analysis showed that surgical resection, site of metastases, number of metastatic lesions ... node or soft tissue involv
122
Prolonged Survival after Complete Resection of Metastases from Intraocular Melanoma Eddy C. Hsueh, M.D. Richard Essner, M.D. Leland J. Foshag, M.D. Xing Ye, M.S. He-Jing Wang, M.S., M.D. Donald L. Morton, M.D. John Wayne Cancer Institute, Saint John’s Health Center, Santa Monica, California.
BACKGROUND. The median survival time is only 2– 6 months after a diagnosis of metastases from intraocular melanoma. Because complete resection of metastatic melanoma from a cutaneous primary tumor can prolong survival, the authors hypothesized that resection also might benefit patients with metastases from an intraocular site. METHODS. From 1971 to 1999, 112 patients with metastatic melanoma from an intraocular site were enrolled in various treatment protocols after informed consent was obtained. Prospectively recorded clinical variables and follow-up information were retrieved from the patient database. Survival curves were estimated using the Kaplan–Meier method. Univariate analysis was performed with the log-rank test. Multivariate analysis was performed using the Cox proportional hazards regression model. Propensity score analysis was used to reduce the imbalance between subgroups and to assess treatment effect. RESULTS. Seventy-eight patients (70%) presented with liver involvement. Twenty-
Presented at the 56th Annual Cancer Symposium of the Society of Surgical Oncology, Los Angeles, California, March 6 –9, 2003. Supported in part by Grants CA87071, CA12582, and CA76489 from the National Cancer Institute and by funding from the Harold J. McAlister Charitable Foundation (Los Angeles, CA). Dr. Hsueh’s clinical research is funded in part by a Career Development Award from the American Society of Clinical Oncology. Address for reprints: Donald L. Morton, M.D., John Wayne Cancer Institute, 2200 Santa Monica Boulevard, Santa Monica, CA 90404; Fax: (310) 4495261; E-mail:
[email protected] Received April 30, 2003; revision received September 10, 2003; accepted September 15, 2003. © 2003 American Cancer Society DOI 10.1002/cncr.11872
four patients (21%) underwent resection of metastatic lesions. At a median followup time of 11 months (range, 1–97 months; ⬎ 36 months for survivors), the median survival period was 11 months and the 5-year survival rate was 7%. Univariate analysis showed that surgical resection, site of metastases, number of metastatic lesions, and disease-free interval were correlated significantly with survival (P ⬍ 0.001, P ⬍ 0.001, P ⬍ 0.001, and P ⫽ 0.031, respectively). Multivariate analysis showed that surgical resection was significant (P ⫽ 0.008) but that the site of metastases was not (P ⫽ 0.146). The median survival and the 5-year survival rate were 38 months and 39%, respectively, for surgical patients, versus 9 months and 0%, respectively, for nonsurgical patients. After adjusting for covariate imbalance by propensity score analysis, surgery remained significant (P ⫽ 0.021) on multivariate analysis. CONCLUSIONS. Complete resection may prolong survival in certain patients with distant metastases from intraocular primary melanoma. However, the overall unfavorable prognosis indicates an urgent need for more effective nonsurgical interventions. Cancer 2004;100:122–9. © 2003 American Cancer Society.
KEYWORDS: prolonged survival, complete resection, intraocular melanoma, metastases.
D
istant metastasis of melanoma is associated with a poor prognosis regardless of the primary site, but the metastatic pattern of an intraocular primary tumor appears to be distinct from that of a cutaneous primary tumor. Fifty to 80% of patients with distant metastases from intraocular melanoma initially present with liver involvement,1,2 whereas approximately 50% of patients with distant metastases from cutaneous melanoma present with distant lymph node or soft tissue involvement and approximately 40% present with pulmonary metastases.3–5 Furthermore, complete resection of disease
Metastatic Intraocular Melanoma/Hsueh et al.
