Lecture 3 Outline

1. Positive Predictive Value

Predictive Values
2. Negative Predictive Value
Asst. Prof. TOL BUNKEA, MD,MSc-Epidemiology
(London School of Hygiene and Tropical Medicine, UK)
Head of Epidemiology Unit of National Centre for Parasitology, Entomology and
Malarial Control (CNM, MoH)
3. Reliability and Validity
Lecturer of Epidemiology and Biostatistics (NIPH, UHS)
Tel: 016 690 999
Email: [email protected], 2

Objectives: Predictive Values

At the end of this session you will be able to:
Truth
Disease present Disease absent
1 Define validity and reliability (Test validity measures)
Positive test True positive False positive
Negative test False negative True negative
2 Define and calculate Positive predictive value and
Negative predictive value
𝑻𝒓𝒖𝒆 𝑷𝒐𝒔𝒊𝒕𝒊𝒗𝒆
Describe the effect of prevalence on sensitivity, 𝑷𝒐𝒔𝒕𝒊𝒗𝒆 𝑷𝒓𝒆𝒅𝒊𝒄𝒕𝒊𝒗𝒆 𝑽𝒂𝒍𝒖𝒆 = x 100%
3 𝑨𝒍𝒍 𝑷𝒐𝒔𝒊𝒕𝒊𝒗𝒆
specificity and predictive values
𝑻𝒓𝒖𝒆 𝑵𝒆𝒈𝒂𝒕𝒊𝒗𝒆
𝑵𝒆𝒈𝒂𝒕𝒊𝒗𝒆 𝑷𝒓𝒆𝒅𝒊𝒄𝒕𝒊𝒗𝒆 𝑽𝒂𝒍𝒖𝒆 = x 100%
4 False positive rate, false negative rate
𝑨𝒍𝒍 𝑵𝒆𝒈𝒂𝒕𝒊𝒗𝒆

Predictive Values Predictive Values

• Positive predictive value (PPV) • Negative predictive value (NPV)
– The proportion of patients who test positive – The proportion of patients who test negative who
are actually free of the disease
who actually have the disease.
– The proportion of negative test results that are
– The proportion of positive test results that true negative
are true positive. – Likelihood of not having the disease when test is
– Likelihood of having the disease when test negative.
is positive. 𝑻𝒓𝒖𝒆 𝑵𝒆𝒈𝒂𝒕𝒊𝒗𝒆
𝑻𝒓𝒖𝒆 𝑷𝒐𝒔𝒊𝒕𝒊𝒗𝒆
𝑵𝑷𝑽 = x 100%
𝑨𝒍𝒍 𝑵𝒆𝒈𝒂𝒕𝒊𝒗𝒆
𝑷𝑷𝑽 = x 100%
𝑨𝒍𝒍 𝑷𝒐𝒔𝒊𝒕𝒊𝒗𝒆
• Note: PPV and NPV are not fixed characteristics of the
test

Sante Publique 1
 Another Interpretation of Predictive Value Reality
• Diagnostic tests are never perfect.
• If a person tests positive, what is the
• False positive and false negative results occur.
probability that he or she has the
• How much of a problem these false results
disease? may cause depends on the clinical context in
which a test is used.
• If that person tests negative, what is Truth
Disease present Disease absent
the probability that he or she does Positive test True positive False positive
not have the disease? Negative test False negative True negative

Behind the Test Results What the Test Shows

Predictive Values Applying Concept of Predictive Values

to Screening Test

Sante Publique 2
 Calculating Predictive Values Calculating Predictive Values

How to interpret?

Interpretation
• Among those who had a positive screening
new test, the probability of disease was 44%.
(PPV=44%)

• Among those who had a negative screening

new test, the probability of being disease-free
was 98%. (NPV=98%)

PPV Primarily Depends On …

• Prevalence of the disease in the
PPV? NPV? population tested,
• Sensitivity and specificity)
• In general, it depends more on
• the specificity
Interpretation: • and less on the sensitivity of the test (if the
•Among those who had a positive screening new test, the
disease prevalence is low)
probability of cervical cancer was 67%. (PPV=67%)
•Among those who had a negative screening new test, the
probability of being cervical cancer-free was 99%. (NPV=99%)

Sante Publique 3
 PPV Primarily Depends On … NPV Primarily Depends On …
• Although sensitivity and specificity are not • Negative predictive values are inversely
related to the prevalence of the disease in the
study population, predictive values do vary
related to the prevalence of the disease
with prevalence. in the population.
• Positive predictive values are directly related
to the prevalence of the disease. • The negative predictive value of a test
• The positive predictive value of a test decreases as the prevalence of disease in
increases as the prevalence of disease in the the population increases.
population increases.

• Prevalence => NPV

Prevalence => PPV

Predictive Values Formula Calculation of PPV and NPV

• For those who are interested
• Construct the table and use the definition to
guide the calculation of PPV and NPV
• [Or, use the formula]
• In a multiple testing situation, PPV and NPV
are calculated for each test (not for the
combined test)
• Use Bayes’ theorem

Relationship of Disease Prevalence to Relationship of Disease Prevalence

Predictive Value to Predictive Value

? ?

