Laboratory data from the 1970s first showed that malignant melanoma

2265 Survival by Radiation Therapy Oncology Group Recursive Partitioning Analysis Class and Treatment Modality in Patients with Brain Metastases from Malignant Melanoma A Retrospective Study Jeffrey C. Buchsbaum, M.D., Ph.D. 1 John H. Suh, M.D. 1 Shih-Yuan Lee, M.S.P.H. 1 Mark A. Chidel, M.D. 1 John F. Greskovich, M.D. 1 Gene H. Barnett, M.D. 2 1 Department of Radiation Oncology, The Cleveland Clinic Foundation, Cleveland, Ohio. 2 Department of Neurosurgery, The Cleveland Clinic Foundation, Cleveland, Ohio. The authors thank The Cleveland Clinic Foundation Department of Scientific Publications and, in particular, Jessica Ancker for editorial assistance with this article. They also thank the numerous support personnel in The Cleveland Clinic Foundation Department of Radiation Oncology for assistance in collecting hospital charts. Address for reprints: John H. Suh, M.D., Department of Radiation Oncology, Desk T-28, The Cleveland Clinic Foundation, 9500 Euclid Avenue, Cleveland, OH 44195; Fax: (216) 445-1068; E-mail: suh@radonc.ccf.org Received March 2, 2001; accepted December 23, 2001. BACKGROUND. In a population of patients with brain metastases from melanoma, the authors sought to determine whether various therapies provided any benefit at all, whether local therapy was better than whole brain radiotherapy (WBRT), and whether combined local therapy and WBRT provided any advantage over local therapy alone. They also analyzed survival according to a Radiation Therapy Oncology Group (RTOG) recursive partitioning analysis (RPA) to determine how well the RTOG RPA classes predicted survival in this patient population and whether treatments varied in effectiveness from category to category. METHODS. A total of 74 patients with brain metastases from melanoma were treated at The Cleveland Clinic Foundation between 1984 and 1998. For this study, the authors reviewed patient charts and confirmed survival status. Survival was compared by treatment modality (surgical resection, WBRT, stereotactic radiosurgery, or WBRT combined with local therapy). Survival also was compared according to the RTOG RPA prognostic classes (Class 1, Class 2, or Class 3), which has not been validated previously in patients with malignant melanoma. RESULTS. The median survival was 5.5 months for all patients. Survival varied significantly by RTOG prognostic class; The median survival was 10.5 months (range, 2.2 99.2 months) for patients in Class 1, 5.9 months (range, 0.2 43.9 months) for patients in Class 2, and 1.8 months (range, 0.1 6.9 months) for patients in Class 3 (P 0.0001). Survival analysis showed that combined treatment offered significantly better survival (P 0.0001; combined vs. other). The median survival was 8.8 months (range, 1.8 99.2 months) for the combined therapy group, 4.8 months (range, 1.2 27.8 months) for the local therapy alone group, 2.3 months (range, 0.2 9.6 months) for the WBRT alone group, and 1.1 months (0.1 3.0 months) for the group that received no therapy. CONCLUSIONS. Adding WBRT to local therapy may improve survival in this group of patients: Combined therapy was superior to WBRT alone. The RPA classification scheme likely has prognostic value for patients with brain metastases from malignant melanoma. Prospective studies are required to overcome selection bias and confirm these results. Cancer 2002;94:2265 72. 2002 American Cancer Society. DOI 10.1002/cncr.10426 KEYWORDS: malignant melanoma, whole brain radiotherapy, radiosurgery, brain metastases, recursive partitioning analysis, selection bias. Laboratory data from the 1970s first showed that malignant melanoma cell lines were highly radioresistant, 1 but clinical data on the role of radiation for patients with brain metastases from malignant 2002 American Cancer Society

