Although new diagnostic tests frequently are introduced to hospital laboratories with enthusiasm, many suffer an early demise or gradually fade from use. Two important explanations have been offered for unwarranted optimism about a diagnostic test.1,2 First, the spectrum of patients in whom the test is studied is crucial to obtaining accurate measures of test performance. The ability of the test to make correct classifications must be challenged by a series of patients suspected of having the disease-not simply a group of "normal" and "diseased" individuals. Second, the interpretation of the test must be independent of the establishment of the ultimate diagnosis; the test result itself must not have been used to help establish the true clinical classification. This independent confirmation has been called a "gold standard" or "benchmark" for diagnosis.
Even if one meticulously selects the proper population and uses an appropriate diagnostic standard, important problems remain. Any test