18/11/2024

Evaluation of Diagnostic Test Accuracy &

Health Screening Programs

Joey Yang, Assistant Professor

JC School of Public Health and Primary Care
The Chinese University of Hong Kong

A practical question in public health

• In the initial stage of the COVID-19 pandemic, it is crucial to detect those

who had been infected rapidly. However, PCR test (核酸檢測), the most
accurate method for detecting the infection, required laboratory analysis
and took several days to get a result. Rapid antigen tests (快速抗原測試)
were widely used to make rapid diagnosis.
• How accurate were rapid antigen tests as compared to a PCR test? For
example, if a person was positive on a rapid antigen test, how likely was
the person truly infected? In other words, what was the chance that the
rapid antigen test result was correct?

1
 18/11/2024

Learning outcomes

 Be able to calculate the indexes for diagnostic test accuracy

 Recognize the inverse relation between sensitivity and specificity

 Interpret the meaning and uses of Receiver Operating Characteristic curve

 Name some of the key criteria for a good screening program

 Explain what study design is needed for evaluating screening programs

 Enumerate the reasons why screening may not be beneficial

 Enumerate the possible harms of screening

Diagnostic test and related terms

2
 18/11/2024

Gold Standard: A test that has been generally accepted, at that moment,
as the standard and the best method for definite diagnosis of that disease

• Hemoglobin for anemia

• Immunoassay test for HIV infection
• PCR test for COVID-19
• Coronary angiography for coronary heart disease
• Oral glucose tolerance test (OGTT) for type 2 diabetes
• Colonoscopy for colorectal cancer
• Histopathology of liver biopsy for liver cancer
• Trans-rectal ultrasound (TRUS)-guided biopsy for prostate cancer

Why may clinicians and patients prefer less accurate tests?

 Simpler
 More rapid
 Safer: less invasive, less adverse effects/ complications
 Cheaper
 More acceptable

3
 18/11/2024

The test being compared to the gold/ presumably better

standard: index test

 Fasting blood glucose test vs. OGTT for type 2 diabetes

 Faecal occult blood test (FOBT) vs. colonoscopy for colorectal cancer
 Prostate-Specific Antigen (PSA) test vs. biopsy for prostate cancer
 Rapid urease test vs. endoscopy for Helicobacter pylori infection
 Blood test of pregnant women vs. needle biopsy to obtain a sample from
within the uterus for Down Syndrome
 Chest x-rays/ sputum smears vs. bronchoscopy/ lung biopsy for pneumonia
 Electrocardiograms (ECGs)/ blood test vs. catheterization/ cardiac MRI for
acute myocardial infarction
7

The simpler tests are used as proxies for more elaborate

but more accurate or precise ways of establishing the
presence of disease, with the understanding that some
risk of misclassification results. This risk is justified by the
safety and convenience of the simpler tests.

But simpler tests are only useful when the risks of

misclassification are known and are acceptably low. This
requires a sound comparison of their accuracy to an
appropriate standard.

4
 18/11/2024

“Truth”
Disease
+ -
Test a b
True positive False positive a+b all positives
+

c d
False negative True negative
- c+d all negatives

Total a+c all ill b+d all healthy

Prevalence: Proportion of persons with disease in the population.
Prevalence = (a+c)/(a+b+c+d)
Sensitivity: the proportion of people with the disease who have a positive
test (i.e., the index test) for the disease. (A sensitive test will rarely miss
people with the disease.)
Sensitivity = True positive/All disease = a/(a+c)
False negative rate: the proportion of people with the disease who
have a negative test.
False negative rate = c/(a+c)  = 1 - Sensitivity
9

“Truth”
Disease
+ -
Test a b
True positive False positive a+b all positives
+

c d
False negative True negative
- c+d all negatives

Total a+c all ill b+d all healthy

Specificity: the proportion of people without the disease who have a

negative test. (A specific test will rarely misclassify people as having
the disease when they do not.)
Specificity = True negative/All healthy = d/(b+d)
False positive rate: the proportion of people without the disease who
have a positive test.
False positive rate = b/(b+d)  = 1 - Specificity
Accuracy: Proportion of persons who are correctly identified by the test
Accuracy= (a+d)/(a+b+c+d)
10

