00:01
So let's go through some of the advantages
and disadvantages of RCTs, because I mentioned
they are gold standard, I mentioned how much
we love the RCT and how much we rely upon
it, but there are some disadvantages. First
of all, the biggest advantage to an RCT is
that because it's controlled, the 'C' is controlled,
or the 'C' is clinical, depending upon your
perspective, I can control the extraneous
variables that are in my study. I can make
sure that I know that nothing else caused
the outcome because I've selected these individuals
for their characteristics. On the other hand, it's
ridiculously expensive, sometimes millions
of dollars to run one of them. Also it doesn't
represent the real world, what do I mean?
Well if I have selected individuals for certain
characteristics and I'm controlling what they're
exposed to, sometimes in literally a clinical
hospital environment that prevents other people
from being involved, that doesn't represent
the true post-market environment that the
drug or therapy is going to be tested in once
it hits the market. So maybe my RCT isn't
giving me sufficient real-life knowledge that
I can apply. Also the great advantage of doing
an RCT is that we can establish a temporal
relationship with absolute clarity. I know
for a fact that the drug came before the change,
I know for a fact that the smoking came before
the cancer, if indeed I did an RCT on smoking
and cancer, which by the way I can't ethically.
01:26
So the temporal relationship is critical to
helping establish causality. Also if I do
blinding, and we'll talk about blinding in
a second, then it's difficult to argue with
the conclusions that I draw within the confines
of the study. The internal validity which
we will define later on, is pretty secure
if the RCT is well-designed and blinded. Ethically
however, RCT is problematic. So think about
again the possibility of running an RCT on
smoking and lung cancer, I'd have to randomly
select individuals who are compelled to smoke
for many years to see if lung cancer manifests.
It's not ethical to do that to people. Also
very often if I'm testing a life-saving drug,
I'm giving it to people who probably need
it, but the control group is not getting that
drug. Is it ethical to have people in a study
that I know need treatment and I'm not giving
them the treatment that I suspect will make
them better? That's required for a control
group, I have to deny some people the treatment.
02:28
Is that ethical?
Let's talk about randomization now, that's
the big 'R' in RCT and that's the thing that
we're so excited by all the time, because
the 'R' gives us a lot of power and a lot
of logical confidence. So in a randomized
controlled trial, participants are assigned
by chance rather than choice, to either treatment
group or the control group. Remember, in a
cohort study they are choosing to do something,
a behavior, an exposure. In an RCT, I'm assigning
them as the researcher to one group or another
and that assignation, that assignment, is
based upon randomization. And again, the classic
randomization design is a coin toss, heads
you go in this group, tails you go in that
group. In reality we don't use coin, we
use far more advanced randomization techniques,
but the coin toss is a good example for helping
you understand how it works. Why do we randomize?
We randomize to help reduce the chances of
selection bias, you recall selection bias
from our lecture on biases. How does this
work? Think about this; think about if I know
that the drug I am testing works differently
on men than on women, I'll make sure that
I have equal numbers of men in my control
group as in my therapy group. If I know that
my drug works differently on old people and
young people, then I'll only test it on old
people, or I'll only test it on young people,
or I'll make sure that I compare my treatment
and control groups using only age as a factor.
03:56
In other words, the factors or variables that
I know are confounding my relationship, I
can control for, but what about the ones that
don't know about, that's the beauty of randomization.
04:08
Things that I don't know about may be affecting
the relationship between my exposure and my
outcome. If I randomly allocate the group
that subject gets put into, then I can be
well assured that those factors that I cannot
control for are also randomly and equally
distributed, I hope. When I say I hope I mean
I'm relying upon the magic mathematics to
give me a very good probability that those
characteristics are equally distributed. That's
the power of randomization. So randomization,
when we do it, each participant that enters
my study has an equal chance of being either
in this group or in that group. If they have
a higher chance of being in another group,
then my allocation process is biased. It does
not presume that all the participants are
the same. Unfortunately randomization doesn't
guarantee equivalence, what does that mean?
Well simple randomization, like flipping
coin, doesn't guarantee I'm going to have
equal numbers of people in both groups. Think
about this, if I'm flipping a coin, heads
goes in this group and tails goes in that
group, it's entirely possible I flip heads
eight times in a row out of 10 people. Now
I've got eight people in this group and 10
people in the other group. That's not great.
05:23
Remember, coin flipping isn't what we always.
There are more complicated ways of randomization
to account for this, but simple randomization
does not guarantee equivalence or balance
between the study arms, as we call them. So
when we randomize, what we're looking for
is to eliminate bias, primarily selection
bias. We hope to have balance between the
arms with respect to certain variables, like
gender and sex and age and all those other
things, but especially the ones we didn't
even know about. And remember that randomization
also forms a basis for a lot of statistical
testing, all the tests we might want to apply,
assume that my data is somewhat randomly distributed.