Transcript A diagnostic test study

Diagnostic Test Studies
Tran The Trung
Nguyen Quang Vinh
Why we need a diagnostic test?
 We need “information” to make a decision
 “Information” is usually a result from a test
 Medical tests:
y To screen for a risk factor (screen test)
y To diagnosse a disease (diagnostic test)
y To estimate a patient’s prognosis (pronostic test)
 When and in whom, a test should be done?
y When “information” from test result have a value.
Value of a diagnostic test
 The ideal diagnostic test:
y Always give the right answer:
x Positive result in everyone with the disease
x Negative result in everyone else
y Be quick, safe, simple, painless, reliable & inexpensive
 But few, if any, tests are ideal.
 Thus there is a need for clinically useful
substitutes
Is the test useful ?





Reproducibility (Precision)
Accuracy (compare to “gold standard”)
Feasibility
Effects on clinical decisions
Effects on Outcomes
Determining Usefulness
of a Medical Test
Question
Possible Designs Statistics for
Results
1. How
Studies of:
reproducible - intra- and inter
is the test? observer &
- intra- and inter
laboratory
variability
Proportion
agreement,
kappa, coefficient
of variance, mean
& distribution of
differences (avoid
correlation
coefficient)
Determining Usefulness
of a Medical Test
Question
2. How
accurate is
the test?
Possible Designs
Statistics for
Results
Cross-sectional, case- Sensitivity,
control, cohort-type
specificity,
designs in which test PV+, PV-,
result is compared
ROC curves,
with a “gold standard” LRs
Determining Usefulness
of a Medical Test
Question
3. How
often do
test results
affect
clinical
decisions?
Possible
Designs
Diagnostic
yield studies,
studies of pre& post test
clinical
decision
making
Statistics for Results
Proportion abnormal,
proportion with
discordant results,
proportion of tests
leading to changes in
clinical decisions; cost
per abnormal result or
per decision change
Determining Usefulness
of a Medical Test
Question
4. What are
the costs,
risks, &
acceptability
of the test?
Possible
Designs
Prospective or
retrospective
studies
Statistics for Results
Mean cost, proportions
experiencing adverse
effects, proportions
willing to undergo the
test
Determining Usefulness
of a Medical Test
Question
Possible Designs
5. Does
doing the
test
improve
clinical
outcome,
or having
adverse
effects?
Randomized trials, cohort
or case-control studies in
which the predictor
variable is receiving the
test & the outcome
includes morbidity,
mortality, or costs related
either to the disease or to
its treatment
Statistics for
Results
Risk ratios, odd
ratios, hazard
ratios, number
needed to treat,
rates and ratios
of desirable
and
undesirable
outcomes
Common Issues for
Studies of Medical Tests
 Spectrum of Disease Severity and Test Results:
y Difference between Sample and Population?
y Almost tests do well on very sick and very well
people.
y The most difficulty is distinguishing Healthy & early,
presymtomatic disease.
 Subjects should have a spectrum of disease
that reflects the clinical use of the test.
Common Issues for
Studies of Medical Tests
 Sources of Variation:
y Between patients
y Observers’ skill
y Equipments
=> Should sample several different institutions to
obtain a generalizable result.
Common Issues for
Studies of Medical Tests
 Importance of Blinding: (if possible)
y Minimize observer bias
y Ex. Ultrasound to diagnose appendicitis
(It is different to clinical practice)
Studies of Diagnostic tests
 Studies of Test Reproducibility
 Studies of The Accuracy of Tests
 Studies of The Effect of Test Results on Clinical
Decisions
 Studies of Feasibility, Costs, and Risks of Tests
 Studies of The Effect of Testing on Outcomes
Studies of Test Reproducibility
 The test is to test the precision
y Intra-observer variability
y Inter-observer variability
 Design:
y Cross-sectional design
y Categorical variables: Kappa
y Continuous variables: coefficient of variance
 Compare to it-self (“gold standard” is not
required)
Studies of the Accuracy of Tests
 Does the test give the right answer?
 “Tests” in clinical practice:
y Symptoms
y Signs
y Laboratory tests
y Imagine tests
To find the right answer.
