|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Originally published as JCO Early Release 10.1200/JCO.2009.23.9806 on August 31 2009 © 2009 American Society of Clinical Oncology.
Reply to J. MehtaDepartments of Medical Oncology and Biostatistics, Dana-Farber Cancer Institute, Boston MA In his careful review of our article,1 Mehta2 mentions several other variables that could affect the outcome of reduced intensity conditioning allogeneic hematopoietic stem-cell transplantation (RIC HSCT). One of those is the remission status of the patients at the time of transplantation. When the multivariable analyses for overall survival (OS) and progression-free survival (PFS) among the patients treated with RIC HSCT were repeated with separate variables for patients in complete remission (CR) and partial remission (PR) at HSCT, the results were unchanged (hazard ratios [HRs] for sirolimus-containing graft-versus-host disease prophylaxis regimen are 0.5, P = .022 for OS, and 0.4, P = .003 for PFS). Therefore, the different proportion of patients in CR versus PR in the two groups did not affect the results. Another factor of potential importance is pre-HSCT lactate dehydrogenase (LDH). We collected LDH from all the RIC patients at the time of transplantation admission. Median LDH was 168 U/I (range, 118 to 290 U/I) in the no-sirolimus group, and 170 U/I (range, 105 to 399 U/I) in the sirolimus group (P = .7). In the no-sirolimus RIC group, 22% of patients had a pre-HSCT LDH above the upper limit of normal, compared with 12% in the sirolimus group (P = .2). When LDH was added to the Cox models, sirolimus retained its significance (HR = 0.5, P = .042 for OS, and HR = 0.5, P = .010 for PFS). CD34 dose was not a significant factor in our prior analysis of RIC HSCT outcome for lymphoma, and comorbidity indices were not available in this data set, thus those variables were not included in the models. Therefore, the inclusion in the analyses of the potentially most important variables mentioned by Mehta had no effect on our results or our conclusions. Regarding the donor cytomegalovirus (CMV) serostatus in the RIC HSCT group, we are indebted to Mehta2 for catching an error in the table1 (which we will report as an erratum). There were in fact eight (out of 23) CMV seropositive donors in the no-sirolimus group, and 38 (out of 101, with two patients of unknown donor serostatus) in the sirolimus group. The two-sided P value for the comparison is 1.0. Therefore, donor CMV serostatus was not different between the two groups; furthermore, as shown in Tables 2 and 3 of our article,1 CMV risk had no significant bearing on transplantation outcome. In our Figures 3A and 3D, the correct P values are .7 (not 0.07) and .6 (not 0.06), respectively.1 Those were typographical errors that occurred during the typesetting process (ie, the manuscript when submitted and reviewed did not contain those errors), and the P value of .6 for Figure 3D is indeed correctly given in the text (Results section). In Table 1, for any given category, a P value for the comparison of the two patient groups can be obtained.1 For categoric variables, this is usually done with Fisher's exact test, which is the method we used with the data in Table 1. Thus, for example, it is possible to compare the proportion of patients in CR at the time of RIC HSCT. In this case, in the sirolimus group, 24% (25 of 103) of the patients were in CR at the time of transplantation, versus 9% (2 of 23) in the no-sirolimus group. By Fisher's exact test, the two-sided P value for this comparison is .16 (the value of .08 used by Mehta2 is the one-sided P value, but the one-sided test is inappropriate in this comparison). Similarly, it is possible to compare the proportion of patients who are chemosensitive at the time of HSCT (ie, patients in CR or PR at transplantation). In our case, the proportion of CR plus PR is almost identical between no-sirolimus patients (78%) and sirolimus patients (76%; two-sided P value = 1.0). Because we found no difference in OS or PFS between patients in CR and patients in PR in our study, we collapsed these two categories in the multivariate model. As the purpose of Table 1 is to examine whether the patient groups are comparable with respect to each baseline characteristic, the significance threshold was kept at 0.05 and not adjusted for multiple comparisons, as is standard in this type of analysis. Mehta2 correctly points out that in Tables 2 and 3, the reference group for "Prior ASCT" was indeed patients who had not had a prior ASCT, and the reference group for "more than 2 lines" was the group of patients who had received up to two lines of prior therapy.1 Those reference groups were not explicitly listed in the tables, and we hope that this did not lead to confusion.
Regarding the questions Mehta2 brings up regarding HRs and P values, we would submit the following. Parameters of proportional hazards models are estimated from partial likelihood functions using the Newton-Raphson algorithm. For each covariate of interest, the analysis yields an estimate of the parameter (βi), which when exponentiated is the estimate of HR. The P value associated with the Wald Mehta2 also questions the estimation of HRs that differ for OS and PFS, such as "age greater than 50." A P value of .3 or .5 (the values for age older than 50 years associated with the OS and PFS HRs, respectively) is nonsignificant, that is, not significantly different from 1 within a random fluctuation. In this case, being greater than 1 or less than 1 is not relevant. Age older than 50 years should not be considered to significantly affect OS or PFS one way or the other. To interpret the results, as Metha invites us to do, to mean that age older than 50 years is associated with inferior OS, but superior PFS imputes statistical significance to the HRs that they do not have. We have repeated the analyses after receiving Mehta's comments, with identical results. Finally, Mehta2 questions the P values of the log-rank tests for comparison of survival between various groups. For the 154 patients without lymphoma who underwent RIC HSCT, the P value is indeed .06. The reason is that the log-rank test takes into account the difference in survivals at all time points. Even though the 3-year survival is similar between the two groups, the OS curve for the sirolimus patients is higher than that of the nonsirolimus patients at early time points, explaining the relatively low (but still nonsignificant) P value. We did not show the curves because they were not central to our manuscript, but have provided them below (see Figs 1 and 2).
In conclusion, we thank Mehta for providing these comments. We are grateful for the chance to correct the mistake and the typographical error that he caught. Other than that, we welcomed the opportunity to again repeat our analyses to ensure their accuracy, and stand by our methods, our results, and our conclusions. AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST The author(s) indicated no potential conflicts of interest. ACKNOWLEDGMENT This work was funded in part by Grant P01 HL070149 from the National Heart, Lung and Blood Institute. P.A. is a recipient of a career development award from the Leukemia and Lymphoma Society. REFERENCES
1. Armand P, Gannamaneni S, Kim HT, et al: Improved survival in lymphoma patients receiving sirolimus for graft-versus-host disease prophylaxis after allogeneic hematopoietic stem-cell transplantation with reduced-intensity conditioning. J Clin Oncol 26:5767–5774, 2008. 2. Mehta J: Sirolimus-containing graft-versus-host disease prophylaxis in lymphoma patients. J Clin Oncol 27:e138; 2009.
Related Article
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2009 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
