So another kind of bias that I want to talk
about is called Berkson's bias and this one is
a little confusing, so bear with me and think
deeply about what the problem really is here.
So Berkson's bias, sometimes called Berkson's
paradox or Berkson's fallacy, and ultimately
it talks about the problem of having a study
population in a clinical environment, in a
hospital and how people in a hospital are
a little bit different from people not in
the hospital. It's named for a 1946 paper
by Berkson, and the definition of Berkson's
bias is that, 'the set of selective factors
that lead hospital cases and controls in a
case-control study are systematically different
from one another', okay what does that mean?
I want to walk you through a pedantic ordinary
example. Let's say we have a stamp collector
and that stamp collector has 1000 stamps. And
among those thousand stamps, he has 100 that
are quite rare and 300 that maybe are quite
attractive, they are pretty. That leaves 600
that don't qualify as either pretty or rare.
However, 30 of them are both pretty and rare,
he's particularly proud of those stamps. So,
what percentage of all the stamps are rare?
Well out of 1000, 100 are rare, that's 10%.
What percentage of all the pretty stamps are
rare? Out of the 300, 30 are, that's also
10%. What does that tell us about any relationship
between prettiness and rarity? Nothing, there
is no relationship between prettiness and
rarity in this sample of 1000 stamps. However,
let's say the stamp collector decides to go
to a stamp show or convention of some kind and
he takes 50 stamps to show off. Amongst those
50, he brings the 30 that he's most proud
of, the ones that are pretty and rare. Now
30 out of those 50 that he brought is 60%,
that's an enormous number. So based upon that
sample of 50 stamps, you might conclude that,
oh there is a great relationship between prettiness
and rareness and you'd be probably wise in
doing so, because based upon the information
in front of you, well it seems that all rare
stands are good-looking. What does this mean,
what does this mean for clinical research?
So what do stamps have to do with epidemiology?
Well in the same way that that selection of
50 stamps gave an artificial sense that there
is a relationship between prettiness and
rarity, in a hospital environment where we're
selecting certain kinds of patients to be
in a clinical environment, you can have an
artificial relationship between different
diseases. That's because people in a hospital
tend to be comorbid. They tend to have more
than one condition and we might conclude erroneously
that these diseases have something in common
with each other. That's not necessarily the
case. That's Berkson's bias.