5
 18/11/2024

Example: Assume a population of 1,000 people of whom 100 have

a disease and 900 do not have the disease

Test to identify the 100 People with the disease

True Characteristics
in the Population
Results of No
Disease Disease Total
Test
Positive TP, a=80 FP, b=100 180
Negative FN, c=20 TN, d=800 820
Total 100 900 1,000

Sensitivity = a / (a + c) = 80 / (80 + 20) = 80.0%

False negative rate = c / (a + c) = 20 / (80 + 20) = 20.0%
Specificity = d / (b + d) = 800 / (100 + 800) = 88.9%
False positive rate = b / (b + d) = 100 / (100 + 800) = 11.1%
Accuracy = (a + d) / (a + b + c + d) = (80 + 800) / (80 + 100 + 20 + 800) = 88.0%
11

How to define the cutoff points?

Tests with dichotomous results---there is no problem in choosing a
suitable cutoff point if the diseased (D1) and non-diseased (D0) populations
are clearly separated according to the test measure:
Negative Positive

D0 D1 Test measure

Tests of continuous variables--- Some overlaps between the two

populations usually occurs

Negative Positive

D0 D1 Test measure
FN FP

12

6
 18/11/2024

Effect of changing cutoff points on

sensitivity and specificity

When the cutoff point is progressively moved up (the criteria

becoming more stringent), FP is progressively reduced (and
eventually eliminated), resulting in an increasing specificity

Negative Positive

D0 FN D1
Test measure

13

Effect of changing cutoff points on

sensitivity and specificity

When the cutoff point is progressively moved down (the criteria

becoming more lenient), FN is progressively reduced (and
eventually eliminated), resulting in an increasing sensitivity

Negative Positive

D0 FP D1
Test measure

14

7
 18/11/2024

Example on setting up cut-off points--

tests of continuous variables

Diabetes Non-Diabetes

High

Blood
Sugar

Low
15

Sensitivity = 5/20 =25%

Specificity= 18/20=90% Test

16

8
 18/11/2024

Sensitivity = 17/20 =85%

Specificity= 6/20= 30%

17

Receiver Operating Characteristic (ROC) curve

100%
True Positive Rate
(sensitivity)

0%
0% 100%
False Positive Rate
(1-specificity)
18

9
 18/11/2024

Receiver operating characteristic

(ROC) curves
 The area under a ROC curve quantifies the overall
ability of the test to discriminate between those
individuals with the disease and those without the
disease.
 A truly useless test (one no better at identifying true
positives than flipping a coin) has an area of 0.5.
 A perfect test (one that has zero false positives and
zero false negatives) has an area of 1.00.
 Most tests will have area between those 0.5 and 1.

19

ROC curve extremes

Best Test: Worst test:

100%
100%
True Positive Rate

True Positive Rate

AUC = 100%
(sensitivity)

(sensitivity)

AUC = 50%
0% 0%
0% 100% 0% 100%
False positive rate False positive rate
(1-specificity) (1-specificity)

The distributions The distributions

don’t overlap at all overlap completely
20

10
 18/11/2024

ROC curve comparison

A good test: A poor test:

100%
100%
True Positive Rate

True Positive Rate

(sensitivity)

(sensitivity)
AUC = 90% AUC = 65%

0%
0%
0% 100% 0% 100%
False positive rate False positive rate
(1-specificity) (1-specificity)

21

Receiver operating characteristic

(ROC) curves
There is a trade-off between sensitivity and specificity (increasing
one will be at the expense of the other)
ROC curves
0.7
The ROC can be used for:
C
0.6 1) finding the best cutoff point
B A for a test: mathematically, the
True positive rate

0.5
optimal cutoff point is the one
(sensitivity)

0.4 closest to the upper left corner

0.3

0.2
2) comparing the performance
0.1 of several tests for the same
disease condition (the larger,
0
0 0.1 0.2 0.3 0.4 0.5 0.6 the better)
False positive rate
(1-specificity) 22

11
 18/11/2024

How to define the cutoff points?

be
There is a yin-yang relationship between sensitivity and specificity.
Changing test cut-off values to increase the sensitivity will reduce
the specificity, and vice-versa.

