| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Editorial |
1 Department of Surgery, Duke University Medical Center, Box 3118, Durham, North Carolina 27710
2 American Society of Clinical Oncology, 1900 Duke Street, Suite 200, Alexandria, Virginia 22314
Correspondence: Address correspondence and reprint requests to: Douglas S. Tyler, MD; E-mail: tyler002{at}acpub.duke.edu
Predicting outcomes for patients diagnosed with primary melanoma, as with any cancer, has several important utilities, including patient counseling, therapeutic decision making, designing and interpreting clinical trials, and optimizing health outcome assessments. Melanoma, more so than many other cancers, has lent itself to the development of predictive models initially based on microstaging methods, such as tumor depth using Breslow thickness and Clark level of invasion, and, more recently, to detecting micrometastases with the sentinel lymph node biopsy technique. In addition, a number of other factors have been associated with varying degrees of prognostic significance that reflect either patient-related factors (such as age, sex, site, or lymphatic drainage pattern) or tumor-related factors (such as ulceration, regression, vertical growth phase, mitotic rate, lymphovascular invasion, tumor-infiltrating lymphocytes, satellitosis, or perineural invasion). What has proven quite challenging, however, has been how to develop a useful staging system that tries to incorporate these multiple factors in a fashion that uses reliable and independent predictors of survival outcome and accurate prognostic modeling in the least invasive way possible.
In 2002, the American Joint Committee on Cancer (AJCC) completely revised its staging system for melanoma to more accurately incorporate independent prognostic factors based on a multivariate analysis.1 This staging system evaluated an array of prognostic factors and ultimately incorporated tumor-node-metastasis and stage grouping criteria that included tumor thickness, level of invasion, ulceration, nodal status, and distant disease status. The staging system was additionally validated by using a newly created 13-institution database called the AJCC Melanoma Staging Database, which contained prospectively collected data on 17,600 patients.2 There has been uniform agreement that the new staging system is a marked improvement over the old staging system with regard to its clinical utility.3 However, there are some concerns that the new staging system, especially with regard to thin melanomas less than 1 mm in depth, may have some limitations in discriminating recurrence and survival differences within the T1N0M0 subgroup. Although one of the important aspects of a staging system is certainly to allow grouping of patients to allow a global assessment of interventions on homogeneous groups of patients, another evolving consideration is the ability to provide an individual patient with specific prognostic information. As the number of independent predictive factors identified increases, future staging systems will need to be flexible enough to allow for incorporation of these various factors and a method for validating them.
In this issue of Annals of Surgical Oncology, Wong et al.4 have constructed a predictive model of sentinel lymph node positivity. Using the Memorial Sloan-Kettering Cancer Center melanoma database, they have constructed a nomogram for melanoma that is based on age, site, thickness, level, and ulceration and that is relatively easy to use; it has an ability to predict the probability of a positive sentinel lymph node. The nomogram was compared with AJCC staging for accuracy in predicting micrometastases in a sentinel lymph node by using a prospectively generated validation data set obtained from the Sunbelt Melanoma Trial. The Memorial Sloan-Kettering nomogram was found to be statistically more accurate than the AJCC staging system in predicting nodal positivity and provided a rapid method of calculating this probability on the basis of easily accessible patient and tumor characteristics. However, for the minor degrees of difference between the AJCC predictions and the Memorial Sloan-Kettering Cancer Center nomogram, one has to question how often information obtained from the nomogram would be different enough from the AJCC staging probabilities that one would recommend a different intervention or form of therapy, especially when there are some limitations in the nomogram’s calculations for patients with very thin and very thick melanomas. Because there is a selection bias for patients with thin melanomas undergoing sentinel lymph node biopsy, the accuracy of the tool seems to fall off at the thin end of the spectrum of melanoma thicknesses. Nevertheless, this approach is a forerunner of how we will be modifying our approach to staging melanomas and other cancers in the coming years.
