Problems of spectrum and bias in evaluating the efficacy of diagnostic tests.

Abstract

To determine why many diagnostic tests have proved to be valueless after optimistic introduction into medical practice, we reviewed a series of investigations and identified two major problems that can cause erroneous statistical results for the "sensitivity" and "specificity" indexes of diagnostic efficacy. Unless an appropriately broad spectrum is chosen for the diseased and nondiseased patients who comprise the study population, the diagnostic test may receive falsely high values for its "rule-in" and "rule-out" performances. Unless the interpretation of the test and the establishment of the true diagnosis are done independently, bias may falsely elevate the test's efficacy. Avoidance of these problems might have prevented the early optimism and subsequent disillusionment with the diagnostic value of two selected examples: the carcinoembryonic antigen and nitro-blue tetrazolium tests.