Where should one ‘draw the line’?

This very much depends on the consequences of being wrong--
whether it is preferable to be labeled false negative or false
positive.

23

High or Low Cut Off Level?

False positives False negatives

- Further unnecessary test - Missing serious disease at
- Unnecessary treatment an early treatable stage
- Patient’s emotional affect - Spreading of infectious
- Burden to the health care system disease like COVID-19

24

12
 18/11/2024

The consequences of FN or FP

If a disease labeled FN has severe consequences (e.g., an HIV-

positive person incorrectly tested negative is given a false sense of
security, and then carries on with unprotected sex, thus endangering
others), one would want FN to be as low as possible, and the
sensitivity as high as possible.

If a disease labeled FP is undesirable (e.g., a healthy woman

incorrectly tested positive on Pap smear may be subjected to
unnecessary surgery and /or radiation treatment), one would want FP
to be as low as possible and the specificity as high as possible

25

Selection of Proper Cut Off Point

High Sensitivity High Specificity

( false negative) ( false positive )
- Severe or fatal disease - Non-fatal disease
- Effective therapy available - Non-curable disease
at an early treatable stage - Dangerous or expensive
- False positive - not so method of treatment
harmful - False positive - harmful !!

26

13
 18/11/2024

Ways to increase sensitivity

1. Lowering the cut-off or making criteria for disease

definition less stringent

2. Parallel Testing – the administration of 2 screening

tests at the same time and persons with a positive
result on any of these tests are considered positive

27

Parallel testing

 Parallel testing To be positive for the combination, a positive for

EITHER one of the tests will suffice.
 This raises sensitivity but lowers specificity.
Example: breast cancer screening frequently employs a combination
of mammography and breast physical exam applied in parallel. If
either test is positive, then further investigation is indicated.

Sensitivity of tests A & B combined in parallel: A+ B-A*B

Specificity of tests A & B combined in parallel : A* B

28

14
 18/11/2024

Ways to increase specificity

1. Making the criteria for disease definition more stringent

or increasing the cut-off level

2. Seral testing – an initial screening test is administered

and then, only persons who are positive on this
preliminary test will undergo a second additional
screening procedure

29

Series testing

 Serial testing: To be positive for the combination, one must be

positive on both the first AND second tests.
 Raises specificity, but lowers sensitivity

Example: HIV is first tested with a sensitive (but not specific)

serological test. This catches all positives, but includes many false
positives. The Western blot is done only on positives. It is very
specific and can exclude the false positives.

Sensitivity of tests A & B combined in series: A* B

Specificity of tests A & B combined in series: A+ B-A*B

30

15
 18/11/2024

Serial test Parallel test

“AND” “OR”

31

High sensitivity vs. high specificity: summary

High sensitivity favored High specificity favored

More diseased people correctly ruled in: More non-diseased people correctly ruled out:
 Severe or fatal disease at early stage;  The following test/treatment is harmful,
 Highly efficacious/effective treatment painful and expensive;
available;
 Infectious, huge harm to others;

More non-diseased people wrongly ruled in: More diseased people wrongly ruled out:
 The following test/treatment is not  Non-fatal disease;
invasive or painful;  No effective treatment available;

How to get higher sensitivity: How to get higher specificity:

 Lower cutoff value;  Higher cutoff value;
 Make the definition less stringent;  Make the definition more stringent;
 Parallel tests;  Serial tests;

32

16
 18/11/2024

Positive and negative predictive values

In the clinical setting, a more important question is:

If the test results are positive (or negative) in a given patient, what is the
probability that this patient has (or does not have) the disease?

In other words:
What proportion of patients who test positive (or negative) actually have (or
do not have) the disease?

