Appendix 2 Comments on the Interpretation of Diagnostic Tests
The interpretation of tests used in clinical practice is important, but studies that define the accuracy of diagnostic tests can be difficult to understand. In chapter 23, the paper by Stein and colleagues1 on the diagnosis of pulmonary embolism is discussed briefly. Let us consider its results in relation to the use of multidetector computed tomographic angiography (CTA) in the diagnosis of pulmonary embolism. This diagnostic procedure was compared with a reference standard. The authors report (in the abstract) that the test was inconclusive in 51 of 824 patients, but in the remaining 773 it had a sensitivity of 83% and a specificity of 96%. They also report that in patients likely (on clinical grounds) to have a pulmonary embolus, the positive predictive value of the test was 96%. What does this mean?
Let us begin by considering just the 773 patients who had both a positive reference diagnosis and a technically successful CTA. More generally, assume that n patients were tested for pulmonary embolus, of whom some had the condition and some did not. The test will have been positive in some and negative in others (Table 1). These results will include true-positives (a) and true-negatives (d), but unless the test was perfect, there will have been some false-positives (b) and some false-negatives (c). These results can be represented in a table in which a + c patients have the disease and b + d do not, and in which a + b tested positive and c + d tested negative.
