Abstract
Diagnostic tests provide information about the presence or absence of disease. However,
even after application of diagnostic tests, significant uncertainty about the state
of the patient often remains. This uncertainty can be quantified through the use of
information theory. The “information” contained in diagnostic tests published in the
medical literature of the years 1982 through 1986 was evaluated using Shannon information
functions. Information content, averaged over all prior probabilities of disease,
ranged from 0.002 bits to 0.720 bits of information; the tests therefore provided
from 0.3% to 100% of the information needed for diagnostic certainty. Median average
information was 0.395 bits, corresponding to only 55% of the information required
for diagnostic certainty. Reclassifying test results into multiple mutually exclusive
outcome categories allowed extraction of a median of 14% and a maximum of 109% more
average information than that obtained using a dichotomous positive/negative classification.
We conclude that the “information” provided by many of the tests published in the
medical literature is insufficient to overcome diagnostic uncertainty. Information
theory can quantify the uncertainty associated with diagnostic testing and suggest
strategies for reducing this uncertainty.
Key-Words
Information Theory - Diagnostic Tests
