See editorial and companion articles on pages 1560-3, 1687-96, and 1697-703, this issue.
Preliminary findings were presented at the Annual Meeting of the American Society of Clinical Oncology; May 29-June 2, 2009; Orlando, FL.
Melanoma patients who develop brain metastases (B-Met) have limited survival and are excluded from most clinical trials. In the current study, the authors attempted to identify primary tumor characteristics and clinical features predictive of B-Met development and post-B–Met survival.
A prospectively accrued cohort of 900 melanoma patients was studied to identify clinicopathologic features of primary melanoma (eg, thickness, ulceration, mitotic index, and lymphovascular invasion) that are predictive of B-Met development and survival after a diagnosis of B-Met. Associations between clinical variables present at the time of B-Met diagnosis (eg, extracranial metastases, B-Met location, and the presence of neurological symptoms) and post-B–Met survival were also assessed. Univariate associations were analyzed using Kaplan-Meier survival analysis, and the effect of independent predictors was assessed using multivariate Cox proportional hazards regression analysis.
Of the 900 melanoma patients studied, 89 (10%) developed B-Met. Ulceration and site of the primary tumor on the head and neck were found to be independent predictors of B-Met development on multivariate analysis (P = .001 and P = .003, respectively). Clinical variables found to be predictive of post-B–Met survival on multivariate analysis included the presence of neurological symptoms (P = .008) and extracranial metastases (P = .04). Ulceration was the only primary tumor characteristic that remained a significant predictor of post-B–Met survival on multivariate analysis (P = .04).
The development of brain metastases (B-Met) heralds a dismal prognosis for melanoma patients. Such patients have a median survival of <6 months,1, 2 and their poor prognosis precludes entry into most clinical trials of novel agents. Surgery, chemotherapy, and radiosurgery can extend survival in select patients, but a definitive cure conferring long-term survival is unattainable for the vast majority of patients with B-Met.2-4
Compared with other solid malignancies, it is more common in melanoma that B-Met represent the only site of metastatic involvement.5 This observation has led to speculation that melanoma has a brain-specific tropism, and investigations using animal models of spontaneous melanoma B-Met to elucidate the molecular mechanisms of site-specific metastases are currently underway.6 However, the mechanism driving this predilection is currently unknown, and we are unable to use molecular markers to predict which primary melanoma patients will ultimately develop B-Met or which patients will have the best chance for extended survival after the development of B-Met. Previous retrospective studies have identified clinical variables present at the time of B-Met diagnosis that are predictive of post-B–Met survival including age, the number of lesions, and the presence of extracranial disease.3, 7, 8 These studies may have clinical implications for the appropriate selection of those B-Met patients most likely to benefit from additional, possibly more aggressive treatment. However, the lack of effective treatment for melanoma B-Met in terms of improving overall survival makes it less likely that variables identified at the time of B-Met diagnosis will ultimately impact mortality in a way that is clinically meaningful. There remains a need for the identification of prognostic variables present early in the melanoma course that can identify patients at highest risk for B-Met who may benefit from increased surveillance and/or prophylactic interventions targeting the central nervous system (CNS).
Mitotic index has been incorporated into the American Joint Committee on Cancer (AJCC) staging system,9 and recent reports suggesting that ulceration may be an independent predictor of better clinical response to both adjuvant interferon and adjuvant vaccination with antigenic peptides emphasize the prognostic relevance of primary tumor characteristics.10-12 Although thickness and ulceration of the primary tumor factor into the staging of melanoma patients with primary and lymph node disease, to our knowledge, the influence of primary tumor characteristics on the clinical course of patients with AJCC stage IV melanoma has not been defined. Staging of patients with distant metastases is dictated primarily by the site of metastatic involvement, with primary tumor characteristics no longer playing a role. Previously published studies assessing the impact of primary tumor characteristics on post-B–Met survival are retrospective.1, 3 In the current study, we identified primary tumor characteristics and clinical features that are predictive of B-Met development and post-B–Met survival in a cohort of 900 prospectively accrued melanoma patients.
