Understanding diagnostic

tests

P.J. Devereaux, MD, PhD

McMaster University
 Goals
• Understand sensitivity and specificity
 inform why they are in general problematic
• Discuss likelihood ratios
• Inform pretest and posttest probabilities of disease
• Understand testing and treatment thresholds
• Go through case examples
 convince you that understanding diagnostic properties is
essential to good clinical practice
 Diagnostic Tests: Definitions
Gold Standard
Disease present Disease absent

Test positive (+ve) (a) (b)

True positive (TP) False positive (FP)

Test negative (-ve) (c) (d)

False negative (FN) True Negative (TN)

Sensitivity: proportion of people with disease who have +ve test

Sensitivity= TP = a .

TP+FN a+c
SNOUT-Sensitive test, Negative result rule OUT disease
Specificity: proportion of people free of disease who have -ve test
Specificity= TN = d .

TN+FP d+b
 Disease present Disease absent

Test positive (+ve) (a) (b)

True positive (TP) False positive (FP)

Test negative (-ve) (c) (d)

False negative (FN) True Negative (TN)

Post-test probability = TP = a .
of disease given a TP+FP a+b
positive result

Post-test probability = FN = c .

of disease given a FN+TN c+d

negative result
 Likelihood ratios
Positive likelihood ratio (+ve LR) = LR for +ve test
- defined as likelihood of +ve test among individuals with disease,
relative to likelihood of +ve test among those without disease

Negative likelihood ratio (-ve LR) = LR for –ve test

- defined as likelihood of -ve test among individuals with disease,
compared to probability of -ve test among those without disease

Cystoscopy (Reference Test)

NMP 22 Result Bladder Cancer No Bladder Cancer
Positive 44 179
Negative 35 1073
Total 79 1252
Likelihood ratio for positive test result = (44/79)/(179/1252) = 3.90
Likelihood ratio for negative test result = (35/79)/(1073/1252)] = 0.52
 Coronary Artery Disease (CAD)
•This example will deal with exercise stress testing
(EST) in diagnosis of CAD
•Before testing we need to make some cut off
points for post-test probability of disease given a
+ve or –ve test
o
such that we will except as being truly positive,
negative, or still indeterminate
• What we are talking about is referred to as testing
and treatment thresholds
 Testing and treatment thresholds
•Reason for this decision is that we are not using gold
standard test which tell us disease present or absent with
certainty
•We are using another test which is not 100% accurate
therefore you will never be 100% confident that disease is
present or absent after test
•However, accurate tests can sometimes get you very close to
100% confident of your diagnosis
•With this in mind lets set some cut off points so that after
the test we can tell patient
a. They have CAD and need treatment with drug X
b. They do not have CAD and can rest assured
c. They still remain in an uncertain category, where more
testing is needed
 Case
35y/o male presents to your office with
3 month hx of substernal chest pain, predictably
brought on with exertion
lasts 45 minutes before resolving with rest

What is his pretest probability of CAD?

 CAD % probability of > 75% coronary
stenosis
Symptoms:
1. Substernal location
2. Brought on by exertion
3. Relieved in <ten minutes by rest or nitroglycerin
Ages: 30-39 40-49 50-59 60-69
Asymptomatic MEN 2% 6% 10% 12%
(0 symptoms) WOMEN .3% 1% 3% 8%
(1 symptom) MEN 5% 14% 22% 28%
WOMEN 1% 3% 8% 19%
(2 symptoms) MEN 22% 46% 59% 67%
WOMEN 4% 13% 32% 54%
Typical angina MEN 70% 87% 92% 94%
(all 3 Sx) WOMEN 26% 55% 79% 91%
 EST
Next, you decide to do an EST and he
gets 1.7 mm of ST depression

What is the specificity?

