Julian, I wanted to send you this comment in a private note, but that option somehow doesn’t work on this post.

To preempt any misunderstanding, what I’m going to say is in absolute professional respect for your opinion, and in honest appreciation of the work that you put into this article. If you sense a big BUT coming, here it is:

I totally disagree with your recommendation to avoid milk. I have written an article earlier this year that supports my contention.

Here is my rationale, and I illustrate it using the paper that you quoted about the correlation between milk intake and prostate cancer (btw, it is not possible to derive causation from observational data, no matter how thorough the statistical calisthenics).

First, there is the reliance on assessing dietary habits with FFQs. The study participants had to retrospectively assess their diet at baseline for the entire preceding year. That is known to introduce systematic error and unreliable data. The participants were then followed for 7 years. As if dietary habits stay the same. But these sources of error aren’t the worst offense.

What kills the validity of any conclusion is what we know in statistics as vibration of effects (VoE). I’ll explain it using this study’s association of dairy with prostate cancer. Of all the known documented potential risk and protective factors to PCa, the ones that the authors corrected for (BMI, exercise, education etc) are only a fraction of that total. They corrected for altogether 12 cofactors, but I could easily quote around 20. Now, let’s assume we consider only 15 of those 20 to be taken seriously, then we can construct 455 different models with 12 cofactors out of the 15. Only if all the models, or at least an overwhelming majority of them confirm the dairy-PCa association by going into the same direction, could we assume that there is a link. In nutrition science, however, in most models the associations range from positive to negative or, if they stay in the same direction, from significant to non-significant. I have written an article about that (Why everything you read about food and health may be misleading). What the authors of the study did not consider, even though their detailed FFQ would have allowed them to tease it out of the data, is the dietary inflammatory index, trans-fatty acid intake, selenium intake, aspirin and NSAIDs, and statins.

Interestingly, if you look at figure 2, the HR for full-fat milk and total milk was non-significant. So, even this methodically flawed investigation does not support the vilification of milk.

I act accordingly. I have been drinking almost a litre of full-fat milk every day for many years. Add to that 1.5 pounds of full-fat yoghurt. That I don’t have PCa qualifies, of course, only as anecdotal evidence. But I can live with that.