Malignant melanoma of the uvea disseminates purely hematogenously, Very Long-Term Prognosis of Patients with Malignant Uveal Melanoma

Transcription

1 Very Long-Term Prognosis of Patients with Malignant Uveal Melanoma Emma Kujala, 1 Teemu Mäkitie, 2 and Tero Kivelä 1,2 From the 1 Ocular Oncology Service and 2 Ophthalmic Pathology Laboratory, Department of Ophthalmology, Helsinki University Central Hospital, Helsinki, Finland. Supported by research grants from the Eye Foundation, the Eye and Tissue Bank Foundation, the Research Foundation of the Orion Corporation, the Instrumentarium Science Foundation, the Lilly Foundation, the Sigrid Jusélius Foundation, the Finnish Medical Foundation, and Helsinki University Central Hospital Research Fund Grants TYH1217 and TYH3203. Submitted for publication May 28, 2003; revised July 8, 2003; accepted July 30, Disclosure: E. Kujala, None; T. Mäkitie, None; T. Kivelä, None The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked advertisement in accordance with 18 U.S.C solely to indicate this fact. Corresponding author: Emma Kujala, Department of Ophthalmology, Helsinki University Central Hospital, Haartmaninkatu 4 C, PL 220, FIN HUS, Helsinki, Finland; emma.kujala@hus.fi. PURPOSE. To investigate the very long-term prognosis of patients with uveal melanoma and the clinical characteristics influencing it. METHODS. Charts, registry data, and histopathologic specimens of 289 consecutive patients with choroidal and ciliary body melanoma treated in the district of the Helsinki University Central Hospital, Finland, between 1962 and 1981 were audited. Definitions for coding the cause of death were adapted from the Collaborative Ocular Melanoma Study (COMS). Competing risks were taken into account by using cumulative incidence analysis and competing risks regression. RESULTS. Of the 289 patients treated, 239 were deceased at the end of follow-up. The audited cause of death was uveal melanoma in 145 (61%) of them. The median follow-up of the 50 survivors was 28 years. The original histopathologic diagnosis of metastasis and second cancer was correct in 91% of all specimens, but immunohistochemical reassessment changed 10% of biopsy and 7% of autopsy diagnoses. Of 45 positive autopsies, 18% were performed without suspicion of melanoma. Uveal melanoma related mortality was 31% (95% confidence interval [CI], 26 37) by 5 years, 45% (95% CI, 40 51) by 15 years, 49% (95% CI, 43 55) by 25 years, and 52% (95% CI, 45 58) by 35 years, according to cumulative incidence analysis. Of patients who died of uveal melanoma, 62%, 90%, 98%, and 100% did so within 5, 15, 25, and 35 years, respectively. Between 15 and 35 years, 20% to 33% of deaths were still due to uveal melanoma. By competing risks regression analysis, the hazard ratio was 1.08 (P ) for each millimeter increase in tumor diameter, 2.27 (P ) for extraocular growth, and 1.89 (P ) for ciliary body involvement. CONCLUSIONS. Metastatic uveal melanoma was the leading single cause of death throughout the study. Cumulative incidences provide a sound basis for patient counseling and design of trials. (Invest Ophthalmol Vis Sci. 2003;44: ) DOI: /iovs Malignant melanoma of the uvea disseminates purely hematogenously, unless it perforates the sclera and infiltrates the conjunctival lymphatics. 1,2 Local recurrence is infrequent, but this cancer often metastasizes before it is diagnosed. 3 5 In fact, approximately 50% of patients with uveal melanoma are thought to die within 10 years after diagnosis, irrespective of the type of treatment. 6 Several case reports describe supposedly unusual malignant uveal melanomas that showed clinical metastasis more than 20 years after enucleation These isolated reports cannot be used to quantitate the frequency of such a very late appearance of progressive metastasis. Few studies report survival data beyond 15 years, and the number of patients who remain under review at that time typically is low In published Kaplan-Meier curves of melanoma-related mortality, latest deaths tend to occur between 10 and 18 years after diagnosis. 