MATERIALS AND METHODS
Patients receiving treatment for primary or metastatic melanoma at New York University (NYU) Langone Medical Center were prospectively enrolled in the Interdisciplinary Melanoma Cooperative Group (IMCG) database13 from August 2002 through October 2008. The study was approved by the Internal Review Board of NYU, and all patients provided written informed consent at time of enrollment.
IMCG patients with evidence of B-Met as documented by the attending clinical oncologist were identified and medical charts were reviewed for clinical variables including location of B-Met (frontal, temporal, occipital lobe, etc), laterality (unilateral vs bilateral), number of metastatic lesions (at time of initial B-Met diagnosis and cumulative), presence of extracranial metastases at the time of B-Met diagnosis, first metastatic site at time of stage IV diagnosis, total number of visceral metastases, presence of neurological symptoms, total number of symptoms at time of B-Met diagnosis, evidence of hydrocephalus, leptomeningeal lesions, and hemorrhage. Information regarding B-Met–directed treatments including craniotomy, radiosurgery, whole brain radiotherapy, and systemic therapy (chemotherapy, immunotherapy, and biotherapy trials) was also obtained. In the case of patients who died during the follow-up period, the attending medical oncologist was contacted to determine whether B-Met was considered to be the cause of death.
Primary tumor characteristics assessed included thickness, histological subtype, anatomic site, mitotic index (few indicates 2 mitoses, moderate indicates 3-5 mitoses, and many indicates ≥5 mitoses/high-power field), regression, ulceration, and lymphovascular invasion (LVI). Cases with reported LVI were reviewed by the IMCG pathologist (F.D.) for histopathologic confirmation and assessed microscopically for LVI, which was defined as the presence of tumor cells within the peritumoral endothelial-lined spaces.
Descriptive statistics were calculated for baseline demographic and clinicopathologic characteristics. For categorical variables, the frequency distribution between patients with and without B-Met was evaluated using the chi-square test. For analyses pertaining to primary tumor characteristics, patients with melanoma of an unknown primary tumor or patients without complete information for a given variable were excluded from the analysis. The Wilcoxon rank sum test was used to compare the median primary tumor thickness between patients with and without B-Met. The independent effects of demographic and clinicopathologic variables on B-Met development were assessed by multivariate logistic regression analysis. Referent categories for categorical (indicator) variables in the multivariate logistic regression model were defined based on the category of each predictor variable that had the lowest risk of B-Met development as observed in the univariate analyses. Post-B–Met survival (defined as the day of B-Met detection to the date of last follow-up or death) was estimated using Kaplan-Meier survival analysis, and univariate associations between clinicopathologic variables and post-B–Met survival were assessed using the log-rank test. The independent effect of predictors of post-B-Met survival identified on univariate analysis was assessed by multivariate Cox proportional hazards regression analysis. Adjusted hazard ratios (HRs) were computed and 95% confidence intervals for the HR are presented to assess the precision of the obtained estimates. The median follow-up time (from the date of the initial melanoma diagnosis) was computed based on survivors in the non-B-Met cohort. Follow-up time for the B-Met patients was calculated based on B-Met survivors and based on the total cohort. All P values are 2-sided, with statistical significance evaluated at the .05 α level. All analyses were performed using SAS (version 9.2; SAS Institute Inc, Cary, NC) and Stata (version 10.0; Stata Corporation, College Station, Tex) statistical software.
