Pediatric Informatics: Computer Applications in Child Health (Health ...

12 Diagnostic Decision Support 163
(where P() = “probability of,” | = “given,” F = finding, D = Disease, and −D =
absence of Disease)
(sensitivity of finding or test for the disease) � (prevalence of disease)
=
(sensitivity� prevalence) + (1 − specificity) �(1
− prevalence)
An illustration that demonstrates the importance of prevalence is the calculation of
avian influenza (AI, “bird flu”) risk when compared with the calculation of a very
common disease, the “common cold.” AI is rare, with 385 human cases reported
by June 2008 and 34 cases in 2008 thru July 2. 7 Using an estimate of the world
population of 6,707,380,479, 8 AI prevalence in humans is approximately 5.1 × 10 −9 .
If disease (D) = avian influenza and the finding (F) is “cough,” the probability of a
patient having AI, given that cough is present AND that 20% of patients without AI
have cough (P(F|−D) = 0.2; specificity (1−P(F|−D) = 80%), and that the sensitivity
of cough in patients with AI (P(F|D) ) is 90%, then:
The PPV of cough ⎫
0.9 � 5.1�10 ⎬ =
⎭
−9 −9
for Avian influenza (0.9�5.1�10 ) + 0.2 �(1 − 5.1�10 )
4.6�10 =
~ 0.2
= 2.3� 10
Therefore, cough alone is not predictive of the diagnosis of avian influenza.
This contrasts markedly with the example of a nonspecific upper respiratory infection
(URI). If the sensitivity of the presence of “cough” is 60% in patients with URI
(sensitivity = P(F|D) = 0.6), the prevalence of URI in the population is 10% (P(D)
= 0.1) and if 10% of patients without URI have cough (P(F|−D) = 0.1; specificity =
90%), then the PPV of cough alone for URI is 0.4 or 40%, given our assumptions.
This has face validity, with the other 60% of diseases presenting with cough alone
including allergic rhinitis, bronchitis, pneumonia, asthma, sinusitis, pertussis, and
others (including avian influenza!).
12.4 Decisions: Influences and Information Support
Diagnostic thinking is influenced by several factors:
Total and recent experience:
A rare disease seen recently may come more
easily to mind than a highly prevalent disease that has never been encountered
Heuristics (rules of thumb): Algorithms or decision rules
Parsimony (Ockham’s razor): The explanation of findings in as few diagnoses
as possible9 Prevalence:
Experienced pediatric clinicians teach that having two common
diagnoses is much more likely than having a single very rare disease10 −9
−9
−8