can prolong survival for select patients with metastatic melanoma from a cutaneous primary site.6 –10 Intraocular melanoma is relatively rare. Its estimated incidence is 4 – 8 per million individuals per year.11,12 The average age at diagnosis is 55 years,13 and equal numbers of males and females are affected. The rarity of intraocular melanoma is reflected in the limited number of published studies, and the small size of these studies prevents statistically significant conclusions. Not surprisingly, treatment guidelines for patients with metastases from intraocular melanoma have been based on clinical trials for patients with metastatic melanoma from a cutaneous primary site. To further define the prognostic determinants of intraocular melanoma, we evaluated the outcomes of patients who were treated for metastatic melanoma from an intraocular primary site during the past 30 years at the John Wayne Cancer Institute (Santa Monica, CA). Our goal was to determine whether complete resection might prolong survival, as has been observed for patients undergoing surgical treatment of metastatic melanoma from a cutaneous primary site.
TABLE 1 Demographic Characteristics Characteristic Gender Male Female Age (yrs) ⬍ 60 ⱖ 60 Surgery Yes No Initial metastatic sitea Liver Lung Soft tissue and lymph nodes Other No. of metastatic sites 1 2 3 No. of metastatic lesions 1 2 3–4 ⱖ5 Initial treatment of metastasis Systemic chemotherapy Surgery Liver perfusion None Immunotherapy Other Disease-free interval (yrs) ⬍2y 2–5 y 5–10 y ⬎ 10 y
MATERIALS AND METHODS Patient Population A retrospective chart review was conducted under a protocol approved by the joint Saint John’s Health Center/John Wayne Cancer Institute institutional review board. The John Wayne Cancer Institute melanoma database, which contains prospectively collected data for ⬎ 11,000 patients treated since 1971, was searched for patients with intraocular melanoma. Of the 176 patients identified, 112 were diagnosed with metastatic melanoma on or before December 1999. Excluded from this study group were patients with a conjunctival primary site and patients with a periorbital primary site that extended into the eye.
Statistical Analysis Overall survival (OS) was defined as the period between diagnosis of metastatic melanoma and death. Disease-free interval (DFI) was defined as the period between diagnosis of primary intraocular melanoma and diagnosis of metastatic melanoma. The median DFI was used as a cutoff for statistical analyses. Survival curves were estimated using the nonparametric Kaplan–Meier method. The log-rank test was used for univariate analysis to determine differences between curves. Multivariate analysis by Cox proportional hazards regression used only the variables that were significant in univariate analysis. P ⬍ 0.05 indicated statistical significance. All statistical analyses were two tailed and were performed using
123
a
No. (%)
53 (47) 59 (53) 71 (63) 41 (37) 24 (21) 88 (79) 78 (70) 32 (29) 27 (24) 16 (14) 81 (72) 21 (19) 10 (9) 19 (17) 10 (9) 5 (4) 78 (70) 36 (32) 24 (21) 21 (19) 13 (12) 12 (11) 6 (5) 27 (24) 42 (38) 24 (21) 19 (17)
Patients with multiple metastatic sites were counted multiple times.
Statistical Analysis System software (SAS Institute, Cary, NC). To reduce possible bias due to imbalance of covariates between treatment groups, a propensity score analysis was performed.14 –16 Propensity scores were estimated using logistic regression. All clinical factors and interaction terms were included in the model. Patients with overlapping propensity scores were selected for a subgroup analysis that balanced background characteristics for surgical and nonsurgical treatment. The log-rank test and Cox regression analysis were applied to the subgroups.
