Correspondence
Gene-Expression Signatures in Breast Cancer
N Engl J Med 2003; 348:1715-1717April 24, 2003
- Article
To the Editor:
The study by van de Vijver et al. (Dec. 19 issue)1 is of considerable interest with respect to the biology of breast cancer. However, from the point of view of the individual patient and her physician, the sensitivity and specificity of prognostic profiles are more important than odds ratios. It is possible to calculate these factors in part from the results. For example, for “all new patients in the consecutive series” (180 patients, listed in Table 2 of the article), the sensitivity and specificity of the poor prognostic signature are 0.93 (95 percent confidence interval, 0.81 to 0.98) and 0.53 (95 percent confidence interval, 0.45 to 0.61), respectively. If a hypothetical treatment were administered on this basis, nearly two thirds of the patients with a poor prognostic signature (65 of 104 patients) would be overtreated. Thus, the reported genetic signatures do not seem to be much sharper weapons for daily practice than other, easier-to-obtain prognostic indicators.
We also question the representativeness of the samples studied by van de Vijver et al. Intratumoral heterogeneity is common in breast cancer, and there is a growing body of evidence indicating that this neoplasm is frequently multiclonal.2-4 Therefore, a priori extrapolation from a limited topographic tumor region to the total tumor seems unacceptable. Van de Vijver et al. present insufficient details and analysis of the intratumoral representativeness of their samples.
4 ReferencesPeter Helmbold, M.D.
Johannes Haerting, Ph.D.
Heinz Kölbl, M.D.
Martin Luther University Halle–Wittenberg, 06097 Halle (Saale), Germany
peter.helmbold@medizin. uni-halle. de 1
van de Vijver MJ ,He YD ,van 't Veer LJ , et al. A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 2002;347:1999-2009
Full Text | Web of Science | Medline2
Symmans WF ,Liu J ,Knowles DM ,Inghirami G . Breast cancer heterogeneity: evaluation of clonality in primary and metastatic lesions. Hum Pathol 1995;26:210-216
CrossRef | Web of Science | Medline3
Barranco SC ,Perry RR ,Durm ME , et al. Intratumor variability in prognostic indicators may be the cause of conflicting estimates of patient survival and response to therapy. Cancer Res 1994;54:5351-5356
Web of Science | Medline4
Going JJ ,Abd El-Monem HM ,Craft JA . Clonal origins of human breast cancer. J Pathol 2001;194:406-412
CrossRef | Web of Science | Medline
To the Editor:
Accurately differentiating breast cancers with an inherently bad prognosis from those with a good prognosis could have a major effect on the provision of tailored and appropriate therapy. Unfortunately, previous, highly promising prognostic indicators have not fulfilled their promise. Furthermore, claims that are not justified by the data can be misleading.1 Without details about tumor size, the data reported by van de Vijver et al. do not support their conclusions concerning tumor size in breast cancer. The dichotomous comparison that they used is misleading,2 since the likelihood of metastatic spread increases with size. I suspect that most of the cancers in their sample were larger than 1 cm, whereas the median size of invasive cancers in our screening program is 9 mm. Tabar et al. have shown that patients with invasive cancers smaller than 1 cm almost universally do extremely well,3 contradicting the assertions of van de Vijver et al.
The data the authors provide do not show that early detection is of little benefit. Their data do not prove that metastatic spread “is an early and inherent genetic property of breast cancer,” and they do not refute the theory “that metastatic potential is acquired relatively late during multistep tumorigenesis.” Randomized, controlled trials of screening have confirmed that down-staging results in an absolute decrease in mortality that is not due to biases such as lead-time bias or length bias. This new genetic test, if confirmed, will help tailor therapy, but the data do not negate the importance of earlier detection.
3 ReferencesDaniel B. Kopans, M.D.
Massachusetts General Hospital, Boston, MA 021141
Kolata G. Early detection of cancer: nothing is black and white. New York Times. December 24, 2002:F1.
