馃攳
Sensitivity and Specificity Explained Clearly (Biostatistics) - YouTube
Channel: MedCram - Medical Lectures Explained CLEARLY
[2]
okay so welcome to another MedCram
lecture we're going to talk about what
[6]
may be considered a dry subject but I
think it's actually very important it's
[10]
the topic of sensitivity and specificity
in relation to statistics now sometimes
[14]
we see this appear on test and we've got
to know what sensitivity and specificity
[17]
means I think it's also very helpful in
interpreting some of the data that we
[22]
see in medical journals and it's
actually very helpful in our everyday
[25]
life in terms of understanding the test
that we're ordering on patients so I
[28]
think it deserves a little bit of a an
introduction and an analysis and how to
[33]
actually do the questions that you need
to do for tests so imagine if you will
[38]
with me that you've got a nice car and
you want to protect your car so you
[43]
install a car alarm okay he's got a
little blinking light there and you've
[47]
got a knob in that car that allows you
to change the sensitivity on that car
[58]
alarm well you've got a couple of
choices you can either have the
[62]
sensitivity high and what does that mean
that means that well if a thief tries to
[69]
break in the car alarm is probably going
to go off but it also means that since
[75]
you're parked on a on a busy street if a
motorcycle goes by that could also set
[79]
it off so if you have a thief and the
car alarm went off then that would be a
[86]
true positive result you would hear that
car alarm go off and sure enough there's
[92]
a thief trying to break into your car
that would be a true positive so the
[96]
disease here is is your car being broken
into or is it not being broken into and
[100]
the test is is the car alarm going off
or is the car alarm not going off and
[106]
you can actually break this up into a
square and have the four possibility so
[118]
you could in terms of our disease you
could have a positive disease or
[124]
negative disease and in terms of a test
you can either have the alarm going off
[129]
or the alarm not going off so in this
box here this would be if there was a
[133]
thief trying to steal your car and the
car alarm was going off that would be a
[138]
true positive okay if there wasn't a
thief trying to steal your car but your
[146]
car alarm went off that would be a false
positive that would be like a motorcycle
[151]
going by or somebody touching the car as
it as it went by and the car alarm goes
[155]
off you go out there the alarms going
off that no one's tried to break in and
[159]
obviously that's not an optimal result
you don't want to have false positives
[162]
because every single time the alarm goes
off you've got to go out there I've got
[167]
to check the car you got to make sure no
one's getting into it
[169]
also the more times this car alarm goes
off the less likely you are of taking it
[174]
seriously so not good
the other possibility is which is also
[178]
not a good possibility is that your car
is being broken into but the alarm
[182]
doesn't go off that would be a false
negative that's probably the worst
[186]
result that you could have you go out to
your parking lot the car's gone okay the
[190]
alarm never went off it didn't disengage
the engine that's a problem and then
[194]
finally the thing that happens you know
all the time is a true negative
[198]
car alarm is not going off and your car
is not being broken into that's that's a
[203]
good result so we like true positives we
like true negatives we want to minimize
[208]
these as much as possible
and you can see how that would happen
[213]
depending on how you sent the
sensitivity of your car alarm if you
[216]
sent the sensitivity of the car alarm up
very high what you would see is you
[221]
would start to have a lot of these false
positives but the chances of having
[225]
false negatives would go down okay
conversely if you set the sensitivity of
[231]
your car alarm very low you probably
wouldn't be getting motorcycles setting
[236]
off the car alarm but on the other hand
it might be easier for a thief to break
[240]
into your car and to steal it without
the car alarm going off so the point
[244]
here is I want to show you that there's
a trade-off in terms of where you set
[247]
the sensitivity for a test it's very
important because depending on where you
[251]
set the sensitivity for a test you're
either going to get more false positives
[254]
or false negatives and depending on what
you're testing for and what kind of
[260]
situation you're in you know you may
want to set the sensitivity of that
[264]
alarm to different areas for instance if
you're living in it
[267]
high-crime area where you know cars are
getting stolen every day and there's not
[273]
a lot of motorcycles sure set that
sensitivity up pretty high
[276]
if on the other hand you're parking your
car in the area where there's not a lot
[280]
of thieves and and there's really been
no reported thieves but there's a lot of
[284]
motorcycles going by you may want to set
that sensitivity low so it just sort of
[288]
depends on where you're living and the
same thing applies to medicine now we're
[294]
going to talk a little bit about these
false positive true positives false
[297]
negatives in this box a little bit but I
wanted to introduce to you the concept
[301]
of what sensitivity is and what
specificity is sensitivity
[306]
mathematically speaking okay is the true
positives divided by the true positives
[315]
plus the false negatives and you can see
here that sensitivity is the true
[322]
positives divided by the true positives
plus the false negatives and so the
[331]
sensitivity as you can see
mathematically will be very high if the
[334]
false negative rate was very low in fact
if something is very sensitive and it's
[343]
negative you can practically rule out
that disease okay on the flip side
[350]
specificity which takes of this column
specificity is the true negative rate