RESULTS As shown in Table 1, the study group had a slight female predominance (53%). The median DFI was 45 months and 38% of patients had a DFI that exceeded
124
CANCER January 1, 2004 / Volume 100 / Number 1
TABLE 2 Clinical Characteristics of Surgical and Nonsurgical Groups
Characteristic Gender Male Female Age (yrs) ⱖ 60 ⬍ 60 No. of metastatic sites ⬎1 1 No. of metastatic lesions ⱖ5 1–4 Metastatic site Livera Other Disease-free interval (mos) ⱖ 45 ⬍ 45 a
No. of surgical patients (%)
No. of nonsurgical patients (%)
TABLE 3 Disease Recurrence Pattern of Patients Undergoing Complete Metastasectomy
Initial site of disease
No. of No. of complete disease resections recurrences
Adrenal Bone Stomach Liver Lung
1 1 1 2 6
P value
10 (42) 14 (58)
43 (49) 45 (51)
0.531
7 (29) 17 (71)
34 (39) 54 (61)
0.393
2 (8) 22 (92)
29 (33) 59 (67)
0.017
2 (8) 22 (92)
76 (86) 12 (14)
⬍ 0.001
5 (21) 19 (79)
73 (83) 15 (17)
⬍ 0.001
19 (79) 5 (21)
37 (42) 51 (58)
0.001
Soft tissue/lymph node 10
1 0 1 1 5 8
Site of recurrent disease (n) Adrenal (1) Spleen (1) Liver (1) Lung (2), brain (1), small bowel (1), soft tissue (1) Liver (6), soft tissue (2)
With or without other metastatic sites.
5 years. Seventy-eight patients (70%) had liver involvement when initially diagnosed with metastatic melanoma. Most patients (70%) presented with at least 5 lesions. The median follow-up time was 11 months (range, 1–97 months; ⬎ 36 months for survivors). The median survival period for the entire group was 11 months and the 5-year survival rate was 7%. Of the 112 patients, 24 (21%) underwent surgery for metastatic melanoma. The clinical characteristics of the surgical and nonsurgical groups are shown in Table 2. Of the 24 surgical patients, 21 (88%) underwent complete resection of metastatic melanoma. Complete resection was not possible in the remaining three patients, all of whom had hepatic metastases (one patient had three lesions, and two patients had five lesions). Twelve (50%) surgical patients received postoperative immunotherapy with an investigational therapeutic polyvalent cancer vaccine (Canvaxin vaccine, CancerVax Corp, Carlsbad, CA): 7 (29%) were followed with observation, 3 (13%) received adjuvant systemic chemotherapy, 1 underwent hepatic perfusion with dacarbazine, and 1 received cytokine treatment. The sample size was too small to evaluate the effect of postoperative adjuvant treatment on survival. Table 3 shows the pattern of disease recurrence in the 21 patients who underwent complete resection. Ten patients underwent complete resection of distant soft tissue and lymph node metastases: eight developed
FIGURE 1. Overall survival curves for patients with initial metastases to the liver or other organ sites. The gray line represents patients with initial liver involvement (n ⫽ 78), and the black line represents patients with initial metastases to other organs (n ⫽ 34; P ⬍ 0.001). aListed values indicate number of patients who died during a given period/total number of living patients at the start of that period.
disease recurrence, and six experienced hepatic recurrence. For patients with hepatic metastasis, the median survival time and 5-year survival rate were 9 months and 2%, respectively, compared with 28 months and 22%, respectively, for patients without liver involvement at initial diagnosis (Fig. 1). Patients without liver involvement at initial diagnosis tended to have a longer DFI (P ⫽ 0.057). For patients with 5 or more metastatic lesions on presentation, the median survival time and 5-year survival rate were 8.9 months
Metastatic Intraocular Melanoma/Hsueh et al.
