More doubts about the Mediterranean diet


I recently criticized a Swedish study of the Mediterranean diet in which the lead author was Gianluca Tognon. Yes. An Italian really did study the Mediterranean diet in Sweden! The paper was titled: "Does the Mediterranean diet predict longevity in the elderly? A Swedish perspective".

He has replied to my comments as follows:

Hi, I’m Gianluca Tognon the first author of this article. On behalf of all the authors of this research, I would like to reply to some issues you mentioned in your article.

First of all thanks for having read it with criticism instead of just reporting the results, science is not based on absolute truths but it can just give the best estimation of the truth after a lot of discussions like this one.

You criticized the fact that we used a refined version of the Mediterranean diet score in order to get an association. The other “unrefined” score did not show an opposite association compared to this one. It was inversely associated with mortality but not statistically significant. This means that what it was measuring was not the whole story. [Why?]

What we did during refinement was to replace total cereals with unrefined cereals based on the fact that low glycaemic index food items are considered healthier than refined one. Then we included eggs, as a possible marker of a “western” diet (together with meat products) and polyunsaturated fats likely to be the kind of unsaturated fats in the healthy Nordic diet. Finally, we included alcohol intake because it’s a hallmark of the Mediterranean diet.

So, since we did not produce several random scores and then report the one that was associated, but based the refinement on previous knowledge, we do not consider this cheating. It is not the first time that this score is modified or adapted to different contexts: development and regional adaptation is a constant and evidence-based process.

Anyway, in the article we reported a sensitivity analysis which showed that none of the factors included in the score we used was able alone to explain our result. This is reported in the article we published.

What is important to say, is that our research was based on food groups (vegetables, cereals, fish products, etc.). Which food items are included in a specific food group can vary from country to country. The important thing is that the dietary pattern inspired by the traditional Mediterranean diet can be exported in other contexts without losing its healthy properties.

We think that explaining this result by just social class would be too simple. Our statistical model was adjusted for smoking, marital status, education and weight status, all possible mediators of the influence of social class on health.

So the association we obtained was independent of potential confounding factors (which, by the way, could also explain the differences in survival between Australian and Greeks since diet is not the only factor affecting survival).

We excluded “implausible” diet reporters by excluding subjects with extreme ratios between the energy calculated from their reported diet and their calculated basal metabolic rate. This did not modify the association making it positive instead of negative, but allowed us to get a better estimation of how much the effect was in terms of longevity increase.

The questionnaire was validated, and I have to tell you that validation of an instrument (like diet history here) with another one is absolutely the rule in epidemiology and not just in psychology as you reported. Diet assessment is subject to errors, validation allows you to reduce them as much as possible.

Finally, regarding antioxidants as a possible explanation, you’re correct that the discussion on this issue is still open. This is an observational study, so we cannot provide a biological explanation of what we observed. But we can speculate what the mechanisms could be. Antioxidants might or might not be part of the story, other researchers will then try to demonstrate if this is the truth or not.

Scientific evidence is not based on a single study, but it’s based on several results, obtained in different contexts by different independent researchers.

I am aware that this will not reply to all possible criticisms about this research, but again thanks for giving us the opportunity to clarify some important issues.

Prof. Tognon is a very civil and polite man so I am reluctant to be too hard on him but his reply really only amplifies my original criticisms. He used an existing index of the Mediterranean diet and found that it predicted nothing. He then dreamt up his own version of the Mediterranean diet and that predicted something. So which is the true measure of a Mediterranean diet? Surely the first one. The second one could more accurately be called the Tognon diet!

The only real social class marker he used was education but education is only a start. Many highly educated people are poor and many rich people are not well educated. Bill Gates never got a degree. A proper study of social class would also require inclusion of income, occupational status, self-perceived class and IQ. So his findings are very much open to explanation in terms of a class effect.

Finally, Prof. Tognon in his bracketed comment above does something I have been waiting for advocates of the Mediterranean diet to do. He admits that national diets tell us nothing certain about lifespans because other factors could explain long life. I have made that point myself in the past but I have mainly gone along with the joke and pretended that the longer lifespan of Australians (when contrasted with Mediterranean lifespans) tells us something. It does not, of course.

But the same applies to the claim of benefits from the Mediterranean diet. Mediterranean people get fewer heart attacks. So what? How do we know that that is due to their diet? We don't. It could be some genetic factor at work, for instance.

Prof. Tognon's admission has thrown the whole body of epidemiological assertions about the benefits of the Mediterranean diet out the window -- including his own assertions. Correlation is not causation.