2266 CANCER April 15, 2002 / Volume 94 / Number 8 melanoma are conflicting and of limited power. For example, one study found that radiation was palliative for a small number of patients, regardless of the number of metastases. 2 Conversely, some studies have suggested that melanomas have better response to radiosurgery than adenocarcinomas. 3,4 Whole brain radiotherapy (WBRT) does not appear to increase the efficacy of stereotactic radiosurgery (SRS) in patients with brain metastases from malignant melanoma. 5 Overall, radiation is effective for symptom control in patients with brain metastases from malignant melanoma. 6 However, a large 1998 review found no significant difference in survival between patients who received postoperative radiotherapy and patients who did not. 7 This finding was confirmed in a review of surgical patients at Memorial Sloan-Kettering. 8 In addition, debate continues about the applicability of the prognostic classification developed by the Radiation Therapy Oncology Group (RTOG) for patients with brain metastases. The RTOG algorithm was based primarily on data from patients with metastatic breast and lung carcinoma, and its prognostic value has not been assessed widely in patients with brain metastases from malignant melanoma. To provide further depth to the literature on this controversial topic, we analyzed the results from our institution s patients with brain metastases from malignant melanoma who were treated with linear accelerator-based radiosurgery (LINAC RS), gamma knife radiosurgery, or surgery. We sought to determine whether these therapies provided any benefit at all, whether local therapy was better than WBRT, and whether combined local therapy and WBRT provided any advantage. We also analyzed survival according to the RTOG recursive partitioning analysis (RPA) classes to determine how well these classes predicted survival in this patient population and whether treatments varied in effectiveness from class to class. MATERIALS AND METHODS Patient Sample We screened our departmental data base and The Cleveland Clinic Foundation s main data base for all records of patients with malignant melanoma who were treated for brain metastases. In all patients, treatment decisions were made by the physician and the patient; no algorithm or randomization was used. All patients received follow-up care at our institution. None had prior cranial malignancies. Diagnoses were confirmed radiographically with computed tomography (CT) scans or magnetic resonance imaging (MRI), except in several patients whose diagnoses were confirmed by surgical biopsy. No patients were excluded. Data Collection Data were collected from departmental and hospital charts and telephone interviews. If survival status was not clear in the medical records, then we contacted patients, families, or physicians by telephone or checked the patient s status in the United States Social Security data base. The data collected included age, gender, Karnofsky performance score (KPS) at diagnosis, symptom profile, location of lesion, number of lesions, size of lesions, histology, staging at initial diagnosis, dates of all procedures, dose of radiation used, type of resection (subtotal or macroscopic total), length of hospital stay, length of steroid usage/dependency, complications, cause of death, and date of death. Patient data were acquired and used following the guidelines set forth by our Institutional Review Board, which approved this study. Statistical Techniques We used the RTOG method 9 to assign prognostic classes to patients with brain metastases. The method uses RPA based on the KPS, primary lesion status, presence of extracranial metastases, and age. Patients were assigned to Class 1 (best prognosis), Class 2, or Class 3 (worst prognosis). For all patients, survival was assessed as of April 1, 2000. Actuarial data were evaluated and plotted using the Kaplan Meier method, and significance was evaluated using Cox Mantel log-rank tests for single variable comparisons and Cox proportional hazards for multivariate comparisons. The median survival is reported with the range in parentheses after the presented value. Statistical significance was defined as P 0.05 (two tailed). All statistical evaluations were conducted with the assistance of the statistical software packages JMP (version 3.2.1) and StatView (version 5.0; SAS Institute, Cary, NC). 