The letters about our article [1] suggest that we provided sufficient information for readers to critically review the results.

Drs. Lederman and van der Horst calculated that patients had "a treatment failure rate that was 71% higher if they received atovaquone." Using an odds ratio (OR), patients assigned to atovaquone had almost double the chance of not responding to therapy, although it was not statistically significant (OR, 1.9; 95% CI, 0.78 to 4.92). This failure rate for atovaquone is not trivial. Likewise, the dose-limiting toxicity associated with pentamidine is not trivial. Side effects of intravenous pentamidine may cause substantial morbidity and mortality [2-5]. If a therapy is intolerable, it cannot be effective. Patients assigned to pentamidine had 15 times the chance of having a dose-limiting adverse event (OR, 15.1; CI, 3.3 to 68.9).

In a study with two end points, it is unknown and unknowable whether patients with one end point might have eventually reached the other end point. For instance, how many of the 19 patients (36%) withdrawn from the study because of adverse events associated with pentamidine might have failed to respond to therapy if the adverse event had not occurred? Because patients were potentially at risk for both end points (but could be counted in only one end-point category), the major comparison in this study involved "successful therapy."

"Successful therapy" was completion of treatment without interruption by either a failure to respond or a dose-limiting toxicity. The equivalency null hypothesis cannot be "proved"; however, we were unable to reject it because of the low power. The rates of successful therapy were not statistically different because pentamidine produced more dose-limiting toxicity and atovaquone had a higher rate of treatment failure. We discussed these points (in the Methods, Results, and Discussion sections and in Tables 4 and 5), and we regret that some readers found this terminology misleading.

Drs. Stoeckle and Tennenberg cite the disparity in deaths at the 4-week follow-up (13% compared with 8%; P = 0.53). Nine patients in each group (16% compared with 17%; P = 1.00) had died by 8 weeks. We cannot conclude anything from the 4-week disparity or the 8-week parity in total deaths.

As Drs. Stoeckle and Tennenberg noted and as we described in the Discussion section, several alternatives are available for the treatment of P. carinii pneumonia. Atovaquone is one of these alternatives, and we hope that our article provides clinicians with useful information about atovaquone efficacy and toxicity.