There were 89 patients who presented with or developed melanoma B-Met during the study period (2002-2008). The mean age at the time of B-Met diagnosis was 60.3 years (standard deviation, 16 years), and 60% of the patients were male (Table 1). The majority of patients (n = 46; 52%) were designated as having AJCC stage I or II disease at the time of the initial melanoma diagnosis (Table 1) (Fig. 1), and the median time from the initial melanoma diagnosis to B-Met was 2.1 years (range, 2 months-13.3 years) (Table 1). The median follow-up time, based on all B-Met patients (n = 89) was 2.7 years (range, 0.4-17.7 years). At the end of the study period, 70 (79%) of the B-Met patients had died with melanoma, 13 (15%) were alive with melanoma, and 6 (7%) were alive without melanoma. The median post-B-Met survival was 5.8 months (range, 0.2-76.8 months). The incidence density for B-Met in the current study cohort was 3.19 cases per 100 person-years at risk, which translates to 8 cases per 100 patients at risk (based on a median length of follow-up of 2.4 years among survivors in the cohort). At the time of B-Met diagnosis, 63% (n = 56 patients) of patients had evidence of extracranial disease. With regard to treatment modalities, approximately 70% of patients (n = 62) received 2 to 4 interventions directed toward B-Met, including craniotomy (n = 32 patients; 36%), gamma knife radiosurgery (n = 43 patients; 48%), temozolomide (n = 58 patients; 65%), or another type of systemic therapy other than temozolomide (n = 70 patients; 79%). Patients who received craniotomy had the highest median survival of the cohort at 7.9 months, which was 2.1 months longer than the overall median survival of 5.8 months.
Table 1. Baseline and Primary Tumor Characteristics of Patients Who Developed B-Met Identified From a Cohort of 900 Melanoma Patients With Prospective Clinical Follow-Up
B-Met indicates brain metastases; SD, standard deviation; NA, not applicable; AJCC, American Joint Committee on Cancer.
Sum may be less than the total cohort because of patients with melanoma of an unknown primary (27 in the non-B–Met cohort and 13 in the B-Met cohort) or a lack of data for a given variable.
Determined using the Wilcoxon rank sum test or chi-square test as appropriate.
Calculated based on all patients in the B-Met cohort (n=89) from the time of initial melanoma diagnosis and based on survivors (n=726) in the non-B–Met cohort (n=811) from the time of initial melanoma diagnosis. The median follow-up time based on survivors in the B-Met cohort (n=19) was 3.3 years (range, 0.4-17.7 years). P = .22 for comparison of the median follow-up time among survivors in both cohorts (median, 2.4 years for the non-B–Met cohort vs 3.3 years for the B-Met cohort, respectively).
Excludes 1 patient with mucosal melanoma in the B-Met cohort.
To assess the clinical and primary tumor characteristics associated with B-Met development, the 89 B-Met patients identified from the cohort of 900 were compared with the 811 patients who did not develop B-Met during the study period. The mean age at the time of the initial melanoma diagnosis was the same for both B-Met and non-B–Met patients (57 years) (Table 1). A higher percentage of B-Met patients were male (60% vs 53%), although the difference was not statistically significant (P = .25). The median follow-up time, based on survivors, for the non-B–Met cohort (726 survivors of 811 patients without B-Met) was 2.4 years (range, 0.1-17.3 years). There were 85 deaths (10.5%) in the non-B–Met cohort: 63 patients died with melanoma, 16 patients died without melanoma, and 6 patients died with their melanoma status unknown. There was no significant difference noted in the median follow-up time between the B-Met and non-B–Met cohort (2.7 vs 2.4 years, respectively; P = .23).
Primary Tumor Characteristics Associated With the Development of B-Met
On univariate analysis, stage, thickness, anatomic site, histologic subtype, ulceration, mitotic index, and LVI were found to be significantly associated with the development of B-Met (Table 1). Patients with B-Met had thicker lesions (1.85 mm vs 0.95 mm; P < .0001) with a higher percentage of tumors located on the head and neck (29% vs 15%; P = .002). The B-Met patients had a higher frequency of nodular melanoma (54% vs 27%; P < .0001) and ulcerated lesions (48% vs 17%; P < .0001), and a higher mitotic index (45% with “many” vs 19%; P < .0001). LVI was twice as prevalent in the B-Met group (18% vs 9%; P = .01). A multivariate analysis demonstrated that a primary anatomic site on the head and neck and ulceration remained independent predictors of B-Met development after adjusting for other variables (adjusted odds ratio [OR], 2.52 [P = .003] and adjusted OR, 3.08 [P = .001], respectively) (Table 2).
Table 2. Multivariate Analysis of Clinicopathologic Variables of Primary Melanoma Associated With the Development of B-Met
Only patients with complete data for a given variable were included in the analysis.
