence of 10-28% (and presumably a commensurate reduction in risk of death), there is another option available to us that offers even greater prevention potential. Screening for the early stages of colon cancer among those people known epidemiologically to be most at risk has been shown to be enormously beneficial. If colorectal cancer is identified early enough, the 5-year survival rate is 90% (compared to a survival rate of 11-12% at much later stages of diagnosis).
There you have it. Epidemiologically speaking, you can probably continue to have a bit of processed meat now and then without exposing yourself to an appreciable risk of cancer, so long as you are conscientious about getting screened when you reach the appropriate age. That’s my take on this, as a non-clinician numbers wonk.
Of course, there are other reasons to eschew the meat –environmental, economic, and ethical– but that’s an entirely different conversation!