Saturated Fat Epidemiology for Math Geeks

A week or so ago, I posted a bit of epidemiology concerning saturated fat intake associated with heart-disease deaths by country. As you saw, it was all over the map. I did speculate, however, that if you were going to try to fit a curve, it would slope downward, meaning: more saturated fat, less heart disease deaths.

Well, owing to my vast network of resources [grin], physicist Robert McLeod offered to fit a curve if I could get him the tabular data, which, thanks to Ricardo, I did. So, here's the graph (see here for the one with the country labels).

Picture 2

Here's what Robert had to say.

All statistics done in MATLAB. I found that if I define

SF = % saturated fat intake

CHD = # heart deaths per year per 100,000 men


CHD = (-4.734 +/- 2.003)*SF + (144.5 +/- 21.4)

+/- errors are standard deviations (i.e. one sigma) with an R^2 =
0.13 (terrible) between the fit data and experimental data.

The plot I provided shows the baseline along
with a top and bottom curve which are the 95 % confidence interval
lines (~1.96 sigmas).

Although the statistics appear fairly poor, we can make one statement
of interest. A positive slope is equivalent to a positive
correlation between CHD and saturated fat (i.e. saturated fat bad!)
and a negative slope is a negative correlation (i.e. saturated fat
good!). Evaluating that statement using confidence intervals we have
a 0.9 % chance of a positive slope and a 99.1 % chance that the slope
is negative.

In other words, increased saturated fat intake is 99 % likely to be
correlated with decreased incidence of death from heart disease.


  1. Robert M. on March 30, 2009 at 14:09


    Yeah, mostly an amusement for the eye. It does look slightly different from Keys' original curve.

  2. Patrik on March 30, 2009 at 22:18

    Looking at just the lower-right quadrant of the graph there seems to be a positive relationship between geographic latitude and deaths (compare France to Finland). This group seems to be largely the wealthy Western European countries.

    Smack dab in the center is the former Eastern Bloc.

    Would be interesting to control for GDP per capita as well. If you take just wealthy Western Europe, the inverse correlation between SF and deaths would be even more pronounced.

  3. Stephan on March 30, 2009 at 15:27

    Nice. It's good to have statistics applied to those data. I wonder why the media haven't picked up on this negative association…

  4. ReachWest on March 30, 2009 at 19:26

    Very interesting – it sure flies in the face of the "conventional wisdom".

  5. Patrik on March 30, 2009 at 22:20

    BTW where is Germany in the graph?

  6. Jedidja on March 30, 2009 at 15:27

    Great stuff! Love seeing studies like this deconstructed, as Dr. Eades and Mark Sisson do as well :)

  7. sam on March 31, 2009 at 01:24


  8. Juha S said... on March 31, 2009 at 01:29

    If this curve would be true by eating more than 30,5 % calories from saturated fat your CHD risk would be less than 0. ;-)

  9. Dave in Ohio on March 31, 2009 at 02:14

    Hyperlipid recently posted some graphs of Keys' original data. Note the much higher percentage of Saturated fats in 1953.

    The percentages of SF for 1996 seem very low by comparison. And the death rate is also lower, most likely due to much improved cardiac interventional procedures, rather than the lower percentage of saturated fat.

  10. on March 31, 2009 at 08:59

    I would like to thank you for the high quality of your blog. I have been reading it with others for 2 months, and i turned "paleo" and lost so much fat without pain, it isn't even funny! But stop about me (it is just to tell you to keep on the good work).

    Thanks for the links to the database.

    by the way, I 've plotted/calculated (matlab user too) occurence of cardiovascular diseases against fat from animals in Kcal/day instead of %, and it shows approximately the same trend: negative correlation (corre=-0.3380, 95% -> [-0.1962,-0.466] )
    Moreover, occurence of cardiovascular diseases against wheat in Kcal/day seems to show a tiny positive correlation (corre=0.2323, 95% -> [0.0833,0.3710])…

    But the data are so scattered that all of this has to be taken lightly.

    I like to plot against "quantity" instead of % to minimize correlation problem with total kcal intakes.

  11. Dave in Ohio on March 31, 2009 at 09:08

    Correction: The 1953 charts used total fat while these use SF, so the 1998 percentages are not much different.

  12. Dr Dan on March 31, 2009 at 03:12

    GREAT POST!!!!!!

  13. Richard Nikoley on March 31, 2009 at 07:10

    Off the scale: 222 x 33

  14. Richard Nikoley on March 31, 2009 at 09:57

    Thank you, Mark. There are plenty of people doing high quality "10 Things to X" sorts of posts, and that's just not me. There's plenty of room, I guess, for my style.

    Yea, I don't want people reading too much into that. Mainly, I look upon it as pretty solid falsification of the fat/cholesterol -> heart disease hypothesis. What I think is that you can have a low or high sat fat intake and be just fine so long as you stay away from frankenfoods.

    Congrats on the success. It's so easy, it's like…natural or something.

  15. Eric on January 12, 2010 at 15:32

    Unfortunately learning these things outside of mainstream ignorance usually is only for smart or desperate people. And anyone who started by learning what the health authorities say may never bother to look further to see the light. Maybe it’s good I have a weight problem as was forced to “jump over the fence” to discover I was in the nutritional dark ages when I got over.

Leave a Comment

You must be logged in to post a comment.