Thread by @GidMK

Oh look, lots of people are Wrong On The Internet today

1. This study was observational
2. The absolute risk difference (WHICH THEY REPORT) was tiny
3. The "organic" group were healthier in lots of ways

What did the scientists do? They asked French adults about what they ate, classified them into groups according to organic food consumption, and then compared these groups risk of cancer

Study is here: jamanetwork.com/journals/jamai…

In observational studies, it's always hard to draw causal conclusions, because it's possible that something external to the study caused the results

I will say, this study did the best job I have yet seen of controlling for confounders

Still there are several that I can immediately see might be an issue (i.e. ethnicity)

The main results are interesting. There was a significant (p<0.001) trend towards organic food being protective at the highest levels of intake

However, it's worth noting that the only real difference was between people with the highest and lowest intakes

Their "organic food score" was based on the responses to whether people ate 16 types of organic food

There is a significant issue with bias here

The issue is that most of these foods are "healthy" options. What this means is that only people who ALREADY EAT the healthy choices are going to eat the organic version

To their credit, the authors did try to control for this in the analysis, but it's still something that is likely to influence their results

Another big issue is that people who ate more organic food were MUCH healthier on every metric than people who didn't

Again, the authors controlled for the factors they know about, but residual confounding is very likely here

And then we come to the absolute risk difference, which as I mentioned was TINY

Remember the Daily Mail? Reporting that 86% reduction in "blood cancer risk"?

Firstly, they got it wrong. The odds ratio was 0.24, so the reduction was 76%

The absolute reduction there was less than 0.1%!

(Also for stats-y people, the 95% CI was 0.04-0.66!)

In fact, if you want to look at individual cancers - rather than the trend for all cancer - organic food was NOT associated with MOST cancers (red = associated, green = not)

We could just as easily write a story about how organic food does nothing for breast, prostate, colorectal, and skin cancers, but that's a much less interesting headline

So, what to take away from this study?

Firstly, there are actually already bigger, arguable better studies on the topic that have found the opposite result!

Secondly, THIS HAS NOTHING TO DO WITH PESTICIDES

This study didn't look at pesticides at all. We have no idea if organic consumption equates to reduced pesticide ingestion - especially considering that there is some evidence to suggest the opposite!

Thirdly, the absolute risk is tiny

The "high organic" group ate MORE THAN 20x the organics of the "low organic" group, and they only saw a 0.6% reduction in risk

That's a huge expenditure for a minuscule benefit, even if this study is correct

Finally, there's a good chance that these results are meaningless. The more factors the authors controlled for, the smaller the statistical difference

There's a good chance that if you could control for everything, the result would disappear entirely

This study probably means very little to your life. Eating organic is ~probably~ better for the environment, but that's about it

I've written about this before medium.com/@gidmk/organic…

Also, this isn't a criticism of the study, the actual research was pretty cool. I would say that the authors were a bit optimistic in their conclusion, but otherwise it was interesting epidemiological research

Something I missed earlier - it's also worth noting that in most of the interesting subgroups the association totally disappeared

What this means is that organics are likely only useful in reducing the cancer risk of elderly women, which to me points to the results being likely down to statistical noise

If the effect disappears when you don't look at a single group of cancers - postmenopausal breast cancer - then it's more than likely it's not there at all

Also worth noting that the results are probably not generalizable, considering that this sample was heavily weighted towards highly-educated French women