10. Diagnostic & Screening Tests
Diagnostic and screening tests attempt to reveal an otherwise
hidden truth about patients (i.e., their health status: diseased
or disease-free).
•Physical examination
•Radiographs/Computed Tomography (CT)
•Blood and urine assays
•Cytology (Paps smear, Oral brush biopsy)
•Saliva (HIV testing)
11. Discrimination & Classification
“The fundamental principle of diagnostic testing [and
screening] rests on the belief that individuals with disease are
different from individuals without disease and that diagnostic
[and screening] tests can distinguish between these two
groups.”
Riegelman, Studying a Study and Testing a Test, 2000
•Valid (i.e., accurate)
Sensitivity, specificity, ROC
Predictive values
Multiple tests
•Reliable (i.e., precise or repeatable)
Percent agreement
Kappa
12. Discrimination & Classification
Disease status comes from an external source of “truth”
regarding the patients in the population:
•Gold standard or reference standard
Adequate
Independent
Unbiased
Representative
13. Interlude: The Gold Standard
Unbiased
• The procedure used to establish the truth should not bias
the truth.
• Surgery or histology the “truth” will consist of the
more advanced cases
Representative
• Cadaver studies of TMJ (older). Patients younger.
• Caries simulations (drilled holes in teeth) versus natural
lesions
14. Interlude: The Gold Standard
Adequate
•Surgery or autopsy (common in imaging studies)
•Time between imaging and surgery/biopsy
•Applies to positive cases
•Negative cases – clinical follow-up
Independent
•Histology provides an independent truth.
•Occasionally all of the available information, including
the test being tested is used to establish the gold
standard. Bone lesion for example (BFO). Creates a bias
in favor of the test
15. Discrimination & Classification
“Appearances to the mind are of four kinds. Things either are
what they appear to be [ ]; or they neither are, nor appear to
be [ ]; or they are, and do not appear to be [ ]; or they are
not, and yet appear to be [ ]. Rightly to aim in all these
cases is the wise man’s task.”
Epictetus (c. 50-120)
Discourses, Bk I, Chp 27
16. Validity: Sensitivity & Specificity
Sensitivity
= Ability of the test to correctly identify those with disease
= Probability of testing positive given the presence of disease
= TP / (TP + FN)
= a / (a + c)
17. Validity: Sensitivity & Specificity
Specificity
= Ability of the test to correctly identify those without disease
= Probability of testing negative given the absence of disease
= TN / (FP + TN)
= d / (b + d)
18. Validity: Sensitivity & Specificity
Assume a population of 1000 people of whom 100 have a
disease. Of these 100 people, the test correctly identifies 80.Of
the 900 disease-free people, the test correctly identifies 800.
Sensitivity = a / (a + c) = 80 / 100 = 80%
Specificity = d / (b + d) = 800/ 900 = 89%
Gordis, 2009, Table 5-1
19. Validity: Sensitivity & Specificity
Sensitivity and Specificity
• Inherent characteristics of the test
• Stable over different populations with different disease
prevalence
• Useful for comparing performance of two tests
(e.g., Digital versus film mammography / Pisano, NEJM 2005)
• Have a reciprocal relationship with one another
20. Validity: Sensitivity & Specificity
Low cutoff High sensitivity
Low specificity
False positives
Moderate cutoff balance
High cutoff Low sensitivity
High specificity
False negatives
Courtesy, S. Fleming, 2011
21. Validity: Receiver Operating Characteristic Curve
X-axis:
False positive ratio
(1-specificity)
Y-axis:
True positive ratio
(sensitivity)
27. Validity: Performance / Predictive Value
Sensitivity and specificity are useful, but
• May be numerically different if obtained on a group of
people with early stages of disease compared with a group
with more advanced disease.
• We do not know ahead of time who has the disease and
who does not. Rather, we get the test results and need to
interpret the findings.
28. Validity: Performance / Predictive Value
Positive Predictive Value
= Ability of the test to correctly identify those who test positive
= Probability of having the disease given a positive test result
= TP / (TP + FP)
= a / (a + b)
29. Validity: Performance / Predictive Value
Negative Predictive Value
= Ability of the test to correctly identify those who test negative
= Probability of not having the disease (i.e., being disease-free)
given a negative test result
= TN / (FN + TN)
= d / (c + d)
30. Validity: Positive & Negative Predictive Values
Assume a population of 1000 people of whom 100 have a
disease. Of these 100 people, the test correctly identifies 80.Of
the 900 disease-free people, the test correctly identifies 800.
Positive PV = a / (a + b) = 80 / 180 = 44%
Negative PV = d / (c + d) = 800/ 820 = 98%
Gordis, 2009, Table 5-7
31. Validity: Predictive Values & Prevalence
Assume a test with a sensitivity of 80% and specificitity of 90%.
What happens to the predictive values when the prevalence of
the disease varies? To fill in the cells, assume a convenient total
population, in this case 1000.
80 90
20 810
Positive PV = a / (a + b) = 80 / 170 = 0.4706 = 47.1%
Negative PV = d / (c + d) = 810/ 830 = 0.9759 = 97.6%
After Kramer Clinical Epidemiology and Biostatistics, 1988
32. Validity: Predictive Values & Prevalence
Assume a test with a sensitivity of 80% and specificitity of 90%.