[358]
down here divided by the true negative
plus the false positive and clearly you
[366]
can see there that if the specificity is
very high this false positive has to be
[371]
very low and so if something is very
specific and it's positive the chances
[377]
of it being falsely positive are
extremely low we'll get into that more
[382]
another way they'll sometimes test you
is they'll look at all of the different
[385]
populations of things okay so this is
what we're dealing with if we can go
[389]
back to the car example here is a
population of all the things that could
[394]
falsely set off an alarm and here is the
population of things
[399]
that could truly set off alarm so this
is what thieves could do okay and so
[408]
what we're trying to do is we're trying
to devise a test that would separate
[412]
these things out so what I mean to say
here is that this would be the disease
[417]
negative state okay so this is the
disease here is your car being stolen
[422]
and here this is all the stuff that
could set off a car alarm but it's not
[426]
really your car being stolen so this
would be like motorcycles boomboxes
[430]
things of that nature and this is all of
the things that thieves could do to
[434]
steal your car so this would be the
disease positive state the problem here
[438]
is that there's some overlap and so no
matter where you set your threshold for
[442]
your sensitivity you're going to have a
problem in that you're going to be
[447]
including one or the other so there's
there's a couple of areas here where you
[450]
could set the sensitivity you could set
the sensitivity right here at point a
[457]
you could set the sensitivity here at
point B or you could set the sensitivity
[463]
here at Point C or any of these points
and now I want you to notice here what
[470]
what's going on so if we were to set set
it here then everything to the right
[473]
would be positive and everything to the
left would be negative everything to the
[478]
right of B would be positive to the left
would be negative everything to the
[481]
right of C would be positive and to the
left to C would be negative so the
[485]
question is and this is the question
that you'll sometimes get on tests is
[489]
which of the following points represents
the most sensitive test and which of the
[495]
following points represents the most
specific test so specificity is the
[498]
opposite of sensitivity in this case so
of course if you are anything that is on
[505]
the right side of a test is going to be
a positive test anything that is on the
[510]
left side of this point is going to be a
negative test depending on which
[515]
population they belong to will tell you
whether it's a true or a false negative
[519]
or positive let's take for example this
point here at C okay so here at C if we
[526]
were to set the alarm at C in terms of
specificity and sensitivity then all of
[531]
this area right
here would be a positive now because it
[537]
belongs to the actual disease it would
be a true positive but all of this back
[542]
here would be a negative now some of
these would be false negatives because
[548]
they still belong to the disease but
then some of these would be true
[553]
negatives because they don't belong to
the disease and so what you notice here
[558]
is that if something is positive it is
definitely part of the disease but
[561]
you're missing out on a lot of falsely
negative but notice here that we've
[568]
moved this C all the way out here so
that there is no more false positives if
[573]
you have a very low false positive rate
that means you're very specific okay
[582]
just for review remember that the
sensitivity is the true positive rate
[588]
through positive rate divided by the
true positive rate plus the false
[593]
negative rate and the specificity is the
true negative rate divided by the true
[601]
negative rate plus the false positive
false positive rate so because of that
[609]
if we were to look at this point here a
notice what we have here at a at a there
[615]
is no more false negatives and because
there is no false negatives negative
[620]
would be on this side of the line okay
so negative there are no disease or
[627]
robbers of the car if you will that are
showing up on this line because we've
[631]
moved it so far to this side so this is
the most sensitive setting that we could
[637]
set it to so imagine that imagine this
in terms of our car if we turn the
[643]
sensitivity of the car alarm up very
very high we're going to catch every
[648]
single thief that's going to try to
steal our car which is represented by
[651]
this curve but we're also going to catch
a lot of false positives here which
[659]
represent motorcycles and you know very
loud cars or loudspeakers that are
[665]
driving by on the converse
if we set a very specific test at right
[672]
here notice here that we're no longer
going to pick up it'll be negative now
[677]
because negative is to the left of the
line all of these things that could make
[681]
the car alarm go off will be treated
appropriately as a true negative but
[686]
we've made it so specific now for the
car alarm to go off that there's things
[689]
that thieves can do to break into your
car okay so so long as you set this up
[695]
with the negative disease on the left
and the positive disease on the right
[701]
always remember that this direction is
the most sensitive and this direction is
[707]
the most specific which actually is
going to help us later that to remember
[710]
that sensitivity is on the left and
specificity is on the right
[716]
I hope this clears things up in terms of
sensitivity and specificity the next
[721]
lecture is going to talk about the
actual mathematics in this lecture I
[724]
want to talk about concepts let's talk
next about the actual mathematics and
[729]
how to calculate things so you can
calculate things quickly on it test
[733]
exams
thanks for joining us
[743]
you
Most Recent Videos:
You can go back to the homepage right here: Homepage