Why we need better test evaluations.

Abstract

A laboratory test is clinically useful only if it successfully answers a question of consequence to patient management. Unfortunately, the results and conclusions of many published test evaluations are misleading or of uncertain validity because common-sense principles of study design are overlooked. This is illustrated by examples from recent literature. We suggest that tests should be evaluated with prospective studies of patients representative of the population for which the test will ultimately be used. The clinical question to be addressed by the test should be clearly stated, and then answered for each patient by means independent of the test being evaluated. When comparing tests with each other, decision levels should be chosen to give either the same sensitivity or specificity for each. The use of soundly designed protocols for the clinical evaluation of tests provides the information needed to select the most effective tests.