11,18,19 The reasons for the scarcity of long-term survival data are that patients with uveal melanoma are mostly middle-aged or older, their life expectancy may be limited because of other illnesses, follow-up data of patients treated long ago are difficult to collect, the founding of national cancer registries is fairly recent, and patients may be difficult to trace long after treatment. We determined the very long-term prognosis, defined as survival 15 years or more after treatment of a primary choroidal and ciliary body melanoma, in a consecutive series of patients from a defined region. Cause of death was audited by using cancer registry data and the patients charts and by immunohistochemical restaining of histopathologic specimens. We also tested statistically the sufficiency of follow-up to conclude whether cured patients were present in the data set. Cumulative incidence analysis and competing risks proportional hazards regression were used to estimate survival proportions most relevant for counseling of patients, most of whom are currently managed conservatively without access to histopathology. PATIENTS AND METHODS Purpose of the Study The primary goal was to determine cause-specific cumulative incidence of death 15 years or more after radical surgery in patients with primary melanoma of the choroid and ciliary body and the proportion of deaths due to melanoma. Secondary goals were to evaluate clinical characteristics associated with death and to compare estimates with those obtained by the Kaplan-Meier method. Eligibility Criteria and Study Design Eligible for analysis in the study were patients who had choroidal and ciliary body melanoma managed a minimum of 20 years earlier. Iris melanomas were ineligible. This investigation was approved by the Institutional Review Board and adhered to the tenets of The Declaration of Helsinki. Files of the Ophthalmic Pathology Laboratory, Department of Ophthalmology, Helsinki University Central Hospital, were searched from May 1962, when it was founded, to December 1981, to enroll all consecutive patients who underwent enucleation or exenteration. During this period, radical surgery was standard treatment for all but the smallest melanomas, which were observed for growth, and all eyes Investigative Ophthalmology & Visual Science, November 2003, Vol. 44, No. 11 Copyright Association for Research in Vision and Ophthalmology 4651

2 4652 Kujala et al. IOVS, November 2003, Vol. 44, No. 11 removed in the district were submitted to this laboratory. Two tumors diagnosed at autopsy and one rediagnosed as a nevus were excluded. Largest basal tumor diameter (LBD) and tumor height were measured from the sections or recorded from pathology reports. Follow-Up and Assessment of Cause of Death A total of 240 patients had died by December 31, Complete data for all patients, except one whose identity could not be verified, were obtained from the Finnish Population and Cancer Registries and all hospitals and histopathology laboratories that had participated in management of uveal melanoma, its metastases, and other malignant tumors. Data for one patient who had emigrated was tracked manually. Records concerning terminal illness were obtained. Death certificates and histopathologic specimens were evaluated by consensus of two investigators to ascertain whether metastatic melanoma was present. Definitions from the Collaborative Ocular Melanoma Study (COMS) were adapted. 20,21 If a specimen represented melanoma metastasis by review or it could not be retrieved but the original report unquestionably mentioned moderate to heavy melanin or HMB-45 immunopositivity, the code was dead with melanoma metastasis (confirmed metastasis). If the report did not mention either characteristic or if only a fine-needle aspiration biopsy had been performed and clinical findings (e.g., hepatomegaly, elevated liver function tests, and abnormal liver imaging) were consistent with hepatic metastases, the code ended suspected melanoma metastasis. If the death certificate specified melanoma, but clinical data were uninformative or the diagnosis was other than cancer when symptoms and clinical findings were consistent with hepatic metastases, it ended possible melanoma metastasis. If a specimen represented another malignancy by review, the code was malignant tumor present, not metastatic melanoma (confirmed second cancer). If there was evidence of a second primary cancer and the clinical course did not suggest hepatic metastasis, the coding ended suspected second cancer, and if the death certificate specified other cancer but clinical data were uninformative it ended possible second cancer. The code was no evidence of malignancy (confirmed nonneoplastic disease) if the autopsy revealed no cancer. If the death certificate specified disease other than cancer, the patient was not registered in the cancer registry for a second cancer, the patient s charts were consistent with the given diagnosis, and liver imagining and liver function test results or both were normal within 6 months before death, the code ended suspected nonneoplastic disease. If neither test had been conducted within 6 months of death, the code ended possible nonneoplastic disease. Insufficient evidence to establish presence of malignancy was used if original charts could not be retrieved. Statistical Analysis Analyses were performed on computer (Stata, ver. 7.0; Stata Software, College Station, TX, and R, ver ; available at org/ provided by The R Foundation for Statistical Computing, Vienna, Austria) software. All probabilities were two-sided, and P 0.05 was considered significant. Univariate analysis of survival was based on the cumulative incidence method, which appropriately handles failures from competing risks. 