|
Correction
to
Philip et al.,
|
| |||||||||||||||||||||||||||||||||
First, the authors state that the base-line characteristics of the two study groups were similar. According to their Table 1, however, there were consistently more patients with adverse prognostic factors in the conventional-treatment group than in the transplantation group: 30 percent versus 20 percent with high-grade histologic features, 17 percent versus 9 percent with a second relapse, 24 percent versus 16 percent with a tumor over 5 cm at the main location, and 15 percent versus 4 percent with other extranodal disease. The nonsignificant P values noted by the authors are not reassuring, since we are concerned not with a chance occurrence of these differences but rather with their effect on the outcomes measured. This effect is best determined with Cox's proportional-hazards model,2 and the authors should report the results of such an analysis.
Second, the authors provide no information on the distributions of age and disease stages in the two study groups. Both age and the stage of disease could be expected to have an effect on the outcome in this population.
Third, the authors do not describe the measures used to ensure concealment of the group assignments. Poor concealment has been shown to yield overestimates of the treatment effect.3 This study was stopped early because of low accrual after the first two and a half years of enrollment, suggesting that there was widespread knowledge of the preliminary results among the participating investigators, which may have increased the temptation to subvert randomization.4
Finally, the most useful information on the treatment effect is the difference in overall survival between the two groups, with some indication of the preciseness of that measure. The best summary statistic for this purpose is the 95 percent confidence interval for the difference in survival between the groups. The authors should provide this information.
Carl D. Atkins, M.D.
242 Merrick Rd.
Rockville Centre, NY 11570
References
Although aggressive non-Hodgkin's lymphoma is known to be quite sensitive to radiation, when only radiotherapy is used, doses over 40 Gy (at around 2 Gy per fraction) are required to obtain local-control rates that exceed 90 percent.1 Unfortunately, the optimal total dose, fractionation schedule, and target volume of radiation in combined-modality treatment are unknown at present. Moreover, in patients with recurrent disease, the response to radiation may differ from the response in patients with primary disease, because of altered cellular radiosensitivity and the rate of proliferation. Probably because of the extrapolation of radiation doses from Hodgkin's disease and low-grade non-Hodgkin's lymphoma to aggressive non-Hodgkin's lymphoma, there is a tendency to use low-dose radiation in patients with the latter disorder.
We hope future trials will be designed to reconsider not only the place of radiotherapy in the treatment of lymphoma but also the schedule and dose patterns and the target volumes in different clinical situations, such as after high-dose chemotherapy with stem-cell support. The detailed results of the trial conducted by the Southwest Oncology Group may soon provide important information on this subject.2
Dirk De Ruysscher, M.D., Ph.D.
Philippe Spaas, M.D
Sint-Norbertus Ziekenhuis
2570 Duffel, Belgium
Emmanuel van der Schueren, M.D., Ph.D.
University Hospital
3000 Leuven, Belgium
References
No information is given about the chief end point of the study (survival or event-free survival) and the difference in the anticipated end point that the investigators wished to determine in comparing the two treatments. The reader is informed about a two-sided P value of less than 0.05 but not about the statistical power considering the hypothesized difference in outcome and the number of patients studied.
Two planned interim analyses were performed. The first was performed "to detect any abnormal toxic effects in the transplantation group (>15 percent)." In the second analysis, "differences [were] considered significant if the P value was less than 0.01." What were the toxic effects the investigators wished to detect, and what kind of end-point difference did the second interim analysis seek? Since the study was closed early because of low patient accrual, before the planned number of patients had been enrolled, what was the power of the preliminary evaluation and the range of the proposed benefit from high-dose chemotherapy and autologous bone marrow transplantation, as compared with conventional therapy? A study is terminated early because of either an interim analysis showing a difference in the chief end point (such as survival) with a high level of statistical significance or a low level of accrual with equivalent end points in the two groups of the study at the time of the analysis. In such a case, the reader is given information about the probability of having missed an anticipated difference or the range of differences that could be ruled out as possible true differences.