2
Kopans DB . The breast cancer screening controversy: lessons to be learned. J Surg Oncol 1998;67:143-150
CrossRef | Web of Science | Medline3
Tabar L ,Fagerberg G ,Duffy SW ,Day NE ,Gad A ,Grontoft O . Update of the Swedish two-county program of mammographic screening for breast cancer. Radiol Clin North Am 1992;30:187-210
Web of Science | Medline
to the Editor:
Van de Vijver and colleagues interpret the results of their innovative study, which focused on gene profiling as a predictor of survival in breast cancer, as supporting the hypothesis that metastatic potential is determined relatively early in tumorigenesis. However, two pieces of evidence, highlighted by Hellman,1 point to the development of metastatic potential in early breast cancer as a relatively late phenomenon. First, the breast-cancer screening program has reduced breast cancer–associated mortality by 30 percent. This implies that early detection by mammography before breast cancers become large enough to be palpable may prevent the development of distant metastases. Second, in the Gustave Roussy series of 3000 patients with breast cancer treated before the use of adjuvant systemic therapy, distant metastases were a continuous function of increasing tumor size.2 In the Dutch study, the genetic profile was not associated with the diameter of the tumor, perhaps because of the relatively small size of that study as compared with the much larger Gustave Roussy series.
2 ReferencesIan H. Kunkler, F.R.C.R.
Western General Hospital, Edinburgh EH4 2XU, United Kingdom
i.kunkler@ed. ac. uk 1
Hellman S . Karnofsky Memorial Lecture: natural history of small breast cancers. J Clin Oncol 1994;12:2229-2234
Web of Science | Medline2
Koscielny S ,Tubiana M ,Le MG , et al. Breast cancer: relationship between the size of the primary tumour and the probability of metastatic dissemination. Br J Cancer 1984;49:709-715
CrossRef | Web of Science | Medline
To the Editor:
In research to validate a prognostic system, the inclusion of 61 patients from the original training group used to derive RNA-expression signatures1 could bias the results, because the validation group is not independent. The prognostic system made possible accurate prediction in the training group, as would be expected: 31 of 49 patients with lymph-node–negative disease who were predicted to have a poor prognosis (63 percent) had one (Table 2 of the article). In the validation group, the signatures performed less well when the patients with lymph-node–negative disease were analyzed independently: 11 of 34 patients predicted to have a poor prognosis (32 percent) had one. In the remaining analyses, the system showed some discrimination in the totally independent, lymph-node–positive group (Figure 2E and 2F). However, for the remaining analyses in Table 3 and Figure 2A, 2B, 2C, 2D, 3A, 3D, 3E, 3F, and 3G, the degree of prognostic discrimination may have been inflated by the inclusion of patients from the training group.
Combining training and validation samples is not justified, as proposed, by the “similarity” of odds ratios for patients with lymph-node–negative disease, for two reasons. First, odds ratios overestimate relative risks, particularly as outcomes become more common. The odds ratios are similar, despite dissimilar underlying relative risks (6.1 and 10.7), because the outcome is more common in the validation group (43 percent vs. 18 percent). Second, and more important, relative risks (whether risk ratios or odds ratios) do not reflect information about absolute rates of the outcome. More informative would be a presentation of data in life tables, after the exclusion of the original patients.
1 ReferencesDavid F. Ransohoff, M.D.
University of North Carolina–Chapel Hill, Chapel Hill, NC 27599-70801
van 't Veer LJ ,Dai H ,van de Vijver MJ , et al. Gene expression profiling predicts clinical outcome of breast cancer. Nature 2002;415:530-536
CrossRef | Web of Science | Medline
Author/Editor Response
Helmbold and colleagues argue that we do not provide accurate, useful tools for patient care. We feel that the survival curves presented in our article, based on a consecutive cohort of patients, provide the most relevant information for patients and physicians. Our findings indicate that nearly 40 percent of patients have a tumor with a good-prognosis profile and a very low frequency of relapse. This is a significant improvement over the currently used prognostic factors (Figure 3 of our article). Tumors may be multiclonal with regard to some characteristics, but they are homogeneous in space and time with respect to histologic features and genetic alterations, including HER2 gene amplification.1 We have profiled several pairs of RNA samples extracted from different sections of the same tumor and have found gene-expression profiles to be similar (data not shown). Thus, despite possible intratumoral heterogeneity, the gene-expression profile accurately predicts disease outcome. The validation of the prognostic value of our profile further underscores the notion that heterogeneity in sampling is not a major factor.