22 This is mandatory when long-term survival is evaluated, because competing events increase with follow-up as patients become older. In the calculation of cumulative incidence, only patients alive at the study s termination and those lost to follow-up were censored. 22 We estimated mortality related to melanoma, second cancer, and nonneoplastic disease. Cumulative incidence between categories was compared with the Gray s K-sample test. 23 Age was divided in quartiles and LBD in three categories ( 10, 10 15, and 16 mm). For comparison, the Kaplan-Meier estimate 24 of melanoma-related mortality was calculated. In the calculation of the Kaplan-Meier estimate, patients who die of unrelated causes are also censored. Deaths that occur after the first competing risk event contribute more to the estimate than is appropriate, and the melanoma-related mortality is overestimated. 22 The magnitude of the discrepancy depends on the timing and number of competing risk events. Survival curves were plotted to show mortality rather than survival to ease comparison with cumulative incidence. 24 Evidence for the presence of cured ( immune ) patients in the data was tested according to Maller and Zhou. 25,26 The null hypothesis was that there are no cured patients. Reverse censoring suggested that censoring distribution was uniform, and the corresponding statistical Table A.2 was used. 25 If the null hypothesis is rejected, cured patients either are present or follow-up is insufficient to be decisive. 25 The sufficiency of follow-up was tested by the q n test of Maller and Zhou, 25,26 which is based on the distance between the largest censored and uncensored failure time. Both tests are nonparametric and make no assumption of the type and shape of the survival distribution. 25,27 Multivariate analysis of melanoma-related survival was based on competing risks proportional hazards regression. 28 Because cumulative incidence analysis suggested that mortality would not differ within the two lowest and two highest age quartiles, age at diagnosis was dichotomized according to its median. Tumor dimensions were modeled as continuous variables. Models were compared with each other by using the deviance test. 29 For comparison, the hazard rate (HR) was calculated by using the more commonly applied Cox proportional hazards regression, 30 in which melanoma-related deaths after the first competing risk event contribute more to the statistics than is appropriate. RESULTS All 289 patients enrolled were white and 47% were male. Their median age was 57 years (range, 13 97). The cumulative frequency of having a primary choroidal and ciliary body melanoma diagnosed increased by 5% per decade between the ages of 20 and 40 years; the rate changed between 40 and 45 years of age, after which 25% of cases were diagnosed per decade until 75 years of age (Fig. 1A). Less than 10% of all patients were diagnosed after the age of 75 years. The median age at death of the 239 (83%) patients who died was 69 years (range, 32 98; Fig. 1A). The median follow-up of the 50 survivors was 28 years (range, 21 39). One patient was lost to follow-up (Table 1). Of the primary tumors, 221 (76%) were limited to the choroid, and 68 involved the ciliary body. Extrascleral growth was present in 29 (10%) eyes. The median tumor height and LBD were 7 mm (range, 1 20) and 13 mm (range, 3 25), respectively, for the 274 tumors for which these measures were known (Fig. 1B). Biopsy and Autopsy Histopathology A surgical or core needle biopsy to diagnose metastases or second primary cancer was performed in 77 (27%) patients. The original report identified 54 (70%) as melanoma metastases, 22 (29%) as second cancer, and 1 (1%) as nonneoplastic. Autopsy was performed in 70 (29%) of the 239 deceased patients. According to the autopsy report, 25 (36%) died without malignancy. Metastatic melanoma was recorded in 42 (60%) and second cancer in 3 (4%) autopsies. Reassessment of 60 biopsy specimens showed the diagnosis to be incorrect in 6 (10%; 95% CI, 8 29). Five were amelanotic, epithelioid cell melanoma metastases to the liver and not hepatocellular and metastatic colonic carcinoma, and one was an anaplastic glioma. The diagnosis in 3 (7%; 95% CI, 1 18) of 45 autopsies in which a malignancy was found was incorrect. One presumed cholangiocarcinoma was in fact metastatic uveal melanoma and two presumed melanoma metastases were mis-