FIGURE 2. Overall survival curves for patients with fewer than or at least five metastatic lesions. The black line represents patients with fewer than 5 metastatic lesions (n ⫽ 34), and the gray line represents patients with at least 5 lesions (n ⫽ 78; P ⬍ 0.001). aListed values indicate number of patients who died during a given period/total number of living patients at the start of that period. and 0%, respectively, compared with 27.7 months and 27.7%, respectively, for patients with fewer than 5 metastatic lesions on presentation (Fig. 2). For patients with a DFI ⬍ 45 months, the median survival time and 5-year survival rate were 10 months and 0%, respectively, compared with 11 months and 14%, respectively, for patients with a DFI ⱖ 45 months (Fig. 3). Univariate analysis demonstrated that surgical treatment, liver metastasis, number of metastatic lesions, and DFI were significant prognostic variables for survival (P ⬍ 0.001, P ⬍ 0.001, P ⬍ 0.001, and P ⫽ 0.031, respectively; Table 4). Age, gender, and number of metastatic sites were not significant. The median survival time and 5-year survival rate for surgical patients were 38 months and 39%, respectively, compared with 9 months and 0%, respectively, for nonsurgical patients (Fig. 4). As expected, patients undergoing surgical treatment of metastatic melanoma had a higher proportion of soft tissue/distant lymph node metastases (P ⬍ 0.001), fewer metastatic lesions (P ⬍ 0.001), a longer DFI (P ⫽ 0.001), and isolated organ sites of involvement (P ⫽ 0.017; Table 2). Therefore, the favorable prognosis of patients undergoing surgery as primary treatment for their metastatic intraocular melanoma
125
FIGURE 3.
Overall survival curves for patients with a longer or shorter disease-free interval (DFI). The black line represents patients with a DFI ⱖ 45 months (n ⫽ 56), and the gray line represents patients with DFI ⬍ 45 months (n ⫽ 56; P ⫽ 0.031). aListed values indicate number of patients who died during a given period/total number of living patients at the start of that period.
may reflect a selection bias. However, multivariate analysis indicated that considering all important clinical variables and interactions among the variables, surgical treatment was the only significant prognostic variable for survival in this cohort (P ⫽ 0.008). Multivariate analysis demonstrated that liver involvement at initial presentation was not significant (P ⫽ 0.146; Table 4). Due to the difference in covariates (metastatic site, number of metastatic sites, number of metastatic lesions, and DFI) between the surgical and nonsurgical groups, multivariate analysis may not be reliable for determining treatment effect. Therefore, propensity score analysis was used to balance the covariates and to reduce bias. Each patient was assigned a propensity score based on age, gender, location, number of metastatic sites, number of lesions, and DFI as estimated by a logistic regression model. Because prognostic factors were not distributed evenly between the two treatment groups, the propensity scores were very different. For example, the mean ⫾ standard deviation and median scores were 0.07 ⫾ 0.17 and 0.004, respectively, for the nonsurgical group, compared with 0.74 ⫾ 0.27 and 0.86, respectively, for the surgical group. Thirty-four patients (21 nonsurgical patients and 13 surgical patients) had overlapping propensity
126
CANCER January 1, 2004 / Volume 100 / Number 1
TABLE 4 Statistical Impact of Clinical Variables on Overall Survival Univariate analysis
Characteristics Gender Male Female No. of metastatic sites ⬎1 1 Age (yrs) ⱖ 60 ⬍ 60 Disease-free interval (mos) ⱖ 45
No. of patients
P value (logrank test)
53 59
0.551
8.5 7.5
31 81
0.524
4.3 8.5
41 71
0.294
4.9 8.8
56
0.031
14.1
⬍ 45 No. of metastatic lesions ⱖ5
56
1–4 Metastatic site Livera
34
Other Surgery Yes
34 24
No
88
78
78
5-year OS (%)
Multivariate analysis P value (Wald test)
Hazard ratio (95% confidence interval)
0.986
1.00 (reference range, 0.66–1.53)
0.329
1.37 (reference range, 0.73–2.57)
0.146
1.46 (reference range, 0.88–2.44)
0.008
0.34 (reference range, 0.15–0.76)
0 ⬍ 0.001
0 27.7
⬍ 0.001
1.5 21.5
⬍ 0.001
39.3 0
OS: overall survival. a With or without other metastatic sites.
scores and were selected for subgroup analysis (Table 5). The log-rank test demonstrated a significant difference between the surgical and nonsurgical subgroups (P ⫽ 0.001). Multivariate analysis using Cox proportional hazards regression showed that surgery remained a significant predictor for OS (P ⫽ 0.012). Liver involvement approached statistical significance (P ⫽ 0.057; Table 6). Propensity scores for patients in the surgical and nonsurgical subgroups were estimated by a logistic model. Cox proportional hazards regression, with adjustment for the propensity scores, confirmed a significant survival benefit for surgical treatment. The hazard ratio was 0.28 (95% confidence interval, 0.10 – 0.83) and the P value was 0.021.