10 Treatment Delivery Radiation therapy The methods of radiation delivery varied over time. LINAC RS was used for SRS until December 1996. The gamma knife has been used for SRS since January 1997. WBRT was performed with megavoltage-based external beam radiotherapy. Patients were simulated using standard techniques and were treated in various fractionation schemes using opposed lateral fields. In some patients, adjunct chemotherapy was added to surgery or radiation therapy. Fraction size varied from 180 centigrays (cgy) to 550 cgy, and fractions were given one per day in all patients. The total dose varied from 22 grays (Gy) to 55 Gy. The majority of patients who received WBRT (29 of 61 patients) were treated with fractions of 300 cgy for a total dose of 30 Gy. SRS

Survival in Malignant Melanoma Brain Metastases/Buchsbaum et al. 2267 TABLE 1 Characteristics of 74 Patients Treated for Brain Metastases from Melanoma Characteristic No treatment (n 3 patients) WBRT alone (n 25 patients) Local therapy alone WBRT with local therapy (n 10 patients) a (n 36 patients) a P value Gender (%) Male 1 (33) 21 (84) 6 (60) 25 (69) 0.1873 Female 2 (67) 4 (16) 4 (40) 11 (31) Age in yrs (%) 65 1 (33) 17 (68) 8 (80) 30 (82) 0.1740 65 2 (67) 8 (32) 2 (20) 6 (18) KPS (%) 70 3 (100) 10 (40) 0 (0) 0 (0) 0.0001 b 70 0 (0) 15 (60) 10 (100) 36 (100) Primary tumor controlled (%) Yes 2 (67) 17 (68) 7 (70) 35 (97) 0.0140 b No 1 (33) 8 (32) 3 (30) 1 (3) Systemic disease (%) Yes 2 (67) 20 (80) 7 (70) 22 (61) 0.4807 No 1 (33) 5 (20) 3 (30) 14 (39) Location of other metastases Liver 1 6 1 7 Lung 1 10 1 12 Bone 0 3 2 3 Other 2 17 6 10 Single lesion (%) Yes 2 (67) 3 (12) 6 (60) 16 (44) 0.0116 b No 1 (33) 22 (88) 4 (40) 20 (56) Solitary lesion (%) Yes 1 (33) 3 (12) 1 (10) 9 (25) 0.4622 No 2 (67) 22 (88) 9 (90) 27 (75) Stage (%) I 1 (33) 2 (8) 1 (10) 4 (11) 0.2501 II 2 (67) 11 (44) 2 (20) 7 (19) III 0 (0) 3 (12) 4 (40) 10 (28) IV 0 (0) 9 (36) 3 (30) 15 (42) RTOG RPA classes 1 0 (0) 0 (0) 1 (10) 11 (30) 0.0001 b 2 c 0 (0) 12 (48) 6 (60) 24 (67) 2 0 (0) 3 (12) 3 (30) 1 (3) 3 3 (100) 10 (40) 0 (0) 0 (0) Median age in yrs (range) 67 (32 84) 54 (40 86) 49 (31 73) 53 (27 78) 0.3175 Median KPS (range) 30 (10 50) 70 (30 100) 90 (70 100) 90 (70 100) 0.0029 b Median lesion numbers (range) 1 (1 10) 3 (1 25) 1 (1 6) 2 (1 6) 0.0216 b Median maximum greatest dimension in cm (range) 1 (0.7 3.0) 2 (0.5 5.0) 2 (0.8 5.0) 2.5 (0.5 7.0) 0.1642 WBRT indicates whole brain radiotherapy; RTOG indicates Radiation Therapy Oncology Group; RPA: recursive partitioning analysis. a Local therapy was either surgery or radiosurgery with either Linac or Gamma Knife. b Represents data with P 0.05. c RTOG Group 2 was broken up into Group 2 and Group 2 with the latter group comprised of patients with controlled primary lesions. dosing varied from 15 Gy to 25 Gy in one fraction, and gamma knife dosing followed standard RTOG guidelines, which are based on the greatest lesion dimension. Surgery Standard craniotomy technique was used to remove the lesions. Resection was either macroscopic total resection or subtotal resection. Biopsy-only procedures of cranial lesions were considered subtotal resection. Complications within 30 days of surgery were attributed to surgery if they appeared to be unrelated to a prior medical condition. RESULTS In 74 of 1154 patients (6.4%) with brain metastases who were treated at The Cleveland Clinic between 1984 and 1998, the histologic diagnosis was malignant melanoma. Table 1 shows the characteristics of this patient sample.