3 IOVS, November 2003, Vol. 44, No. 11 Very Long-Term Prognosis of Uveal Melanoma 4653 autopsies and 5% (95% CI, 2 10) of all deaths due to uveal melanoma. FIGURE 1. Cumulative frequency distribution plot of age at surgery for the primary tumor and age at death (A), and a jittered scatterplot of the height and largest basal diameter of the primary tumor (B) in 289 patients with choroidal and ciliary body melanoma. diagnosed metastatic mucocellular and anaplastic carcinoma of unknown origin. Of 40 autopsies in which metastatic uveal melanoma was confirmed, 7 (18%; 95% CI, 7 33) were performed without suspicion of melanoma metastasis (e.g., because of presumed cerebrovascular accident, renal failure after hip surgery, and second cancer). This represented 10% (95% CI, 4 20) of all Audited Cause of Death The cause of death on the death certificate was melanoma in 132 patients (55%), second cancer in 27 (11%), and nonneoplastic disease in 79 (33%; Table 1). The audited cause of death was uveal melanoma in 145 patients (61%), second cancer in 17 (7%), and no evidence of malignancy in 75 (31%). Of the original diagnoses of metastatic melanoma and second cancer, 98% and 52%, respectively, were correct (Table 1). The audited cause of death was considered confirmed in 128 (54%) patients, suspected in 28 (12%), and possible in 63 (23%; Table 1). Of melanoma metastases and second cancers, 63% and 65%, respectively, were confirmed. All-Cause Mortality For all-cause mortality, Kaplan-Meier (Fig. 2A) and cumulative incidence estimates are identical. 22 The all-cause mortality by 15, 25, and 35 years was 65%, 79%, and 88%, respectively, Table 2 shows the corresponding survival estimates. Melanoma-Related Mortality The Kaplan-Meier method progressively exaggerated melanoma-related mortality by 5, 7, and 10 percentage points at 15, 25, and 35 years, respectively (Fig. 2A, Table 2). The cumulative incidence estimate increased until 34 years from surgery to 45% (95% CI, 40 51) by 15 years, 49% (95% CI, 43 55) by 25 years, and 52% (95% CI, 45 58) by 35 years (Fig 2B; Table 2 gives the corresponding survival estimates). Cured individuals may have been present among survivors (P 0.01), but the odds leaned toward insufficient follow-up (P 0.10, Table 3). Of the 145 patients who died of uveal melanoma, 62% (95% CI, 54 70) did so within 5 years, 80% (95% CI, 73 86) within 10 years, 90% (95% CI, 84 95) within 15 years, 98% (95% CI, ) within 25 years, and 100% within 35 years (95% CI, ). Competing Causes of Death The cumulative incidence of dying of a second cancer or nonneoplastic disease by 25 years was 6% (95% CI, 3 8) and 24% (95% CI, 19 29), respectively (Fig. 3A, Table 2 gives the corresponding survival estimates). The Kaplan-Meier method exaggerated mortality by 6 and 21 percentage points, respectively (Table 2). The proportion of melanoma-related deaths was on average 83% during the first 5 years after surgery (Table 4). Between 10 TABLE 1. Audited Cause of Death in 239 Patients with Choroidal and Ciliary Body Melanoma Melanoma Metastasis Second Cancer No Evidence of Malignancy Undefined All Deaths n % n % n % n % n % Certainty of coding Confirmed Suspected Possible Insufficient evidence Cause of death in death certificate Melanoma metastasis Second cancer No evidence of malignancy Unknown Total audited

4 4654 Kujala et al. IOVS, November 2003, Vol. 44, No. 11 FIGURE 2. Kaplan-Meier (A, C, E) and cumulative incidence (B, D, F, G) estimates of melanoma-related mortality, and cumulative incidences of dying of second cancer and nonneoplastic disease (B, D, F, G) according to characteristics of the patients treated. Mortality is plotted for all patients (A, B) and according to gender (C, D) and age at surgery (E G). Data below the graphs correspond to patients at risk in each category. Note, for example, that the cumulative incidence of dying of melanoma by 35 years was 2.8 times higher than the cumulative incidence of dying of nonneoplastic disease for women but only 1.3 times higher for men (D). Crossbars: 95% confidence intervals; ticks: censored observations. and 14 years after diagnosis, the proportion was 43% (95% CI, 26 61), and thereafter it fluctuated between 33% and 20%. The proportion of deaths due to nonneoplastic disease increased from 14% during the first 5 years to between 62% and 70% after 14 years (Table 4). Death caused by a second primary cancer did not increase. Seven times and two times as many patients died of melanoma as of a second cancer or nonneoplastic disease, respectively. Mortality According to Clinical Covariates The Kaplan-Meier (Fig. 2C) and cumulative incidence estimate of melanoma-related death was comparable for males and fe-

