A while back, when Dr. Perlmutter was doing interviews to promote "Grain Brain," I participated in a spirited online discussion about his ideas and the sources he used to support his ideas. One particular comment I made was this:


Just yesterday, I received a reply to this comment:

I understand well the basic limitation of observational studies: they can't establish cause and effect although the correlation of the analyses cited look quite suggestive. But I think that the commenter's point about looking at individual rates of death rather than country-level rates is valid.

Does anyone else have any thoughts on this commenter's argument?

But it might mean that smoking by itself is insufficient to cause cancer--or that some other factor greatly increases the effect of smoking on cancer.



If you look at the cholesterol conundrum video RawNut posted yesterday--the presenter showed data suggesting that a high particle count, or high triglycerides, might only be risk factors for cardiovascular disease on the condition that insulin be elevated past a certain point as well. A diet and lifestyle that raises all of these might increase heart disease, where one that raised all but insulin might not. Something could be a risk factor in one setting, not in another.



On the individual data point--well, okay. But looking at Western Europe and North America, we're looking at a relatively homogenous culture. What if something about these cultures--high sugar intake? Years of access to trans fats? Maybe something leading to high insulin--makes high cholesterol a risk factor (especially if high cholesterol travels with a high particle count, high triglycerides, low hdl, as it often does when insulin is high).

Pretty long-winded way of saying that context is everything. Maybe tweaking cholesterol can improve things some, in a certain context. Maybe there are other options so effective that they make cholesterol and statins practically irrelevant. Hope so.

If you concede the point about needing to use studies of individuals rather than whole countries, than that well-known graph becomes meaningless, doesn't it?

If you concede the point. I sure don't. Something I remember from a well-known vegan advocate--I think it was McDougall, of all people--applies here, I think. He was talking about epidemiological studies that showed very weak associations of disease with meat intake vs plant intake. What he basically said, was that if you look at meat intake in some part of America, you won't find much advantage to decreased meat intake. Basically because veganism is so wildly unpopular on a population basis, that when you split things into quintiles, there often just isn't enough difference in the diet and lifestyle between even the top and bottom quintiles to make all that much difference. Looking at data from different countries, instead of individuals--or looking at what people ate in the 17th century, and how they did atherosclerosis wise vs people today, if you could get reliable data to make the comparison--might give us some valuable information.

If you think about it, this applies to paleo as well. It sidesteps the idea that eating like people who didn't get metabolic syndrome, type II diabetes, etc. might be a good idea.

Treat a thousand with statins, how many benefit? As many as if you find a way to lower insulin, increase insulin sensitivity, etc.? I doubt it.

You lost me after "If you concede the point. I sure don't." How does what you said connect with asserting that using country-wide statistics to prove a point is invalid?

But the person making the argument about country wide statistics him/herself then posts a study using statistics from North America plus Western Europe. How is it valid to lump all these countries together, but not to compare country to country? It's very convenenient to try and define the rules of a debate that you're right in the middle of.

I'm not saying that you can't get important data from the approach suggested. But that isn't to say that you can't get important data from a country vs. country approach. The one doesn't necessarily trump the other--we might just have a hold of another part of the elephant or something.

What?! The critic specifically said "A meta-analysis of such individual-level data...". The title of the study is Blood cholesterol and vascular mortality by age, sex, and blood pressure: a meta-analysis of individual data from 61 prospective studies with 55 000 vascular deaths. How does that become country-level data?

I think a valid rebuttal would be along the lines of "how do you think country-level data is compiled? It's from a collection of individual-level data."

Okay then... how is it reasonable to lump individual data from all these countries together but not to compare country to country data? How is the one approach reasonable and useful, and the other not? How does individual data from say Framingham invalidate a study comparing the Swedish to the Kitivans? Why exactly shouldn't we look at differences between countries?

To me, those WHO graphs, or the "22 country" graphs meant to counter Ancel Keys infamous study don't show that cholesterol isn't a factor in heart disease or mortality. It's entirely possible for cholesterol to be a major factor in one country, and not another. I don't know if this is the case, but it's possible. So in a way, I end up agreeing. I just disagree that the WHO graph is entirely useless.

Two rat cages. In one the rats are exposed to aflatoxin from rancid peanuts, in another, they aren't. In one rat cage--a high casein diet causes high liver cancer. In the other, it seems to be harmless. Yes we ought to pay attention to any apparent effects of excess casein in one of the cages. But we ought to pay attention to the cage where casein doesn't seem to matter, as well.



My root argument is that the graph isn't meaningless. It's maybe not what the author intended (the author of the WHO graph, that is), in that it's not a slam-bam refutation of a relationship of cholesterol to heart disease. What I see in this and in the 22-countries type graphs is that at a given average national cholesterol level, you can have a wide variation in heart disease mortality.

It would be interesting to see a less Western-centric study like the one posted with data from mostly Western Europe and North America. In places where the WHO data shows high mortality from heart disease paired with lower average cholesterol, I wonder what the relationship between changes in cholesterol and heart disease looks like?

That, I think, is a reasonable rejoinder.

That doesn't address one of critic's main thrusts: the graph, while allegedly drawn from WHO data, isn't specific about the methodology behind the graph, nor has that methodology been peer reviewed. That immediately makes the graph suspect and invalid to use for advocating anything.

That would be a reasonable line of thinking except for the fatal problem noted earlier.

Someone or some group with reputable credentials (e.g. a Cochrane Collaboration group) would have to validate the data/the data collection process and the graphing methodology. Then, yes, a comparison would be interesting.