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 ).

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

then

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.