Paul Jaminet has a cholesterol series going and I was intrigued by some curvilinear graphs showing how total cholesterol (TC) is correlated with mortality for each country: http://perfecthealthdiet.com/?p=3836
What's interesting in these graphs is that the lowest mortality level is associated with TC of 200-240. By contrast, mortality rises sharply as cholesterol falls below 200, more than it does when TC rises above 240. Clearly, low TC is more dangerous than high TC.
Overall, this seems like a great graphical presentation and it affords some of us whose TC falls within that range some sense of relief. The original graphs are from Ricardo Carvalho of Canibais e Reis: http://www.canibaisereis.com/2009/03/21/nutrition-and-health-database/
But think about the TC for a minute. This is a misleading and stupid benchmark, one that would produce more noise than signal in any statistical analysis. It's obvious if you examine the formula:
TC = HDL + LDL + Trigs/5
So you're adding HDL, supposedly a "good cholesterol" number to an arguably "bad number" (LDL) and adding 1/5th of the Trigs which reveal the metabolic status of your LDL and denote its particle size. Take a TC of 227, for example, for 2 patients:
TC(1) = 65 (HDL) + 150 (LDL) + 60 (Trigs) = 227 TC(2) = 35 (HDL) + 150 (LDL) + 210 (Trigs) = 227
They have exactly the same TC but the components tell something quite different. TC(1) is a low carber who minimized his Trigs to 60. We don't need a particle size test here because his 150 LDL will be the bouncy, fluffy kind. Plus he has high HDL, a marker of cardiovascular health and is inversely associated with heart disease.
Take the second patient: he is not a low carber and is consuming around 250-350g carbs, a fair amount of which is probably processed. HDL is low because of his diet and sedentary lifestyle, yet his LDL is identical. From the Framingham study, we know that the Trigs are correlated with particle size; in this case, a fair amount would be the smaller ones that are atherogenic and associated with heart disease.
This is just an example, but what results are you hoping for when you do analysis based on TC?
If you want to measure relationships, shouldn't HDL be subtracted from LDL (up to, say, 85, above which level it is meaningless)? And perhaps 1/2 or 1/3 of Trigs should be added to LDL to equal TC? How about doing "right" statistical analyses based on recalculated TC numbers (which should be readily available)?
I'm not criticizing the analysis done by PHD, which is valid: the TC number is what it is. That's why we examine its components and ratios (TC/HDL, Trigs/HDL). But if you're the American Heart Association, and you keep saying TC is too high, you immediately lose credibility.
Correlation does not equal causation, even when the hypothesis is consistent with paleo biases.
First, if cholesterol had that much to do with it, the curve would be a tighter fit -- i.e., you would NOT have some dots so far from the line (on both ends).
Second, if you exclude those cultures with very low cholesterol (on the theory that something like poverty/limited fat intake might be behind extreme low cholesterol values), then the graph looks much less compelling.
That said, it is a neat graph because it shows that the original evidence for the lipid hypothesis as presented by Ancel Key actually would have STRONGLY suggested the opposite conclusion from the one he reached if he didn't cherry-pick the evidence. Nonetheless this type of evidence is wholly insufficient to prove anything...