2268 CANCER April 15, 2002 / Volume 94 / Number 8 FIGURE 1. Kaplan Meier analysis of survival by treatment group (survival is outlined fully in Table 2). The difference between patients in the group that received combined therapy and patients in any of the other three treatment groups was statistically significant (P 0.0001). The diamonds represent patients who were living at the time of this report. The median survival for each therapy group is shown. The differences between the combined treatment group and the other groups were statistically significant (P 0.0001). WBRT: whole brain radiotherapy. There were more male patients than female patients. Stage was distributed as shown. The distribution of other sites of metastases is shown, and some patients had more than one site of metastasis. Chemotherapy was not used in the vast majority of the patients, and, when it was, a large number of different agents were used. Survival The median survival for all patients was 5.5 months (range, 0.1 99.2 months). Patients were stratified into four treatment groups for survival analysis: those who received no therapy (n 3 patients), those who received local therapy only (surgery or SRS; n 10 patients), those who received WBRT alone (n 25 patients), and those who received local therapy (surgery or SRS) combined with WBRT (n 36 patients). In the combined group, 14 patients underwent surgery and WBRT; 19 patients underwent SRS and WBRT; and 3 patients underwent SRS, surgery, and WBRT: Survival of patients in these three subgroups of the combined arm were 11.3 months, 12.3 months, and 13.1 months, respectively. Figure 1 shows that survival in the combination therapy patients was significantly better compared with survival in the other three treatment categories grouped together (P 0.001). Comparison of the SRS and WBRT group with the surgery and WBRT group demonstrated no significant difference (P 0.5128). The median survival for patients with melanoma who did not receive any adjuvant therapy was 1.1 months (range, 0.1 3.0 months). For patients who received either surgery or SRS without WBRT, the median survival was 4.8 months (range, 1.2 27.8 months). For patients who received WBRT only, the median survival was 2.3 months (range, 0.2 9.6 months). For patients who received both WBRT and local therapy (SRS or surgery), the median survival was 8.8 months (range, 1.8 99.2 months). Fraction size and total radiation dose did not significantly affect survival in the patients who received WBRT alone or in combination with other therapy. Fraction size did not affect survival in patients who received either SRS or WBRT. A second analysis was performed to compare survival by RTOG RPA class. Survival differences were statistically significant among the groups (P 0.0001). The data are shown in Figures 2 and 3. The median survival was 10.5 months (range, 2.2 99.2 months) for RPA Class 1 patients, 5.9 months (range, 0.2 43.9 months) for RPA Class 2 patients, and 1.8 months (range, 0.1 6.9 months) for RPA Class 3 patients (P 0.0001 for Class 1 vs. Class 2 vs. Class 3 in all combinations). The role of chemotherapy was not evaluated in this study and was used in a minority of our patients, often as part of a protocol. Log-rank univariate and proportional hazards multivariate analyses of the data are summarized in Tables 2 4. Significant factors on log-rank analysis included control of the primary lesion, single lesion status, solitary lesion status, RTOG RPA class, surgery, and treatment group. On univariate log-rank analysis, control of the primary lesion, single lesion status, solitary lesion status, RPA class, surgery group, SRS group, and treatment group were statistically significant. On univariate proportional hazards analysis, it was found that KPS was significant, whereas greatest lesion dimension (almost significant), fraction size during WBRT, and the number of lesions were not significant. On multivariate proportional hazards analysis, control of systemic disease, KPS, and treatment group were significant. There was a significant difference between patients who were treated with local therapy alone and patients who were treated with local therapy plus WBRT. There was a significant correlation between treatment group and RPA class, as shown in Table 1.

Survival in Malignant Melanoma Brain Metastases/Buchsbaum et al. 2269 FIGURE 2. Kaplan Meier analysis of survival by Radiation Therapy Oncology Group recursive partitioning analysis class (survival is outlined fully in Table 2). The difference between the three curves was statistically significant (P 0.0024). The diamonds represent patients who were living at the time of this report. FIGURE 3. Kaplan Meier analysis of survival according to the modified Radiation Therapy Oncology Group (RTOG) recursive partitioning analysis (RPA) Class. RPA Class 2 was split into two groups, patients with controlled primary lesions (Class 2 ) and patients without controlled primary lesions (Class 2 ). The diamonds represent patients who were living at the time of this report. Only eight patients did not have control of their primary lesion at the time they were diagnosed with brain metastasis and otherwise met the criteria for RTOG RPA Group 2. The differences between Classes 1, 2, 2, and 3 were not formally significant as a whole according to a Wald analysis (see Table 4) due to the small numbers, although Class 1 differed significantly from Class 3 (P 0.0012). Local and Regional Control Of 74 patients, 25 patients had failure of local control within the brain confirmed by CT or MRI scan data after the treatment was delivered. A total of 52 patients received follow-up CT or MRI scans. In all patients, local recurrence was accompanied by concurrent systemic recurrence or progression. Local cranial recurrence occurred in 2 of 10 patients (20%) who received local therapy only (surgery or SRS), in 4 of 24 patients (17%) who received WBRT only, and in 18 of 37 patients (49%) who received local therapy combined with WBRT. Scans were ordered based on clinical suspicion of recurrent disease without set guidelines. These data likely reflect the fact that patients who are treated aggressively may be more likely to be scanned with the onset of new symptoms. In our series, 18 of 25 follow-up scans were performed for patients who received local therapy combined with WBRT.