33

Positive and negative predictive values

“Truth”
Disease
+ -
Test a b
True positive False positive a+b all positives
+

c d
False negative True negative
- c+d all negatives

Total a+c all ill b+d all healthy

 Positive predictive value (PV+) is the proportion of patients with
positive test results who are correctly diagnosed a/(a+b)
 Negative predictive value (PV-)is the proportion of patients with
negative test results who are correctly diagnosed d/(c+d) 34

17
 18/11/2024

Example: Assume a population of 1,000 people of whom 100 have

a disease and 900 do not have the disease

Test to identify the 100 People with the disease

True Characteristics
in the Population
Results of No
Disease Disease Total
Test
Positive TP, a=80 FP, b=100 180
Negative FN, c=20 TN, d=800 820

Total 100 900 1,000

Positive predictive value = a / (a + b) = 80 / (80 + 100) = 44.4%

Negative predictive rate = d / (c + d) = 800 / (20 + 800) = 97.6%

35

Determinants of predictive values:

• Sensitivity (from research publications)

• Specificity (from research publications)
• Prevalence of the disease in YOUR patient population

• With all other parameters remaining unchanged:

• Sensitivity↑  PPV↑
• Specificity↑  PPV↑
• Prevalence↑  PPV↑

• Hard to predict the direction of change of PPV when sensitivity and

specificity change together (which is common), because they often
go in opposite directions
36

18
 18/11/2024

Relation between results of liver scan and diagnosis in

344 patients
(Drum and Christacapoulus,1972)

Liver scan Pathology

Sensitivity 90% Abnormal Normal Total
Specificity 63% (+) (-)
Abnormal (+) 231 32 263
Normal (-) 27 54 81
Total 258 86 344

Prevalence of disease ( liver abnormality) = 258/344 = 0.75 or 75%

Positive predictive value = 231/263 = 0.88 or 88%

37

Relation between results of liver scan and diagnosis in

344 patients
(Drum and Christacapoulus,1972)

Liver scan Pathology

Sensitivity 90% Abnormal Normal Total
Specificity 63% (+) (-)
Abnormal (+) 77 96 173
Normal (-) 9 162 171
Total 86 258 344

Prevalence of disease ( liver abnormality) = 86/344 = 0.25 or 25%

Positive predictive value = 77/173 = 0.45 or 45%

38

19
 18/11/2024

Relationship between prevalence & predictive values

As prevalence decreases, positive predictive value

decreases, while negative predictive value increases

Assuming sensitivity = 95%, specificity = 95%

Prevalence (%) 99 90 80 70 60 50 40 30 20 10 1

PPV (%) 99.9 99.4 99 98 97 95 93 89 83 68 16

NPV (%) 16 68 83 89 93 95 97 98 99 99.4 99.9

39

Screening program and its evaluation

40

20
 18/11/2024

Screening

 “Screening is the identification of asymptomatic diseases or risk

factors.”

 “Medical screening is the systematic application of simple test(s)

or inquiry to apparently well persons to detect people having a
specific pre-clinical disorder or at increased risk of disease for
early treatment or prevention with an aim to improve the health
of the screened.”

41

Diagnostic vs. Screening tests

• Tests performed in persons without symptom or sign of an illness are usually

termed screening, whereas those done in individuals with a symptom or sign
are referred to as diagnostic.

• Screening tests use quick and simple testing procedures to identify and
separate persons: who may have a disease from those that may not
• Diagnostic tests help to make diagnosis in symptomatic disease or to follow-
up on screening test

42

21
 18/11/2024

Diagnostic vs. Screening tests

Normal colon Adenoma (10-15 years) Colorectal cancer

Normal Pre-clinical phase Clinical phase

Biologic onset of Detectable Onset of Formal Treatment

Disease by screening symptoms diagnosis

Fecal occult blood test Colonoscopy

43

Diagnostic vs. Screening tests

Screening test Diagnostic test

When/to whom Preclinical phase, asymptomatic or at risk Clinical phase, symptomatic

Characteristics Usually simple/ rapid/ cheap/ safe/ less accurate Accurate, ‘gold standard’

If positive Diagnostic test, prevention, or early treatment Treatment

44

22
 18/11/2024

Disease Screening Test Diagnostic Test

Cervical cancer Pap smear Colposcopy

Liver cancer Alpha fetoprotein Liver biopsy

HIV infection ELISA for HIV Western blot

Coronary heart ECG Coronary

disease angiography

45

Colorectal Cancer Screening Program

46

23
 18/11/2024

FOBT in high-risk individuals

Test positive Test negative

Colonoscopy

Cancer Cancer
present Absent

Treatment Follow-up

(Health improved?)

