Assessment of the Diagnostic Accuracy of Laboratory Tests Using Receiver Operating Characteristic Curves; Approved Guideline
|Publication Date:||1 November 2011|
This guideline outlines the steps and principles of prospectively planned and retrospective studies to evaluate the intrinsic diagnostic accuracy of a clinical laboratory test, defined as its fundamental ability to discriminate correctly among alternative states of health. It is not intended to help determine how best to use a diagnostic test in clinical practice, but instead to determine how accurate a laboratory test is in terms of diagnostic sensitivity and specificity.
Receiver operating characteristic (ROC) curve methodology arose in response to needs in electronic signal detection and problems with radar in the early 1950s.2 It is derived from conditional probabilities, as originally formulated by Bayes.3 This guideline aims to define ROC curves and to explain how to design, construct, interpret, and apply the information from ROC studies to evaluate diagnostic tests. For simplicity, only continuous scales, such as those typical for in vitro diagnostic tests, are discussed. The clinical condition that the test is intended to detect must be verifiable through some means other than the test under investigation. In other words, there must be an independent clinical reference standard against which one can compare the test. By selecting cutoffs between positive and negative diagnoses along the continuous scale of the test, the diagnostic outcomes for these decision levels are compared to the true clinical condition, which, in turn, generates the ROC curve.
This guideline will be of value to a wide variety of possible users, including:
• Investigators who are developing new tests for specific applications
• Manufacturers of reagents and devices for performing tests who are interested in assessing or validating test performance in terms of diagnostic accuracy
• Regulatory agencies interested in establishing requirements for claims related to diagnostic accuracy
• Clinical laboratorians who are reviewing data or the literature, and/or generating their own data, to make decisions about which tests to employ in their laboratories
• Health care or scientific workers interested in critical evaluation of data being presented on clinical test performance