00:01
Hello and welcome to epidemiology. In this
lecture we are going to start the process
of understanding bias and in my opinion bias is
probably the most important idea in epidemiology,
because if you don't end up being a health
researcher or any kind of scientist for that
matter, understanding bias will help you be
a smarter, better citizen, it's that important.
00:23
So in this lecture we are going to understand
the basic idea of bias and also start the
process of understanding selection bias, which
is one type of very common bias. There are
lots of different types of biases and we will
get through some of them, but right now we're
going to talk about selection bias.
00:37
I want to start by exploring some definitions
of bias. One of them is, bias is any systematic
error in the design, conduct or analysis of
a study that results in a mistaken estimate.
00:50
And the important qualities of that definition
are the systematic aspects of it, the error
aspects of it and the fact that it ends up
giving you a mistaken or problematic or erroneous
conclusion. Another definition is, bias is
a systematic error in the epidemiologic study
that results in an incorrect estimate. Again,
key concepts there, the systematicity and
the incorrect conclusion or estimate that
results from it all. In my opinion, the best
kind of definition has to do with the fact
that biases lead to erroneous conclusions.
01:23
It's impossible to eliminate all aspects of
bias from any study, but we strive to minimize
that small extent, as well it's not usually
possible to control for biases in a post hoc
analysis statistically, so it is important
that we try to make sure that bias doesn't
exist as much as we can in the design of a
study. So we care about bias because biases
can mask an association between variables
that really are related. For example, maybe
you are studying whether or not a risk factor
is associated with certain outcome, some
kind of behavior may cause the disease, a bias can
prevent you from detecting that relationship.
02:04
A bias can also create a false or spurious
relationship between two variables, maybe
it shows that this behavior causes or is associated
with a certain kind of disease or outcome
and that's incorrect, you don't want that
either. Sometimes the bias can cause us to
overestimate a real relationship. Sure, maybe
a risk factor is truly associated with a certain
kind of outcome, but not as much as we think
it is. And similarly, a bias can cause us to
underestimate the size of a real relationship.
But remember bias happens when something
is systematically wrong with the way a study
has been designed, usually in how we've selected
the participants, which is what systematic
selection bias is all about and as a result,
it's mostly avoidable. I would argue it isn't
usually entirely avoidable, but almost entirely
avoidable. So remember when we're looking
at a typical study, we are trying to analyze
a sample to learn something about a greater
population, almost all medical research eventually
has to be applicable to a broader population,
otherwise why bother examining a handful of
patients. So typically a study involves identifying
a reference or total population that we want
to derive some wisdom about. We sample that
population and that sample is where we conduct
our analyses. Now in order to get that sample
we apply a sampling scheme, some people aren't
going to be included in our sample. Those
who are eligible go on to be studied and we
apply inclusion criteria to identify those
people that are going to end up being studied
in our sample. And we're going to ask those
people to participate, that's the process
of informed consent, "Would you like to be
a part of my study?" Some will say yes, and
some will say no. At the end of all this,
we're going to have a set of individuals that
are part of my study and it's their data that
I use to infer some wisdom about the total
population. Now look at this flowchart and
understand that at any given point in selecting
individuals, there could be bias, the inclusion
criteria I apply, the sampling criteria, whether
or not they agree to participate or not, these
are all opportunities for something systematic
to be applied that causes a difference between
my sampled population and the total population.
04:25
Even after I've selected my participants,
some will be lost to follow-up,
and that loss may bias my final sample.
04:33
So there are a host of different kinds of
biases that we're going to explore over several
lectures, but in today's lecture we're going
to talk mostly about selection bias and selection
bias is when an erroneous conclusion can arise
from how we select our subjects.