2270 CANCER April 15, 2002 / Volume 94 / Number 8 TABLE 2 Univariate Kaplan Meier Survival Analysis in 74 Patients Variable No. of patients Medial survival in months (range) Standard error Log-rank P value Gender Men 53 5.7 (0.2 88.6) 0.364 0.7621 Women 21 3.7 (0.1 99.2) 1.945 Age (yrs) 65 56 5.9 (0.8 99.2) 0.561 0.0172 a 65 18 2.5 (0.1 35.5) 0.530 KPS 70 13 6.3 (0.1 6.9) 0.449 0.0001 a 70 61 1.8 (0.2 99.2) 0.781 Stage I 8 4.8 (0.1 32.3) 0.778 0.1394 II 22 3.7 (0.2 28.9) 2.111 III 17 5.8 (1.8 88.6) 0.823 IV 27 6.9 (0.8 99.2) 1.291 Control of primary lesion Yes 61 5.9 (0.1 99.2) 0.586 0.0007 a No 13 2.4 (0.8 8.5) 0.270 Systemic (with other metastases) Yes 51 5.5 (0.2 43.9) 0.714 0.0519 No 23 5.7 (0.1 99.2) 2.036 Single lesion Yes 27 7.9 (0.1 99.2) 1.688 0.0057 a No 47 4.6 (0.2 60) 0.798 Solitary lesion Yes 14 7.9 (0.1 99.2) 4.677 0.0092 a No 60 5.2 (0.2 43.9) 1.356 RTOG RPA class 1 12 10.2 (2.2 99.2) 3.637 0.0001 a 2 42 6.1 (0.2 43.9) 0.648 2 7 3.7 (1.2 8.5) 2.226 3 13 1.8 (0.1 6.9) 0.449 WBRT Yes 61 5.8 (0.2 99.2) 0.390 0.3003 No 13 4.6 (0.1 35.5) 1.168 SRS Yes 32 7.5 (1.8 99.2) 1.556 0.0041 a No 42 2.9 (0.1 88.6) 1.458 SRS type Gamma knife 16 6.5 (3.5 5.5) 0.900 0.2445 Linac 16 11.2 (1.8 99.2) 4.800 Surgery Yes 25 10.2 (1.2 99.2) 3.997 0.0002 a No 49 4.1 (0.1 35.5) 1.048 Extent of resection Subtotal 6 5.7 (4.1 28.9) 4.715 0.1987 Macroscopic total 19 10.2 (1.2 99.2) 3.482 Treatment group No treatment 3 1.1 (0.1 81) 1.102 0.0001 a WBRT only 25 2.3 (0.2 9.6) 0.250 Local therapy (surgery or SRS) only 10 4.8 (1.2 27.8) 0.791 Local therapy plus WBRT 36 8.8 (1.8 99.2) 2.850 KPS: Karnofsky performance status; RTOG: Radiation Therapy Oncology Group; RPA: recursive partitioning analysis; WBRT: whole brain radiotherapy; SRS: radiosurgery. a Represents data with P 0.05. Complications Ten of 70 patients who were treated with either SRS or WBRT experienced an acute complication: swelling requiring increased steroid dosing in 9 patients and seizure in 1 patient, which was controlled with medication. No radiation patients had symptomatic radiation necrosis as a side effect. Of 25 patients who underwent surgery alone or with WBRT, 6 patients had acute complications requiring admission to hospital: One patient had an infection, 2 patients had hemorrhages, and 3 patients had new central nervous system deficits. No cerebrospinal fluid leaks were caused by surgery. No surgical patient had long-term complications. At 1 month after treatment, 6 of 70 living patients still were using steroids. By 3 months, two of these patients still were using steroids, and two additional patients had begun using steroids. DISCUSSION The current data suggest that 1) patients in RTOG RPA Class 1 were more likely to be treated aggressively with combination WBRT and a local therapy (SRS or surgery), 2) combination therapy in the context of patient selection is superior to either local therapy alone or WBRT alone, 3) local therapy increases survival relative to WBRT if selection bias is ignored, and 4) patients with brain metastases from melanoma have poor survival if they present with poor RTOG RPA class. This is in agreement with a recent evaluation 11 of the RTOG RPA classification that validated the approach for historic comparisons based on RTOG 91-04. The data in Tables 1 and 2 show clearly that treatment was correlated significantly with RPA class and that RPA class was correlated significantly with survival. Inherent in any nonrandomized study is the problem of selection bias. Much of the survival benefit of combination therapy (Fig. 1) likely was due to this bias. This critical point was explored by Gaspar et al., 9 who came to a similar conclusion: Patient selection plays a critical role in survival in brain metastases research series. This conclusion was reinforced indirectly in the RTOG RPA validation article due to the small numbers of Class 3 patients that were included. 