00:01
Now let's look at the case control study.
The case control
study sort of works backwards in time, we
begin by ascertaining the outcome, who has
the disease, who doesn't have the disease
and we look backwards in time to see who had
the exposure. In other words, who had smoking
behavior in this case. So we have those individuals
who are cases, the cases are those who have
the disease, in this case lung cancer. The
controls are those who don't have a disease,
in this case people without lung cancer. We
look backwards in time to see which members
of both groups had a smoking behavior, how
do we look backwards in time? We could ask
them, were you a smoker? We can look at their
medical records; those are the two commonest
ways of doing so. So again, the path of the
cross-sectional inquiry is to go backwards
in time, chronologically we look backwards,
that's the direction of inquiry. We determine
the proportion who were exposed, in this case
to smoking amongst our cases and the proportion
who were exposed amongst the controls and
we compare those two numbers. So again, here's
our scenario, we have people with lung cancer
and people without lung cancer. We ask how
many were smokers, and how many weren't smokers.
01:18
We compare those two proportions, it's as
simple as that. We get our comparative figure
and we determine if that's large or small
and that tells us if there is likely an association
between smoking and lung cancer. Now case-control
studies are useful in a variety of contexts,
especially if the outcome is rare. Think about
it, if the disease that I'm looking for is
very rare, I can go out and find some of those,
I can go to a clinic that specializes in that
disease and find those cases, that can identify
some other controls quite easily, people who
don't have that disease and I look backwards
to see what the exposure was. The problem
is I can't measure incidence rates with a
case-control approach, the reason is that
I went out and I found those individuals with
the disease, because I found them I can't
say they manifested on their own. I control
the numbers, so incidence is not possible,
because incidence is you'll call, is the proportion
of new cases of a certain disease in a population.
02:16
And the major challenge in a case-control
study is how to select the controls, how to
do so without introducing new kinds of bias.
Now in our lecture on bias, we cover this
particular example, a 1981 study in the New
England Journal of Medicine that showed an
association between drinking coffee and getting
pancreatic cancer. Now that was a case-control
study, the cases were people with pancreatic
cancer and the controls were people in the
hospital that didn't have pancreatic cancer,
but had other G.I. Issues, gastrointestinal
issues. The problem of course is that those
with gastrointestinal issues were less likely
to have drunk coffee and so by comparing the
two groups, we have an artificially exaggerated
association between coffee drinking and pancreatic
cancer. That was a case of selection bias.