5 IOVS, November 2003, Vol. 44, No. 11 Very Long-Term Prognosis of Uveal Melanoma 4655 TABLE 2. All-Cause and Cause-Specific Survival Proportions of 289 Patients with Choroidal and Ciliary Body Melanoma All-Cause Melanoma Metastasis Second Cancer No Evidence of Malignancy Time from Surgery (y) Survival 95% CI Survival 95% CI Survival 95% CI Survival 95% CI Cumulative incidence estimate Kaplan-Meier estimate males (45% vs. 53% at 25 years, P 0.16, Gray s K-sample test, difference between curves; Fig. 2D). The former method exaggerated especially male mortality (9 vs. 6 percentage points). Mortality due to second cancer was also comparable (P 0.15), whereas males were at higher risk of dying of nonneoplastic disease (28% vs. 20%, P 0.020; Fig. 2D). The Kaplan-Meier method grossly exaggerated mortality of the highest age quartile (2, 6, 5, and 17 percentage points per quartile, Fig. 2E). The cumulative incidence estimate was comparable for the two lowest (42% vs. 40% at 20 years) and highest age quartiles (56% vs. 56%), suggesting change around the median age (57 years; P 0.050; Fig. 2F). In all quartiles in which patients survived beyond 25 years, melanoma deaths still occurred. Death of nonneoplastic disease became increasingly common with age (5%, 18%, 24%, and 38%, respectively; P ; Fig. 2G). By tumor location, the Kaplan-Meier method exaggerated mortality by 6 and 7 percentage points, respectively (Fig. 3A). Melanoma-related deaths were more frequent (71% vs. 43% at 25 years, P ; Fig. 3B) and noncancer deaths less frequent (P ) when the tumor involved the ciliary body than when it was limited to the choroid. After 20 years, melanoma-related deaths occurred only among patients with choroidal melanoma. With extraocular extension, Kaplan- Meier estimates were 7 and 13 percentage points higher than cumulative incidences, respectively (Fig. 3C). Melanoma-related deaths were more frequent if extraocular extension was present (72% vs. 46%; P , Fig. 3D). The Kaplan-Meier method exaggerated mortality by tumor size by 1, 9, and 8 percentage points, respectively (Fig. 3E). The cumulative incidence estimates increased with increasing LBD (18%, 52%, 59% for small, medium-sized, and large tumors at 25 years, respectively, Fig. 3F; P ). Noncancer mortality was not associated with tumor size (P 0.80, Fig. 3G). Cured patients were likely to be present in all subgroups by tumor location, extrascleral growth, and tumor size (Table 3). There was sufficient follow-up to be confident of their presence among those who had extrascleral extension (P 0.95) and nearly enough among those in whom the tumor involved the ciliary body (Table 3). Follow-up was insufficient (P 0.10) in patients with small and large melanomas for us to be confident of the presence of cured patients. TABLE 3. Test for Immunes and Sufficient Follow-Up to be Confident of Their Presence Among 289 Patients with Choroidal and Ciliary Body Melanoma Presence of Immunes* Sufficient Follow-Up B p n Critical Value P q n Value Critical P All patients Tumor location Choroidal Ciliary body involved Extrascleral growth No Yes 4/ Largest basal diameter 10 mm mm mm 2/ * When P 0.10 there is evidence for presence of cured patients in the population. When P 0.10 and P 0.90 there is evidence against and for sufficient follow-up, respectively. Mean of adjacent Tables B of Maller and Zhou 25 (B fell between two tables). Mean of adjacent columns in Table A.2 of Maller and Zhou 25 (sample size fell between two columns).

6 4656 Kujala et al. IOVS, November 2003, Vol. 44, No. 11 FIGURE 3. Kaplan-Meier (A, C, E) and cumulative incidence (B, D, F, G) estimates of melanoma-related mortality, and cumulative incidences of dying of second cancer and nonneoplastic disease (B, D, F, G) according to ciliary body involvement (A, B), extraocular growth (C, D), and largest basal diameter of the primary tumor (E G). Data below the graphs correspond to patients at risk in each category. Note, for example, that the cumulative incidence of dying of melanoma by 35 years was 2.0 times that of dying of nonneoplastic disease for medium-sized melanomas, whereas the incidence of dying of nonneoplastic disease was 1.3 times higher than that of dying of melanoma when the tumor was small (F, G). Crossbars: 95% confidence intervals; ticks: censored observations. Multivariate Analysis By multivariate competing risks and Cox regression, tumor height lost significance in models that included LBD. Gender, ciliary body involvement, extraocular growth, and LBD were significantly associated with risk of melanoma-related death (Table 5). In contrast to Cox regression, age at diagnosis (HR 1.01) lost significance and was omitted from the final competing risk model, which included LBD (HR 1.08 for each 1-mm increase; P ), extraocular growth (HR 2.27; P ), and ciliary body involvement (HR 1.89; P ) as