11 The current data are similar to those reported by Kondziolka et al., who found that combined SRS and WBRT controlled disease better and possibly improved survival compared with WBRT alone. 12 This is in contrast to results published by Mori et al., who found that WBRT combined with SRS/surgery offers no benefit. 13 Although our series was not randomized like the series reported by Kondziolka et al., we had similar results and had relatively large numbers of patients who were treated with combined modalities

Survival in Malignant Melanoma Brain Metastases/Buchsbaum et al. 2271 TABLE 3 Univariate Proportional Hazards Analysis of Continuous Variables Affecting Survival in Patients with Brain Metastases from Melanoma Category Coefficient ( ) Standard error Wald Chi-square Hazard ratio (95% CI) P value Age at diagnosis (increase in risk for each year increase) 0.013 0.009 1.885 1.013 (0.994 1.032) 0.1698 KPS at diagnosis (decrease in risk for each 10-point increase) 0.419 0.083 25.286 0.659 (0.560 0.775) 0.0001 a No. of lesions at diagnosis (increase in risk with each additional lesion) 0.037 0.025 2.198 1.038 (0.988 1.090) 0.1382 Greatest dimension of largest lesion (decrease in risk for each additional cm in greatest dimension) 0.196 0.101 3.745 0.822 (0.674 1.003) 0.0530 WBRT fractions (decrease in risk for each number of fraction increase) WBRT alone 0.117 0.070 2.798 0.889 (0.775 1.020) 0.0944 WBRT and local therapy 0.021 0.028 0.565 0.979 (0.927 1.034) 0.4523 95% CI: 95% confidence interval; KPS: Karnofsky performance status; WBRT: whole brain radiotherapy. a Represents data with P 0.05. TABLE 4 Multivariate Proportional Hazards Analysis of Variables Affecting Survival in Patients with Brain Metastases from Melanoma Parameter Coefficient ( ) Standard error Hazard ratio (95% CI) P value Wald globle test (P value) Model 1 Age in yrs ( 65 vs. 65) 0.615 0.339 1.849 (0.952 3.592) 0.0696 0.0001 KPS ( 70 vs. 70) 0.981 0.461 0.375 (0.152 0.926) 0.0335 a Control of primary tumor (no vs. yes) 0.381 0.360 1.464 (0.723 2.963) 0.2891 Systemic (no vs. yes) 0.728 0.338 0.483 (0.249 0.937) 0.0314 a Single (no vs. yes) 0.430 0.341 1.537 (0.830 2.846 0.1715 Treatment group (vs. local therapy and WBRT) No treatment 1.847 0.807 6.342 (1.305 30.822) 0.0220 a WBRT only 0.826 0.368 2.284 (1.111 4.694) 0.0246 a Local therapy (SRS or surgery) only 0.422 0.391 1.525 (0.709 3.281) 0.2805 Model 2 RTOG RPA class (vs. RPA Class 3) 1 1.582 0.602 0.206 (0.063 0.669) 0.0086 a 0.0001 2 0.884 0.437 0.620 (0.175 0.974) 0.0433 a 2 0.499 0.532 0.607 (0.214 1.742) 0.3486 Single (no vs. yes) 0.494 0.310 1.639 (0.893 3.007) 0.1110 Treatment group (vs. local therapy and WBRT) 2.070 0.791 7.928 (1.680 37.409) 0.0089 a No treatment 0.872 0.369 2.392 (1.161 4.929) 0.0180 a WBRT only 0.365 0.407 1.440 (0.648 3.197) 0.3703 Local therapy (SRS or surgery) only 95% CI: 95% confidence interval; KPS: Karnofsky performance status; WBRT: whole brain radiotherapy; SRS: radiosurgery; RTOG: Radiation Therapy Oncology Group; RPA: recursive partitioning analysis. a Represents data with P 0.05. (n 37 patients) using the current technology. We also had more patients with melanoma than