1. Principles of Epidemiology Lecture 12 Dona Schneider, PhD, MPH, FACE
Screening
Principles of Epidemiology
Lecture 12
Dona Schneider, PhD, MPH, FACE

2. Principles Underlying Screening Programs
Validity – the ability to predict who has the disease and who does not
Sensitivity – the ability of a test to correctly identify those who have the disease
A test with high sensitivity will have few false negatives
Specificity – the ability of a test to correctly identify those who do not have the disease
A test that has high specificity will have few false positives

3. Principles Underlying Screening Programs (cont.)
An ideal screening test would be 100% sensitive and 100% specific – that is there would be no false positives and no false negatives
In practice these are usually inversely related
It is possible to vary the sensitivity and specificity by varying the level at which the test is considered positive

4. Calculating Measures of Validity
True Diagnosis
Test Result
Disease
No Disease
Total a b
a+b
Positive
Negative
Sensitivity = a/(a+c); the probability of having a positive test if you are positive
Specificity = d/(b+d); the probability of having a negative test if you are negative
Positive Predictive Value = a/(a+b); the probability of having the disease if you test positive
Negative Predictive Value = d/(c+d); the probability of not having the disease if you test negative
Prevalence = (a+c)/(a+b+c+d)
Accuracy (efficiency of test) = (a+d)/(a+b+c+d) c d
c+d
Total
a+c
b+d
a+b+c+d

5. Note the Following Screening Relationships
Specificity + false positive rate = 1
d/(b+d) + b/(b+d) = 1
If the specificity is increased, the false positive rate is decreased
If the specificity is decreased, the false positive rate is increased
Sensitivity + false negative rate = 1
a/(a+c) + c/(a+c) = 1
If the sensitivity is increased, the false negative rate is decreased
If the sensitivity is decreased, the false negative rate is increased

6. Probability of Disease
Pre-test probability of disease = disease prevalence
Post-test probability of disease =
If normal, c/(c+d)
If negative, a/(a+b)

7. Interrelationship Between Sensitivity and Specificity

8. Sensitivity and Specificity of a Blood Glucose Level
Sensitivity and Specificity of a Blood Glucose Level of 110 mg/100 ml for Presumptive Determination of Diabetes Status
Blood Glucose Level
(mg/100 ml)
Diabetics
(Percent)
Nondiabetics
(Percent)
All those with level over 110 mg/100 ml are classified as diabetics
92.9
(true positives)
51.6
(false positives)
All those with level under 110 mg/100 ml are classified as nondiabetics
7.1
(false negatives)
48.4
(true negatives)
100.0
100.0

9. Adjusting Sensitivity and Specificity by Adjusting Cut Points

10. Which is Preferred: High Sensitivity or High Specificity?
If you have a fatal disease with no treatment (such as for early cases of AIDS), optimize specificity
If you are screening to prevent transmission of a preventable disease (such as screening for HIV in blood donors), optimize sensitivity

11. Remember….
Sensitivity and specificity are functions of the screening test
If you use a given screening test on a low prevalence population, you will have a low positive predictive value and potentially many false positives

12. Translated into Real Life…..
Elisa is about 90% sensitive and 99% specific
Population
Prevalence of HIV
PV+
PV-
NJ (7 million)
1.5%
58%
99.8%
Disease Yes
Disease No
Total
Test +
94,500
68,950
163,450
Test -
10,500
6,826,050
6,836,550
Total
105,000
6,895,000
7 million
Efficiency of test = (TP + TN)/Total tested = 98.9%
But, 10,500 people who are HIV+ think they are disease free
Another 68,950 are frightened into believing they have the disease and require more testing

13. If You Change To a High Risk Population, You Get Better Results….
PV+
PV-
Prevalence of HIV
IV Drug User
50%
98.9%
90.8%
Total
Disease Yes
Disease No
Test +
3,150 35
3,185
Test -
350
3,465
3,185
Total
3,500
3,500
7,000
Efficiency of test = (TP + TN)/Total tested = 94.5%
Now 350 people who are HIV+ think they are disease free
But only 35 are frightened into believing they have the disease and require more testing

14. Suppose You Have a Very High Prevalence?
HIV seropositivity is 90% among IV drug users in Newark
PV+ = 99.9%
PV- = 52%
But, why bother to screen?

15. Example: Breast Cancer Screening
Mammogram Results
Disease
No Disease
Total
Positive
132
983
1,115
Sensitivity = 132/177 = 74.6%
Specificity = 63,650/64,633 = 98.5%
Positive Predictive Value = 132/1,115 = 11.8%
Negative Predictive Value = 63,650/63,695 = 99.9%
You can improve sensitivity and specificity by using more than one screening test, using multiphasic screening and targeting high risk populations
If prevalence of a disease is low, even a highly valid test will yield a low predictive value
Negative 45
63,650
63,695
Total
177
64,633
64,810

16. Example: Disease X (prevalence = 2%) True Diagnosis of Disease X
Test Results
Disease
No Disease
Total
Positive 18 67 49
Negative 2
933
931
The prevalence of Disease X is 2% (1,000 x 0.02 = 20)
Sensitivity = 18/20 = 90%
Specificity = 931/980 = 95%
Positive Predictive Value = 18/67 = 27%
Negative Predictive Value = 931/933 = 99.8%
Total
1000 20
980

