Melanoma patients who develop brain metastasis have a very poor prognosis, in part because of relative resistance to external beam radiotherapy and in part because of the low activity and poor central nervous system (CNS) penetration of the available systemic agents. Three articles in this issue of Cancer summarize retrospective analyses of melanoma patients with brain metastases.1-3 Although the articles investigate similar questions, each considers different patient characteristics, and the results of the articles provide different and diverging conclusions. Herein, we summarize the results and the authors' conclusions from each of the 3 articles, examine potential limitations of each article's conclusions, attempt to synthesize the results, and conclude with ideas for future research.
The 3 articles reviewed in this issue of Cancer add to and update previously reported experience regarding the prognosis of melanoma patients diagnosed with brain metastases, but do not answer the important question of identifying which patients diagnosed with early melanoma will go on to develop this complication that most frequently determines the outcome in metastatic melanoma and how and when to best intervene therapeutically. If clinicians wish to use retrospective databases to help improve outcomes for these patients, then better biologic discriminators, based on careful clinico-pathologic correlational observations, need to be incorporated and then validated in large datasets to define useful prognostic subsets.
A Summary of the 3 Articles
Zakrzewski et al1 reviewed data from 900 patients enrolled in the Interdisciplinary Melanoma Cooperative Group at New York University between August 2002 and October 2008 and attempted to identify the clinical and pathological features of primary melanoma that increase the probability of developing brain metastases. They also examined clinical and pathologic factors that impacted the survival of melanoma patients who were diagnosed with brain metastases. The authors found that patients with primary melanoma on the head or neck were more likely to develop brain metastases than patients with primary melanomas elsewhere and that an ulcerated primary tumor was associated with an increased risk of developing brain metastases. With respect to survival after the development of brain metastases, the presence of extracranial metastases and the presence of neurological symptoms at the time of diagnosis of brain metastases and ulceration of the primary tumor were associated with poor survival.
Eigentler et al2 reported data from 9 cancer centers of the German Cancer Society from 1986 through 2007. The authors found that >1 brain metastasis and elevated lactate dehydrogenase (LDH) at the time of diagnosis of brain metastases were associated with poor survival after brain metastasis diagnosis. Among patients with a single lesion, patients treated with whole-brain irradiation or chemotherapy were found to have poorer survival after a diagnosis of brain metastases than patients who underwent neurosurgery or stereotactic radiosurgery.
Davies et al3 analyzed data from patients enrolled in clinical trials in the Melanoma Medical Oncology department at The University of Texas M. D. Anderson Cancer Center between 1986 and 2004. The authors found that being diagnosed with brain metastasis before January 1, 1996, having >3 brain metastases or leptomeningeal involvement, and a diagnosis of brain metastases after a diagnosis of extracranial metastases were associated with poor survival after brain metastases diagnosis. Among patients who received systemic therapy as the initial treatment for brain metastases, patients who previously responded to systemic therapies were found to have longer survival than patients who did not respond.
Does Ulceration of the Primary Tumor Warrant Brain Prophylaxis?
Zakrzewski et al1 concluded that primary tumor ulceration is the strongest independent prognostic factor for eventual diagnosis of brain metastatis. The authors suggested that this may warrant heightened surveillance protocols for patients with ulcerated lesions and that adjuvant therapy directed at the CNS, such as prophylactic cranial irradiation, may be useful for patients with ulcerated primary lesions. However, our interpretation of the report suggests that primary tumor ulceration may be less useful in predicting brain metastases than the authors suggest. Among the 900 patients reported, 168 had ulceration of their primary melanoma. However, during the time of follow-up and inclusion in this report, only 36 (21%) patients were diagnosed with brain metastasis. The 36 patients with brain metastasis and an ulcerated primary tumor represent approximately 40% of the 89 patients who developed brain metastasis.1
The analyses performed on this large (nearly 1000) patient cohort provide evidence that ulceration of the primary tumor is independently associated with the development of brain metastases (multivariate odds ratio, 3.08). Unfortunately, a strong association does not always suggest good predictive ability. Alternative statistical methods are necessary to develop and evaluate predictive models. In this study, 79% of patients with ulcerated primary tumors had not developed brain metastases, and 60% of patients with brain metastases did not have an ulcerated primary tumor.1 These numbers suggest that an ulcerated primary lesion may in fact be a poor predictor of eventual metastases.
