Grant's study is interesting since there is so little cross-cultural data
on Alzheimer's disease (AD) in relation to diet. If the noted statistical
relation to fat intake (or total energy intake) can be confirmed, I would
be interested to see more research on intake of micronutrients (e.g.
antioxidants) which is negatively related to fat intake (or total energy
intake) in most cross-cultural settings.

At 02.12 97-09-30, Art De Vany wrote:
>There are few degrees of freedom because of the
>small sample size.  Consequently, outliers are influential.

As you can see on the graph at
http://www.coa.uky.edu/ADReview/Grant_Images/grantf1.gif there are no
obvious outliers.

>The
>equation for fat is: AD prevalence rate = -0.203+(0.0312*fat
>(grams/day)).  A large negative intercept term and a positive slope,
>leading to the author's conclusion.  A large negative intercept
>often indicates a non-linear relationship (which these data surely
>exhibit).

The intercept is actually close to zero and could be negative by chance. If
we change only the Chinese AD prevalence rate from 1.2 to 1.4 per cent the
intercept becomes positive.

>Because fat intake varies so much among countries,
>compared to total caloric intake, the differences among countries
>are expanded in the regressions using fat as the dependent variable.
> This exaggerates the influence of outliers.

This argument is applicable in any similar epidemiological study.

>Take a look at this graph from the ariticle
>http://www.coa.uky.edu/ADReview/Grant_Images/grantf1.gif and you
>will see that there are two groups: developed Northern American and
>European countries and Asian and African countries.  There are no
>points in the middle and a vast gulf between the data clusters.

I would say the data fit the regression line better than in many important
surveys.

>Within the developed cluster there is no relation between fat
>consumption and Alzheimer's.

Yes there is. Within this cluster R2=0.77 (p=0.0062) for fat intake versus
AD, meaning that fat intake explains 77 per cent of the variation of AD
between these countries. Actually, in the Asian/African cluster the
correlation is not far from significant either (R2=0.81, p=0.065).

>Measurement error is all over these data: different reporting
>rates, more thorough surveys, and genetic composition of the
>populations are just a few.

Quite true. If the paper was on some disease which had been studied more in
relation to nutrition it would not bring much new input. But in this case
it is definitely interesting.

>Grant's article is excellent on other grounds, but the statistics
>show nothing, or they hint at a genetic disposition to Alzheimer's
>more than an environmental one.

A rather tendentious statement it seems. What hypothesis are you in love with?

-------------------------------------------------------------------
Staffan Lindeberg M.D. Ph.D. Dept of Community Health Sciences, Lund
University, Mailing address: Dr Staffan Lindeberg, Primary Health Care
Centre, Sjobo, S-22738 Sweden, +46 416 28140, Fax +46 416 18395
<[log in to unmask]> http://www.panix.com/~paleodiet/lindeberg/
-------------------------------------------------------------------