98%
What is the sensitivity?
9%
 What do you tell him?
1. He has CAD and needs drug X
2. He does not have CAD and can rest assured
3. He is in an uncertain category and needs
further testing
 Post-test probability of CAD
• To determine his post-test probability of
disease start by setting up 2X2 table
 with knowledge that his pretest likelihood of
disease was 21.8%
• this value corresponds to # of people with disease
 CAD
Present Absent
218 782
EST + ve TP FP
EST – ve FN TN

Sensitivity (9%) = TP Therefore, TP=19 and FN=218-19=199

TP+FN
(218)

Specificity (98%) = TN Therefore, TN=766 and FP=782-766=16

TN+FP
(782)
 Now table looks like

Present Absent
218 782
EST+ve TP FP
(19) (16)

EST-ve FN TN
(199) (766)

Post-test probability = TP = 19 = 54%

of disease given this TP+FP 19+16
positive results
 Case
45 y/o female with history of epigastric pain
undergoes endoscopy and this reveals 2 cm
duodenal ulcer
biopsy taken for clo test, comes back -ve
What is her pretest probability of H. pylori?
What is the sensitivity of clo test?
What is the specificity of clo test?
 What would you tell her?

1. She needs Tx for H. pylori

2. She does not need Tx for H. pylori
3. She needs further investigations
 Post-test probability of H.pylori
To determine her post-test probability of H.
pylori start by setting up a 2X2 table
with knowledge that her pretest likelihood was
95%
this value would therefore correspond to the #
of people with disease
 H. PYLORI
Present Absent
950 50
Clo test +ve TP FP
Clo test –ve FN TN

Sensitivity (95%) = TP Therefore, TP = 902 and FN = 950-902=48

TP+FN
(950)

Specificity (98%) = TN Therefore, TN = 49 and FP = 50-49=1

TN+FP
(50)
 Now table looks like
Present Absent
950 50
Clo test +ve TP FP
(902) (1)
Clo test –ve FN TN
(48) (49)

Post-test probability = FN = 48 = 49%

of disease given this FN+TN 48+49
positive results
 Now use likelihood ratios

For question 1 the finding of 1.7 mm of ST

Depression has LR=4.2

Start off by finding pretest probability of

disease on left-hand column

Then place a ruler through LR in center column

and read off post-test likelihood in right hand
column
 Summary
• In clinical decision making
• sensitivity and specificity have substantial limitations
• likelihood ratios are practical
• Pretest probability is important to inform post-
test probability
• Understand testing and treatment thresholds can
facilitate consistency in clinical care
• Understanding diagnostic properties is essential
to good clinical practice
 Searching for studies evaluating
diagnostic tests
 OVID click on “limits” and find “clinical queries”
 then click on “diagnosis (sensitivity)” or “diagnosis
(specificity)” depending on your goal
 PubMed on left hand side of page find PubMed Services
 click on “clinical queries” and then click on “search by clinical
study category”
• next under category click on “diagnosis”
 and then under scope click “specific search” or “sensitive search”
depending on your goal
 Resources for evaluating diagnostic
tests
 Empirical evidence of design-related bias in
studies of diagnostic tests
 JAMA September 1999;282:1061-1066
 MOST IMPORTANT RESOURCE FOR YOU TO
BUY
 Diagnostic Strategies for Common Medical
Problems. Editor: Robert J. Panzer
 Likelihood ratios
1. Pulmonary embolism: V/Q scanning
High probability: 18.3
Intermediate probability: 1.2
Low probability: 0.36
Normal/near normal 0.10

2. DVT: IPG Duplex U/S

-L.R. = 0.05 -L.R. = 0.07
+L.R. = 19+L.R. = 47

3. Exudative pleural effusion:

LDH ratio > 0.6 Protein ratio > 0.5
-L.R. = 0.14 -L.R. = 0.10
+L.R. = 43+L.R. = 45
 Disease present Disease absent

Test positive (+ve) (a) (b)

True positive (TP) False positive (FP)

Test negative (-ve) (c) (d)

False negative (FN) True Negative (TN)

 Positive predictive (+PV) value: proportion of people with

+ve tests who have disease
+PV= TP = a .
TP+FP a+b
+PV = post-test probability of disease given +ve result
 Negative predictive (-PV) value: proportion of people with
-ve test who are free of disease.
-PV= TN = d .
FN+TN c+d
100-(-PV)= post-test probability of disease given -ve test