A simplified flow of a colorectal cancer screening program

47

WHO’s criteria of a good screening program

1. The disease being sought should be an important public health
problem.
2. There should be an acceptable form of treatment for patients with
recognizable disease.
3. The natural history of the disease, including its development from
latent to declared disease, should be adequately understood.
4. There should be a recognizable latent or early symptomatic stage.
5. There should be a suitable screening test or examination for
detecting the disease at the latent or early symptomatic stage, and
this test should be acceptable to the population.

48

24
 18/11/2024

WHO’s criteria of a good screening program

6. The facilities required for diagnosis and treatment of patients
revealed by the screening program should be available.
7. There should be an agreed policy on whom to treat as patients.
8. Treatment at the pre-symptomatic, borderline stage of a disease
should favorably influence its course and prognosis.
9. The cost of case-finding (which should include the cost of diagnosis
and treatment) needs to be economically balanced in relation to
possible expenditure on medical care as a whole.
10. Case-finding should be a continuing process, not a ‘once and for all’
project.

49

Unfortunately, these criteria are rarely carefully examined by

people who propose various screening programs. In fact,
many people are even not aware of these criteria or even
don’t realize that there should be some criteria for screening.

“Early detection & early treatment (entailed by screening) are

always beneficial”

50

25
 18/11/2024

Summary of the Criteria

1. Disease
• Important public health problem
• Long detectable pre-clinical phase

2. Screening test
• Performance: sensitivity, specificity, positive predictive value
• Feasibility: simplicity, cost, safety, acceptability
• Backed by accurate diagnostic methods

3. Treatment
• Effectiveness
• Early treatment after screening being more effective than later
treatment without screening, when the patient becomes symptomatic
• Safety
• Cost-effectiveness
51

Low positive predictive value in screening

 Because of the low prevalence of most diseases in

asymptomatic people, the positive predictive value of most
screening tests is low, even for tests with high specificity.
 Clinicians who perform screening tests on their patients must
accept the fact that they will have to work up many patients who
have positive screening test results but do not have disease.
 However, they can minimize the problem by concentrating their
screening efforts on people with a higher prevalence for disease.

52

26
 18/11/2024

Simplicity of screening

 An ideal screening test should take only a few minutes to perform,

require minimum preparation by the patient, depend on no special
appointments, and be inexpensive.
 Simple, quick examinations such as blood pressure
measurements are ideal screening tests. Conversely, tests such
as colonoscopy, which are expensive and require an appointment
and bowel preparation, are best suited for diagnostic testing in
patients with symptoms and clinical indications.
 Other tests fall between these two extremes.

53

Determinants of (financial) cost of screening

 The test/ procedure itself

 Subsequent evaluations performed on patients with positive test
results (Thus sensitivity, specificity, and predictive value affect
cost)
 Whether the test requires a special visit to the physician
 Extra time off work
 Additional transportation
 How often a screening test must be repeated

54

27
 18/11/2024

Safety of screening tests

It is reasonable and ethical to accept a certain risk for diagnostic tests

applied to sick patients seeking help for specific complaints. The physician
cannot avoid action when the patient is severely ill, and does his or her
best. It is quite another matter to subject presumably well people to risks.
In such circumstances, the procedure should be especially safe.
 When colonoscopy, with a rate of two perforations per 1,000
examinations, is used to screen for colorectal cancer in people in
their 50s, perforations occur more often than cancers are found.
 One estimate of risk projected 29,000 excess cancers as a result of
70 million CT scans per formed in the United States in a single year.

55

Acceptability of screening tests

 If a screening test is associated with discomfort, it usually takes

several years to convince large percentages of patients to obtain
the test. This has been true for Pap smears, mammograms,
sigmoidoscopies, and colonoscopies.
 The acceptability of the test to clinicians can determine which tests
patients receive.