Positive PV = a / (a + b) = 400 / 450 = 0.8888 = 88.9%
Negative PV = d / (c + d) = 100/ 550 = 0.8181 = 81.8%
After Kramer Clinical Epidemiology and Biostatistics, 1988
33. Validity: Predictive Values & Prevalence
Assume a test with a sensitivity of 80% and specificitity of 90%.
Positive PV = a / (a + b) = 720 / 730 = 0.9863 = 98.6%
Negative PV = d / (c + d) = 90/ 270 = 0.3333 = 33.3%
After Kramer Clinical Epidemiology and Biostatistics, 1988
34. Validity: Predictive Values & Prevalence
Assume a test with a sensitivity of 80% and specificitity of 90%.
Some additional terms:
• Pretest probability = prior probability = prevalence
• Post-test probability = posterior probability =
positive/negative predictive value
• Bayes Theorem (Thomas Bayes, 1702-61)
45. Multiple Tests: Simultaneous
Suppose in a population of 1000
people, 200 have the disease and
Test A sensitivity = 80%
Test B sensitivity = 90%
Net sensitivity = A+, B+ or both
Step 1: 0.8 x 200 = 160 who are A+
Step *: 0.9 x 200 = 180 who are B+
Step 2: 0.9 x 160 = 144 who are A+B+
Step 3: 160 – 144 = 16 who are A+ only
Step 4: 180 – 144 = 36 who are B+ only
Step 5: 144 + 16 + 36 = 196 = A+,B+, or
both
Step 6: 196/200 = 98%
Courtesy, S. Fleming, 2011
46. Multiple Tests: Simultaneous
Suppose in a population of 1000
people, 800 don’t have the disease
Test A specificity = 60%
Test B specificity = 90%
Net specificity = A- and B-
Step 1: 0.6 x 800 = 480 who are A-
Step *: 0.90 x 800 = 720 who are B-
Step 2: 0.9 x 480 = 432 who are A-
and B-
Step 3: 432/800 = 54%
Courtesy, S. Fleming, 2011
49. Reliability
Reliability (aka repeatability or precision) is the ability of the
test to give consistent results when performed more than once
by on the same individual under the same conditions, even if
conducted by different examiners.
Sources of variability (the antithesis of repeatability)
•Subjects
BP reading (throughout day, sitting/standing, R/L arm)
Serum glucose (throughout day, day of the week)
•Instrumentations
PSA assay (5% variability even when measuring identical blood
sample)
•Observer
Intra-observer
Inter-observer
50. Reliability: Percent Agreement
Percent agreement
= number of tests that agree / total number of tests
= (a + d) / (a + b + c + d)
= 35 / 40
= 0.875 = 87.5%
51. Reliability: Kappa
Measure agreement beyond that expected from chance alone:
Kappa = (percent agreement – chance agreement)
(1 – chance agreement)
Kappa varies between 0 (no agreement) and 1 (perfect agreement)
< 0.40 Poor agreement
0.40 - 0.75 Fair to good agreement
> 0.75 Excellent
In example, chance agreement = 0.695
Kappa = (0.875 – 0.695)/(1 – 0.695) = 0.180/0.305 = 0.590
53. Reliability: Calculating Kappa
Two pathologists independently read and score 75
histopathology slides using their own criteria to subtype the
lesion as Grade II or Grade III
Gordis, 2009, Figure 5-17
58. Screening
“Screening is defined as the presumptive identification of
unrecognized diseasese or defects by the application of tests,
examinations, or other procedures that can be applied rapidly.”
Friis and Seller, 2009
“For screening to be of benefit, treatment given during the
detectable preclinical phase must result in a better prognosis
than therapy given after symptoms develop.”
Hennekens and Buring, 1987
59. Screening
Nature of the Disease
•Important health problem
Morbidity/Mortality
•Treatable
Unethical to screen if untreatable, except to prevent transmission
(e.g., early cases of AIDS versus protecting blood supply)
•Relatively high prevalence
Rare disease PPV is low & cost per case detected is high
Exceptions: Phenylketouria (PKU), 1 in 15,000 births, but
consequences are severe (mental retardation), treatment is
simple (dietary restriction), screening tests are simple.
•Detectable preclinical phase (long latency period)
Biological Symptoms
Onset Appear
Clinical
Clinical Outcome
Screening
Diagnosis
60. Screening
Nature of the Test
•Simple
Easy to learn and perform
No complicated patient preparation
•Rapid
To administer
To yield results
•Safe
Screened populations are overwhelmingly healthy – keep them
that way
•Valid and reliable
High sensitivity
Relatively high specificity – accept some FP as there will be
follow-up confirmatory tests, but what is the cost and morbidity
of the follow-up, the cost of mislabeling someone, etc.
61. Screening
Societal Factors
•Cost
Relatively inexpensive
Benefit/cost ratio favorable versus other health care expenditures
•Acceptable
Unpalatable or difficult tests refusal to participate
62.
63. Resources
Langlotz, Radiology 2003 – supplement to Gordis, especially
for ROC curves.
Pisano et al. NEJM 2005 – example of an application of
concepts.
Linker, AJPH 2012 – and interesting historical perspective of
screening, specifically for scoliosis.
US Preventive Services Task Force (USPSTF) – the source of
many guidelines (and some controversy) regarding screening:
< http://www.uspreventiveservicestaskforce.org/>.