DISCUSSION Several studies have demonstrated a grim prognosis for patients with metastatic melanoma from an intraocular site versus a cutaneous primary site.17–20 However, because most of these studies are limited to patients enrolled in chemotherapy trials or patients with disseminated metastasis involving the liver, the
number and outcome of patients with resectable disease are not clear. Reports of surgical intervention for Stage IV melanoma include very few patients with an intraocular primary site. Feun et al.6 reported only 3 cases among 102 patients, and Karakousis et al.8 reported only 5 cases among 114 patients. Autopsy series have demonstrated liver involvement in almost all patients who died of the disease,21 but no studies with adequate sample size have evaluated the initial presentation and treatment of metastatic intraocular melanoma. Therefore, the role of surgery for metastatic melanoma from an intraocular primary site cannot be assessed from the current literature. In the current study, the median survival and 5-year OS rate of 38 months and 39%, respectively, for the surgical group were much higher than the 18 months and 18%, respectively, reported after surgical resection of metastatic melanoma from cutaneous primary sites.6 –10 This difference probably reflects different sample sizes. However, our findings suggest a role for metastasectomy in the treatment of melanoma from an intraocular primary site. Although only 21% of
Metastatic Intraocular Melanoma/Hsueh et al.
127
TABLE 5 Subgroups of Surgical and Nonsurgical Patients Who Had Overlapping Propensity Scores
Characteristic
FIGURE 4. Overall survival curves for surgical and nonsurgical patients. The black line represents surgical patients (n ⫽ 24), and the gray line represents nonsurgical patients (n ⫽ 88; P ⬍0.001). aListed values indicate number of patients who died during a given period/total number of living patients at the start of that period. our study population underwent surgery, the surgical cohort had a significantly longer OS than the nonsurgical cohort. Thirty percent of patients presented with metastasis to nonhepatic sites. These patients were more likely to undergo surgical resection (56% [19 of 34]) than were patients with liver involvement (6% [5 of 78]). This finding is not surprising, because patients without liver involvement were more likely to have less extensive metastasis. Ninety-two percent of patients without liver involvement had fewer than 5 metastatic lesions, compared with 14% of patients with liver involvement. We found that the DFI was a statistically significant prognostic factor on univariate analysis. A long DFI between diagnosis of the primary tumor and detection of metastatic disease probably indicates a slow-growing and less aggressive tumor with a more favorable postoperative prognosis. In contrast, a short DFI probably would correspond to aggressive tumor growth and poorer prognosis. A patient whose intraocular melanoma has metastasized to the liver may be a candidate for surgery if DFI is ⬎ 45 months, assuming that this patient has fewer than 5 metastatic lesions and no other evidence of disease on a full-body metastatic workup. The use of propensity score analysis merits further
Gender Male Female Age (yrs) Mean ⫾ SD Median (range) No. of metastatic sites 1 2 No. of metastatic lesions 1 2 3 4 5 Liver involvement No Yes Disease-free interval (mos) Mean ⫾ SD Median (range)
Surgical patients (n ⴝ 13)
Nonsurgical patients (n ⴝ 21)
7 6
7 14
0.238a
52.5 ⫾ 10.8 53.7 (30.9–66.0)
54.1 ⫾ 12.5 53.1 (32.5–84.9)
1.000b
11 2
18 3
0.930a
9 1 1 0 2
6 3 1 1 10
0.029c
9 4
10 11
0.217a
105.4 ⫾ 81.7 96.8 (13.1–300.4)
83.3 ⫾ 66.8 67.9 (5.1–248.1)
0.483b
P value
SD: standard deviation. a Chi-square test. b t test. c Wilcoxon rank sum test.
