Dr. Inciardi believes that our report [1] left several issues unanswered. We take this as a compliment, given that the opposite often seems true when few data are tortured until they yield their secrets. The extensive data assembled for this study were far greater than we could hope to include in a single report. In choosing our priorities, we omitted logistic regression analysis because variables such as age, underlying disease, and depth and duration of granulocytopenia were well balanced, as can be seen in Table 1 of our original report, suggesting that any possible effect would be slight. We concur with Dr. Inciardi, however, that such an analysis might provide a more useful assessment for patient-related factors. One member of our group has already completed an analysis on prognostic factors in the development of superinfection [2], and we plan to extend this analysis to outcome in general.

Regarding Dr. Inciardi's second point, we only slightly exceeded the 770 episodes necessary for sufficient power to draw our main conclusion; therefore, we clearly cannot reduce the type II errors surrounding other nonsignificant differences to an acceptable level because these concerned subgroups. We do agree, however, that presenting the odds ratio on an arithmetic scale magnifies in a direction that favors the test regimen. Perhaps a log scale would more appropriately reflect the fact that these two regimens were expected to be equivalent (Figure 1).

View larger version (16K): [in this window] [in a new window]   Figure 1. Comparison of satisfactory responses by region. Odds ratios () are shown with their 95% CIs (—) after conversion to the natural logarithm, indicating that neither regimen was favored.

Rightly, statistical analysis of clinical studies receives more attention than it did a decade ago. Then, life was simple (or we were naive), and a regimen either cured or failed to cure. We are still ascending the learning curve, but the goal posts are being continually moved. For example, we are now told to include only the first treatment episode for each patient, despite substantial arguments to the contrary. We also accept the need to recruit many patients to ensure sufficient predictive power. This necessarily enjoins us to mount a multicenter trial, thereby introducing perhaps the largest single confounding variable of all—major differences in perception and practice among participating centers. Yet, we doubt whether all these endeavors are sufficiently convincing to the skeptical clinician who is left to wonder whether the claims made in reports such as ours will be fulfilled in clinical practice. Rather, we believe that the points we addressed in our discussion are more likely than any statistical analysis, no matter how sophisticated, to temper the clinician's judgment.