Limitations to the Clinical and Pathologic Analysis of Risk Factors and Prognostic Factors of Melanoma Brain Metastases
Beyond statistical issues, the authors' conclusions do not address any aspect of the etiology of death in melanoma patients. Although older data suggest that the short survival of melanoma metastatic to the brain is directly because of death from inadequately controlled brain disease, newer series in the era of stereotactic radiosurgery, which is currently used for most patients with limited metastatic disease in the brain, may demonstrate that these patients are now more likely to die of extracranial metastatic disease, calling into question the value of brain prophylaxis. Further, in advanced disease, the control by whole brain irradiation is so poor that it has largely been supplanted by SRT. While WBI may be more effective as prophylaxis based on activity against micro metastatic disease, its uncertain effects, and long-term toxicities in a population of patients, most of whom do not go on to develop metastatic disease, is very unlikely to have a favorable therapeutic index.
Any analysis should also consider the biology and natural history of melanoma: if ulcerated melanoma gives rise to a high risk of eventual brain metastases, is that a direct effect of ulceration that could be mitigated by treating the brain early (if effective prophylactic therapy to the brain were available), or is the route to the brain actually via the development of earlier or more aggressive systemic disease? The latter explanation would call for expanded research efforts to find effective local and systemic therapies for melanoma, avoiding the development of brain metastases. The data provided in Zakrzewski et al1 thus do not provide a statistically sound approach with which to describe the development and prevention of brain metastases, nor do the authors provide a biologically based rationale for the surveillance or interventions proposed.
Should S-100B Protein and LDH Be Used in Treatment Planning for Patients With Brain Metastasis?
Eigentler et al2 considered several potential prognostic factors for survival after a diagnosis of brain metastasis, including LDH and serum S-100B protein. LDH is an established risk factor for stage IV melanoma and has been included in the staging guidelines for distant metastatic melanoma.4 Over the last decade, many groups have investigated the prognostic value of S-100B protein,5, 6 including a meta-analysis suggesting its independent prognostic value for patients with stages I to III melanoma but not stage IV.6
Eigentler et al had follow-up data from 672 patients.2 Only 40% of these patients (N = 270) had information regarding serum S-100B protein. The multivariate analysis that included S-100B protein was based on approximately 22% of patients (N = 151). Among this subset of patients, multivariate analysis controlling for performance status, number of brain metastases, and LDH did not reveal S-100B to have an independent value.2
Furthermore, the generalizability of any analysis suggesting the value of S-100B protein must be called into question. The patients who have S-100B data available may be different from patients who do not have available S-100B data. For example, the authors performed 2 multivariate analyses: 1 including S-100B protein data based on 151 patients and 1 without S-100B protein data based on 354 patients. In the former model, the hazard ratio (HR) for performance status was 2.7, with a P value of .04. In the model without S-100B protein data, the HR for performance status was 1.2, with a P value of .17.2 This large difference in the HR for 1 factor in the multivariate analysis suggests that patient characteristics in the 2 subsets were qualitatively different. This is a clear example of the general principle that selected prognostic factors may work best when applied to defined subpopulations, because the factors applied in broad multivariate analysis may not be of practical application to subsets defined for clinical management and translational research strategies.
