Drs. Cass, Cantor, and Clover's first concern is the generalizability of the study results. Although the study site is university affiliated, it serves as the primary care site (not tertiary care) for a population of primarily local residents. We agree that the study population was atypical in that only men were included; however, no known gender differences exist in the clinical presentation or clinical course of sinusitis.

Their second concern is the high prevalence of sinusitis (38%) and possible misclassification by the criterion standard radiographs. The epidemiologic data Cass and colleagues use to suggest a lower prevalence of sinusitis is suspect because diagnoses were made without objective clinical criteria or criterion standard confirmation. Incidence data on complications of common colds should not be extrapolated to the prevalence of sinusitis in symptomatic patients whose antecedents include allergic rhinitis, upper respiratory tract infections, and anatomic defects. Nevertheless, we too were surprised by the high prevalence. Few quality data on sinusitis in general medical practice exist; however, studies in subspecialty populations [1, 2] and in British general practices [3] showed an even higher prevalence in symptomatic patients. We suggest that the 0.5% incidence rate has been misinterpreted and that the prevalence of sinusitis in patients meeting entry criteria for our study is indeed high.

Cass and colleagues' third concern is that roentgenogram misclassification may have been increased by including patients with chronic sinusitis, which by definition includes patients with symptoms lasting longer than 3 months [4]; these patients were excluded from our study. Compared with sinus aspiration and culture, radiographs do misclassify a small proportion of patients. However, repeat radiographs 3 to 6 weeks later might not decrease the misclassification rate because information about the relative time course for radiographic resolution of acute sinusitis is incomplete.

We agree wholeheartedly that independently validated results yield much greater confidence. However, given the sample size and number of clinical variables, split sample training and validation sets were not feasible. We elected not to use bootstrap or jackknife validation methods; instead we continue to collect data to validate our prospective findings.