Screening tests: parameters
and interpretation of results
Al Katz, MD, MPH
Dept. of Public Health Sciences
University of Hawaii
Definition of Screening (in the
context of secondary prevention)
• The presumptive identification of
unrecognized disease by the application of
screening tests.
• Screening tests sort out apparently well
persons who probably have disease from
those who probably do not.
Definition of Screening (cont.)
• A screening test is not intended to be
diagnostic
• Persons with positive screening tests must
be referred to their physicians for diagnosis
and necessary treatment
Adapted from: US Commission on Chronic Illness,
1951
Characteristics of Screening
Tests
• Reliability = “Repeatability” or precision
• Validity = “Truth” or accuracy; does the
screening test really measure what it is
supposed to measure? To assess validity,
need an external standard (“gold standard”)
for comparison
Which shooter has the most validity?
The most reliability?
Shooter “A”
Shooter “B”
Validity parameters
• Sensitivity: The ability of the screening test
to correctly identify (as “screen positive”)
those with the disease
• Specificity: The ability of the screening test
to correctly identify (as “screen negative”)
those without the disease
Relationship between sensitivity
and specificity
• In many instances, there is a “reciprocal”
relationship between sensitivity and
specificity: one is increased only at the
expense of the other
Non-diseased
Diseased
Random serum glucose levels
With overlapping curves of
normal and diseased, need to
select a screening “cut-off”
Non-diseased
value: above which will be
“screen positive”; below
which will be “screen
negative”
Diseased
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
Where do we place the “cut-off” if we want 100%
sensitivity?
Non-diseased
Diseased
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
100% sensitivity = may
have “false positive”
Non-diseased results but no “false
negative” results
Diseased
Screen negative Screen positive
Random serum glucose levels
Where do we place the “cut-off” if we want 100%
specificity?
Non-diseased
Diseased
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
Non-diseased
Diseased
Screen negative Screen positive
Random serum glucose levels
100% specificity = may have
“false negative” results but no
Non-diseased “false positive” results
Diseased
Screen negative Screen positive
Random serum glucose levels
How to set “cut-off” value
• How bad is it to miss a case: What is the
cost of a “false negative” result?
How to set “cut-off” value
• How bad is it to miss a case: What is the
cost of a “false negative” result?
• How bad is it to label a normal person as
having a disease: What is the cost of a
“false positive” result?
How to set “cut-off” value
• How bad is it to miss a case: What is the
cost of a “false negative” result?
• How bad is it to label a normal person as
having a disease: What is the cost of a
“false positive” result?
• What is the prevalence of the disease in the
population (is it common or rare)?
2 X 2 contingency table
True disease present
Yes No
a a+b
+ b
Screening
test result
_ c d c+d
a+c b+d a+b+c+d
2 X 2 contingency table
True disease present
Yes No
True +
a a+b
+ b
Screening
test result
_ c d c+d
a+c b+d a+b+c+d
2 X 2 contingency table
True disease present
Yes No
True + False +
a a+b
+ b
Screening
test result
_ c d c+d
a+c b+d a+b+c+d
2 X 2 contingency table
True disease present
Yes No
True + False +
a a+b
+ b
Screening
test result True -
_ c d c+d
a+c b+d a+b+c+d
2 X 2 contingency table
True disease present
Yes No
True + False +
a a+b
+ b
Screening
test result False - True -
_ c d c+d
a+c b+d a+b+c+d
2 X 2 contingency table
True disease present Total
Yes No screen
True + False +
positive
a a+b
+ b
Screening
test result False - True - Total
_ c d c+d screen
negative
a+c b+d a+b+c+d
Total with disease Total without disease
Sensitivity
True positives = a
Total with disease a+c
Specificity
True negatives d
=
Total without disease b+d
Positive Predictive Value
True positives a
=
Total screen positives a+b
Predictive value
• Positive predictive value = proportion of
positive screening tests that actually have
the disease
• “Consumer’s” value: A patient gets
screened for a disease. If the test result is
positive, what does it mean? Does he/she
have the disease?
Predictive value of a screening
test
• In general, positive predictive value (PPV) is
influenced by how common or rare the disease is,
in the population being screened.
