We continue our examination of diagnostic tests that are used to assign
individuals to one of two mutually exclusive categories, those having or
not having a particular disorder. Our last article showed how the
information yielded by a test depends on the choice of cutoff and on the
patient's pre-test probability of having the disorder. In this article, we
demonstrate a method of optimizing a test's information yield that allows
us to determine which of several tests gives the most information. Our
technique allows us to compare a test's performance with that of a perfect
diagnostic test.Abstract Teaser