7 IOVS, November 2003, Vol. 44, No. 11 Very Long-Term Prognosis of Uveal Melanoma 4657 TABLE 4. Percentage of Deaths Due to Specific Causes Relative to All Deaths for 237 Patients Who Died of Choroidal and Ciliary Body Melanoma Interval (y) Melanoma Metastasis Second Cancer No Evidence of Malignancy All Deaths n % 95% CI n % 95% CI n % 95% CI n % Total * * Audited cause of death remained undefined for two patients. independent predictors of melanoma-related death. Gender (HR 1.36 for females, P 0.095) was not excluded as an independent predictor of melanoma-related mortality (Table 5). DISCUSSION Metastatic melanoma was the leading single cause of death throughout the follow-up period. Our 5- and 10-year audited melanoma-specific Kaplan-Meier survival estimates of 68% and 57% agree with previous estimates, which range from 70% to 74% and from 56% to 63%, respectively. 6,15,19,31 The few available estimates of 15-year survival range from 50% to 55%, 15,19,31 compared with 50% in our study. Between 10 and 15 years after diagnosis of melanoma, annual deaths from competing causes gradually outnumbered melanoma-related deaths. Nevertheless, the 14 melanoma-related deaths that occurred 15 years or more after treatment amounted to one third of annual deaths in the long term. The proportion of deaths caused by second cancers was small and relatively stable over time. Of 1205 deaths caused by uveal melanoma in the Registry of Ophthalmic Pathology, Armed Forces Institute of Pathology, Washington, D.C., only 59 (4.9%) occurred after 15 years, 16 and the corresponding numbers were 8 (4.1%) of 197 deaths in the Moorfields Eye Hospital, London, 32 and 3 (4.5%) of 67 deaths registered at the University of Iowa, Iowa. 13 All these figures are roughly half of our estimates (9.6%), suggesting that late melanoma-related deaths have been underestimated in previous series that were based on nonaudited registry data. 6,13,17,32 A histopathologic audit showed that from 7% to 10% of original cancer diagnoses that had not been based on immunohistochemistry were incorrect. Most often, amelanotic melanoma metastasis had been confused with second cancer. Biopsy and autopsy were almost equally likely to result in TABLE 5. Cox and Competing Risks Proportional Hazards Regression of Melanoma-Related Mortality among 289 Patients With Choroidal and Ciliary Body Melanoma Observed Long Term Cox Regression Competing Risks Regression Variable P HR Coefficient SE 2 P HR 95% CI Univariate regression Gender* Age Tumor location Extraocular growth LBD Tumor height Multivariate regression Gender* Age Tumor location Extraocular growth LBD Final multivariate model Gender* Tumor location Extraocular growth LBD Abbreviations: SE, standard error; HR, hazard ratio; CI, confidence interval; LBD, largest basal diameter. * Categories: male 0, female 1. Categories: two lowest age quartiles 0, two highest age quartiles 1. Categories: no ciliary body involvement 0, ciliary body involvement 1. Categories: no 0, yes 1. Continuous variable in millimeters.

8 4658 Kujala et al. IOVS, November 2003, Vol. 44, No. 11 misdiagnosis, but biopsy was more likely than autopsy to underestimate metastatic melanoma. These results extend our previous observations 33 and cast doubt on the appropriateness of using nonaudited registry data in prognostic studies. Our very long-term Kaplan-Meier estimates were similar to those reported in a unique 10-year nationwide cohort of 292 Danish patients treated from 1943 to and subsequently observed for 25 to 35 years. 35,36 This study found survival proportions at 20 and 25 years to be 42% and 40%, respectively, compared with 45% and 44% in our study. The similarity of the results of our regional series based on a much later 20-year cohort with 20 to 40 years of follow-up suggests that the figures are robust. They are consistent with the opinion that the prognosis of uveal melanoma has not appreciably improved over time, 37 despite the fact that the number of patients with extraocular growth has decreased from 17% in the Danish study 34 and 10% in ours. The Kaplan-Meier estimates reflect the theoretical situation that melanoma would be the only possible cause of death. 22,38 Cumulative incidence estimates, which take competing risks into account, 38 are more accurate when the actual risk of dying has to be cited in patient counseling and as an aid in calculating accrual for prospective studies. Kaplan-Meier estimates are too pessimistic for the former and too optimistic for the latter purpose. Our study provides estimates that apply in the long term and to melanomas of all sizes. During the first decade, the difference between the estimates was minor, but eventually Kaplan-Meier analysis exaggerated melanoma-related mortality by 25% (10% units). Actual melanoma-related mortality reached 50% by 30 years rather than by 15 years, as estimated by the Kaplan-Meier method. 19,33 The earlier and the more frequent the competing risk events are, the larger is the difference between the two estimates. 22 This was evident especially when analyzing the effect of age at diagnosis. In the highest age quartile, competing risks were very frequent and the Kaplan-Meier analysis grossly overestimated mortality. According to competing risks regression, age was not an independent predictor of melanoma-related death, in striking contrast to our Cox proportional hazards regression, suggesting confounding from competing risks in previous Cox regression analyses that identified age as an independent prognostic factor. 13,38,39 Gender was of borderline significance. It may be that, for example, earlier and more frequent cardiovascular deaths in men in part prevented an equal number of melanoma deaths from occurring compared with females. Although we did not observe melanoma deaths more than 34 years after diagnosis, metastasis after 40 years has been reported. 7,8 Care must be taken when ascribing very late deaths to the original primary cancer, because of the possibility of a second primary melanoma developing in the other eye, skin, and mucous membranes. 8 Statistical tests recently developed for assessing the presence of cured patients among longterm survivors 26,27 failed to provide conclusive evidence of their presence in our series, notwithstanding the 20- to 40-year follow-up, supporting the concept that very late metastases indeed are part of the spectrum of disseminated uveal melanoma. This field of survival statistics is in its infancy, and the tests available are relatively crude and conservative. 25 It is still unlikely that a follow-up for patients with uveal melanoma will ever be available that is sufficiently long to allow the conclusion that all remaining survivors are cured. Our study confirms that uveal melanoma typically can lead to death caused by delayed metastasis several decades after the primary tumor was definitively treated. Where the metastatic cells reside, in apparent dormancy, and the events leading to delayed progressive clinical metastasis are important to identify. Evidence for the presence of cured patients was found in subgroups that had a short rather than long median survival. One independent epidemiologic study has provided evidence that mechanisms governing cure of uveal melanoma are not identical with those that determine survival time among uncured patients. 6 These observations support the idea that the ability of uveal melanoma cells to escape from the eye is not strictly linked with their ability to grow progressively. References 1. Tojo D, Wenig BL, Resnick KI. Incidence of cervical metastasis from uveal melanoma: implications for treatment. Head Neck. 1995;17: Dithmar S, Diaz CE, Grossniklaus HE. Intraocular melanoma spread to regional lymph nodes: report of two cases. Retina. 2000;20: Eskelin S, Pyrhönen S, Summanen P, Hahka-Kemppinen M, Kivelä T. Tumor doubling times in metastatic malignant melanoma of the uvea: tumor progression before and after treatment. Ophthalmology. 2000;107: The Collaborative Ocular Melanoma Study Group. The Collaborative Ocular Melanoma Study (COMS) randomized trial of preenucleation radiation of large choroidal melanoma II: initial mortality findings. COMS report no. 10. Am J Ophthalmol. 1998;125: Diener-West M, Earle JD, Fine SL, et al. The COMS randomized trial of iodine 125 brachytherapy for choroidal melanoma, III: initial mortality findings. COMS Report No. 18. Arch Ophthalmol. 2001; 119: Gamel JW, McLean IW, McCurdy JB. Biologic distinctions between cure and time to death in 2892 patients with intraocular melanoma. Cancer. 1993;71: Coupland SE, Sidiki S, Clark BJ, McClaren K, Kyle P, Lee WR. Metastatic choroidal melanoma to the contralateral orbit 40 years after enucleation. Arch Ophthalmol. 1996;114: Shields JA, Augsburger JJ, Donoso LA, Bernardino VJ, Portenar M. Hepatic metastasis and orbital recurrence of uveal melanoma after 42 years. Am J Ophthalmol. 