Caveats on the Clinical Application of Historical and Incomplete Data
Unfortunately, the use of serum biomarkers such as S-100B protein and others currently under investigation suffers from a lack of sufficient association with outcomes to support their routine collection and subsequent analysis or their prospective investigation as independent biomarkers with prognostic or predictive value. Ideal biomarkers have recognized functions in the cancer of interest; even lacking this characteristic, a convincing biomarker has usually been selected from observational analyses followed by validation with prospective investigations, including rigorous multivariate analyses of all important factors also applicable to the dataset. Those markers proven to have independent effects may be accepted for clinical use in appropriate settings, and those with an overarching impact on prognosis may be sufficiently powerful to be incorporated into staging systems (for example, the recent incorporation of mitotic rate into the staging system for early stage melanoma and the persistence of serum LDH levels in substaging metastatic melanoma).
Has Survival for Patients With Brain Metastasis Improved Over the Past 20 Years?
Davies et al reported that, although still dismal, overall survival after a diagnosis of brain metastasis has improved over the last 26 years.3 The authors base this conclusion on the results of univariate and multivariate analyses applied to patients diagnosed before January 1, 1996 versus patients diagnosed after that date.
We concur with the authors that survival after a diagnosis of brain metastasis was always poor and remains so today.7, 8 On the basis of the data in this report, we are not able to concur that 1996 represented a change point in survival. This is attributable to the concern regarding lead time bias, based in large part on the infrequent use of magnetic resonance imaging before 1996 and its high frequency subsequently (both because of its superiority to computed tomography and its wider application to determine which patients might benefit from early surgery or radiotherapy as well as participation in clinical trials, which have often excluded these subjects). Stereotactic radiotherapy methods most likely have made a greater overall contribution to the management of melanoma than for any other single malignancy. The reasons include melanoma's relative radioresistance with dose response effects that can be addressed with stereotactic radiotherapeutic techniques; its chemoresistance, rendering control of brain metastases almost completely dependent on radiotherapy; and its very high rate of CNS involvement compared with overall cases, which may exceed half of patients diagnosed with advanced disease and be responsible for death in the majority of these patients.
Unfortunately, the retrospective nature of the data used in the survival analysis is not able to rule out this bias. To get a better sense of whether prognosis after a diagnosis of brain metastases has improved over the last 26 years, an analysis using a nonsurvival endpoint could be used.
What Can We Learn From These 3 Articles?
The 3 articles reviewed herein add to and update previously reported experience8, 9 regarding the prognosis of melanoma patients diagnosed with brain metastases. However, the conclusions add little to answer the important question of identifying which patients diagnosed with early melanoma will go on to develop this complication that most frequently determines the outcome in metastatic melanoma and how and when to best intervene therapeutically.
Melanoma patients with brain metastasis, whether treated in New York, Texas, Germany, or anywhere else, have a very poor prognosis. Ignoring potential lead time bias, Davies et al report that the median survival for patients diagnosed before 1996 was 4 months and, for those diagnosed after 1996, the median survival was 6 months.3 This incremental improvement in survival for patients with brain metastases implies that general medical advances and non-disease–specific therapeutic interventions such as stereotactic radiosurgery have made only a small difference for melanoma patients with brain metastases. If we wish to use retrospective databases to help improve outcomes for these patients, we need to incorporate better biologic discriminators, based on careful clinicolaboratory correlational observations, and validate them in large datasets to define useful prognostic subsets.
Now that it is possible to analyze many aspects of the molecular pathogenesis of melanoma subsets, a few tantalizing leads have already been reported. For example, human brain metastatic melanoma cells have shown increased signal transducer and activator of transcription 3 (STAT3) and AKT activity.10, 11 These results will need validation in larger studies and, ultimately, in multivariate analyses. Most importantly, a large genomic analysis is to be performed by investigators from the same group that contributed the report of Davies et al.3 The results of this analysis, combined with meticulous clinical annotation performed in a prospective fashion, will be a welcome addition to the existing literature, as a bridge to patient management and new clinical trial design. Incorporating this information, along with new information from molecular biology and the many ongoing “omics” experiments, could help us better understand brain metastasis development.
CONFLICT OF INTEREST DISCLOSURES
The authors made no disclosures.