Determining and interpreting the accuracy of a test.

摘要

When a clinician asks "is this test accurate?" the real question of interest is "does this test give the right answer?" From a statistical standpoint, there is not a mathematical expression for accuracy; in evaluating the accuracy of a test, one must be familiar with the sensitivity and specificity of the test. A test is highly accurate if it is both highly sensitive and highly specific. From a practical standpoint, a test is accurate if it gives the "right" answer more often than not, which requires comparison with some gold standard for diagnosis. Most of the time, such a gold standard is not available, and the clinician must decide just how accurate the test must be in a given situation.For example, the sensitivity and specificity of a quantitative culture of a bronchoalveolar lavage (BAL) specimen in the diagnosis of pneumonia might be 75% and 68%, respectively, depending on the threshold set for the number of organisms present in the specimen. As empiric treatment for pneumonia is often begun simultaneously with obtaining BAL and cultures, then discontinued if the cultures are negative, this test is acceptably accurate. In testing for cancer, where the diagnosis must be certain, a sensitivity and specificity of 75% and 68%, respectively, would be unacceptable.