1985;100: Midena E, de Belvis V, Dei Tos AP, Antonini C. Isolated brain metastasis of malignant choroidal melanoma 27 years after enucleation. Arch Ophthalmol. 1999;117: Chisholm JF. A long term follow-up on malignant melanomas of the choroid based on Terry and John series. Am J Ophthalmol. 1952;1953: Char DH, Kroll S, Phillips TL. Uveal melanoma: growth rate and prognosis. Arch Ophthalmol. 1997;115: McLean IW, Ainbinder DJ, Gamel JW, McCurdy JB. Choroidalciliary body melanoma: a multivariate survival analysis of tumor location. Ophthalmology. 1995;102: Folberg R, Rummelt V, Parys-van Ginderdeuren R, et al. The prognostic value of tumor blood vessel morphology in primary uveal melanoma. Ophthalmology. 1993;100: Gamel JW, McLean IW, Greenberg RA. Interval-by-interval Cox model analysis of 3680 cases of intraocular melanoma shows a decline in the prognostic value of size and cell type over time after tumor excision. Cancer. 1988;61: Seddon JM, Albert DM, Lavin PT, Robinson N. A prognostic factor study of disease-free interval and survival following enucleation for uveal melanoma. Arch Ophthalmol. 1983;101: McLean IW, Foster WD, Zimmerman LE, Martin DG. Inferred natural history of uveal melanoma. Invest Ophthalmol Vis Sci. 1980;19: Abrahamsson M. Malignant melanoma of the choroid and the ciliary body in Halland and Gothenburg: incidence, histopathology and prognosis. Acta Ophthalmol (Copenh). 1983; 61: Pe er J, Rummelt V, Mawn L, Hwang T, Woolson RF, Folberg R. Mean of the ten largest nucleoli, microcirculation architecture, and prognosis of ciliochoroidal melanomas. Ophthalmology. 1994; 101: Seregard S, Kock E. Prognostic indicators following enucleation for posterior uveal melanoma: a multivariate analysis of long-term

9 IOVS, November 2003, Vol. 44, No. 11 Very Long-Term Prognosis of Uveal Melanoma 4659 survival with minimized loss to follow-up. Acta Ophthalmol Scand. 1995;73: Moy CS, Albert DM, Diener-West M, et al. Cause-specific mortality coding: methods in the Collaborative Ocular Melanoma Study. COMS report no. 14. Control Clin Trials. 2001;22: The Collaborative Ocular Melanoma Study Group. Assessment of metastatic disease status at death in 435 patients with large choroidal melanoma in the Collaborative Ocular Melanoma Study (COMS). COMS report no. 15. Arch Ophthalmol. 2001;119: Gooley TA, Leisenring W, Crowley J, Storer BE. Why Kaplan-Meier fails and cumulative incidence succeeds when estimating failure probabilities in the presence of competing risks. In: Crowley J, ed. Handbook of Statistics in Clinical Oncology. New York: Marcel Dekker; 2001; Gray RJ. A class of K-sample tests for comparing cumulative incidence of a competing risk. Ann Stat. 1998;16: Parmar MKB, Machin D. Survival Analysis: a Practical Approach. Chichester, UK: John Wiley & Sons; Maller RA, Zhou X. Survival Analysis with Long-Term Survivors. Chichester, UK: John Wiley & Sons; Maller RA, Zhou X. Testing for the presence of immune or cured individuals in censored survival data. Biometrics. 1995;51: Ghitany ME, Maller R, Zhou X. Estimating the proportion of immunes in censored samples: a simulation study. Stat Med. 1995; 14: Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc. 1999;94: Kleinbaum DG. Survival Analysis: a Self-Learning Text. New York: Springer-Verlag; Therneau TM, Grambsch PM. Modeling Survival Data: Extending the Cox Model. New York: Springer-Verlag; McLean IW, Foster WD, Zimmerman LE. Uveal melanoma: location, size, cell type, and enucleation as risk factors in metastasis. Hum Pathol. 1982;13: Packard RB. Pattern of mortality in choroidal malignant melanoma. Br J Ophthalmol. 1980;64: Mäkitie T, Summanen P, Tarkkanen A, Kivelä T. Microvascular loops and networks as prognostic indicators in choroidal and ciliary body melanomas. J Natl Cancer Inst. 1999;91: Jensen OA. Malignant melanomas of the uvea in Denmark : a clinical, histopathological, and prognostic study. Acta Ophthalmol (Suppl). 1963;75: Jensen OA. Malignant melanomas of the human uvea: recent follow-up of cases in Denmark, Acta Ophthalmol (Copenh). 1970;48: Jensen OA. Malignant melanomas of the human uvea: 25-year follow-up of cases in Denmark, Acta Ophthalmol (Copenh). 1982;60: Singh AD, Topham A. Survival rate with uveal melanoma in United States: Ophthalmology. 2003;110: Kroll S, Char DH, Quivey J, Castro J. A comparison of causespecific melanoma mortality and all-cause mortality in survival analyses after radiation treatment for uveal melanoma. Ophthalmology. 1998;105: Seregard S, Spångberg B, Juul C, Oskarsson M. Prognostic accuracy of the mean of the largest nucleoli, vascular patterns, and PC-10 in posterior uveal melanoma. Ophthalmology. 1998;105: