Hani Tamim, PHD Associate Professor Department of Internal Medicine American University of Beirut

Hani Tamim, PhD
Associate Professor
Department of Internal Medicine
American University of Beirut
 A graphical plot which illustrates the performance of a
binary classifier system as its discrimination threshold
is varied (3).
 Effective method of evaluating the performance of

diagnostic tests (3).
 ROC analysis is related in a direct and natural way to

cost/benefit analysis of diagnostic decision making (3).
 First developed by electrical engineers and radar
engineers during World War II for detecting enemy
objects in battlefields (3).
 ROC analysis since then has been used in medicine,

radiology, biometrics, and other areas for many decades
and is increasingly used in machine learning and data
mining research (3).
Suppose we have a test statistic for predicting the
presence or absence of disease.
Test criterion
True disease status
Pos Neg
Pos
Neg
Test criterion
True disease status
Pos Neg
Pos TP ☺
Neg
Test criterion
True disease status
Pos Neg
Pos FP 
Neg
Test criterion
True disease status
Pos Neg
Pos
Neg FN 
Test criterion
True disease status
Pos Neg
Pos
Neg TN ☺
Test criterion
True disease status
Pos Neg
Pos TP FP
Neg FN TN
P N P+N
Test criterion
True disease status
Pos Neg
Pos TP FP
Neg PN TN
P N P+N
Accuracy = Probability that the test yields a
correct result (2).
= (TP+TN) / (P+N)
Test criterion
True disease status
Pos Neg
Pos TP FP
Neg FN TN
P N P+N
Sensitivity = Probability that a true case will test
positive.
= TP / P
Also referred to as True Positive Rate (TPR)
or True Positive Fraction (TPF).
Test criterion
True disease status
Pos Neg
Pos TP FP
Neg FN TN
P N P+N
Specificity = Probability that a true negative will test
negative.
= TN / N
Also referred to as True Negative Rate (TNR)
or True Negative Fraction (TNF).
Test criterion
True disease status
Pos Neg
Pos TP FP
Neg FN TN
P N P+N
1- Specificity = Probability that a true negative will test
positive.
= FP / N
Also referred to as False Positive Rate (FPR)
or False Positive Fraction (FPF).
Test criterion
True disease status
Pos Neg
Pos TP FP
FN TN
Neg
P N P+N
Positive Predictive = Probability that a positive test
Value (PPV) will truly have disease.
= TP / (TP+FP)
Test criterion
True disease status
Pos Neg
Pos TP FP
Neg FN TN
P N P+N
Negative Predictive = Probability that a negative test
Value (NPV) will truly be disease free.
= TN / (TN+FN)
Test criterion
True disease status
Pos Neg
Pos 27 173 200
Neg 73 727 800
1000
Se = 27/100 = .27
Sp = 727/900 = .81
FPF = 1- Sp = .19
Acc = (27+727)/1000 = .75
PPV = 27/200 = .14
NPV = 727/800 = .91
 Of these properties, only Se and Sp (and hence FPR)
are considered invariant test characteristics.
 Accuracy, PPV, and NPV will vary according to the

underlying prevalence of disease.
 Se and Sp are thus “fundamental” test properties and

hence are the most useful measures for comparing
different test criteria, even though PPV and NPV are
probably the most clinically relevant properties.
 Now assume that our test statistic is no longer binary , but
takes on a series of values (for instance how many of five
distinct risk factors a person exhibits).
 Clinically we make a rule that says the test is positive if the

number of risk factors meets or exceeds some threshold (#RF
> x).
 Suppose our previous table resulted from using x = 4.
 Let’s see what happens as we vary x.

Test criterion
True disease status
Pos Neg
Pos 245
45 200
Neg 755
55 700
1000
Se = 27/100 = .45
Sp = 727/900 = .78
FPF = 1- Sp = .22
Acc = (27+727)/1000 = .75
PPV = 27/200 = .18
NPV = 727/800 = .93
Se , Sp , and interestingly both PPV and NPV .
‘‘-’’
‘‘+’’
without the disease

with the disease
‘‘-’’
‘‘+’’
without the disease

with the disease
Threshold TPR FPR As we relax our threshold for
6 0.00 0.00 defining “disease,” our true positive
rate(sensitivity) increases, but so
5 0.10 0.11 does the false positive rate
(FPR).
4 0.27 0.19
3 0.45 0.22
The ROC curve is a way to visually
2 0.73 0.27 display this information.
1 0.98 0.80
0 1.00 1.00
Threshold TPR FPR
6 0.00 0.00
5 0.10 0.11
4 0.27 0.19
3 0.45 0.22
2 0.73 0.27
1 0.98 0.80
0 1.00 1.00
What might an even better The diagonal line shows what we would expect
ROC curve look like? from simple guessing (i.e., pure chance).
Threshold TPR FPR
6 0.00 0.00
5 0.10 0.11
4 0.27 0.19
3 0.45 0.22
2 0.73 0.27
1 0.98 0.80
0 1.00 1.00
Note the immediate sharp rise in

sensitivity. Perfect accuracy is
represented by upper left corner
 The ROC curve allows us to see, in a simple visual
display, how sensitivity and specificity vary as our
threshold varies (2).
 The shape of the curve also gives us some visual clues

