Test Characteristics. How good is that test?

Abstract

Clinical information is used to develop differential diagnoses and narrow the possibilities. The history and physical examination motivate our initial diagnostic impression. Clinical laboratory results and further diagnostic maneuvers are used to confirm or refute our initial impressions. These tests and procedures are not perfect; they sometimes yield false results, which taken alone, might lead to misguided therapeutic interventions. Concepts of sensitivity, specificity, and predictive value provide the tools needed to integrate new information with initial diagnostic impressions. The concept of overall accuracy provides a quantitative estimate of how well a particular maneuver will perform. ROC curve analysis provides yet another quantitative and visual assessment of a test's performance. Additionally, the statistical tools described by Hanley and McNeil provide objective, quantitative means for comparing the various tests from which the clinician must select. Electronic spreadsheets and dedicated computer software provide relatively user friendly methods for engaging these techniques. Although individual clinicians might not choose to perform such analyses themselves, the discussion in this article should help the busy physician answer the question, "How good is that test?"