• If screening test has < 100% specificity, PPV will
vary directly with prevalence: higher PPV when
test is applied to higher prevalence populations;
lower PPV when test is applied to lower
prevalence populations
Effect of prevalence on PPV
• Testing for antibodies to Herpes simplex
virus type 2 (HSV-2)
Copas AJ, Cowan FM, Cunningham AL, Mindel A. An
evidence based approach to testing for antibody to herpes
simplex virus type 2. Sex Transm Infect 2002;78:430-4.
HSV-2 antibody screening
Focus Technologies HerpeSelect-2 ELISA IgG:
type-specific antibody test for glycoprotein G2
Sensitivity: 96.1% (95% CI: 88.9-99.2%)
Specificity: 97.0% (95% CI: 93.0-99.0%)
Package insert and technical sheet available from URL:
[Link]
Example 1: 30 y/o female family practice
(FP) clinic patient with 3-9 lifetime sexual
partners, Seattle
• HSV-2 prevalence = 28.1%
• If 100,000 women in this population have
definitive testing for HSV-2:
• 28,100 will have true HSV-2 infection
• Sensitivity of screening test = 96.1%, so
28,100 x 0.961 = 27,004 of the infected
women will be correctly identified by the
screening test as being screen +
2 X 2 contingency table
True disease present
Yes No
27,004 b
+
Screening
test result
_ c d
28,100 100,000
30 y/o female FP clinic patient, Seattle
(cont.)
• 100,000 – 28,100 = 71,900 women will be
uninfected.
• Specificity of the screening test = 97%, so
71,900 x 0.97 = 69,743 of the uninfected
women will be correctly identified by the
screening test as being screen -
2 X 2 contingency table
True disease present
Yes No
27,004 b
+
Screening
test result
_ c 69,743
28,100 71,900 100,000
2 X 2 contingency table
True disease present
Yes No
27,004 2,157
+
Screening
test result
_ 1,096 69,743
28,100 71,900 100,000
2 X 2 contingency table
True disease present
Yes No
27,004 2,157 29,161
+
Screening
test result
_ 1,096 69,743 70,839
28,100 71,900 100,000
Example 1: Positive predictive value
True positives a
Total screen positives =a+b
27,004
29,161 = 92.6%
Example 2: 24 y/o female FP clinic patient
with 1-2 lifetime sexual partners, Seattle
• HSV-2 prevalence = 2.7%
• If 100,000 women in this population are
definitively tested for HSV-2:
• 2,700 will have true HSV-2 infection
• Sensitivity of screening test = 96.1%, so
2,700 x 0.961 = 2,595 of the infected
women will be correctly identified by the
screening test as being screen +
2 X 2 contingency table
True disease present
Yes No
2,595 b
+
Screening
test result
_ c d
2,700 100,000
24 y/o female FP clinic patient, Seattle
(cont.)
• 100,000 – 2,700 = 97,300 women will be
uninfected.
• Specificity of the screening test = 97%, so
97,300 x 0.97 = 94,381 of the uninfected
women will be correctly identified by the
screening test as being screen -
2 X 2 contingency table
True disease present
Yes No
2,595 b
+
Screening
test result
_ c 94,381
2,700 97,300 100,000
2 X 2 contingency table
True disease present
Yes No
2,595 2,919
+
Screening
test result
_ 105 94,381
2,700 97,300 100,000
2 X 2 contingency table
True disease present
Yes No
2,595 2,919 5,514
+
Screening
test result
_ 105 94,381 94,486
2,700 97,300 100,000
Example 2: Positive predictive value
True positives a
Total screen positives =a+b
2,595
5,514 = 47.1%
Effect of prevalence on PPV
• In the higher prevalence (older, greater
number of sex partners) population, 92.6%
of those testing positive are truly HSV-2
infected
• In the lower prevalence (younger, lower
number of sex partners) population, less
than 50% of those testing positive are truly
HSV-2 infected
Final thoughts
May need to be “selective” in applying screening
tests.
Cannot correctly interpret a screening test result
without an understanding/appreciation of the
“predictive value” of the screening test.
Even tests with excellent sensitivity and specificity,
may have low positive predictive values when
applied to populations with low prevalence.
