Sensitivity and Specificity Handout

Sensitivity and Specificity
A Caries Example
Sensitivity is a conditional probability; it Specificity is a conditional probability; it is

is the probability that a diagnostic test is the probability that a diagnostic test is
positive, given that the disease is truly negative, given that the disease is truly
present. absent.
The term “given that” reflects that sensitivity and specificity are conditional probabilities
The true presence or absence of disease is measured by the best possible test one can
come up with. This could be derived from autopsies, extracted teeth, or invasive,
expensive or harmful diagnostic tests. Diagnostic test results are then evaluated against
this gold standard. For instance, for caries the gold standard diagnosis could be
histology. Teeth are extracted, microradiographs taken, and the presence/absence of
caries determined. The gold standard diagnostic can be used to classify teeth or people in
two groups; those with or without disease and to calculate sensitivity and specificity.
Calculating sensitivity Calculating specificity
The sensitivity of a diagnostic test is the The specificity is calculated as the
proportion of correct positive diagnoses in proportion of correct negative diagnoses in
a diseased population. For instance, if 45 a non-diseased population. For instance, if
surfaces truly have caries and bitewing 60 surfaces truly have no caries and
radiographs identify 24 out of the 45 bitewing radiographs identify 58 out of the
lesions correctly, the sensitivity is 24/45 or 60 correctly, the specificity is 58/60 or 97%.
54%.
Why are sensitivity/specificity meaningless statistics for a clinician?
When a clinician performs a diagnostic test, s/he does not know whether a tooth or
individual has the disease or condition of interest. For instance, for caries, the available
information is typically bitewings. Clinicians do not extract teeth to see if the bitewing
information is accurate. The questions of interest for a clinician is:
1) What is the probability that a surface truly has caries, when the bitewing leads to a
diagnosis of caries (a.k.a. the positive predictive value)?
2) What is the probability that a surface truly is sound, when the bitewing leads to a
diagnosis of an absence of caries (a.k.a. the negative predictive value)?
.
1
Sensitivity and specificity
A Caries Example
Gold Standard (e.g., microradiography)

Caries No Caries
Diagnostic test (e.g., bitewing) Caries a (24) b (2)
No Caries c (21) d (58)
a+c (45) b+d (60)
Sensitivity: a/(a+c) or 24/45 or ~0.54
Specificity: d/(b+d) or 58/60 or ~0.97
• Data adapted from J Dent. 1990 Jun;18(3):130-6. (Mileman PA, van der
Weele LT.)
• For caries considered to be 'probably' in the dentin mean sensitivity was 54 per
cent (s.d. 14 per cent) and specificity was 97 per cent (s.d. 5 per cent).
False Positive and False Negative Error Rates

What is the probability that a surface truly has early dentinal caries if you make such a
diagnosis based on the bitewing information? This probability depends on something
you do not know, namely the probability of early dentinal caries in your patient
population. Intuitively, it may be clear that the higher this probability (e.g., caries level),
the more likely a positive diagnostic test is a true positive. For instance, if every surface
examined truly has caries, then every time a clinician classifies a tooth has having caries
from the bitewing, the diagnosis will be correct. Conversely, the lower the disease
probability, the more likely a positive diagnostic test (e.g., a caries lesion as diagnosed on
a bitewing) is a false positive.
Now, if you had PROBABILITY 101 in college or high school, calculating the
probability that disease is present, given that the test is positive is straightforward. For
those who did not have PROBABILITY 101 (or forgot that they had it) it is also
relatively straightforward (Phhhhewwww).
2
Calculating false positive and false negative error rates.
Step 1: Assume the probability of early dentinal caries in your Bellevue practice is 0.3%.
That means for every 10,000 surfaces you examine on your bitewings (250-500 patients),
30 will truly have caries. Thus, if you could extract the teeth and do the histology on the
surfaces, you would come to the conclusion that 30 surfaces truly do have caries.
microsection
Caries No Caries
bitewing Caries a b
No Caries c d
a+c= 30 b+d = 9970 10,000
Step 2: Complete the 2x2 table sensitivity (0.54) and specificity (0.97)
microsection
Caries No Caries
bitewing Caries a=> 30*0.54=16 B
No Caries C D=> 9970*0.97 =9671
a+c (30) b+d (9970) 10,000
Step 3 Fill in “c” and “b” by simple subtraction

microsection
Caries No Caries
bitewing Caries 16 b=>9970-9671=299
No Caries c=> 30-16=14 9671
a+c (30) b+d (9970) 10,000
Step 4: Calculate the number of positive and a negative test results with bitewings by
adding the row entries.
microsection
Caries No Caries
bitewing Caries 16 299 16+299=315
No Caries 14 9671 1+967=9685
a+c (30) b+d (9970) 10,000
In this example, with caries probability of 30/10000, you would conclude, based on the
bitewings, that there are 315 dentinal cavities. However, only 16 of these 315 cavities are
truly cavities as diagnosed by histology. The false positive error rate is 299/315 or
95%. The positive predictive value is 1 – 0.95 or 0.05. For every 100 diagnoses of
dentinal caries, 95 would be histologically sound surfaces and 5 on early dentinal caries
lesions.
Similarly, 9685 times the bitewing would lead to the conclusion of no caries. However,
for 14 surfaces this would be a false negative and the false negative error rate is
14/9685 or 0.001. The negative predictive value is 1 – 0.001 or 0.999.
3
False positive and false negative error rates
Positive and Negative Predictive Value
microsection
Caries No Caries
bitewing Caries 16 299 16+299=315
No Caries 14 9671 14+9671=9685
a+c (30) b+d (9970) 10000
The following statistics can be derived from this 2x2 table
a) Prevalence of disease: 30/10,000 = 0.3%

b) Sensitivity; 16/30 or 0.53
c) Specificity; 9671/9970 or 0.97
d) False positive error rate; 299/315= 0.95
e) Positive Predictive Value; 16/315=0.05
f) False negative error rate; 14/9685=0.001
g) Negative predictive value; 9671/9685=0.99
How can false positive and false negative error rates be reduced?
The more subtle the disease or condition, the higher the error rates. For instance, the
more advanced the caries lesion, the easier it is to diagnose. One does not need a
bitewing radiograph to determine whether there is a D3 lesion on the approximal surface
of a tooth. What is more challenging is to diagnose the presence/absence of early enamel
lesion or early dentinal lesions.

Sensitivity and Specificity Handout

Uploaded by

Copyright:

Available Formats

Sensitivity and Specificity Handout

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sensitivity and Specificity Handout

Uploaded by

Copyright:

Available Formats

Sensitivity and Specificity

Sensitivity is a conditional probability; it Specificity is a conditional probability; it is

Gold Standard (e.g., microradiography)

False Positive and False Negative Error Rates

Step 3 Fill in “c” and “b” by simple subtraction

The following statistics can be derived from this 2x2 table

a) Prevalence of disease: 30/10,000 = 0.3%

You might also like