Relative risk

Now, smoking & lung cancer does fulfill the Bradford Hill criteria. Without going into detail, one of the reasons is the effect size. While a small observed effect does not exclude the possibility of causality, a massive effect makes it more likely that there is a causal link.

Let’s talk about red meat for a minute, since red meat is associated with bowel cancer (and a bunch of other cancers). You’ve likely read something in the news like “Red meat increases risk of cancer by 28%!”. Not knowing about what “relative risk” means, this sounds scary.

If for example we assume that 3% of people who do not consume red meat develop colorectal cancer, a 28% increase is NOT 3%+28% = 31%! 28% means a hazard ratio of 1.28, so the result is 3%*1.28 = 3.84%. Your absolute risk in your entire lifetime goes up by only 0.84%.

In contrast: Smoking and lung cancer has a relative risk of 1500%-3000%, meaning that smoking increases your risk of lung cancer by 15-30 times.

That is why it’s a pretty safe bet to say smoking causes cancer, even though the research to actually proof it does not exist. For meat however, the difference over a lifetime is way too small. For the most part, anything that’s below 500% relative risk ratio is a nothing burger. But that’s exactly what we’re getting, 7% here, 16% there,… it may as well be statistical noise.

Replication crisis

For a study to be considered valid, its results must be reproducible. This is currently not the case for an estimated 70% of studies. Especially in nutrition science there is a lot of flip-flopping. Eggs are bad, they’ll kill you. Never mind, eggs are good. Never mind again, eggs are neutral at best. This leads to a bunch of conflicting evidence… which one is it now?

But one thing is certain:
If there was a definitive, irrefutable answer, this would not be such a hot topic for debate. This simply indicates that we do not have the answer at all, at least not in form of a study.

Significant results

A result in a published study is often referred to as “significant”. What is meant here is simply statistical significance, but it is often misinterpreted as being clinically significant. Statistical significance merely refers to the reliability of the results, meaning the results did not happen by chance. This is usually denoted by the P-Value being smaller than 0.05.

Apart from the fact that researchers can employ so-called “P-hacking” to force the value below 0.05 in order to get published in high quality science journals, it is a meaningless significance for an individual persons choices.

My post about legumes showcases this pretty well, check it out!

The data of nutrition science

Science in general relies on the collection of data. The method of collection can very much impact the results of the study.

Ethnicity, age & gender

Data collected only on Thai women aged 65-85 can only be applied to that demographic. It cannot be applied to a 25-year old European male. Different sex, different age, different genetics.

Study duration

Data collected cannot inform on anything that happens after the studies’ duration. If the study lasted 3 years, you cannot extrapolate it to 5 or 10 years.

Studied population/subjects

If the studied subjects are mice/rats/rabbits/monkeys, you cannot apply the results to humans. Humans are not rats (well, maybe some are).

Also, if your study was done on only a small number of people, it is possible that the results are simply due to chance.

Confounding variables

This is a big one. Good epidemiological studies try to control for these as much as they can, but it’s simply impossible. People have the freedom to do whatever they want, and all of it can influence the outcomes. People take drugs, drink alcohol, smoke, sit on the couch in front of the TV all day, take medications, have different genetics, (don’t) work out,… The list goes on and on and on pretty much indefinitely.

Further, in order to get any kind of statistical significant result, authors like to do something called adjustments. The idea is that somehow these confounders can be “calculated out” of the data. To be honest, I have no idea how exactly that works in practice, I’m not a statistician. But the point is we do not know how much impact each confounder actually had. How do you know how much smoking, alcohol and medication influenced the development of cancer? You don’t, it’s a guess at best. The best analogy I’ve heard for it is “Bake a cake, and once you’re done, you attempt to extract the eggs from it.”

Biases

There are many biases that can impact the results. Just to list a few important ones:

Publication bias: Studies with positive results are more likely to be published than studies with negative results.
Healthy user bias: For example, a study on the benefits of exercise could be skewed because people who regularly exercise are probably health conscious. They are also are less likely to smoke/drink/eat junkfood and are more likely to spend time outdoors.
Confirmation bias: Researchers may have preconceived notions about the outcome of the study and interpret the data to support their believe or disregard data that does not support it.

Worthy to note here, I am biased too of course.

Food frequency questionnaires

Often people are simply asked what they ate. Let alone the fact that most people have no idea what they ate for breakfast yesterday, how would you accurately estimate what you ate for the past year? How many cups of bananas did you eat in the past 6 months? And people supposedly also like to lie on these questionnaires.

Conflicts of interest

A lot of industries have a vested interest in funding studies that support their specific product. The classic one is Pharma Industry, but this applies to any industry out there. While this does not automatically mean the study can be discredited, but it should be taken into consideration.

It would not be the first time studies and researchers have been abused to put certain products into a positive light. I’m looking at you, sugar industry.

As such, authors must declare these so-called conflicts of interest not without good reason. Actually, I’m gonna put an example here because I think it’s hilarious how vast it is. Mind you: The study is simply a “consensus statement”, nothing more than a bunch of people who got together and said “Jup, we agree on this.” So let’s see who paid for this marvelous piece of work: