Predictors of Survival, Neurologic Death, Local Failure, and Distant Failure After Gamma Knife Radiosurgery for Melanoma Brain Metastases

Abstract

OBJECTIVE:

This study sought to assess clinical outcomes in patients receiving gamma knife radiosurgery (GK) for treatment of brain metastases from melanoma and evaluate for potential predictive factors.

METHODS:

We reviewed 188 GK procedures in 129 consecutive patients that were treated for brain metastases from melanoma. The population consisted of 84 males and 45 females with a median age of 57 years. Fifty-five patients (43%) had a single metastasis. Seventy-one patients (55%) received chemotherapy, 58 patients (45%) received biologic agents, and 36 patients (28%) received prior whole brain radiation therapy (WBRT). The median marginal dose was 18.8 Gy (range 12 to 24 Gy).

RESULTS:

Actuarial survival was 52%, 26%, and 13% at 6, 12, and 24 months, respectively. The median survival time was 6.7 months. Local tumor control was 95%, 81% 53% at 6, 12, and 24 months, respectively. The median time to LBF was 25.2 months. Freedom from distant brain failure was 40%, 29%, and 10% at 6, 12, and 24 months, and the median time to DBF was 4.6 months. At the time of data analysis, 108 patients (84%) had died. Fifty-eight patients (52%) died from neurologic death. The median time to neurologic death from GK treatment was 7.9 months. Multivariate analysis revealed that hemorrhage of metastases prior to GK (P = .02) and LBF (P = .03) were the dominant predictors of neurologic death.

CONCLUSIONS:

GK achieves excellent local control and may improve outcomes as a component of a multidisciplinary treatment strategy. Distant brain failure and neurologic demise remain problematic and prospective trials are necessary.

INTRODUCTION

Melanoma is the third most common cancer to metastasize to the brain (12, 22) and brain metastases from melanoma contribute significantly to death and reduction in quality of life in this patient population (3, 25). Current treatment options are insufficient; the median survival with treatment is typically less than 1 year after the development of brain metastases, and patients commonly die of brain metastases.

Controversy has arisen regarding the optimal management of patients with brain metastases from melanoma. Although local control rates are high for melanoma brain metastases treated with radiosurgery (9), melanoma histology has been identified as one of the major risk factors for distant brain failure (DBF) after primary radiosurgical management of brain metastases (26). However, whole-brain radiotherapy (WBRT) yields suboptimal results, in part secondary to subsets of radioresistant phenotypes (2, 4). Further adding to the clinical dilemma is the fact that several classes of novel treatment strategies have evolved over time, including immunotherapy (11, 13), BRAF inhibitor therapy (19), and cytotoxic chemotherapy that penetrates the blood-brain barrier (8, 19). It is unclear what effect these agents have had on the natural history of melanoma brain metastases, although recent literature suggests that systemic agents may alter the natural history of disease of brain metastases (8, 18).

The purpose of this study is to review our series of patients with melanoma brain metastases treated with gamma knife radiosurgery (GK) to assess survival data, local brain failure (LBF), and DBF. We have focused on prognostic factors for dying of neurologic death because this end point provides a measure of the efficacy of central nervous system—directed therapies.

METHODS

Data Acquisition

This study was approved by the Wake Forest University Institutional Review Board. The Wake Forest University Medical Center Department of Radiation Oncology Gamma Knife Tumor Registry was searched for all patients who underwent GK and had a diagnosis of melanoma. One hundred twenty-nine patients with melanoma were identified who were treated with GK between April 2000 and July 2009 at Wake Forest University Baptist Medical Center in Winston-Salem, North Carolina. Patient outcomes were determined using the patients’ electronic medical records.

Patient characteristics are summarized in Table 1. Patient factors such as age, recursive partitioning analysis (RPA) class, status of extracranial metastatic disease, and previous systemic therapeutic regimens were determined from patients’ electronic medical records. The RPA class was defined per the Radiation Therapy Oncology Group analysis reported by Gaspar et al. (20). The status of extracranial metastatic disease was reported as none, oligometastatic, or widespread. Oligometastatic disease was defined as 5 or fewer nonbrain metastases.

Table 1.

Patient Characteristics

Total patients	129
Male/female ratio	84:45
Median age (years)	57.0
Median interval from primary to brain metastasis (months)	28.6
Brain metastases at presentation (number, %)
1 lesion	55 (43)
2–4 lesions	53 (41)
>4 lesions	14 (16)
Location of metastases (number, %)
Supratentorial	123 (95)
Infratentorial	39 (30)
Brainstem	12 (9)
Disease status at first gamma knife radiosurgery (number, %)
None	15 (12)
Oligometastatic	45 (35)
Widespread	69 (53)
Karnofsky Performance Status class (number, %)
60	7 (5)
70	17 (13)
80	85 (66)
90	16 (12)
100	4 (3)
Recursive partitioning analysis class (number, %)
I	10 (8)
II	111 (87)
III	7 (5)

Gamma Knife Procedures

Radiosurgery was performed on the Leksell Gamma Knife model, B, C, or Perfexion (Elekta AB, Stockholm, Sweden). A total of 188 GK procedures were performed to treat a total of 550 brain metastases. The median number of tumors treated during the first GK session was 2. Eighty-five patients (66%) had only 1 gamma knife session, whereas 44 patients (34%) had multiple procedures.