Both Kopans and Kunkler argue against our hypothesis that metastatic capability is an early property of breast cancer, since tumor size is an established prognostic factor. We also found that size is an independent prognostic factor (Table 4 of our article). However, small tumors can also cause distant metastases.2 Kopans criticizes the dichotomy (≤20 vs. >20 mm) used in Table 1 to describe the relation between tumor size and gene-expression profile. However, as can be seen in Table 4, in the analysis of the risk of distant metastases we did not use this dichotomy.
We believe that the genetic makeup or gene-expression profile of a tumor is a strong determinant of its propensity to develop distant metastases. In addition to this, stochastic processes influence the likelihood that distant metastases will actually develop. We agree that smaller tumor size is associated with better outcome but argue that the intrinsic capacity of some small tumors to metastasize reduces the benefit of early detection. We do not argue against the need for early detection of breast cancer.
Ransohoff argues that combining training and validation samples is not justified and led to inflated discriminating power. We did not use the “similarity” of odds ratios for patients with lymph-node–negative disease as a justification for the inclusion of 61 patients from the training group. The justification of our strategy is thoroughly discussed in the “Validation Strategy” section of our article.
2 ReferencesMarc J. van de Vijver, M.D., Ph.D.
Netherlands Cancer Institute, 1066 CX Amsterdam, the NetherlandsYudong D. He, Ph.D.
Rosetta Inpharmatics, Kirkland, WA 98034Laura J. van 't Veer, Ph.D.
René Bernards, Ph.D.
Netherlands Cancer Institute, 1066 CX Amsterdam, the Netherlands
rbernards@nki.nl 1
Gancberg D ,Di Leo A ,Cardoso F , et al. Comparison of HER-2 status between primary breast cancer and corresponding distant metastatic sites. Ann Oncol 2002;13:1036-1043
CrossRef | Web of Science | Medline2
Tabar L ,Chen HH ,Duffy SW , et al. A novel method for prediction of long-term outcome of women with T1a, T1b, and 10-14 mm invasive breast cancers: a prospective study. Lancet 2000;355:429-433[Erratum, Lancet 2000;355:1372.]
CrossRef | Web of Science | Medline
- Citing Articles (6)
Citing Articles
1
Axel Walther, Elaine Johnstone, Charles Swanton, Rachel Midgley, Ian Tomlinson, David Kerr. (2009) Genetic prognostic and predictive markers in colorectal cancer. Nature Reviews Cancer 9:7, 489-499
CrossRef2
Zsofia K. Stadler, Steven E. Come. (2009) Review of gene-expression profiling and its clinical use in breast cancer. Critical Reviews in Oncology/Hematology 69:1, 1-11
CrossRef3
David F. Ransohoff. (2007) How to improve reliability and efficiency of research about molecular markers: roles of phases, guidelines, and study design. Journal of Clinical Epidemiology 60:12, 1205-1219
CrossRef4
Abhijit Mazumder, Yixin Wang. (2006) Gene-expression signatures in oncology diagnostics. Pharmacogenomics 7:8, 1167-1173
CrossRef5
D. F. Ransohoff. (2005) Lessons from Controversy: Ovarian Cancer Screening and Serum Proteomics. JNCI Journal of the National Cancer Institute 97:4, 315-319
CrossRef6
David F. Ransohoff. (2005) Opinion: Bias as a threat to the validity of cancer molecular-marker research. Nature Reviews Cancer 5:2, 142-149
CrossRef
