Correspondence
A Predictive Model for Non-Hodgkin's Lymphoma
N Engl J Med 1994; 330:574-575February 24, 1994
- Article
To the Editor:
The model of Shipp et al. for predicting survival in patients with aggressive non-Hodgkin's lymphoma (Sept. 30 issue)1 may represent an improvement over staging alone, but the approximations that the authors use in simplifying their model may limit its value in comparing different groups of treated patients or in assigning risk to an individual patient. The generalizability of their results is limited by several factors.
First, the authors treat the relative risks of the five independent variables as equal, whereas the true values vary by as much as 14 percent from the mean.
Second, in the model the cutoff point for lactate dehydrogenase (LDH) is set at the upper limit of normal because it was “easier to use” than the cutoff point that the authors found to be most predictive (a level 20 percent above normal).
Third, the univariate data (Table 2 of the article) showed that gastrointestinal involvement had little effect on survival, as compared with involvement at other extranodal sites, and that central nervous system involvement carried a somewhat worse prognosis. The effect of excluding patients with these types of involvement was, however, not addressed.
Fourth, the authors chose 60 years of age as a cutoff point because “this dichotomy was most commonly used in previous analyses,” but they give no information on the most predictive cutoff in the training sample.
Finally, the authors dichotomized all their variables. To the extent that the true survival curves for each variable differ from step functions, which change at the chosen cutoff, this method may represent a gross approximation.
If, instead of using the above approximations, the authors described survival according to a formula incorporating the independent variables (with the caveat given above about extranodal sites), the data would be more fully used and the accuracy of predictions improved. This technique has been successful in separating patients with testicular cancer into good-risk and poor-risk groups2. It would minimize the tendency of small differences among patients to cause large differences in risk assignment.
2 ReferencesCarl D. Atkins, M.D.
South Shore Hematology-Oncology Associates, Rockville Centre, NY 115701
The International Non-Hodgkin's Lymphoma Prognostic Factors Project
Full Text | Web of Science | Medline2
Bosl GJ ,Geller NL ,Cirrincione C , et al. Multivariate analysis of prognostic variables in patients with metastatic testicular cancer. Cancer Res 1983;43:3403-3407
Web of Science | Medline
To the Editor:
Shipp et al. of the International Non-Hodgkin's Lymphoma Prognostic Factors Project recommend that an international index and an age-adjusted international index be used in the design of future therapeutic trials involving patients with aggressive non-Hodgkin's lymphoma. It was puzzling to us that no mention was made of patients with AIDS-associated non-Hodgkin's lymphoma.
This subgroup tolerates chemotherapy poorly, and their prognosis is considerably worse than that of patients with lymphoma that is not HIV-related1. The intensity of therapy in these patients is limited by their susceptibility to infections and severe cytopenia. Carbone et al. identified a subgroup of patients with AIDS-associated non-Hodgkin's lymphoma whose tumors evidenced latent expression of the gene for Epstein-Barr virus2. Such patients may be expected to have different therapeutic outcomes from patients without such expression.
If the intent was to exclude patients with HIV, then it would have been appropriate to say so and to suggest that the prognostic models developed, as well as the international indexes, would be applicable only to patients with non-Hodgkin's lymphoma not associated with HIV infection. Adam M. Myers, M.D.
2 ReferencesGeorge E. Moore, M.D.
Denver General Hospital, Denver, CO 80204-45071
Levine AM ,Wernz JC ,Kaplan L , et al. Low-dose chemotherapy with central nervous system prophylaxis and zidovudine maintenance in AIDS-related lymphoma: a prospective multi-institutional trial. JAMA 1991;266:84-88
CrossRef | Web of Science | Medline2
Carbone A ,Tirelli U ,Gloghini A ,Volpe R ,Boiocchi M . Human immunodeficiency virus-associated systemic lymphomas may be subdivided into two main groups according to Epstein-Barr viral latent gene expression. J Clin Oncol 1993;11:1674-1681
Web of Science | Medline
Author/Editor Response
The authors reply:
To the Editor: Dr. Atkins questions our assumptions regarding relative risks, serum LDH values, and age at diagnosis in the international index described in our article. For simplicity and interpretability, we treated the relative risks as approximately equal in our prognostic-factor model. We believe that our simplification is amply supported by the data and leads to a more useful model. Atkins correctly notes that in our training sample we found a cutoff value for LDH of 1.2 times normal to be slightly more predictive than a cutoff of 1 times normal. In our training sample, the most predictive cutoff value for age was 58 years at diagnosis.
When the cutoff values for LDH and age are changed to 1.2 times normal and 58 years, a Cox regression analysis of data from the training sample produces results very similar to those in Table 3 of our paper. In fact, the largest change is in the relative-risk coefficient for age, which increases from 1.96 to 2.04. We examined the choice of cutoff values for both age and LDH very carefully during the original analysis and concluded that using 1 times normal for LDH and 60 years for age resulted in a more widely usable model. Since any statistical method is based on a model that can only approximate reality, it is important not to overinterpret details of an analysis that may be dependent on the model.
The three additional variables in the international index were categorical variables. We looked at all reasonable ways to subdivide each of the categorical variables, and in each case the dichotomy chosen (e.g., Ann Arbor stage I or II vs. stage III or IV) produced the most predictive model.
Drs. Myers and Moore correctly note that patients with AIDS-associated non-Hodgkin's lymphoma have less favorable prognoses and that such patients tolerate chemotherapy poorly because of HIV-related infections and cytopenia. For these reasons, patients with AIDS-associated non-Hodgkin's lymphoma were not included in the international index.
Margaret A. Shipp, M.D.
David P. Harrington, Ph.D.
Dana-Farber Cancer Institute, Boston, MA 02115- Citing Articles (2)
Citing Articles
1
Henry B Koon, Gregory C Ippolito, Alison H Banham, Philip W Tucker. (2007) FOXP1: a potential therapeutic target in cancer. Expert Opinion on Therapeutic Targets 11:7, 955-965
CrossRef2
P De Paepe, C De Wolf-Peeters. (2007) Diffuse large B-cell lymphoma: a heterogeneous group of non-Hodgkin lymphomas comprising several distinct clinicopathological entities. Leukemia 21:1, 37-43
CrossRef