The decision to treat melanoma brain metastases was made by a multidisciplinary team consisting of medical oncologists, radiation oncologists, and neurosurgeons. An individualized therapeutic plan was formulated based on a patient’s degree of symptoms, extent of brain metastases, and systemic disease status. Melanoma patients with minimally symptomatic oligometastatic central nervous system disease generally received GK initially. When melanoma lesions were resected initially, GK treatment to the resection cavity routinely followed. GK was used as a salvage therapy for DBF after initial surgical resection, GK, or WBRT. GK was also offered in select cases as a boost after WBRT.

The Gamma Plan treatment planning system (Elekta AB, Stockholm, Sweden) was used to develop a conformal treatment plan with a median marginal dose of 18.8 Gy (range 12 to 24 Gy). The dose selected was based on the lesion size and volume as previously described by Shaw et al. (27). Table 2 summarizes GK treatment data.

Table 2.

Gamma Knife Radiosurgery and Other Treatment Data

Total number GK procedures	188
Total number metastases treated with GK	550
Median number GK treatments per patient	1.0
Median number isocenters at first GK treatment	6.0
Median number metastases treated at first GK	2.0
Median minimum marginal dose (Gy) at first GK	19
Number patients receiving GK with:
1 treatment (number, %)	85 (66)
2 treatments (number, %)	35 (27)
3 treatments (number, %)	5 (4)
4 treatments (number, %)	3 (2)
>4 treatments (number, %)	1 (1)
Patients receiving therapy with:
Chemotherapy, initial (number, %)	52 (40)
Chemotherapy, salvage (number, %)	19 (15)
Biologic agents, initial (number, %)	43 (33)
Biologic agents, salvage (number, %)	15 (12)
WBRT, initial (number, %)	36 (28)
WBRT, salvage (number, %)	25 (19)
GK, boost after WBRT (number, %)	4 (3)
GK, salvage after WBRT (number, %)	32 (25)
Craniotomy, initial (number, %)	40 (31)
Craniotomy, salvage (number, %)	13 (10)

GK, gamma knife radiosurgery; WBRT, whole-brain radiation therapy.

Patient Follow-Up

Follow-up assessments were typically conducted at 4 to 6 weeks after GK treatment, and subsequently every 3 months for the first year. In cases with intracranial progression of disease after GK treatment, additional GK treatment was often offered if patients had previously received WBRT or if patients without previous WBRT did not experience early or numerous (>4) DBF.

Time intervals that were recorded include time to local failure, distant failure, and death. Neurologic death was defined as previously described by Patchell et al. in patients who had stable systemic disease and progressive neurologic dysfunction or if they had both advancing systemic and neurologic function simultaneously (23).

Local failure was defined as either a pathologically proven recurrence of melanoma within the GK treatment field or a combination of imaging and clinical characteristics of local treatment failure. Patients with suspected treatment failure were generally followed up with serial imaging and treated conservatively with either steroid therapy or a combination of vitamin E and pentoxifylline before determination of a treatment failure. Imaging characteristics of treatment failure included serial increases in size of enhancement and/or increased perfusion on perfusion-weighted imaging. Hemorrhage was detected using signal patterns on T2- and T1-weighted MRI sequences, and on gradient-echo sequencing in many cases. All patients had MRI scans interpreted by fellowship-trained neuroradiologists.

Statistical Analysis

Descriptive analyses of overall survival (OS), time to neurologic death, and time to LBF and DBF were depicted with Kaplan-Meier plots and compared with the log rank test. Times were calculated from the date of initial GK treatment. Univariate logistic and Cox proportional hazards regression analyses were used to describe covariates of interest. Covariates included patient and disease parameters (stage, size, comorbidities), time to event (time to local, distant failure), treatment-related parameters (adjuvant chemotherapy, whole-brain radiation, and biologic agents), and imaging characteristics (hemorrhage). Each covariate was tested for the relevant regression assumptions in the logistic regression model describing predictors of neurologic death and including absence of multicollinearity and overly influential outliers. Backward stepwise selection was used for the final selection of covariates for the multivariate model. A required threshold of a P value of .2 was used for initial inclusion in the model. A receiver-operator characteristic model was constructed to estimate the fit of the model. All computations were carried out using SAS version 9.2 (SAS Institute, Cary, North Carolina, USA) and Microsoft Excel 2010.