TABLE 6 Multivariate Analysis of Surgical and Nonsurgical Subgroups with Overlapping Propensity Scores Characteristic
P value
Hazard ratio (95% confidence interval)
Age No. of metastatic lesions DFI Liver involvement Surgery
0.978 0.643 0.560 0.057 0.012
1.00 (0.97–1.03) 1.06 (0.82–1.39) 1.00 (0.99–1.00) 2.26 (0.98–5.24) 0.27 (0.10–0.75)
DFI: disease-free interval.
discussion. The propensity score can reduce bias for observed covariates.14 –16 If the covariates that are believed to be related to the treatment assignment can be measured, then the propensity score method can produce approximately unbiased estimates of treatment effect. When two groups of patients have similar propensity scores, they tend to have balanced prognostic factors. In our observational study, patients with favorable prognostic factors were more likely to undergo
128
CANCER January 1, 2004 / Volume 100 / Number 1
surgical treatment. This selection bias could skew an estimate of treatment effect. Because a traditional multivariate analysis (such as the Cox model) might not eliminate this bias, we used a propensity score to determine the conditional probability of being treated with surgery given the observed covariates. Most patients with very low propensity scores were in the surgical group and had better prognostic factors, whereas patients with very high propensity scores were in the nonsurgical group and had poor prognostic factors. The propensity scores ranged from 0.011 to 0.950 for the surgical cohort and from 0.141 to 0.999 for the nonsurgical cohort. Subsequently, patients whose propensity scores were ⬎ 0.950 or ⬍ 0.141 were excluded. The propensity scores for the two treatment groups overlapped for the remaining patients. The background characteristics of these two groups were more similar, allowing an unbiased estimate of treatment effect. When propensity scores were reestimated for this subset of patients, the new propensity scores were similarly distributed between the two treatment groups. There was no specific assumption for the propensity score method. The analysis again showed a significant survival benefit for the surgical group. There were no 5-year survivors in the nonsurgical group. This finding may be due to ineffective treatment regimens for this disease. Alexander et al.22 reported encouraging results for isolated hepatic perfusion of melphalan with or without tumor necrosis factor. In their study of 21 patients, the response rate of 62% included 2 complete responses. Chemobiotherapy (intrahepatic or intraportal fotemustine followed by interleukin-2 and interferon) also yielded significant response rates in patients with ocular melanoma metastatic to the liver.23 Because of the poor prognosis of patients with metastatic ocular melanoma, these strategies may merit evaluation as potential adjuvant therapy for patients whose intraocular primary site is a highrisk factor for metastasis. The retrospective nature of the current study prevents any definitive recommendation for surgery as a treatment alternative in patients with metastatic intraocular melanoma. The observation of long-term survival in the surgical group but not in the nonsurgical group strongly supports the use of surgery as a part of the treatment regimen for patients with this disease. Nonetheless, even if surgery can prolong survival in a select subset of patients, effective systemic treatment is needed urgently. One-half of the surgical patients in the current study received postsurgical adjuvant immunotherapy, but its impact on survival
could not be analyzed due to the limited number of patients. Further study is needed to optimize the postoperative treatment of patients whose intraocular melanoma has metastasized to hepatic and/or nonhepatic sites.
REFERENCES 1.
2. 3.
4.
5.
6.
7.
8. 9.
10.
11.
12.
13.
14.
15.
16.
17.
