2. OBJECTIVES
Be able to:
Define sensitivity and specificity
Construct a contingency table for a given test and condition
What factors influence positive predictive value and negative
predictive values?
3. PREVALENCE AND
INCIDENCE
Measures of how common a disease is:
Prevalence = cases per population at risk
Incidence = number of new cases per population at risk
4. SENSITIVITY
Among patients who have the disease, how often will the test
be positive
High Sensitivity = almost never misses someone with disease
Low sensitivity = can often be negative in someone with
disease
5. SPECIFICITY
Among patients who do not have the disease, how often will
the test be negative?
High specificity = test is almost always negative if the person
does not have disease
Low specificity = test can often be positive in someone who
does not have the disease
6. EXAMPLE 1:
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 29%
• Am J Respir Crit Care Med. 1997 Aug;156(2 Pt 1):492-6
How should we interpret a positive or negative result?.
7. CONTINGENCY TABLE
Possibilities of disease states and test results
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+)
Test negative (T-)
8. CONTINGENCY TABLE
Possibilities of disease states and test results
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
9. PREVALENCE AND
INCIDENCE
Prevalence = cases per population at risk
Incidence = number of new cases per population at risk
• How does high vs low prevalence affect the contingency
table?
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
10. PREVALENCE AND
INCIDENCE
Prevalence = cases per population at risk
Incidence = number of new cases per population at risk
• How does high vs low prevalence affect the contingency
table? Amount in each column
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
11. EXAMPLE 1:
Has Disease (D+)
= 2900
Does Not Have
Disease (D-) =
7100
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 29%
• Am J Respir Crit Care Med. 1997 Aug;156(2 Pt 1):492-6
Let’s consider 10,000 hypothetical patients evaluated for PE:
12. SENSITIVITY
Among patients who have the disease, how often will the test
be positive
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
13. SENSITIVITY
Among patients who have the disease, how often will the test
be positive
= True Positive / (True Positive + False Negatives)
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
14. SPECIFICITY
Among patients who do not have the disease, how often will
the test be negative
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
15. SPECIFICITY
Among patients who do not have the disease, how often will
the test be negative
=True Negative / (False Positive + True Negative)
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
16. EXAMPLE 1:
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 29%
• Am J Respir Crit Care Med. 1997 Aug;156(2 Pt 1):492-6
Specificity = True Negative / All Negative
Sensitivity = True Positive / All Positive
Has Disease (D+)
= 2900
Does Not Have
Disease (D-) =
7100
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
17. EXAMPLE 1:
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 29%
• Am J Respir Crit Care Med. 1997 Aug;156(2 Pt 1):492-6
41% = True Negative / 7100
99.5% = True Positive / 2900
Has Disease (D+)
= 2900
Does Not Have
Disease (D-) =
7100
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
18. EXAMPLE 1:
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 29%
• Am J Respir Crit Care Med. 1997 Aug;156(2 Pt 1):492-6
True Negative = 7100 * .41 = 2911
True Positive = 2900 * .995 = 2885
Has Disease (D+)
= 2900
Does Not Have
Disease (D-) =
7100
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
19. EXAMPLE 1:
Has Disease
(D+)
= 2900
Does Not Have
Disease (D-) =
7100
Test positive (T+) 2885 4189 (=7100-2911)
Test negative (T-) 5 (=2900–2885) 2911
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 29%
• Am J Respir Crit Care Med. 1997 Aug;156(2 Pt 1):492-6
True Negative = 7100 * .41 = 2911
True Positive = 2900 * .995 = 2885
20. POSITIVE PREDICTIVE
VALUE
Among patients who test positive for the disease, how often
they have the disease
Why do we care?
Because we don’t know whether the patient has the disease
or not!
21. POSITIVE PREDICTIVE
VALUE
Among patients who test positive for the disease, how often
they have the disease
*requires that we know the prevalence
Unlike the sensitivity and specificity, it will vary with
prevalence. Sensitivity and specificity are characteristics of
the test.
