association between coffee drinking and pancreatic
cancer. That was a case of selection bias.
Case-control studies are susceptible to that kind
of bias if you select the controls inappropriately.
Now let's look at the cohort study, using
the same example of smoking and lung cancer.
In the case of the cohort example we ascertain
the exposure first and then we look forward
in time to see who gets the outcome, in this
case lung cancer. So the direction chronologically
of the inquiry is going to be forwards in
time, remember the case-control looks backwards
in time, cohort looks forward in time. We
begin by finding individuals who smoke, some
individuals who don't smoke or some other
kind of exposure and we wait as time passes
to see who gets the outcome and who doesn't
have the outcome. Now from the cohort example,
we can compute incidence rates because I didn't
select these individuals, I just watched them
and saw how they started their behaviors. And
then I saw if the outcome manifested on its
own naturally, that's a true incidence rate.
So cohort designs allow us to compute incidence
rates, case controls do not. That's an important
consideration when we attempt to measure actual
ratios of association. So remember, the direction
of inquiry in a cohort study is to look forward
in time and in our lung cancer smoking example
we begin by finding some smokers and finding
some people who don't smoke and waiting until
some of them develop cancer and we compare
the proportions of those who get cancer in
the smoking group to those who get cancer
in the non-smoking group. The cohort study
has a lot of advantages; the first obvious
advantage is, we can compute incidence rates.
That's huge. We love that. It's very important.
Next is that we can yield more information
on the incidence of other kinds of diseases.
So we can look at a variety of things that
may manifest from my exposure. We have a clear
temporal relationship between exposure and
disease, I know that smoking came before the
cancer, remember in our cross-sectional example,
I didn't have that ability because I was ascertaining
exposure status and outcome status simultaneously.
In the cohort example I know for a fact the
smoking came before the lung cancer because
I saw it happen. And lastly we can measure
rare exposures using the cohort study. Think
about it, let's say something is really rare,
like doing a certain activity, walking on
the moon, traveling in space, I don't know,
think of a variety of things you could probably
think of that are rare. I can select those
individuals, find some people who didn't have
those experiences and then wait in time to
see if certain diseases manifest. Other advantages
include that the cohort study allows us to
look at a variety of exposures, I can look
at smoking, I can look at eating behaviors,
where you live, all in the same study pool.
I can look at multiple outcomes of a particular
exposure as well if I choose to. I can minimize
bias by choosing my individuals carefully,
I have a lot of control here. It's also the
strongest design in the observational category
for establishing cause-and-effect, I can't
really establish cause-and-effect, for reasons that
I will discuss when we talk about experiments,
but to the extent that observational studies
are able to offer a degree of wisdom, cohort
studies are the best.
So the disadvantages though are quite extreme.
It's quite time-consuming. If I have to wait
to see if someone develops cancer from smoking,
I'm going to wait years, if not decades, possibly
a lifetime. So it's not particularly useful
in that sense, therefore it can be expensive.
Often it requires a large sample size, because
maybe my outcome that I care about is only
manifesting in a certain small percentage
of the cases of individuals who engage in
this behavior, so I need a lot of people doing
this behavior before a certain outcome will
manifest. Let's say 0.1% of smokers developed
lung cancer, I just made up that number, don't
look it up. In that case I'll need a great
many smokers to study to get even a single
case of lung cancer manifesting. And as I mentioned,
all this can be quite expensive. Other disadvantage
is, we can't really measure rare diseases
using a cohort study. Think about that, if
my disease is quite rare I may wait forever
for it to manifest if I'm looking at it in
a prospective, forward in time process. I
may have a lot of losses to follow up.
In our lecture on biases, we learned that loss
to follow-up is a major consideration, so
is I'm following a bunch of people through
time, perhaps years, perhaps decades, I'm
going to lose a fair amount, most likely. And lastly,
changes over time in how I'm ascertaining
an outcome may in fact lead to some biases
as well. So maybe when I started my study,
I had a certain test for determining if you
get the disease I'm interested in. Let's say
it's not lung cancer, let's say it's something
cognitive, psychological test, a quiz maybe,
and five years later, some psychologists develop
a better test that I could've used and I introduce
that test halfway through my study, I've now
created a real problem in my study, because
I've changed the way that I diagnose my outcome.
So that's always something we try to look
to and control if we can.
I want to talk a little bit now about the
verbiage we use to talk about the flow of
time, prospective and retrospective. These
terms refer to the time frame, not necessarily
the direction of inquiry, but the timeframe.
The direction of inquiry describes when I'm
going to get my exposure versus my outcome,
but the timeframe is how in real time I'm
going to conduct my study. So prospective
study moves forward in time. We have other
words for it as well, longitudinal studies
or concurrent studies. Some people say cohorts
are naturally longitudinal or concurrent.
Retrospective on the other hand means backwards
in time, other words to use it, historical
or non-concurrent. So a cohort study is typically
prospective, we begin by looking at exposure
statuses and we move in forward in time to see if
there's an outcome. Is there such a thing
as a retrospective cohort? Yes there is, this
would confuse a lot of people, but it isn't
that difficult to understand. A retrospective
cohort is the opposite of a prospective cohort.
In a prospective cohort, you ascertain the
exposure status and you wait to see if the
outcome manifests. In a retrospective cohort,
you do this in the past, using historical
records. I look up medical records in a certain
hospital, I find some individuals who had
an exposure, I look forwards to the medical
records to see if they manifested an outcome.
It's still moving forwards in time, but it
happened in the past. That's what makes it
a retrospective cohort.
So, when are you going to use which designs?
This is kind of fun. So remember where I mentioned