Transcript No Slide Title - Michigan State University

Diagnosis: EBM Approach
Michael Brown MD
Grand Rapids MERC/
Michigan State
University
Scenario



1 day colicky pain with nausea
diffuse to RLQ
mild tenderness, T 37
Step 1: Clinical Question

In the patient presenting to the ED with
suspected appendicitis, what is the accuracy of
helical CT ?
Step 2: Search

MeSH Browser
–
–
–

appendicitis AND
computerized tomography AND
sensitivity and specificity
Clinical Query
–
diagnosis
Step 3: Critical Appraisal


Internal Validity
Results (focus today)
–

even if critical appraisal not your bag
External Validity
Evidenced-based Medicine



stresses methodology
de-emphasizes statistics
simplify: NNT, LR
Likelihood Ratio:
How to use
How to calculate
Examples


Appendicitis
Pulmonary embolism
–
–
JAMA series
current slant
Threshold Approach to Clinical
Decision Making

Treatment threshold
–

if above, start therapy
Test threshold
–
if below, no further testing
Pauker NEJM 1980
Diagnostic Testing

Treatment threshold for PE?
–

Test threshold for PE?
–

If above: heparin
If below: discharge home
If between?
–
Further testing
Test/Treatment Threshold
prior
probability
(prevalence)
LR
post-test prob
(predictive
value)
Test/Treatment Threshold
Pretest Probability

experience in your setting
–

prevalence of condition in literature
–

patient population
Oxford web site
scoring systems
Estimate Pretest Probability for PE




history
risk factors
physical exam
initial screening tests
Estimate Pretest Probability for PE

Not exact science
–
–


usually a range 40-60%
low, intermediate, high
done daily in clinical practice
clinical prediction rules
–
physicians estimate very close
Wicki 2001
Bayesian Analysis?



Thomas Bayes 1702-1761
English clergyman
Doctrine of Chances
Bayesian Analysis

pretest probability
–


prevalence
LR for diagnostic test result
post-test probability
–
predictive value
Interpretation



convert pretest prob to odds
odds x LR = post-test odds
convert odds back to prob
Interpretation



convert pretest prob to odds
odds x LR = post-test odds
convert odds back to prob
Fagan
Nomogram
Effect on pretest probability:



>10 or <0.1
5-10 and 0.1-0.2
approach 1
large changes
moderate
no effect
Advantages of LR:




combines sensitivity and specificity
interpret test result on individual patient
multiple cut-offs
sequential testing
Shortcut: LR for + test =
sensitivity
1 - specificity
Calculate: LR=
prob (test result) with disease
prob (test result) without disease
2 x 2 Table
Target Disorder
Present
Diagnostic
Test Result
Totals
Absent
Positive
a b
a+b
Negative
c d
c+d
Totals
a+c b+d
a+b+c+d
CT and Appendicitis
Target Disorder
Diagnostic
Test Result
Totals
Positive
Present
29
Absent
4
33
Negative
1
66
67
Totals
30
70
100
Funaki et al
CT and Appendicitis

probability of + CT with appendicitis
29/30 = .97
CT and Appendicitis
Target Disorder
Diagnostic
Test Result
Totals
Positive
Present
29
Absent
4
33
Negative
1
66
67
Totals
30
70
100
CT and Appendicitis



Likelihood of + CT with appendicitis
29/30 = .97
Likelihood of +CT without appendicitis
4/70 = .057
LR for + CT = 17
Effect on pretest probability:



>10 or <0.1
5-10 and 0.1-0.2
approach 1
large changes
moderate
no effect
Scenario

1 day colicky pain with nausea
diffuse to RLQ
mild tenderness, T 37

Pretest probability 30%


–
range 20 - 40%
Helical CT
Helical CT
Helical CT
Effect on pretest probability:



>10 or <0.1
5-10 and 0.1-0.2
approach 1
large changes
moderate
no effect
Advantages of LR:



combines sensitivity and specificity
interpret test result on individual patient
multiple cut-offs
–

don’t have to lump!
sequential testing
Multiple cut-offs
appendicitis
No disease
a
b
x
y
CT positive
CT equivocal
CT negative
Total
Total
Calculate: LR=
prob equivocal CT with disease
prob equivocal CT without disease
Multiple cut-offs
appendicitis
No disease
a
b
x
y
CT positive
CT equivocal
CT negative
Total
Total
Multiple cut-offs: V/Q scan




normal
low prob
intermediate prob
high prob
LR 0.1
LR 0.4
LR 1
LR 18
JAMA series
Sequential Testing



post-test probability 1st test
new pretest probability for 2nd test
assume independence
Helical CT : Diagnosis of PE

CT +
LR 8

CT -
LR .2
Rathbun, 2000
ELISA D-dimer: Diagnosis of PE

> 500
LR 2

<500
LR .1
Brown, Bermingham 2001
Helical CT
D-dimer
Test/Treatment Threshold
Questions?
Scenario

1 day colicky pain with nausea
diffuse to RLQ
mild tenderness, T 37

Pretest probability 30%


Discussion: CT and appy



Internal Validity
Results
External Validity
CT and Appendicitis

Flaws?
–
–
–
Minor
Major
Fatal
CT and Appendicitis (Rao)
Target Disorder
Diagnostic
Test Result
Totals
Positive
Present
52
Absent
1
53
Negative
1
46
47
Totals
53
47
100
+LR = .98/.02 = 46
-LR = .019/.98= .02
Rao et al
CT and Appendicitis (Funaki)
Target Disorder
Diagnostic
Test Result
Totals
Positive
Present
29
Absent
4
33
Negative
1
66
67
Totals
30
70
100
Funaki et al
CT and Appendicitis



Likelihood of + CT with appendicitis
29/30 = .97
Likelihood of +CT without appendicitis
4/70 = .057
LR for + CT = 17
Funaki: CT and Appendicitis


LR for a positive CT 17
LR for a negative CT 0.03
CT and Appendicitis



Likelihood of + CT with appendicitis
29/30 = .97
Likelihood of +CT without appendicitis
4/70 = .057
LR for + CT = 17
Helical CT
Sources of LR:

PE: + CT LR = 8
- CT LR = .2
–

Pharyngitis: neg rapid strep
- LR = 0.2
–

meta-analysis: Rathbun et al
hosptial data
Alcoholism: CAGE >3
LR = 250
–
web
Combinations (LRxLR)

D-dimer <500, CT -

LR .02