17. Example: Disease X (prevalence = 1%) True Diagnosis of Disease X
Test Results
Disease
No Disease
Total
Positive 9
49.5
58.5
Negative
940.5
941.5 1
The prevalence of Disease X is 1% (1,000 x 0.01 = 10)
Sensitivity = 9/10 = 90%
Specificity = 940.5/990 = 95%
Positive Predictive Value = 9/58.5 = 14.5%
Negative Predictive Value = 940.5/941.5 = 99.9%
Total 10
980
1000
To increase positive predictive value increase prevalence by screening high risk populations

18. Importance of Prevalence in Screening
Assume we have a test for AIDS which has a sensitivity of 100% and a specificity of %. We wish to apply it to female blood donors who have an HIV prevalence of 0.01% and we wish to apply it to male homosexuals in San Francisco, in whom the prevalence is 50%. For every 100,000 screened we find:
Female Donors
True Diagnosis of HIV
Test Results
Disease
No Disease
Total 10 5 15
Positive
99,985
Negative
99,985
Total 10
99.990
100,000
PV+ =
True Diagnosis of HIV
Male Homosexuals
Disease
No Disease
Total
50,003
Positive
50,000 3
49,997
Negative
49,997
Total
50,000
50,000
100,000
PV+ =

19. Relationship of Specificity to Predictive Value
500
250
Prev = 50%, Sens = 50%, Spec = 50%,
PV = 250/500 = 50%
1,000 -
Test +
Disease
400
Prev = 20%, Sens = 50%, Spec = 50%,
PV = 100/500 = 20%
1,000
800
200
500
100 -
Test +
Disease -
400 20
Prev = 20%, Sens = 90%, Spec = 50%,
PV = 180/520 = 31%
1,000
800
200
420
Test
580
180 +
Disease
Prev = 20%, Sens = 50%, Spec = 90%,
PV = 100/180 = 56%
1,000
800
200
820
100 -
720
Test
180 80 +
Disease

20. Suppose You Are Faced With the Following Brain Teaser
In a given population of 1,000 persons, the prevalence of Disease X is 10%. You have a screening test that is 95% sensitive and 90% specific.
What is the positive predictive value?
What is the efficiency of the test?

21. Suppose You Are Faced With the Following Brain Teaser (cont.)
Set up a 2x2 table
True Diagnosis of Disease X
Test Results
Disease
No Disease
Total
Positive
True Positive
False Positive
Given that the prevalence is 10% and the population is 1,000, then 100 people are sick
This means 900 are not
Since the sensitivity is 95%, 95% of the 100 truly sick will test as true positives (n=95)
Since the specificity is 90%, 90% of the 900 truly well will test as true negatives (n=180)
Fill in the blanks
Negative
False Negative
True Negative
Total
100
900
1000

22. Suppose You Are Faced With the Following Brain Teaser (cont.)
True Diagnosis of Disease X
Test Results
Disease
No Disease
Total
Positive 95 90
185
Negative
PV + = True positives / All positives
a / (a+b) = 95 / 185 = 51.4%
PV - = True negatives / All negatives
d / (c+d) = 810 / 815 = 99.4%
Efficiency of the Test = Total correct / Total predictions
(a+d) / (a+b+c+d) = ( ) / 1000 = 90.5% 5
810
815
Total
100
900
1000

23. Principles Underlying Screening Programs
Reliability – the ability of a test to give consistent results when performed more than once on the same individual under the same conditions
Variation in the method due to variability of test chemicals or fluctuation in the item measured (e.g., diurnal variation in body temperature or in relation to meals)
Standardize fluctuating variables
Use standards in laboratory tests, run multiple samples whenever possible
Observer variation
Train observers
Use more than one observer and have them check each other

24. Principles Underlying Screening Programs
Yield – the amount of previously unrecognized disease that is diagnosed and brought to treatment as a result of the screening program
Sensitivity
You must detect a sufficient population of disease to be useful
Prevalence of unrecognized disease
Screen high risk populations
Frequency of screening
Screening on a one time basis does not allow for the natural history of the disease, differences in individual risk, or differences in onset
Diseases have lead time
Participation and follow-up
Tests unacceptable to those targeted for screening will not be utilized

25. Conditions for Establishing Screening Programs
The condition should be an important health problem
There should be an accepted treatment for patients with recognized disease
If there is no treatment, it is premature to institute screening
Facilities for diagnosis and treatment should be available
It is unethical to screen without providing possibilities for follow-up
There should be a recognizable latent or early symptomatic stage
If early detection does not improve survival, there is no benefit from screening

26. Conditions for Establishing Screening Programs (cont.)
There should be a suitable test for examination, with sufficient sensitivity and specificity to be of use in identifying new cases
The test should be acceptable to the population
The natural history of the condition, including development from latent to declared disease, should be adequately understood
There should be an agreed-upon policy concerning whom to treat as patients

27. Conditions for Establishing Screening Programs (cont.)
The cost of case-finding should be economically balanced in relation to possible expenditure on medical care as a whole
Case-finding should be in a continuing process and not a one-time project

28. Biases in Screening Referral Bias (volunteer bias) Length Bias
Screening selectively identifies those with a long preclinical and clinical phase (i.e., those who would have a better prognosis regardless of the screening program)

29. Biases in Screening (cont.)
Lead Time Bias
The apparently better survival that is observed for those screened is not because these patients are actually living longer, but instead because diagnosis is being made at an earlier point in the natural history of the disease

30. Biases in Screening (cont.)
Overdiagnosis Bias (a misclassification bias)
Enthusiasm for a new screening program may result in a higher rate of false positives and give false impression of increased rates of diagnosis and detection
Also, false positives would result in unrealistically favorable outcomes in persons thought to have the disease