Evaluating Diagnosis Flashcards
What do studies of a diagnostic test usually evaluate the relationship between?
Being positive for standard diagnostic test vs being positive for evaluated screening test
What are the 4 groups in diagnostic test studies
True positive = A
False positive = B
False negative = C
True negative = D
First column for table for diagnostic test studies
Test positive
Test negative
Total
Second column
Disease (gold standard positive)
A
C
A+C
3rd column
No disease (gold standard -ve)
B
D
B+D
4th column
A+B
C+D
A+B+C+D
First row
Diseased, No disease
2nd row
Test positive, A, B, A+B
3rd row
Test negative, C, D, C+D
4th row
Total, A+C, B+D, A+B+C+D
Define sensitivity
Usefulness of a test for truly diseased population
Define specificity
Usefulness of a test of non-diseased population
Calculation for sensitivity
A/A+C
true positive/total diseased
Calculation for specificity
D/B+D
true negative/total non-diseased
Calculation for accuracy
A+D/A+B+C+D
true positive + true negative/total population
Characteristics of a highly sensitive test
Will pick up cases that show even slightest of evidence
Makes more false positives
When is a highly sensitive test preferred?
If diagnosis should not be missed but overdiagnosis is not harmful
Characteristics of highly specific test
Will pick up cases only if definitive evidence noted
Makes more false negatives
When is a highly specific test preferred?
If missing some cases is not bad but wrong labelling can be costly/harmful
Impact of higher cut off on specificity and sensitivity
Low sensitivity
High specificity
When is a receiver operator curve (ROC) useful?
Choosing between two diagnostic tests of different sensitivity and specificity
Choosing a cut off point for making a diagnosis
X and Y axis of ROC curve
X: (1-specificity)
Y: sensitivity
How do interpret a ROC curve
Elbow/knee of the curve is cut-off point with balance between sensitivity and specificity.
Among 2 curves, the one closer to the left upper corner is better screening test
What is ROC analysis used for?
To select optimal cut off point
To dichotomize a continuous scale
What does accuracy of ROC analysis depend on?
Quality of gold standard test
Impact of increase in sensitivity on specificity?
Decrease in specificity
How do we know if a test is not accurate in a ROC curve?
The closer the curve comes to the chance line
What is the area under the ROC curve?
Accuracy
What is Likelihood ratios for a positive test (LR+)?
Likelihood for testing positive rightly/wrong
What is Likelihood ratio for a negative test (LR-)?
Likelihood of testing negatively rightly/wrongly
Calculation for LR+
(A/A+C)/(B/B+D)
Calculation for LR-
(C/A+C)/(D/B+D)
Name another calculation for LR+
Sens/1-spec
Name another calculation for LR-
1-sens/spec
What is LR+ for most clinically available tests?
> 1
What value of LR is used to use a diagnostic tool
> 10
What LR value is used to avoid a diagnostic tool
<0.1
What does likelihood of a diagnosis depend on?
Prevalence or prior probability of disease before applying the test
What is the pretest probability
Prevalence in studied population
Calculation for pretest probability
(A+C)/(A+B+C+D)
How to calculate post-test probability
Convert pretest probability to prest odds
Convert pretest odds to post test odds
Convert posttest odds to post-test probability
Calculation to convert pretest probability to pretest odds
Prob/1-prob
(A+C)/(B+D)
i.e. diseased/not diseased
Calculation to convert pretest odds to post-test odds
LR x Pretest odds
Calculation to convert post-test odds to post-test probability
Post-test odds / 1+ posttest odds
What is the Bayesian nomogram to calculate post-test probability
Draw a straight line across pretest probability and likelihood ratio
Post-test probability s the value obtained on the other side of the nomogram
Who created the Bayesian Nomogram
Fagan
What does positive predictive value (PPV) tell us?
The chance of having the disease if test +ve
Calculation for PPV
A/A+B
What does PPV increase with
Prevalence
What does the negative predictive value (NPV) tell us?
Chance of not having the disease if test -ve
Calculation for NPV
D/C+D
What happens to NPV as prevalence increases?
NPV decreases