For melanoma patients, the most difficult decisions come at each end of the melanoma thickness spectrum. Clinicians have been increasingly using the staging information from a sentinel lymph node biopsy in patients with thin and thick lesions without clearly validating whether information obtained from sentinel lymph node status is the most important factor predicting survival as compared with data obtained from noninvasive pathologic analysis of the primary tumor. For thin melanomas, the concern with developing a system that tries to predict sentinel lymph node status in this patient population is that up to half of the patients with thin melanoma who die from their disease develop distant metastasis initially before they have a regional nodal failure.5 In addition, for all melanomas, approximately 10% of patients who have negative sentinel lymph nodes will go on to develop metastatic disease.6–8 Two recent studies have developed prognostic trees for thin malignant melanomas that were not based on sentinel lymph node status but clearly identified subgroups of thin melanomas that did poorly. The first of these studies examined growth phase, mitotic index, and sex to develop a new classification system for thin melanomas that yielded four subgroups that have a minimal (.5%), low (4.1%), moderate (12.5%), and high (31.1%) risk of developing metastatic disease at 10 years.9 The second study examined a cohort of 12,728 patients with thin melanomas by using the central malignant melanoma registry of the German Dermatologic Society and found that tumor thickness, sex, age, anatomical site, and histopathologic subtype had a higher prognostic significance for thin melanomas, whereas ulceration and Clark level—two of the five variables in the Memorial Sloan-Kettering Cancer Center nomogram—did not.10 For thick melanomas, it has been argued that sentinel lymph node status may not be predictive of overall survival.11 These studies illustrate how optimal nomograms for melanoma will need to be individualized for various disease stages. For example, factors that may be important in determining which patients may have thin melanomas that are lethal may be different from the factors that are important in determining which patients have thick lesions that are not lethal.
As the next version of the AJCC melanoma staging is developed, we will need to consider modifying some of our traditional approaches that use only a few independent prognostic factors to derive a tumor-node-metastasis classification and stage grouping and do not as yet allow the use of multiple prognostic variables, including molecular and genomic criteria. Although anatomical staging is important and will continue to be a central part of staging classifications, we will have to use a data-based electronic process that integrates multiple prognostic variables to enhance its accuracy and reliability, and we will have to incorporate molecular staging criteria as they are validated. The AJCC has been exploring a collaborative staging system that collects common data elements into an electronic database so that staging systems can be revised and updated from a common set of data elements in a more automated fashion. Nomograms, such as that developed by the Memorial Sloan-Kettering group, will become useful tools for predicting metastatic risk more precisely because they can integrate more predictive features of the melanoma itself or the melanoma patient into a common end point, such as 5-year survival rates.
What does the future hold for developing prognostic modeling as we enter into a new era of predictive or prognostic factors that use multiple factors, including molecular markers and genomic profiling? The recent advances in DNA array–based gene profiling and tissue proteomics may markedly change how we predict the natural history of specific lesions. Both of these techniques are in their infancy but have started to demonstrate an ability to delineate tumors of varying biologies.12–14 For example, DNA micro-array analysis of breast tumors can predict very accurately which tumors will be associated with axillary lymph node metastasis.15 Surgeons have been instrumental in the translational application of these techniques and should continue to do so as we challenge our conventional ways of staging with these new technologies.
Received for publication January 28, 2005. Accepted for publication February 14, 2005.
REFERENCES
This article has been cited by other articles:
 |
|
 |
|
  C. R. Rossi, G. L. De Salvo, E. Bonandini, S. Mocellin, M. Foletto, S. Pasquali, P. Pilati, M. Lise, D. Nitti, E. Rizzo, et al. Factors Predictive of Nonsentinel Lymph Node Involvement and Clinical Outcome in Melanoma Patients With Metastatic Sentinel Lymph Node Ann. Surg. Oncol., April 1, 2008; 15(4): 1202 - 1210. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. I. Ross Early-stage melanoma: staging criteria and prognostic modeling. Clin. Cancer Res., April 1, 2006; 12(7): 2312s - 2319s. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
