We appreciate the insightful letters from the perspective of a practicing clinician (Dr. Gelfand) and clinical epidemiologists (Dr. Heller and colleagues). Regarding Dr. Gelfand's questions, our study included consecutive patients who were being evaluated for a possible vasculitis by rheumatologists. In our patients, a spectrum of conditions that make up the differential diagnosis of vasculitis (for example, infections and cancer) were ultimately diagnosed. Overall, the prevalence of vasculitis in our study sample was 26%, but the ACR criteria performed poorly for diagnosis. Their performance as diagnostic criteria would probably be even worse in clinical situations in which a diagnosis of vasculitis is not likely.

Heller and colleagues suggest that clinicians use Bayesian approaches to account for pretest probabilities of diagnosis when evaluating patients with a suspected vasculitis. This is an intriguing idea. A similar approach was used by Khan and colleagues to clarify the use of HLA-B27 tests in the diagnosis of ankylosing spondylitis (1). The challenge will be to see how well the Bayesian models perform prospectively in suspected vasculitis and whether clinicians will use such approaches in actual practice.

We agree that inappropriate application of 1990 ACR classification criteria for vasculitis and indiscriminate antineutrophil cytoplasmic antibody testing done to diagnose vasculitis are important clinical problems (2). Given the wide dissemination of the ACR classification criteria in clinical texts and research articles, the fact that these criteria were developed to describe patient populations in research studies (3) is often missed by busy clinicians and housestaff who are looking for authoritative guidelines for diagnosis. This problem is seen in application of all classification criteria in rheumatology. For example, the ACR criteria for rheumatoid arthritis identify persons who may have either a self-limited process or a severely progressive disease, and this has contributed to confusion regarding outcomes in rheumatoid arthritis (4). We encourage the ACR and other organizations that propose "classification" criteria or diagnostic guidelines to consider these clinical issues when developing and disseminating such criteria.