In the case of positive results obtained in a prematurely closed trial, we are worried that the statistical significance of the difference between the treatment groups is a result of multiple testing when the difference appeared significant.
Claus-Henning Koehne, M.D.
Peter Daniel, M.D.
Humboldt University
13122 Berlin, Germany
This discussion detracts from the real question: How can therapeutic outcomes be compared when the disease treated is unknown? By chance, it is known that at least one patient probably did not have intermediate- or high-grade non-Hodgkin's lymphoma (depending on the correct diagnosis). What about the other patients? This article clearly shows that pathologists are not as irrelevant as either the authors of this report or the editors at the Journal appear to think they are.
J. Adelia McBride, M.D.
William C. Pugh, M.D.
University of Texas M.D. Anderson Cancer Center
Houston, TX 77030
References
To the Editor: We can supply Dr. Atkins with the following answers. The mean age at recruitment was 44.2 years (range, 24 to 59) in the transplantation group and 43.8 years (range, 19 to 59) in the conventional-treatment group. The Karnofsky score was unknown in five cases. The score was greater than or equal to 80 in 44 patients in the transplantation group and in 44 patients in the conventional-treatment group. The score was less than 80 in seven patients in the transplantation group and in nine in the conventional-treatment group (in all of whom it was 60 to 70) (the difference is not significant).
Group assignments were centrally managed in Lyon and were known by the physicians who participated in the trial before inclusion of the patients. All the patients were preregistered before the first course of dexamethasone, cisplatin, and cytarabine (DHAP), and all patients were enrolled and randomized in Lyon. No patient was excluded after randomization.
It is true that randomization was performed without stratification on the basis of prognostic factors (stratification was performed only by the centers). However, we have performed a Cox proportional-hazards analysis in which all the main available prognostic factors were included. This analysis confirmed the superior results in the transplantation group in terms of both event-free survival (relative risk of an event in the conventional-treatment group, 2.24; 95 percent confidence interval, 1.72 to 2.75; P = 0.002) and overall survival (relative risk of death in the conventional-treatment group, 1.90; 95 percent confidence interval, 1.33 to 2.47; P = 0.028).
The study was stopped after seven and a half years, and only the two previously planned interim analyses were performed. The first one, at three years, showed that the rate of death from toxic effects in the transplantation group was under 15 percent (no analysis of the end point was performed). The second analysis was performed two years later to determine whether there was a 20 percent advantage in disease-free survival in one group, without disclosure of the group with the survival advantage, if the difference was not statistically significant (which was the case).
We agree with the comments by De Ruysscher et al. on the role of radiotherapy in the treatment of lymphomas.
The term noted by Drs. McBride and Pugh, on page 1543 of our article, was an error ("small lymphocytic follicular lymphoma" should have read "small lymphocytic lymphoma").
In response to Drs. Koehne and Daniel, the primary end point of the study was event-free survival, and only the two previously scheduled analyses were performed. The statistical hypothesis was a rate of event-free survival of 35 percent in the transplantation group and 15 percent in the conventional-treatment group at two years. The calculated sample size for a power of 80 percent and three years of accrual was 71 patients per group. The study was stopped in June 1994 by the policy board without any additional statistical testing. The power of the final analysis with the initial expected difference is 78 percent.1
We would also like to clarify a statement in our article. On page 1541, in the right-hand column, the first sentence of the second paragraph under Results, "The rate of response to the initial doxorubicin-containing regimen was 58 percent," should have read as follows: "The rate of response to the two courses of DHAP given before randomization was 58 percent."
Thierry Philip, M.D.
Frédéric Gomez, M.D.
Franck Chauvin, M.D.
Centre Léon Bérard
69373 Lyon, France
References
| |||||||||||||||||||||||||||||||||
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | TERMS OF USE | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2009 Massachusetts Medical Society. All rights reserved. |
Previous