22. POSITIVE PREDICTIVE
VALUE
Among patients who test positive for the disease, how often
they have the disease
= True Positives / (True Positive + False Positives)
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
23. NEGATIVE
PREDICTIVE VALUE
Among patients who test negative for the disease, how often
they not have the disease
= True Negative / (False Negative + True Negative)
Has Disease (D+) Does Not Have
Disease (D-)
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
24. EXAMPLE 1:
Has Disease (D+)
= 2900
Does Not Have
Disease (D-) =
7100
Test positive (T+) 2885 4189
Test negative (T-) 5 2911
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 29%
• Am J Respir Crit Care Med. 1997 Aug;156(2 Pt 1):492-6
• PPV = 2885 / (2885 + 4189) = 41%
• NPV = 2911 / ( 2911 + 5) = 99.8%
25. EXAMPLE 1B:
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 75%
• Hypothetical population with very convincing signs and symptoms
26. EXAMPLE 1B:
Has Disease (D+)
= 7500
Does Not Have
Disease (D-) =
2500
Test positive (T+) True Positive False Positive
Test negative (T-) False Negative True Negative
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 75%
• Hypothetical population with very convincing signs and symptoms
27. EXAMPLE 1B:
Has Disease (D+)
= 7500
Does Not Have
Disease (D-) =
2500
Test positive (T+) 7463 1475
Test negative (T-) 37 1025
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 75%
• Hypothetical population with very convincing signs and symptoms
28. EXAMPLE 1B:
D-dimer over 500 for pulmonary embolism
• Sensitivity = 99.5%
• Specificity = 41%
• Prevalence in tested outpatients = 75%
• Hypothetical population with very convincing signs and symptoms
• PPV = 7463 / (7463 + 1475) = 83%
• NPV = 1025 / ( 1025 + 37) = 96.5%
Has Disease (D+)
= 7500
Does Not Have
Disease (D-) =
2500
Test positive (T+) 7463 1475
Test negative (T-) 37 1025
29. EXAMPLE 2:
Testing for Rare disease
• E.g. 24h Urine Metanephrines for Pheochromocytoma
(warning: made up numbers) = prevalence of 1% (among
tested patients)
• Sensitivity = 99%
• Specificity = 99%
• PPV of a positive result?
30. EXAMPLE 2:
Testing for Rare disease
• E.g. 24h Urine Metanephrines for Pheochromocytoma
(warning: made up numbers) = prevalence of 1% (among
tested patients)
• Sensitivity = 99%
• Specificity = 99%
Has Disease (D+)
=100
Does Not Have
Disease (D-) =
9900
Test positive (T+) 99 99
Test negative (T-) 1 9801
31. EXAMPLE 2:
Testing for Rare disease
• Test positive => 99 True positive, 99 False positive!
• PPV 50%
Even good tests are not reliable in low prevalence conditions
Has Disease (D+)
=100
Does Not Have
Disease (D-) =
9900
Test positive (T+) 99 99
Test negative (T-) 1 9801
32. EXAMPLE 3
Lung Ca Screening w/ Low dose CT: (Br J Cancer. 2008 May
20;98(10):1602-7)
• Sensitivity 88.9, Specificity 92.6
• Prevalence: age-adjusted incidence rate of lung cancer is 62
per 100,000 men and women per year in the United States
(Clin Chest Med. 2011 Dec; 32(4): 10.1016/j.ccm.2011.09.001.)
33. OBJECTIVES
Be able to:
Define sensitivity and specificity
Construct a contingency table for testing for a given
condition
What factors influence positive predictive value and negative
predictive values?
34. OBJECTIVES
Be able to:
Define sensitivity and specificity
Sensitivity = If a patient has the disease, how often will the test be positive?
Specificity = if the patient does not have the disease, how often will the test
be negative?
Construct a contingency table for testing for a given condition
Calculate # w/ and w/o dz, use sensitivity on pts w/ and specificity on pts w/o
What factors influence positive predictive value and negative
predictive values?
PPV increases w/ higher specificity+sensitivity and prevalence
NPV increases w/ high specificity+sensitivity and lower prevalence
35. NEXT STEPS
How can we use these characteristics?
• Likelihood ratio
• Bayes Theorem
• Expected Value Theorem
• Questions? Brian.locke@hsc.utah.edu