We agree that the leukocyte esterase test (like any other diagnostic test) should be judged on its ability to provide additional information for evaluating the total clinical picture. Our study's entry criterion was performance on a urinalysis specifically to diagnose or exclude UTI, based on the patient's history and physical examination, the context of the test's use in clinical practice.

We also agree with White and Kunin's description of the potential clinical value of the leukocyte esterase test. We disagree, however, that it is inappropriate to use a clinically diverse population to assess a test with predictive value for a range of conditions.

In fact, diagnostic tests do not detect disease, but rather a marker that is imperfectly associated with disease (and often with other diseases). This distinction is the basis of spectrum bias. A patient's clinical status will affect diagnostic performance precisely because diagnostic tests are imperfect markers for disease. This fundamental issue should be recognized, and, as White and Kunin correctly note, total clinical information should be considered so that diagnostic tests can be used effectively.

It is important to distinguish between the specific information that a diagnostic test provides and how this information is used clinically. The leukocyte esterase test may be very good for detecting pyuria and clinically useful in confirming the presence of an inflammatory process in the genitourinary tract, but because pyuria is imperfectly associated with UTI, the test may be of more limited value for diagnosing UTI (a use for which it has been widely advocated). Our study further showed that the performance of the test for this purpose will vary with the patient's clinical presentation.

Dr. Boyko raises the universal and important issue of evaluation of a diagnostic test with an imperfect "gold standard." He notes that the criterion of 10⁵ CFU/mL or more to define a positive urine culture is imperfect and suggests that the results observed may be due to this artifact of reference-test misclassification. Although the potential biases he notes are correct, spectrum bias persists irrespective of the cutoff selected for a positive reference test. For example, with a culture threshold of 10⁴ CFU/mL, the prevalence of UTI increases (predictably) from 19.5% to 28.8%, and the sensitivity of the test in all patients falls from 0.83 (95% CI, 0.73 to 0.91) to 0.73 (CI, 0.63 to 0.82). In patients with a high prior probability of UTI, however, the sensitivity of the test was 0.92 (CI, 0.82 to 0.97), and in patients with a low prior probability, it was 0.40 (CI, 0.24 to 0.58).

With the use of any diagnostic test, a subjective decision must be made about where to demarcate "disease" from "nondisease" when a reference test can take on a range of values. Although our paper focused on the test cutoff point most familiar to clinicians, the phenomenon of spectrum bias persists no matter where the boundary is drawn.