56

28
 18/11/2024

Summary of the Criteria

1. Disease
• Important public health problem
• Long detectable pre-clinical phase

2. Screening test
• Performance: sensitivity, specificity, positive predictive value
• Feasibility: simplicity, cost, safety, acceptability
• Backed by accurate diagnostic methods

3. Treatment
• Effectiveness
• Early treatment after screening being more effective than later
treatment without screening, when the patient becomes symptomatic
• Safety
• Cost-effectiveness
57

Two common but wrong ways to evaluate the

effectiveness of a screening program
1. 10,000 people were screened. 100 lung cancer cases were detected. The
mean survival time of these 100 screen-detected cases was longer than that
of the cases detected in routine clinical practice.  the screening is
effective in prolong patients’ survival (there could be lead time bias)
2. 10,000 people were screened. 100 lung cancer cases were detected and
provided early treatment. In 5 years, the lung cancer mortality in this group
was 0.5% (=50/10,000), lower than the 1.5% observed in the unscreened
population (the control group).  the screening is effective in reducing
mortality (there could be length-time bias)

58

29
 18/11/2024

Lead-time bias

 Lead-time bias: the systematic error of

apparent increased survival from detecting
disease in an early stage
 What is lead time? How does lead-time bias
occur?

59

Lead-time bias
 Here the disease starts in 1985, is diagnosed in
1992 and the person dies of that disease in
1995. How long is his survival? Three years.

60

30
 18/11/2024

Lead-time bias
 Now we institute a screening program. The disease
starts in 1985 and is detected by the screening program
in 1989. The person dies of the disease in 1995. How
long was the survival? Six years. Screening seems to
have increased their survival time, correct?

61

Lead-time bias

62

31
 18/11/2024

Lead-time bias
 You have also noted that in either situation, it is 10
years from the time the disease started until the
person dies.
 If our measure is survival time, we can easily
produce a lead time bias.
 In this example, there is actually no benefit of the
screening process, in terms of survival. The person
still died in 1995. They know about the disease for
three years longer; that is the effect of the
screening. This example demonstrates a lead-time
bias of three years.
63

A
No true difference
True B
difference

It is indeed possible that the observed increased survival of cases

detected by screening is attributable to effective early treatment, but the
survival increased by treatment cannot be differentiated from lead time.

Source: Clinical Epidemiology, The Essentials, 5th Edition (p.160), by Robert H. Fletcher, Suzanne W. Fletcher, Grant S. Fletcher, 2012, Lippincott
Williams & Wilkins. 64

32
 18/11/2024

Lead-time bias
 An effective screening program for a life-threatening
disease should extend life.
 Screening studies with an outcome of survival time are
subject to lead-time bias that can favors the screening
process when there is no actual benefit from the
program.

65

Lead-time bias
 Instead, we need to look at the mortality rates
from the disease, i.e. the mortality rates in the
exposed group and the non-exposed group.
Disease-specific mortality rates have been the
most commonly used measure of disease
frequency.

 Mortality rates are the 'gold standard' for

measuring the effect of early screening and
treatment, not survival time.

66

33
 18/11/2024

Length-time bias
 Let's use this graph to consider the effect of length bias:

67

Length-time bias
 Disease onset is at zero and each line represents an
individual. The bottom person, for instance, has a very slow
growth rate of disease. The top line, with the steepest slope,
represents someone with aggressive disease. This person
has rapid growth and dies (D). The individuals with slower
growth lived to the point where they get screened (S).
 Length-time bias occurs because a screening initiative is
more likely to detect slow-growing disease. The proportion
of slow-growing lesions diagnosed during screening is
greater than the proportion of those diagnosed during usual
medical care. As a result, outcome appears better in
screened group because more cancers with a good
prognosis are detected.
68

34
 18/11/2024

Solution to length-time bias

Randomization makes the two groups comparable

Screening group

Fast Medium-rate Slow Free of

Growing Growing Growing cancer
Randomization

Tumor Tumor
Tumor

Fast Medium-rate Slow Free of

Growing Growing Growing cancer
Tumor Tumor
Tumor

Control group 69

Randomized controlled trial with mortality rate (or other

“hard” outcomes caused by the screened condition) as
the outcome measure is the best design for evaluating
the effectiveness of a screening program

