We appreciate Dr. Prybylski's insights. He interprets the findings as we do but arrives at them by way of a different path. We agree that it is striking that 14 self-reported symptoms were more prevalent in patients with previous Lyme disease than in those without. Among the 26 symptoms about which patients were queried, we chose these 14 symptoms to highlight in tabular form because they were more prevalent. We similarly noted the striking deficits in health-related quality of life, as measured by the SF-36, among patients with Lyme disease and believe this to be one of the study's main findings. It is true that the prevalence of knee pain was higher when the joint was taken through its range of motion. In an extensive neurologic and musculoskeletal examination and neurocognitive test battery, the groups overall had strikingly similar findings.

We believe that the lack of significant differences between groups on the psychometric tests is a true finding rather than a result of the lack of sensitivity of the measures. In previous studies, these tests have been sensitive to changes in Lyme encephalopathy. The California Verbal Learning Test and the Trailmaking Tests, in particular, can detect the subtle deficits in attention and learning that occur when Lyme disease affects the brain (1). To maximize the chance of detecting differences between groups, we selected population controls and adjusted our analysis for age and sex. In so doing, small differences between groups would be more evident, especially when the groups have "higher than normal" cognitive performance.

The univariate analyses of the risk factors for incomplete recovery were exploratory but important enough to include in the manuscript. Small numbers called for conservative inference. Much work must be done to understand the relative contribution of treatment delay, type of antibiotic therapy, and initial Lyme disease symptoms in relation to long-term recovery. This would be work for future study.

With regard to the use of stepwise logistic regression, we severely limited the stock of predictors in stepwise models and did not have collinearity. We applied a Bonferroni correction, however, because we tested many possible relationships. By doing so, we created the impression for readers that the differences were small and possibly spurious, but we retained the idea that case-patients had many more conditions. Selvin's viewpoint about stepwise procedures is useful but by no means universal (2). Mosteller and Tukey (3) advocate stepwise procedures when used with restraint. We did so by applying a Bonferroni correction.