their total of five patients (three in the WBRT alone arm and two in the combined arm). Despite concerns regarding the radioresistance of melanoma, WBRT in our patient sample may be important, because it affected subclinical disease when local therapy (SRS/surgery) was used to control areas of macroscopic disease. Looking at WBRT results alone in log-rank fashion (see Table 2), WBRT was significant for survival. However, it is clear from our data that combined therapy was superior to WBRT alone, and the Kaplan Meier curves for survival suggest that combined therapy is better than local therapy alone. Our multivariate data show a difference between combined therapy and local therapy. If surgery and SRS are considered therapies that boost the effectiveness of WBRT, then the combined treatment simply modulates dose relative to tumor cell density. This perspective raises the question of using intensitymodulated radiation therapy (IMRT) in patients who are eligible for local therapy combined with WBRT,

2272 CANCER April 15, 2002 / Volume 94 / Number 8 designing treatment plans that boost the dose to larger lesions to SRS-like doses. It also raises the question regarding the timing of WBRT: up front or in reserve? 14,15 In finding prognostic significance in a population of patients with melanoma, our study addresses previous criticism that the RPA categories are based primarily on patients with carcinoma of the breast and lung. 4,11,16,17 When we performed an additional analysis of our patients using the modification of the RPA proposed by Nieder et al. at the M. D. Anderson Cancer Center, we found that, just as Nieder et al. suggested, RPA Class 2 could be divided into two subgroups according to survival (Fig. 3). Multivariate analysis (Table 4) showed that control of the primary lesion was significant in its effects on survival. This is in agreement with the data reported by Nieder et al., who showed that the RPA classification identified patients in RPA Class 2 who should be treated aggressively. 18 There were seven patients in the no control at primary site category, as shown in Table 1 (RTOG RPA Class 2) and could not be shown to differ from RTOG RPA Class 3 (see Table 4, Fig. 3). Our data do lend credence to the assertion that the survival of patients in RPA Class 2 may be bimodal. CONCLUSIONS The most striking finding in the data is that clinicians as a whole had selected patients for treatment in a fashion that correlated with the RTOG RPA schema. In a similar fashion, at first, the observation is troubling that the patients who survived the longest had an almost 50% cranial failure rate, whereas the other groups had cranial failure rates 20%. This finding may be explained by the fact that patients who survive longer may have more time to develop failure of local control. Alternately, the estimate of the local recurrence rate may have been inflated by the fact that these patients were evaluated more rigorously than other patient groups, such as hospice patients. It is unclear whether WBRT should be held in reserve or given at the same time as local therapy, and this issue needs to be evaluated in a prospective fashion. The RTOG RPA classes are likely to be valid for patients with metastatic melanoma based on our data review. The role of local control of the primary lesion and how it is correlated strongly with RPA Group 2 survival, as discussed by Nieder et al., 18 will need to be examined further in a prospective fashion. The use of IMRT to mimic both WBRT and SRS also is an interesting issue to be addressed in future studies. REFERENCES 1. Barranco S, Romsdahl M, Humphrey R. The radiation response of human malignant melanoma cells grown in vitro. Cancer Res. 1971;31:830. 