>2 vs ≤2
Head/neck vs extremity/axial
NM vs SSM/other
Mitotic index, mitoses/mm2
Many vs none/few/moderate
Clinical Variables Predictive of Post-B–Met Survival
Eight clinical variables present at the time of B-Met diagnosis were found to be significantly associated with post-B–Met survival on univariate analysis including age >65 years, the presence of extracranial disease, the number of B-Met lesions (at the time of initial B-Met diagnosis and cumulative), the specific location (frontal lobe vs other), the localization of disease (unilateral vs bilateral), the presence of neurological symptoms, the number of neurological symptoms, and the presence of weakness and fatigue (Table 3). Patients with solitary lesions (19%) were found to have the longest median post-B–Met survival (11.9 months) compared with patients with ≥4 cumulative lesions (58%) who survived only 5.1 months (P = .01). Patients without involvement of the frontal lobe survived twice as long as patients with frontal lobe involvement (10 months vs 4.9 months; P = .01), and patients presenting with weakness or fatigue at the time of B-Met diagnosis had the lowest median survival of the cohort (2.2 months; P <.0001). Variables indicative of systemic disease burden such as the presence of extracranial disease at the time of B-Met diagnosis, the number of B-Met lesions, the presence of neurological symptoms at the time of B-Met diagnosis, and the presence of weakness and/or fatigue were found to be significant predictors of post-B–Met survival (Table 3).
Table 3. Univariate Analysis of Clinical Variables Present at B-Met Diagnosis Associated With Post-B–Met Survival (n=89)
Primary Tumor Characteristics Predictive of Post-B–Met Survival
On univariate analysis, ulceration, mitotic index, and LVI were found to be significant predictors of post-B–Met survival (Table 4). Thickness of the primary melanoma, anatomic site, histologic subtype, and regression were not found to be associated with post-B–Met survival. Patients with ulcerated primary tumors had a median post-B–Met survival of only 4.0 months compared with 9.1 months for patients without ulcerated lesions (P = .004). The presence of LVI was found to be strongly correlated with survival on univariate analysis. The median post-B–Met survival for patients with a primary melanoma positive for LVI was only 1.3 months compared with 6.7 months for patients without LVI (P < .0001) (Table 4). Kaplan-Meier survival curves stratified by ulceration, the presence of LVI, and mitotic index are shown in Figure 2.
Table 4. Univariate Analysis of Clinicopathologic Variables of the Primary Tumor Present at the Time of the Initial Melanoma Diagnosis Associated With Post-B–Met Survival (n=76)a
Excludes 13 patients with melanoma of an unknown primary tumor.
Only statistically significant variables are shown.
Percentages may not sum to 100% due to rounding.
Determined using the log-rank test.
Mitotic index, mitoses/mm2
A multivariate Cox proportional hazards regression model inclusive of 2 clinical variables (the presence of neurological symptoms and the presence of extracranial metastases) and 2 primary tumor characteristics (ulceration and LVI) revealed 3 factors to be independently predictive of post-B–Met survival. Clinical variables that remained statistically significant on multivariate analysis included the presence of neurological symptoms (HR, 2.5; P = .008) and the presence of extracranial metastases at the time of diagnosis (HR, 2.0; P = .04) (Table 5). The presence of ulceration was the only primary tumor characteristic that remained a significant predictor of post-B–Met survival on multivariate analysis (HR,2.5; P = .04).
Table 5. Multivariate Analysis of Clinical Variables and Primary Tumor Characteristics Associated With Post-B–Met Survival
Our study of 900 prospectively enrolled melanoma patients indicates that primary tumor characteristics maintain their prognostic relevance in patients with advanced melanoma. Patients with primary ulcerated melanoma on the head and neck may be at higher risk for the development of B-Met and shorter post-B–Met survival. Consistent with prior studies, the presence of neurological symptoms and the presence of extracranial disease at the time of B-Met diagnosis were the most significant clinical features predictive of post-B–Met survival.