“Gold standard” is required
How accurate is the test?
 Validating tests against a gold standard:
 New tests should be validated by comparison
against an established gold standard in an
appropriate subjects
 Diagnostic tests are seldom 100% accurate
(false positives and false negatives will occur)
Validating tests against a gold
standard
 A test is valid if:
y It detects most people with disorder (high Sen)
y It excludes most people without disorder (high Sp)
y a positive test usually indicates that the disorder is
present (high PV+)
 The best measure of the usefulness of a test is
the LR: how much more likely a positive test is
to be found in someone with, as opposed to
without, the disorder
A Pitfall of Diagnostic test
A test can separate the very sick from the very
healthy does not mean that it will be useful in
distinguish patients with mild cases of the
disease from others with similar symptoms
Sampling
 The spectrum of patients should be
representative of patients in real practice.
 Example: Which is better? What is the limits?
y Chest X-ray to diagnose aortic aneurism (AA). Sample
are 100 patients with and 100 without AA that
ascertained by CT scan or MRI.
y FNA to diagnose thyroid cancer. 100 patients with
nodule > 3cm and had indication to thyroidectomy
(biopsy was the gold standard).
“Gold standard”
 “Gold standard” test: often confirm the presence
or absence of the disease : D(+) or D(-).
 Properties of “Gold standard”:
y
y
y
y
Ruling in the disease (often doing well)
Ruling out the disease (maybe not doing well)
Feasible & ethical ? (ex. Biopsy of breast mass)
Widely acceptable.
The test result
 Categorical variable:
y Result: Positive or Negative
y Ex. FNA cytology
 Continuous variable:
y Next step is: find out “cut-off point” by ROC curve
y Ex. almost biochemical test: pro-BNP, TR-Ab,..
Analysis of Diagnostic Tests
How accurate is the test?
 Sensitivity & Specificity
 Likelihood ratio: LR (+), LR (-)
 Posterior probability (Post-test probability) /
Positive, Negative Predictive value (PPV, NPV);
given Prior probability (Pre-test probability)
Sensitivity and Specificity
a
Sens 
ac
d
Spec 
bd
Disease D
Test
Result
+
-
“Gold standard”
+
a
c
b
d
Positive & Negative Predictive Value
 PV (+): positive
predictive value
 PV (-): negative
predictive value
a
PV () 
ab
d
PV () 
cd
Test
Result
+
-
Disease D
+
a
b
c
d
a /(a  c)
LikelihoodRatio( LR) 
b /(b  d )
Posterior odds
When combined with information on the prior
probability of a disease*, LRs can be used to
determine the predictive value of a particular test
result:
Posterior odds = Prior odds x Likelihood ratio
*expressing the prior probability [p] of a disease as the prior odds [p/(1-p)] of
that disease. Conversely, if the odds of a disease are x/y, the probability of the
disease is x / (x + y)
Choice of a cut-off point
for continuous results
Consider the implications of the two possible
errors:
 If false-positive results must be avoided (such as
the test result being used to determine whether
a patient undergoes dangerous surgery), then
the cutoff point might be set to maximize the
test's specificity
 If false-negative results must be avoided (as
with screening for neonatal phenylketonuria),
then the cutoff should be set to ensure a high
test sensitivity
Choice of a cut-off point
for continuous results
 Using receiver operator characteristic (ROC)
curves:
y Selects several cut-off points, and determines the
sensitivity and specificity at each point
y Then, graphs sensitivity (true-positive rate) as a
function of 1-specificity (false-positive rate)
 Usually, the best cut-off point is where the ROC
curve "turns the corner”
RECEIVER OPERATING
CHARACTERISTIC (ROC) curve
 ROC curves (Receiver
Operator Characteristic)
 Ex. SGPT and Hepatitis
SGPT
D+
D-
Sum
< 50
10
190
200
50-99
15
135
150
100-149
25
65
90
150-199
30
30
60
200-249
35
15
50
250-299
120
10
130
>300
65
5
70
Sum
300
450
750
Sensitivity
1
1-Specificity
1