2. Konefal JB, Emami B, Pilepich MV. Analysis of dose fractionation in the palliation of metastases from malignant melanoma. Cancer. 1988;61(2):243 246. 3. Shiau CY, Sneed PK, Shu HK, et al. Radiosurgery for brain metastases: relationship of dose and pattern of enhancement to local control. Int J Radiat Oncol Biol Phys. 1997; 37(2):375 383. 4. Seung SK, Sneed PK, McDermott MW, et al. Gamma knife radiosurgery for malignant melanoma brain metastases [see comments]. Cancer J Sci Am. 1998;4(2):103 109. 5. Young RF. Radiosurgery for the treatment of brain metastases. Semin Surg Oncol. 1998;14(1):70 78. 6. Carella RJ, Gelber R, Hendrickson F, Berry HC, Cooper JS. Value of radiation therapy in the management of patients with cerebral metastases from malignant melanoma: Radiation Therapy Oncology Group Brain Metastases Study I and II. Cancer. 1980;45(4):679 683. 7. Sampson JHC Jr., Friedman AH, Siegler HF. Demographics, prognosis and therapy in 702 patients with brain metastases from malignant melanoma. J Neurosurg. 1998;88(1):11-20. 8. Wronski M, Arbit E. Surgical treatment of brain metastases from melanoma: a retrospective study of 91 patients. J Neurosurg. 2000;93(1):9 18. 9. Gaspar L, Scott C, Rotman M, et al. Recursive partitioning analysis (RPA) of prognostic factors in three Radiation Therapy Oncology Group (RTOG) brain metastases trials. Int J Radiat Oncol Biol Phys. 1997;37(4):745 751. 10. SAS Institute. JMP, edition 3.2.1. Cary: SAS Institute, 1999. 11. Gaspar LE, Scott C, Murray K, Curran W. Validation of the RTOG recursive partitioning analysis (RPA) classification for brain metastases. Int J Radiat Oncol Biol Phys. 2000;47(4): 1001 1006. 12. Kondziolka D, Patel A, Lunsford LD, Kassam A, Flickinger JC. Stereotactic radiosurgery plus whole brain radiotherapy versus radiotherapy alone for patients with multiple brain metastases. Int J Radiat Oncol Biol Phys. 1999;45(2):427 434. 13. Mori Y, Kondziolka D, Flickinger JC, Kirkwood JM, Agarwala S, Lunsford LD. Stereotactic radiosurgery for cerebral metastatic melanoma: factors affecting local disease control and survival. Int J Radiat Oncol Biol Phys. 1998;42(3):581 589. 14. Chidel MA, Suh JH, Reddy CA, Chao ST, Lundbeck MF, Barnett GH. Application of recursive partitioning analysis and evaluation of the use of whole brain radiation among patients treated with stereotactic radiosurgery for newly diagnosed brain metastases. Int J Radiat Oncol Biol Phys. 2000;47(4):993 999. 15. Chidel MA, Suh JH, Greskovich JF, Kupelian PA, Barnett GH. Treatment outcome for patients with primary nonsmall-cell lung cancer and synchronous brain metastasis. Radiat Oncol Invest. 1999;7(5):313 319. 16. Williams J, Enger C, Wharam M, Tsai D, Brem H. Stereotactic radiosurgery for brain metastases: comparison of lung carcinoma vs. non-lung tumors. J Neurooncol. 1998;37(1): 79 85. 17. Wronski M, Maor MH, Davis BJ, Sawaya R, Levin VA. External radiation of brain metastases from renal carcinoma: a retrospective study of 119 patients from the M. D. Anderson Cancer Center. Int J Radiat Oncol Biol Phys. 1997;37(4):753 759. 18. Nieder C, Nestle U, Motaref B, Walter K, Niewald M, Schnabel K. Prognostic factors in brain metastases: should patients be selected for aggressive treatment according to recursive partitioning analysis (RPA) classes? Int J Radiat Oncol Biol Phys. 2000;46(2):297 302.