Ulceration was the primary tumor characteristic found to be most strongly associated with the development of B-Met and was the only feature of the primary melanoma significantly associated with post-B–Met survival on multivariate analysis. Increasing evidence demonstrated that the presence of ulceration portends a particularly poor prognosis independent of tumor thickness, prompting investigations into the biology of ulceration and the mechanism by which it adversely influences the clinical course. One such study revealed that elevated serum levels of proinvasive matrix metalloproteinase-8 were significantly associated with the presence of LVI, ulceration, and bleeding.14 Thus, it is possible that ulceration is a marker of an increased propensity for hematogenous dissemination, extracellular matrix invasion, and the subsequent development of B-Met independent of tumor thickness. Previous studies also suggested that only a small degree of ulceration is necessary to adversely affect outcome, which is consistent with our data indicating that ulceration remained a significant adverse prognostic factor independent of thickness.15
Conversely, some investigations of recurrent melanoma suggest that, by the time B-Met develop, primary tumor characteristics such as ulceration are no longer relevant and the subsequent trophic interactions with the neural microenvironment ultimately dictate the kinetics of tumorigenesis and the ultimate clinical course.16 The data from the current study suggested that ulceration of the primary tumor does maintain its prognostic capacity even for patients with advanced disease, and thus it is possible that the presence of ulceration is indicative of a unique underlying biologic signature that promotes invasion and metastasis. Further investigation is required to determine the biologic characteristics of the primary melanoma that confer the propensity to metastasize to the brain. However, given our findings, it is possible that the biologic mediators of ulceration may also play a role in the subsequent penetration of the blood-brain barrier. A recent study of B-Met in breast cancer patients reported that expression levels of the novel gene ST6GALNAC5 in primary breast tumors specifically promote metastasis to the brain by enhancing adhesion and passage through the blood-brain barrier.17 The association between ulceration and B-Met noted in the current study suggests that similar investigations are warranted in patients with melanoma. Toward this end, our group is currently using gene expression array, single nucleotide polymorphism array, microRNA array, and other genomic tools to identify a specific genetic signature present in the primary melanoma of patients with B-Met compared with patients with other sites of distant metastases.
Our prospectively collected survival data indicating a median post-B–Met survival of 5.8 months is consistent with contemporary studies reporting median post-B–Met survival ranging from 3.8 months to 5.2 months.1, 3, 7, 18 A previous study of a cohort of 175 patients with melanoma B-Met accrued between 1972 and 1978 demonstrated an overall median post-B–Met survival of only 2.3 months.19 Although our results indicate an increase in post-B–Met survival of approximately 3.5 months over the past 30 years, it may be a reflection of lead time bias conferred by the introduction of imaging techniques such as magnetic resonance imaging. However, it is also possible that advances in the treatment of CNS metastases including stereotactic radiosurgery (SRS) account for the modest increase observed in the current study data. Previous studies have demonstrated that aggressive treatment comprised of SRS followed by a biochemotherapy regimen consisting of dacarbazine or temozolomide plus interferon-α-2B or interleukin-2 can achieve a median post-B–Met survival of >2 years in patients with limited CNS involvement and relatively well-controlled extracranial disease.20-22 These studies also suggest that patients with B-Met that are amenable to SRS or neurosurgery before the administration of combined biochemotherapy have response rates equivalent to those of patients without B-Met and thus should not be excluded from clinical trials.20, 22 The success of SRS and neurosurgery depends largely on patient selection, and survival rates are better for patients with controlled extracranial disease and a single B-Met.21 Proponents of multidisciplinary treatment approaches for patients with B-Met emphasize the need for early detection and subsequent aggressive treatment.20