RESULTS

Survival Data

The median follow-up time after GK was 4.4 months. At the time of our analysis, 108 patients (84%) were dead and 21 (16%) remained alive. Median survival was 52.8 months from the diagnosis of the primary tumor and 6.7 months after initial GK treatment (Figure 1). Actuarial survival rates were 52% at 6 months, 26% at 12 months, and 13% at 24 months after GK. Of the 108 patients who had died at last follow-up, 58 patients (53%) suffered a neurologic death. The median time to neurologic death from initial GK treatment was 7.9 months (Figure 2).

Figure 1.

Kaplan-Meier plot of overall survival.

Figure 2.

Kaplan-Meier plot of freedom from local failure.

Univariate analysis was performed to determine factors that predicted survival after GK. Having a decreased extent of disease at initial GK (none vs. oligometastatic, P = .02), having fewer intracranial metastases at initial GK (P = .02), having an increased interval to DBF (P = .0001), receiving chemotherapy before GK (P = .02), and receiving treatment with biologic agents before GK (P = .017) were associated with prolonged survival.

Univariate analyses revealed that patients having a lower RPA class (0 vs. 2, P = .02), having an increased interval to DBF (P = .0002), and receiving biologic agents before GK (P = .01) were less likely to suffer a neurologic death. Patients that had an increased number of metastases treated at initial GK (P = .002), a lower marginal dose at the first GK treatment (P = .04), and hemorrhagic metastases before GK (P = .0005) were more likely to suffer a neurologic death. In a multivariate model, hemorrhage of metastases before GK (P = .02) and LBF (P = .03) were predictors of neurologic death (Table 3).

Table 3.

Predictors of Neurologic Death: Multivariate Logistic Regression Model

Parameter	Odds Ratio Estimate	95% Confidence Interval	P Value
Number of initial metastases treated	1.07	0.86–1.34	.54
Supratentorial location of metastases	2.36	0.15–36.90	.54
Infratentorial location of metastases	3.51	0.90–13.69	.07
Hemorrhage of metastases prior to GK	6.67	1.31–33.98	.02
WBRT before GK	0.68	0.16–2.80	.59
Local brain failure	12.33	1.34–113.40	.03
Distant brain failure	1.91	0.55–6.61	.31

GK, gamma knife radiosurgery; WBRT, whole-brain radiation therapy.

Local Tumor Control

Figure 3 shows the Kaplan-Meier curve for local control, showing a median time to LBF of 25.2 months. Actuarial freedom from local failure was 95% at 6 months, 81% at 12 months, and 53% at 24 months. Univariate analysis was performed to determine factors that predicted local tumor control. Patients who had hemorrhage of metastases after GK (P = .02) and WBRT before GK (P = .05) were less likely to have local tumor control.

Figure 3.

Kaplan-Meier plot of freedom from distant failure.

DBF

Figure 3 shows the Kaplan-Meier curve for DBF with a median time to DBF of 4.6 months. Actuarial freedom from DBF was 40% at 6 months, 29% at 12 months, and 10% at 24 months. Univariate analysis was performed to determine factors that predicted distant tumor control. An increased number of lesions treated at first GK session (P = .05) was predictive of distant failure. Hemorrhage of metastases after GK trended toward significance for DBF (P = .06).

Hemorrhage-Associated Morbidity

Twenty-eight (21.7%) patients had metastases that hemorrhage before GK treatment. Of the 40 patients who had craniotomies before GK, 16 patients had craniotomies to resect hemorrhagic metastases. Forty-one patients demonstrated evidence of intratumoral hemorrhage after treatment. In 31 cases, hemorrhage was associated with GK-treated metastasis. In 13 cases, hemorrhage was associated with newly diagnosed, untreated metastases. In 7 cases, hemorrhage involved both treated and untreated metastases. In the 13 patients who required salvage craniotomies after GK, 12 patients had craniotomies to resect hemorrhagic metastases. Pre-GK hemorrhage did not predict post-GK hemorrhage (P = .08).

DISCUSSION

The median survival time of 6.7 months seen in the current series is consistent with OS reported in other series of patients with brain metastases from melanoma (15). In this series, 52% of patients died of neurologic death. It is clear from such survival data that improvements in the treatment of metastatic brain disease from melanoma are needed. One of the possible future improvements in the treatment of melanoma patients with metastatic disease is the proper triage and sequencing of brain-directed therapies. For example, identification of factors that predict for shorter overall survival may dictate that patients with poor prognostic factors receive palliative WBRT. Furthermore, elucidation of factors that lead to death from brain metastases may lead to more aggressive management of brain metastases in those populations, such as the combination of surgery or WBRT with radiosurgery.