? ?

Sante Publique 4
 Prevalence of Disease So If a Person Tests Positive …
• The probability that he or she has the
disease depends on the prevalence of
the disease in the population tested and
the validity of the test (sensitivity and
specificity)

• In general, specificity has more impact on

predictive values

The Relationship of Specificity to The Relationship of Specificity to

Predictive Value Predictive Value

The Relationship of Specificity to The Relationship of Specificity to

Predictive Value Predictive Value

Sante Publique 5
 Note:
• So we have seen that the implication of a
positive test result varies according to the
prevalence of the disease.
• The chances of a positive test indicating a
truly positive result may be considerably
different in populations with different
prevalences of the disease in question.
• This point is also important in making
decisions about screening for diseases in a
population.

Age-Specific Breast Cancer Incidence Rates Results of First Screening Mammography by Age
U.S., All Races (SEER 1984-88) Group —UCSF Mobile Mammography Program

PPV of First Screening Mammography by Age

and Family History of Breast Cancer
Reliability (Repeatability)
• Reproducibility, Repeatability, Reliability
– all mean that the results of a test or measure
are identical or closely similar each time it is
conducted
– The extent to which a test will produce the
same result if it is repeated.
• Because of variation in laboratory procedures,
Consequence of Different PPVs observers, or changing conditions of test subjects
(such as time, location), a test may not
consistently yield the same result when repeated

Sante Publique 6
 Reliability (Repeatability) Validity and Reliability
• First graph shows the
distribution of the results of
a reliable test, with little
variation around the true
value.
• Compare this with second
graph, showing the
distribution of the results of
a less reliable test, which
show wider variation around
the central value.

Validity and Reliability

Validity and Reliability

Reliability is the degree to which a research instrument produces consistent results

Validity is how accurate an instrument is at measuring what it is trying to measure.

Reliability (Repeatability) Intra-Subject Variation

• Is a variation in the results of a test
Different types of variation conducted over (a short period of) time
on the same individual.

• The difference is due to the changes

Intra- Inter-
Intra-subject
observer observer
(such as physiological, environmental,
variation etc.) occurring to that individual over that
variation variation
time period

Sante Publique 7
 Variation in Blood Pressure Readings: Inter-Observer and Intra-Observer Variation
A 24-Hour Period • Inter-observer variation is a variation in the
result of a test due to multiple observers
examining the result (inter = between)
• Intra-observer variation is a variation in the
result of a test due to the same observer
examining the result at different times (intra =
within) For example, blood sugar levels vary during the
course of the day.
• The difference is due to the extent to which
observer(s) agree or disagree when
interpreting the same test result

Agreement between Two Observers

(Or Two Observations)
Percent Agreement
• A perfect agreement occurs when:
b=0
c=0

Observer or Instrument Variation:

Example
Overall Percent Agreement

Sante Publique 8
 Exercise
Outcome of Test Results • The director of the blood bank wants to know
whether blood donors who test positive should be
informed of their positive HIV status on the basis of
this test.
• What do you need to calculate in order to help you
advise the director?

Exercise Exercise
• So in this situation you need to calculate the • In a low-prevalence population like this
positive predictive value (PPV) of the EIA in the one we usually employ a two-stage
blood donor population. procedure to test for disease.
• First a screening test with very high
PPV = %
sensitivity is used to identify as many true
positives as possible (in this example, so
Gold Standard Test that the risk of transfusing infected blood
Positive Negative is as low as possible).
Positive test 38 1,999
EIA
Negative test 2 97,961

Exercise
Exercise
• By only doing confirmatory tests on individuals with
• If the aim is also to inform positive individuals positive screening tests, the prevalence of HIV infection
of their status, then a second test is in the population being tested is increased (compared
performed, but only on individuals who had a with the original population.
positive screening test. • The prevalence of HIV infection in the blood donor
population was 0.04%.
• This second test should have maximum
• If we select out all those who had a positive EIA test,
specificity, and may therefore have lower
the HIV prevalence would now be 38 / 1999 = 1.9%.
sensitivity than the screening test.
• This is still very low, but is 475 times higher than 0.04%.
• It is often referred to as a confirmatory test • This, and the high specificity of the confirmatory test,
(although it too is unlikely to be 100% combine to make the positive predictive value of the
accurate). confirmatory test as high as possible.

Sante Publique 9
 Example
• A new diagnostic test to determine the presence of
malaria parasites has been tested on all patients. The
performance of the test has been assessed by
comparison of the results with those obtained by
microscopy (the ‘Gold Standard’).
• The results obtained are shown in the following table:
(Calculate: Prevalence, Sensitivity, Specificity, PPV, NPV)

Summary

Sante Publique 10
 Diagnostic tests: summary

Diagnostic tests: summary Diagnostic tests: summary

Diagnostic tests: summary

Sante Publique 11
 Diagnostic tests: summary Diagnostic tests: summary

Diagnostic tests: summary Diagnostic tests: summary

Sante Publique 12