70

35
 18/11/2024

Effectiveness of various screenings:

evidence from RCTs

71

Effectiveness of various screenings:

evidence from RCTs

72

36
 18/11/2024

Effectiveness of various screenings:

evidence from RCTs

73

Effectiveness of various screenings:

evidence from RCTs

74

37
 18/11/2024

Effectiveness of various screenings:

evidence from RCTs

75

Effectiveness of various screenings:

evidence from RCTs

76

38
 18/11/2024

Reasons why screening (early detection & early

treatment) may not be effective

1. Early treatment is somehow not more effective than routine

treatment (after patients develop symptoms and get diagnosed at
routine timing)
2. Early treatments are indeed more effective, but the compliance or
adherence rate of participants is low (i.e., many people do not
receive adequate treatments) for whatever reasons, e.g., financial,
adverse effects, etc.

77

Reasons why screening (early detection & early

treatment) may not be effective
 The prevalence of the targeted disease (e.g., lung cancer) in the
presumably healthy people is typically low. That means, in the first
place, only a small number out of a large population could theoretically
benefit from the screening.
 Because of the low prevalence, the positive predictive value of most
screening tests is typically low. In other words, most of the positive
results are false-positives. Clinicians will have to work up many patients
who have positive screening test results but do not actually have the
disease or only have “indolent” diseases only. This may cause some
undesirable effects that will offset the beneficial effects at the population
level:

78

39
 18/11/2024

Reasons why screening (early detection & early

treatment) may not be effective
3. Adverse events arising from the screening procedures per se: perforation
during colonoscopy (大腸鏡檢查過程中發生的腸穿孔); long-term
radiation effects (e.g., gene mutation that causes cancer) after exposure
to radiographic procedures such as repeated CT scans
4. Overdiagnosis: the diagnosis of an abnormality that bears no substantial
health hazard and no benefit for patients to be aware of (some of the
following treatments may cause severe complications/ deaths, e.g.,
various therapies for [indolent] cancers)

79

Reasons why screening (early detection & early

treatment) may not be effective

5. False-positive test results: resulting needless workups and treatments

(some of these may cause severe complications/ deaths, e.g., needle
biopsy of suspected prostate cancer, surgery after ovarian cancer
screening). Example:
In a study of ovarian cancer screening, 8.4% (3,285) of 39,000 women
had a false-positive result and one-third of those underwent surgery
as part of the diagnostic evaluation of the test result. Because of false-
positive screening tests, five times more women without ovarian
cancer had surgery than those with ovarian cancer!

80

40
 18/11/2024

Other adverse effects of screening

 Discomfort during the test procedure

 e.g., the majority of women undergoing mammography say that
the procedure is painful, although usually not so severe that
patients refuse the test

 Negative labelling effect:

 Being told that the screening test result is abnormal and more
testing is necessary may have an adverse psychological effect,
particularly in cancer screening. Some people with false-positive
tests continue to worry even after being told everything was
normal on follow-up tests.

81

Case example:
Low-dose CT scan for
lung cancer screening

82

41
 18/11/2024

Benefits
• RCT 1: “Number needed to screen” to prevent 1 lung cancer death:
323 over 6.5 years of follow-up with 3 rounds of annual LDCT
screening for high-risk current and former smokers aged 55 to 74
years.
• RCT 2: “Number needed to screen” to prevent 1 lung cancer death:
130 over 10 years of follow-up with 4 rounds of LDCT screening
for high-risk current and former smokers aged 50 to 74 years.

83

Harms of screening
• radiation-induced cancer, false-positive results leading to unnecessary
tests and invasive procedures, overdiagnosis, incidental findings, and
increases in distress
• For every 1000 persons screened, false-positive results led to 17
invasive procedures (number needed to harm, 59) and fewer than 1
person having a major complication.
• Overdiagnosis estimates varied greatly (0%-67% chance that a lung
cancer was overdiagnosed). Incidental findings were common, and
estimates varied widely (4.4%-40.7%of persons screened).

• Is the screening program worthwhile? Would you like to receive

the screening?

84

42
 18/11/2024

Thank You

85

43