00:00
So, the first thing that we're going to compute
is this thing called sensitivity, how sensitive
is the test to detecting if you have a truly
diseased case. It's a probability of correctly
diagnosing a case, remember a case is somebody
with a disease; we sometimes call this the
positive fraction. On the other hand, specificity
is a probability of correctly rejecting a
case, in other words, if you don't have a
disease, will the test truly find that you
don't have a disease, we call this the negative
fraction. So again, if a patient really is
diseased, the sensitivity tells us the probability
that the test will show that they are really
diseased and if the patient really isn't diseased,
the specificity tells us the probability of
the test will be negative and it will show
that they're not diseased. Which one do we
care about more? Remember a really high sensitivity
tells us that the test is throwing a wide
blanket, it's open to anyone, it's really
likely it's going to find a positive case,
but it is also really likely that it's going
to find a lot of negative cases and think
they're positives, we call those false positives.
Now when we set up a contingency table to
analyze and to compute the important indicators
of a sensitivity test, that table must be
set up exactly the way that I'm showing you,
otherwise the formulas that I'm talking about
won’t apply. So at the horizontal axis,
we want the screening test results with a
positive case on top and the negative on the
bottom. On the vertical axis, we're going
to have our measurements of truth, again
truth is the result of the additional test,
the ultrasound, biopsy, etc. And the first
column are those who test positive on the
truth test and the second column is those
who test negative on the truth test, in other
words those are truly diseased and those who
are truly not diseased. In other words, our
screening test is set up in such a way that
we're going to test which ones are true positives
and false positives and so forth. Now let's
calculate sensitivity, what is sensitivity
again? Sensitivity is going to be the number
of people who test positive on my screening
test who are also truly positive, we call
that A. And we're going to divide that by the
total number of people who are truly positive.
In other words, the proportion of truly diseased
people who test positive, A over A + C. How
do I calculate specificity, similarly, I figure
out the number of people who are truly negative,
as in truly non-diseased, B + D and I count
the number of people who test negative on
my test, that proportion is going to be my
specificity. So the probability of correctly
diagnosing case is sensitivity, the probability
of correctly rejecting case, i.e. finding
that someone doesn't have disease and it's
true, they don't have a disease, that's specificity.
How do I compute prevalence from a contingency
table? Well prevalence is always going be
the total number of actual cases, that's going
to be A + C divided by my entire sample, so
A + C divided by everyone else, that's prevalence.