77
or
with levamisole or with levamisole plus BCG for malignant a Southwest Oncology Group study J Clin Oncol
lymphoma 10
11
12
13
14
15
1985,3 1318-24 Coltman CA, Dahlberg S, Jones SE, et al CHOP is curative in thirty percent of patients with large cell lymphoma a twelve year Southwest Oncology Group follow-up Proc Am Soc Clin Oncol 1986,15 197 Feinstein A The Will Rogers phenomenon Stage migration and new diagnostic techniques as a source of misleading statistics in cancer N Engl J Med 1985,312 1604-8 Sacks H, Chalmers TC, Smith H Randomized versus historical controls for clinical trials Am J Med 1982,72 233-40 Cooper IA, Ding JC, Matthews JP, et al , for the Australia and New England Lymphoma Group A randomized comparison of MACOP-B and CHOP in intermediate grade non-Hodgkin’s lymphoma Proc Am Soc Clin Oncol 1991,10 942 Freedman B Equipoise and ethics of clinical research N Engl J Med 1987,317141-5 Miller TP, Dana BW, Weick JK, et al Southwest Oncology Group clinical trials for intermediate- and some high-grade non-Hodgkin’s lymphomas Semm Haematol 1988,25 17-22 (suppl 2)
Mathematical Modeling or Waiting Decades for an Empirical Answer? In reply:-Dr. Stockler expresses concern that our model showing that third-generation regimen (MACOP-B) may offer survival advantage over first-generation regimens (CHOP) is not correct because of difficulties in: 1) comparability of data
between two arms (CHOP vs MACOP-B), 2) validation of model; and 3) new data about remission rates for MACOP-B regimen. We agree that results of a mathematical model will be better if the data used in the comparison of two treatment strategies come from groups of patients with similar clinical and biological characteristics As specified in our paper, we dealt with this problem by extracting only those data that were comparable in relation to the most important prognostic factors for non-Hodgkin’s lymphoma: pathologic subtype, stage of disease, and sex and age of the patient The ages of typical patients (median ages) differ among various studies, making a direct comparison of data more difficult We chose the age of 57 years for both treatment groups because that was median age for diffuse large-cell lymphoma reported in the largest study of its clinical course so far (see Simon et al Ann Intern Med 1988;109:939). This study was based on 1,175 patients from four institutions treated between 1971 and 1975, before the introduction of either CHOP or MACOP-B. By adopting this value we ensured the same age-specific mortality in both arms of our model. In this way the difference in the results could be ascribed only to the
prognostic factors may be important, do not know how to integrate them into this model However, the same argument can be applied to clinical trials, because our knowledge of prognostic factors is changing, and what is applicable today may not be of equal importance treatment effect. Other
but
we
tomorrow
We agree there is no substitute for empirical data. Any mathematical model should be compared against available data. We compare our model not against data from which the model was derived, but against a final outcome the model was designed to predict-survival. Our model included 12 states (first remission, first relapse, etc.) and it was compared with available data about survival for chemotherapeutic regimens without knowledge whether all of the patients actually went through all the states assumed by the model Our model accurately predicted survival (life expectancy) and thus gave us reassurance that the model could be realistic We are eager to learn how it will behave compared with data that are prospectively collected. MACOP-B remission rates were adopted from a single institution’s experience because we felt that these data were more mature than data from other institutions, the majority of which had been reported only in abstract form. While it may be true that the remission rate for MACOP-B is not as high as the 84% assumed in our model, it does not invalidate the results of the model, i.e., that the first-remission rate and functional status after the first remission is achieved are the most important determinants of life expectancy Our analysis does not exclude CHOP as a better treatment, should it be shown that the remission rate with MACOP-B is less than the CHOP remission rate (fig 3. in the paper), as it seems that ANZ data suggest. Furthermore, our analysis indicates that the model is a quite robust to variation in acute toxicities and relapse rates, as well as to salvage-therapy remission rates-a rather unexpected finding. This may be of critical significance in everyday clinical practice, where the key issue so far as the efficacy of a particular treatment is concerned relates to the trade-off between the remission rate (believed to be increased by adding more aggressive drugs) and the toxicity of treatment. Clinicians are not sure whether they should start treatment with less aggressive therapy and administer salvage therapy after relapse, or begin treatment with more aggressive but more toxic treatment hoping for long-lasting remission. When no empirical data exist regarding such an issue, we believe that it is better to offer an analytic approach to the problem than to passively wait for an answer, which has been expected by the oncology community for almost two decades with regard to lymphoma treatment.- BENJAMIN DJULBEGOVIC, MD, PhD, and THOMAS M WooDCOCK, MD, Division of Medical Oncology and Hematology} Department of medicine, School of Medicine,
University of Louisville, Louismlle,
Downloaded from mdm.sagepub.com at UNIV OF CHICAGO LIBRARY on July 10, 2015
KY 40292
