Day 017 - 25 Jul 94 - Page 16
1 in our case-control studies than in our correlational
analyses, which is also consistent with most of the
2 literature. A possible explanation for this discrepancy
is that associations are easier to demonstrate when they
3 are based on international and interethnic variation,
which exceeds the variation within nations or ethnic
4 groups." Is that what you were telling us a moment ago?
A. Indeed, yes.
5
Q. "An alternative explanation is that correlational analyses
6 based on aggregate data do not control for potential
confounding variables and thus may exaggerate true
7 associations". Can I pause there, because we will come
across it again in the future. What are "confounding
8 variables"?
A. Well, it depends on the study that one is looking at,
9 but these could be that you do not have, for example, the
same age, structure -- they are looking at general members
10 of the public -- so age and sex structure; they may not be
the same in the two groups. It does not take any account
11 of other known risk factors, for example. Is one
particular group, do they have a greater proportion of
12 factors which we know to be associated with a greater risk
of breast cancer, for example, is that more predominant in
13 one group than the other?
14 There are a number of these factors -- you have to take
into account all of them before you can necessarily
15 conclude that there is a possible relationship between one
thing and the other, and these are called confounding
16 variables.
17 Q. Does one draw this conclusion from what is written here,
what you have told us? That one must treat the results of
18 correlational population studies with a good deal of
caution?
19 A. I think so because, as I say, you are not sure that if
you look at population studies in this way and correlate
20 them with a particular group that you are looking at, the
age and sex structure, is exactly the same. Also it is
21 much easier to determine international variations in
disease because we were talking about the problems
22 associated with international studies earlier on, and just
one of them is the problem of cancer registry.
23
I am not saying for a minute that that is the only problem
24 that is associated with international studies, but it is a
problem and it highlights the sorts of problems one may
25 encounter when one looks at international data.
26 Q. Then Kolonel and his colleagues go on: "The association
of dietary fat with breast cancer in our case-control
27 study is certainly weaker than the reported associations
of many other risk factors for this cancer. Although
28 dietary fat may indeed be a weaker risk factor, we cannot
at present eliminate the possibility that greater
29 misclassification in the assessment of dietary exposure
has substantially reduced the apparent case-control
30 differences".