Lorigan JG, Wallace S, Mavligit GM. The prevalence and location of metastases from ocular melanoma: imaging study in 110 patients. AJR Am J Roentgenol. 1991;157:1279 – 1281. Rajpal S, Moore R, Karakousis CP. Survival in metastatic ocular melanoma. Cancer. 1983;52:334 –336. Barth A, Wanek LA, Morton DL. Prognostic factors in 1,521 melanoma patients with distant metastases. J Am Coll Surg. 1995;181:193–201. Balch CM, Soong SJ, Murad TM, et al. A multifactorial analysis of melanoma. IV. Prognostic factors in 200 melanoma patients with distant metastases (Stage III). J Clin Oncol. 1983;1:126 –134. Sirott MN, Bajorin DF, Wong GY, et al. Prognostic factors in patients with metastatic malignant melanoma. A multivariate analysis. Cancer. 1993;72:3091–3098. Feun LG, Gutterman J, Burgess MA, et al. The natural history of resectable metastatic melanoma (Stage IVA melanoma). Cancer. 1982;50:1656 –1663. Harpole DH Jr., Johnson CM, Wolfe WG, George SL, Seigler HF. Analysis of 945 cases of pulmonary metastatic melanoma. J Thorac Cardiovasc Surg. 1992;103:743–748. Karakousis CP, Velez A, Driscoll BA. Metastasectomy in malignant melanoma. Surgery. 1994;115:295–302. Ollila DW, Essner R, Wanek LA, et al. Surgical resection for melanoma metastatic to the gastrointestinal tract. Arch Surg. 1996;131:975–980. Hsueh EC, Essner R, Foshag LJ, et al. Prolonged survival after complete resection of disseminated melanoma and active immunotherapy with a therapeutic cancer vaccine. J Clin Oncol. 2002;20:4549 – 4554. Mahoney MC, Burnett WS, Majerovics A, Tanenbaum H. The epidemiology of ophthalmic malignancies in New York State. Ophthalmology. 1990;97:1143–1147. Huerta C, Rodriguez LA. Incidence of ocular melanoma in the general population and in glaucoma patients. J Epidemiol Community Health. 2001;55:338 –339. Grin JM, Grant-Kels JM, Grin CM, Berke A, Kels BD. Ocular melanomas and melanocytic lesions of the eye. J Am Acad Dermatol. 1998;38:716 –730. Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for casual effects. Biometrika. 1983;70:41–55. Rosenbaum PR, Rubin DB. Reducing bias in observational studies using subclassification on the propensity score. J Am Stat Assoc. 1984;79:516 –524. D’Agostino RB Jr. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med. 1998;17:2265–2281. Flaherty LE, Unger JM, Liu PY, et al. Metastatic melanoma from intraocular primary tumors: the Southwest Oncology Group experience in Phase II advanced melanoma clinical trials. Am J Clin Oncol. 1998;21:568 –572.
Metastatic Intraocular Melanoma/Hsueh et al. 18. Bedikian AY, Legha SS, Mavligit G, et al. Treatment of uveal melanoma metastatic to the liver: a review of the M. D. Anderson Cancer Center experience and prognostic factors. Cancer. 1995;76:1665–1670. 19. Rajpal S, Moore R, Karakousis CP. Survival in metastatic ocular melanoma. Cancer. 1983;52:334 –336. 20. Leyvraz S, Spataro V, Bauer J, et al. Treatment of ocular melanoma metastatic to the liver by hepatic arterial chemotherapy. J Clin Oncol. 1997;15:2589 –2595. 21. The Collaborative Ocular Melanoma Study Group. Assessment of metastatic disease status at death in 435 patients
129
with large choroidal melanoma in the Collaborative Ocular Melanoma Study (COMS): COMS Report No. 15. Arch Ophthalmol. 2001;119:670 – 676. 22. Alexander HR, Libutti SK, Bartlett DL, et al. A Phase I-II study of isolated hepatic perfusion using melphalan with or without tumor necrosis factor for patients with ocular melanoma metastatic to liver. Clin Cancer Res. 2000;6:3062–3070. 23. Becker JC, Terheyden P, Kampgen E, et al. Treatment of disseminated ocular melanoma with sequential fotemustine, interferon ␣, and interleukin 2. Br J Cancer. 2002;87: 840 – 845.