about the overall strength of association between the
underlying test statistic (in this case #RFs that are
present) and disease status (2).
 The ROC methodology
easily generalizes to test
statistics that are continuous
(such as lung function or a
blood gas). We simply fit a
smoothed ROC curve
through all observed data
points (2).
 The total area of the grid
represented by an ROC
curve is 1, since both
TPR and FPR range from 0
to 1.
 The portion of this total area

that falls below the ROC
curve is known as the area
under the curve,or AUC.
 The AUC serves as a quantitative summary of the
strength of association between the underlying test
statistic and disease status (2).
 An AUC of 1.0 would mean that the test statistic could

be used to perfectly discriminate between cases and
controls (2).
 An AUC of 0.5 (reflected by the diagonal 45° line) is

equivalent to simply guessing (2).
 The AUC can be shown to equal the Mann- Whitney U
statistic, or equivalently the Wilcoxon rank statistic, for
testing whether the test measure differs for individuals
with and without disease (2).
 It also equals the probability that the value of our test

measure would be higher for a randomly chosen case
than for a randomly chosen control (2).
AUC~ 0.540
FPR
ROC curve
AUC~0.95
FPR
ROC curve
 What defines a “good” AUC?
 Opinions vary
 Probably context specific
 What may be a good AUC for predicting COPD may be
very different than what is a good AUC for predicting
prostate cancer
 .90-1.0 = excellent
 .80-.90 = good
 .70-.80 = fair
 .60-.70 = poor
 .50-.60 = fail
Remember that <.50 is worse than guessing (4).
 .97-1.0 = excellent
 .92-.97 = very good
 .75-.92 = good
 .50-.75 = fair (5).
 Suppose we have two candidate test statistics to use to
create a binary decision rule. Can we use ROC curves
to choose an optimal one?
 We can formally compare AUCs for two competing test
statistics, but does this answer our question?
 AUC speaks to which measure, as a continuous

variable, best discriminates between cases and
controls?
 It does not tell us which specific cutpoint to use, or

even which test statistic will ultimately provide the
“best” cutpoint.
2 methods:
 The first method assumes that the best cut-off point for
balancing the sensitivity and specificity of a test is the point on
the curve closest to the (0, 1) point. In this method, optimal
sensitivity and specificity are defined as those yielding the
minimal value for (1 − sensitivity)2 + (1 − specificity)2. The
cut-off point corresponding to these sensitivity and specificity
values is the one closest to the (0, 1) point and is taken to be
the cut-off point that best differentiates between people with
disease and those without disease (1).
 The second method that may be used to determine the optimal
cut-off point for a test is the Youden index (J) . J is defined as
the maximum vertical distance between the ROC curve and
the diagonal or chance line and is calculated as J = maximum
(sensitivity + specificity −1). Using this measure, the cut-off
point on the ROC curve which corresponds to J, that is, at
which (sensitivity + specificity − 1) is maximized, is taken to
be the optimal cut-off point. An intuitive interpretation of J is
that it corresponds to the point on the curve farthest from
chance (1).
 Open the excel file.
 The first column is the patient coding number that you do not really need.
The second column gives the age as a continuous variable The third
column is a categorical variable about having thrombosis (1) and not
having thrombosis (0)
 Exercise:
1. Calculate the sensitivity and 1-specificity Considering that age is the test
variable and thrombosis is the state variable. Calculate the ROC curves for
and plot it. Comment on the obtained results.
 Open the excel file.
 The first column is the patient age, The second column gives the sensitivity
and the column 2 the 1-specificity.
 Exercise:
1. Calculate the youden index for each age and choose according to it the
optimal cut off point.
 1- Akobeng AK. Understanding diagnostic tests 3: Receiver operating
characteristic curves. Acta Paediatrica 2007, vol 96, issue 5.
 2- Kaiser permanente center for health research. Receiver operating

characteristic (ROC) curves: assessing the predictive properties of a test
statistic – Decision theory 2009.
 3- http://en.wikipedia.org/wiki/Receiver_operating_characteristic
 4- http://gim.unmc.edu/dxtests/roc3.htm
 5- www.childrens-mercy.org/stats/ask/roc.asp

Hani Tamim, PHD Associate Professor Department of Internal Medicine American University of Beirut

Uploaded by

Copyright:

Available Formats

Hani Tamim, PHD Associate Professor Department of Internal Medicine American University of Beirut

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hani Tamim, PHD Associate Professor Department of Internal Medicine American University of Beirut

Uploaded by

Copyright:

Available Formats

Hani Tamim, PhD

 Effective method of evaluating the performance of

 ROC analysis is related in a direct and natural way to

 ROC analysis since then has been used in medicine,

 Accuracy, PPV, and NPV will vary according to the

 Se and Sp are thus “fundamental” test properties and

 Clinically we make a rule that says the test is positive if the

 Suppose our previous table resulted from using x = 4.

 Let’s see what happens as we vary x.

without the disease

without the disease

Note the immediate sharp rise in

 The shape of the curve also gives us some visual clues

 The portion of this total area

 An AUC of 1.0 would mean that the test statistic could

 An AUC of 0.5 (reflected by the diagonal 45° line) is

 It also equals the probability that the value of our test

 AUC speaks to which measure, as a continuous

 It does not tell us which specific cutpoint to use, or

 2- Kaiser permanente center for health research. Receiver operating

You might also like