We agree with Spiera and colleagues that the larger number of deaths in the cyclophosphamide-containing groups in our study is intriguing and that a survival analysis is the optimal statistical method to determine whether there is a true difference in mortality. However, we believe that combining the mortality data from both cyclophosphamide-containing groups and comparing them with the methylprednisolone group is flawed because this method assumes that any increased mortality in the combination group is attributable to cyclophosphamide. Comparing the methylprednisolone-only group with the cyclophosphamide-only group in a Breslow-Gehan-Wilcoxon survival analysis showed no significant difference in risk for death. We agree that borderline P values should be interpreted cautiously with such a small number of patients because small differences from the expected number of events (a few more or less) may cause large changes in the P value. To distinguish chance statistical oddities from biologically plausible trends, these results should be analyzed in the context of previous experience. To determine whether there was a biologically plausible connection to cyclophosphamide treatment, we analyzed the individual causes of deaths. Only one of five deaths in the cyclophosphamide group could reasonably be attributed to immunosuppression (Pneumocystis carinii pneumonia). One patient died of a clearly unrelated cause: bleeding after a kidney biopsy at another center. Removing this patient from the survival analysis changed the P value to 0.13. We have not seen an increased risk for death in patients treated with cyclophosphamide in previous studies, either at the end of the study or during long-term follow-up (1, 2). Spiera and colleagues also point out that in an intention-to-treat analysis, the rate of end-stage renal disease did not differ among the groups. As discussed in our paper, this apparent equality can be explained by the design of the original study and the practice at our institution to use cyclophosphamide in patients in whom pulse methylprednisolone has failed.

Our major finding was that the addition of methylprednisolone pulses to pulse cyclophosphamide leads to better long-term responses in moderate to severe lupus nephritis without conferring additional risk for adverse events. Dr. Spiera and colleagues conclude that pulse methylprednisolone alone should be the preferable option to treat mild lupus nephritis. However, this is not an appropriate extrapolation from our study because neither the original protocol nor the follow-up was designed to address this question.