The data from the current study do not include details regarding the timing or sequence of treatment modalities, and therefore a direct comparison of survival rates based on the type of intervention was not undertaken. However, we do report that only 36% of the current study patients accrued between 2002 and 2008 underwent craniotomy, which is the same percentage reported in a previous retrospective study of B-Met patients accrued 10 years earlier (1991-2001).1 Although it is likely that the highly significant association between craniotomy and improved outcome observed in previous studies is related to selection bias, the results of the current study suggest that there has been no improvement in increasing the number of patients eligible to undergo surgical resection. It is possible that heightened surveillance protocols for patients with ulcerated lesions on the head and neck may be warranted, but we recognize that our data cannot be extrapolated to make specific recommendations regarding screening or secondary prevention. Although to the best of our knowledge there are currently no prospective data to support specific follow-up regimens for patients with melanoma, there are many who believe that more structured recommendations might result in the earlier detection of solitary B-Met that can be cured with SRS or neurosurgery. Surveillance guidelines for melanoma remain controversial and are typically left to the discretion of the treating physician as dictated by clinical signs and symptoms.23 Radiographic surveillance for asymptomatic patients with localized melanoma is not recommended because of low sensitivity for the detection of occult metastases and the high rate of nonspecific findings that have the potential to result in a series of further unnecessary and potentially risky diagnostic procedures.24, 25 Similarly, the cumulative effect of long-term radiation exposure in a patient with completely excised thin melanoma who is most likely cured of disease must also be considered.26
Given the limited progress of treatment development for melanoma B-Met patients over the past 30 years, it is possible that adjuvant, preventive strategies targeting the CNS may be warranted for patients at very high risk for the development of B-Met. The results of the current study suggest that patients with ulcerated lesions may fall into this category. The current standard of care for melanoma does not include treatments designed to prevent CNS involvement. By contrast, recent data demonstrating that prophylactic cranial irradiation (PCI) reduces the number of symptomatic B-Met and improves both overall and cancer-specific survival for patients with small cell lung cancer has led to the recommendation that all patients who demonstrate a complete response to primary treatment be considered for PCI.20, 21 Considering that the mean patient age at the time of melanoma diagnosis is relatively young, the benefit in terms of life-years lost may justify the consideration of prophylactic treatment in patients at high risk for B-Met who would not otherwise be considered candidates for adjuvant therapy. A recent study using real-time in vivo imaging to capture the single events of the metastatic cascade in a mouse model of melanoma B-Met revealed discrete points when cells are the most inefficient at proliferating and thus likely the most vulnerable to inhibition.27 In melanoma cells, the most inefficient step occurred early in the cascade, during the transition from single perivascular cell to micrometastases. This was in contrast to lung cancer cells, which were most inefficient at a later point in the cascade, during the transition from micrometastases to macrometastases, suggesting that melanoma cells en route to the brain might be targeted most effectively at an early point in the metastatic cascade.27 Similarly, a subset of melanoma cells remained dormant over the course of the experiment yet demonstrated a high degree of motility when they ultimately reached the brain via pre-existing vessels, whereas lung cancer cells demonstrated no such motility.27 These results offer insight into the possible molecular mechanism underlying the delayed presentation of melanoma B-Met in some patients and may provide a rationale for treating patients at high risk for B-Met with prophylactic therapy early in the clinical course. Similarly, another animal study in breast cancer found that mice whose orthotopic tumors responded to treatment with a toll-like receptor agonist before direct intracranial injection of tumor cells demonstrated reduced tumor viability and eventual clearing of brain tumors compared with mice that had not been pretreated.28 Thus, rather than subjecting melanoma patients to PCI, as is done in lung cancer, an alternative strategy could be to design a clinical trial in which patients with localized melanoma designated as being at high risk for B-Met based on clinicopathologic variables and the presence of a specific molecular signature are offered adjuvant treatment to evaluate the possibility of preventing future B-Met.
In summary, the results of the current study revealed clinical and primary tumor characteristics associated with the development of B-Met and post-B–Met survival. Ulceration of the primary tumor was found to be the strongest predictor of B-Met development and was an independent predictor of reduced post-B–Met survival. The results of the current study emphasize the importance of considering primary melanoma tumor characteristics, especially ulceration, in the development of risk stratification and surveillance protocols early in the clinical course. Further investigations into the biology of ulceration and the molecular mechanism underlying its adverse impact on clinical outcome may be warranted.
We acknowledge the data management assistance provided by Ting Tu, BS, and Daniel Krich, BA.
CONFLICT OF INTEREST DISCLOSURES
Supported by the Chemotherapy Foundation, the National Cancer Institute Cancer Center Support Grant (5 P30 CA 016087-27), and the Marc Jacobs Campaign to support melanoma research.