Previously published series have suggested that worse RPA class (6, 17), worse Karnofsky Performance Status score (9, 17), active extracranial disease (17, 21), increased intracranial tumor volume (17), and the presence of hemorrhagic lesions before radiosurgery (17, 24) may be factors that predict for worsened survival in the melanoma population with brain metastases. In the current series, worse RPA class, increased number of metastases, lower marginal dose, and presence of hemorrhagic metastases were associated with worsened survival time. These prognostic factors can be interpreted based on 2 distinct outcomes: 1) death from systemic disease, and 2) death from brain disease. It is likely that worse RPA class is associated with death from systemic disease because RPA class is based on the presence of active systemic disease and patient performance status. In poor RPA class, patients who are likely to die quickly of systemic disease, it may be that palliative WBRT is appropriate. With the other factors that predicted worse survival, neurologic deterioration was likely a significant cause of death.

Multivariate analysis in the current series identified the presence of a pre-GK hemorrhagic metastasis and local brain failure as the dominant factors that predict for dying of neurological causes. Factors that predicted for local control in the present series included prior WBRT and pre-GK hemorrhage of metastasis. A possible mechanism for patients experiencing local failure from GK after prior local failure from WBRT is that the prior WBRT served to select for more radioresistant tumor cells. This possibility would argue for the utility of radiosurgical boost after primary WBRT as an attempt to sterilize these resistant clonogens at the time of first presentation. Prior series have shown that SRS boost after WBRT can improve local control (1). That lower marginal dose predicts worsened survival suggests that larger metastases are more difficult to control with GK alone and predispose to local failure and death. With larger or hemorrhagic metastases, resection followed by radiosurgery to the resection cavity is a technique that leads to excellent local control, even in lesions as large as 3 cm (14). This technique depends on surgical accessibility of lesions.

Recently, there has been a trend in the management of brain metastases of all histologies to avoid WBRT in patients with <5 brain metastases in favor of primary radiosurgery. This trend is due to the toxicities of WBRT, which have been reported to be detectable by 4 months after WBRT in a randomized controlled trial published by Chang et al. (5). Given that survival was not affected by withholding WBRT, and that there is a high local control rate with radiosurgery alone, these patients are thought to benefit from withholding whole-brain irradiation until time of multiple DBF. Our data suggest that the question remains unanswered for melanoma histology. In the study by Chang et al. (5), only 7 patients with melanoma were enrolled, essentially making it impossible to interpret for melanoma histology. Median survival in the current series was short, and although there was a detectable tail to the survival curve, 2-year actuarial survival was only 13%. Melanoma histology previously has been described as one of the dominant factors predicting for DBF after primary radiosurgical management of brain metastases (26). The DBF rate in the current series was 68% at 1 year. An increased number of lesions at the time of GK predicted for shorter interval to DBF in the current series. Other series have cited active extracranial disease as another factor that predicts for DBF (10, 16). Rapid multiple DBF represents a scenario for which use of WBRT is useful, and one which, if predictable, would potentially indicate use of adjuvant WBRT with primary radiosurgery.

It is clear that better data are necessary to potentially aid in predicting which patients are best managed with primary radiosurgical management without either surgery or adjuvant WBRT. Predictive models are currently being developed to help predict outcomes for brain metastases of specific histologies. Sperduto et al. reported a histology-specific graded prognostic assessment of patients with brain metastases (28). For the melanoma-specific graded prognostic assessment, the characteristics reported to predict survival are performance status and number of brain metastases. In our series, a decreased extracranial tumor burden, fewer intracranial metastases, and a longer interval from diagnosis of primary tumor to brain metastasis were factors associated with a more prolonged OS. It is likely that predictive models for brain metastases will continue to evolve over time. It will be important that future models not only address survival, but also address the likelihood of early DBF.

There are several limitations to the current series. As a retrospective review, the conclusions are limited to hypothesis generation given the possibility of patient selection bias. Melanoma histology seems to have distinct biologic behaviors from metastases from other histologies, and it is unlikely that phase III data from brain metastases of all histologies can really be extrapolated to brain metastases from melanoma. Prospective trials may need to focus on the combining of modalities to attempt to optimize the therapeutic ratio. Finally, it may be that the novel systemic agents that have shown activity in metastatic melanoma may ultimately play a role in the multidisciplinary management of melanoma. Novel agents have recently shown benefits when used in conjunction with brain-directed therapies in other cancers, such as renal cell carcinoma (7).

CONCLUSIONS

GK achieves excellent local control and may improve outcomes as a component of a multidisciplinary treatment strategy. DBF and neurologic demise remain problematic, and prospective trials are necessary.

Abbreviations and Acronyms

DBF

Distant brain failure

GK

Gamma Knife radiosurgery

LBF

Local brain failure

OS

Overall survival

RPA

Recursive partitioning analysis

WBRT

Whole-brain radiotherapy