Friday, December 30, 2011

My Observations on Denise Minger's Review of Ancel Keys

Denise Minger of recently wrote a blog post titled "The Truth About Ancel Keys: We've All Got It Wrong".

She had five "key" points about Keys' work near the end of the post which I have summarized here:

  1. The cherry picked six country graph showed a strong positive correlation between heart disease mortality and dietary fat, but the original twenty two country graph also showed a positive correlation, just weaker due to more variation in the data.
  2. Animal fat (saturated) was strongly correlated with heart disease, but plant fat (polyunsaturated) was inversely correlated. These relationships were reversed for other causes of death, which was reasoned that countries with higher standards of living consume more animal fat than plant fat, but also have better health care resulting in less "other" deaths.
  3. Heart disease deaths were inaccurately categorized by some countries.
  4. Food consumption was inaccurate.
  5. Correlation isn't causation.

I want to discuss her first two key points.

Key Point #1

I've heard from others that Keys cherry picked data, but thanks to Denise, her post was the first time that I saw the original data. And I was surprised at how convincing the original data looked showing a positive correlation between dietary fat and heart disease mortality. Here's the screen shot from Denise's page:

And here it is again highlighting Keys' cherry picking:

Without having the actual numbers for this data to do any statistical calculations, it's easy to visually see that Keys improved the fit of the data, but the overall trend is similar to the original data. This is not the same story that others have painted. In a Chris Kesser interview with Chris Masterjohn, this graph was a topic of discussion where it was said that other countries could be picked to show the inverse correlation. This was also stated in chapter five of Anthony Colpo's "The Great Cholesterol Con". Colpo cites Malcolm Kendrick that the opposite result can be obtained by picking Finland, Israel, Netherlands, Germany, Switzerland, France and Sweden. Here's what Kendrick's cherry picking looks like:

I think the lesson here is that a discussion about data is incomplete if there is no visual representation of that data. My conclusion is that Keys' cherry picking is a much more accurate representation of the original data than Kendrick's cherry picking.

Key Point #2

Regarding Minger's second point about countries with higher standards of living eating more animal fat, having higher heart disease mortality but less "other" mortality, Minger chooses to end this topic with her comment, "There’s so much confounding involved with this subject that I don’t even wanna touch it with a ten-foot statistical pole." Personally,  I'm curious to better understand if this correlation is important. Afterall, I believe in the perfect dataset about as much as I believe in Santa Claus. So the searching never stops.

Minger introduces Keys' infamous cherry picking by citing his Wikipedia page. Here, it states that Keys' interests in the relationship between diet and heart disease originated with his 15 year study of 281 Minnesota businessmen (Circulation, 1963). He monitored their body weight, skinfolds, blood pressure and serum cholesterol. The table below from the journal article categorizes serum cholesterol vs. CHD (coronary heart disease) for the Minnesota study as well as two other studies (Framingham, Massachuesetts and Albany, New York). Notice the positive correlation between serum cholesterol levels and CHD, in all three studies.

Furthermore, Keys' statistical analysis given in the table below shows that serum cholesterol levels (X) are a very strong predictor (r = 0.9) of CHD (Y). The five regression models that Keys used were between orders 2 to 3, so the cholesterol effect was not only positive (b coefficient), but it was also greater than just a linear relationship.

So the Minnesota study (and Framingham and Albany) shows a strong correlation between cholesterol and CHD, but is there also a correlation with dietary fat? A Google Scholar search on "effect of dietary fat on serum cholesterol" provides 233,000 results. Notice that the top three results are not papers by Keys, the often described mastermind of the cholesterol conspiracy.

The third result by Mensink et al is interesting. Dr. Mensink is a professor of molecular nutrition at Maastricht University in The Netherlands. His 1992 paper was a meta-analysis of 27 trials between 1970 to 1991 to calculate the effects of carbohydrates and fats on serum cholesterol. A multi variate regression analysis of the trials provided the following relationships between changes in dietary fats (∆S = saturated fatty acids, ∆M = monounsaturated fatty acids and ∆P = polyunsaturated fatty acids) and lipoproteins:

∆ Total Cholesterol = 1.51∆S - 0.12∆M - 0.60∆P

∆ HDL = 0.47∆S + 0.34∆M + 0.28∆P

∆ LDL = 1.28∆S - 0.24∆M - 0.55∆P

∆ Triglycerides = -2.22∆S - 1.99∆M - 2.47∆P

∆ HDL/LDL Ratio = 0.000∆S + 0.003∆M + 0.005∆P

Saturated fats have the largest effect on increasing both HDL and LDL, but mono and poly unsaturated fats increase HDL and decrease LDL. The ratio of HDL to LDL is not effected by saturated fats, but is increased by mono and poly unsaturated fats. Triglycerides were lowered by any kind of fat increase.

So, it seems reasonable that saturated animal fats could be a greater CHD risk than unsaturated plant fats. I'm not becoming a vegan, but in paleo speak, I'm planning on eating more gathered type foods and less hunted type foods.

1 comment:

  1. Ratio of HDL to LDL is not affected by saturated fat because it increases HDL and type A LDL. Type B LDL is the bad guy, and is associated with low fat/low cholesterol diets, much to the confusion of researchers.