or
with levamisole or with levamisole plus BCG for malignant a Southwest Oncology Group study J Clin Oncol
lymphoma 10
11
12
13
14
15
1985,3 1318-24 Coltman CA, Dahlberg S, Jones SE, et al CHOP is curative in thirty percent of patients with large cell lymphoma a twelve year Southwest Oncology Group follow-up Proc Am Soc Clin Oncol 1986,15 197 Feinstein A The Will Rogers phenomenon Stage migration and new diagnostic techniques as a source of misleading statistics in cancer N Engl J Med 1985,312 1604-8 Sacks H, Chalmers TC, Smith H Randomized versus historical controls for clinical trials Am J Med 1982,72 233-40 Cooper IA, Ding JC, Matthews JP, et al , for the Australia and New England Lymphoma Group A randomized comparison of MACOP-B and CHOP in intermediate grade non-Hodgkin’s lymphoma Proc Am Soc Clin Oncol 1991,10 942 Freedman B Equipoise and ethics of clinical research N Engl J Med 1987,317141-5 Miller TP, Dana BW, Weick JK, et al Southwest Oncology Group clinical trials for intermediate- and some high-grade non-Hodgkin’s lymphomas Semm Haematol 1988,25 17-22 (suppl 2)
Mathematical Modeling or Waiting Decades for an Empirical Answer? In reply:-Dr. Stockler expresses concern that our model showing that third-generation regimen (MACOP-B) may offer survival advantage over first-generation regimens (CHOP) is not correct because of difficulties in: 1) comparability of data
between two arms (CHOP vs MACOP-B), 2) validation of model; and 3) new data about remission rates for MACOP-B regimen. We agree that results of a mathematical model will be better if the data used in the comparison of two treatment strategies come from groups of patients with similar clinical and biological characteristics As specified in our paper, we dealt with this problem by extracting only those data that were comparable in relation to the most important prognostic factors for non-Hodgkin’s lymphoma: pathologic subtype, stage of disease, and sex and age of the patient The ages of typical patients (median ages) differ among various studies, making a direct comparison of data more difficult We chose the age of 57 years for both treatment groups because that was median age for diffuse large-cell lymphoma reported in the largest study of its clinical course so far (see Simon et al Ann Intern Med 1988;109:939). This study was based on 1,175 patients from four institutions treated between 1971 and 1975, before the introduction of either CHOP or MACOP-B. By adopting this value we ensured the same age-specific mortality in both arms of our model. In this way the difference in the results could be ascribed only to the
prognostic factors may be important, do not know how to integrate them into this model However, the same argument can be applied to clinical trials, because our knowledge of prognostic factors is changing, and what is applicable today may not be of equal importance treatment effect. Other
but
we
tomorrow
We agree there is no substitute for empirical data. Any mathematical model should be compared against available data. We compare our model not against data from which the model was derived, but against a final outcome the model was designed to predict-survival. Our model included 12 states (first remission, first relapse, etc.) and it was compared with available data about survival for chemotherapeutic regimens without knowledge whether all of the patients actually went through all the states assumed by the model Our model accurately predicted survival (life expectancy) and thus gave us reassurance that the model could be realistic We are eager to learn how it will behave compared with data that are prospectively collected. MACOP-B remission rates were adopted from a single institution’s experience because we felt that these data were more mature than data from other institutions, the majority of which had been reported only in abstract form. While it may be true that the remission rate for MACOP-B is not as high as the 84% assumed in our model, it does not invalidate the results of the model, i.e., that the first-remission rate and functional status after the first remission is achieved are the most important determinants of life expectancy Our analysis does not exclude CHOP as a better treatment, should it be shown that the remission rate with MACOP-B is less than the CHOP remission rate (fig 3. in the paper), as it seems that ANZ data suggest. Furthermore, our analysis indicates that the model is a quite robust to variation in acute toxicities and relapse rates, as well as to salvage-therapy remission rates-a rather unexpected finding. This may be of critical significance in everyday clinical practice, where the key issue so far as the efficacy of a particular treatment is concerned relates to the trade-off between the remission rate (believed to be increased by adding more aggressive drugs) and the toxicity of treatment. Clinicians are not sure whether they should start treatment with less aggressive therapy and administer salvage therapy after relapse, or begin treatment with more aggressive but more toxic treatment hoping for long-lasting remission. When no empirical data exist regarding such an issue, we believe that it is better to offer an analytic approach to the problem than to passively wait for an answer, which has been expected by the oncology community for almost two decades with regard to lymphoma treatment.- BENJAMIN DJULBEGOVIC, MD, PhD, and THOMAS M WooDCOCK, MD, Division of Medical Oncology and Hematology} Department of medicine, School of Medicine,
University of Louisville, Louismlle,
Downloaded from mdm.sagepub.com at UNIV OF CHICAGO LIBRARY on July 10, 2015
KY 40292
