Population-Based Assessment of Clinical Characteristics Predicting Outcome of Conjunctival Melanoma in Whites

Abstract

purpose. To identify the clinical determinants of prognosis and the incidence of malignant conjunctival melanoma in whites.

methods. A nationwide search identified 85 patients in whom primary conjunctival melanoma was diagnosed in Finland between 1967 and 2000, all of whom were enrolled. Data were collected from the Finnish Cancer and Population Registries and from patients’ charts in all involved hospitals. The age-specific and age-adjusted incidences were calculated. Clinical characteristics of the tumors were recorded and time to local recurrence and melanoma-specific survival were analyzed by Kaplan-Meier analysis and univariate and multivariate extended Cox regression.

results. The annual crude incidence of conjunctival melanoma in Finland was 0.51 per million inhabitants, and the age-adjusted incidence (mean, 0.54) increased from 0.4 to 0.8 during the 34-year study period. The median age at diagnosis was 60 years (range, 20–90). Clinically detectable primary acquired melanosis preceded or accompanied the primary tumor in 61% of patients. The 5-year cumulative proportion of cases with local recurrence was 0.36 (95% confidence interval [CI], 0.25–0.48). The melanoma-specific 5-and 10-year mortalities were 0.20 (95% CI, 0.12–0.32) and 0.38 (95% CI, 0.26–0.53), respectively. By multiple-event Cox regression, nonlimbal location of the primary tumor predicted a short time to local recurrence (hazard ratio [HR] 1.81, P = 0.024). Nonlimbal location of the primary tumor (HR 4.08, P = 0.023) and increasing tumor thickness (HR 1.19 for each millimeter change, P=0.063) were associated with increased mortality. Local recurrence, analyzed as a time-dependent covariate, also increased mortality (HR 1.39 for each recurrence, P = 0.014).

conclusions. The incidence of conjunctival melanoma in the white population of Finland increased analogous to cutaneous melanoma. Nonlimbal tumors recur more often and are associated with decreased survival, independent of their greater thickness. Local recurrence contributes to mortality, whereas primary acquired melanosis was not associated with either outcome.

Conjunctival melanoma accounts for 2% to 3% of all ocular tumors in whites, ^{1 2} but this rule of thumb does not define its incidence clearly, and it cannot be applied to all populations. In Africans and Asians, conjunctival melanoma is almost unknown. ^{3 4} Even though several studies have analyzed the clinical features and outcome of conjunctival melanoma, few of them have been population-based. ^{5 6 7 8}  

Most conjunctival melanomas originate from primary acquired melanosis (PAM) with atypia and affect the limbal conjunctiva. ^{5 7 9 10 11} Clinically, PAM appears as superficial pigmentation of the conjunctiva, and invasive malignant melanoma develops in approximately half of patients with PAM with atypia. ⁹ A smaller number of conjunctival melanomas evolve from a preexisting nevus and appear de novo. ^{12 13}  

Currently, local excision combined with cryotherapy and, more recently, adjuvant topical mitomycin C is widely accepted as a primary treatment. ^{5 10 14 15 16} Little evidence exists to support any benefit from exenteration, which was the treatment of choice a few decades ago. ^{7 8 9 17}  

Conjunctival melanoma resembles more its cutaneous than uveal counterpart. Both often metastasize first to the regional lymph nodes, ^{5 9 18 19} unlike uveal melanoma. When distant metastases develop, they have no predilection for liver as uveal melanoma has, and they are capable of spreading to nearly any site in the body. ⁹ Cutaneous melanoma is rapidly increasing in white populations, ²⁰ but whether this is true of conjunctival melanoma is unknown. 

We designed a nationwide, population-based study of conjunctival melanoma in Finland spanning the years from 1967 to 2000, to reliably evaluate the clinical characteristics relevant to local recurrence and survival after development of this relatively rare disease in a white population and to determine its exact incidence. A specific purpose was to determine which clinical factors can be assessed to estimate prognosis before surgical treatment of the primary tumor is undertaken. 

Patients and Methods

Study Design

The primary purpose was to determine the clinical characteristics related to local recurrence and mortality caused by primary conjunctival melanoma. The secondary purpose was to determine the incidence rate of primary conjunctival melanoma in a white population. 

Inclusion Criteria and Data Collection

Eligible for study were all patients in whom primary malignant melanoma of the conjunctiva was diagnosed in Finland between January 1967 and December 2000. 

The patients were identified from The Finnish Cancer Registry and double checked from the records of the Helsinki University Central Hospital and its Ophthalmic Pathology Laboratory, a national referral laboratory where the diagnosis of most patients treated elsewhere had been confirmed. In files of the Cancer Registry, 20 patients carried the diagnosis of melanoma of the eye with unspecified location, and presence of conjunctival melanoma was excluded by reviewing charts from the hospitals where they had been treated; all were uveal melanomas. Similarly, those patients with PAM who had been registered as having invasive conjunctival melanoma were excluded from the present study after review of their charts. 

The clinical and follow-up data for 85 consecutive patients with primary melanoma of the conjunctiva who were thus identified were assembled from the Finnish Population Registry; from patients’ charts at hospitals where they had been treated either for conjunctival melanoma, its metastases, or secondary cancers; from pathology laboratories, and from death certificates. 

The study was approved by the Institutional Review Board, and it adhered to the tenets of the Declaration of Helsinki. 

Assessment of Tumor Characteristics

We were able to retrieve for review and histopathologic confirmation of diagnosis specimens from 83 (98%) of the 85 primary tumors. Of the two missing specimens, one had been reviewed by us earlier when the original diagnosis was made, and we retrieved a specimen from a recurrence in the second case and confirmed it to be a malignant conjunctival melanoma. 

Presence or absence of prior conjunctival nevus and PAM were recorded from patients’ charts and photographs. A nevus was coded as being present when a history of a nonprogressive conjunctival tumor at the same location for 10 years or more was present. PAM ⁹ was coded as present when previous or concurrent nonelevated conjunctival pigmentation outside the invasive tumor was recorded in the chart or could be verified from photographs. 

The tumors were classified into six categories according to location: corneally displaced (invasive melanoma growing on the cornea without evidence of conjunctival tumors other than PAM), ²¹ limbal (melanoma involving the corneoscleral limbal area), bulbar (melanoma in the bulbar conjunctiva separated from the limbus), caruncular (melanoma mainly involving the caruncle), palpebral (melanoma involving tarsal and forniceal conjunctiva, with or without extension to the skin), and diffuse melanoma (melanoma comprising more than two quadrants of conjunctiva or involving more than one location). ^{5 11 22} To allow comparison with previous reports, we combined in the analysis limbal tumors with four corneally displaced conjunctival melanomas, which probably originated from atypical melanocytes that had migrated into the cornea from the limbus. ²¹  

The largest basal diameter of the tumor was preferably measured from scanned color photographs by image analysis software (Olympus DP-SOFT, ver. 3.0; Soft Imaging System GmbH, Hamburg, Germany), which were available from 31 (36%) of 85 primary tumors, assuming that the average cornea was 11 mm in diameter. ²² The episcleral contact area of the tumor was also measured from the scanned photographs. When no photograph was available, the largest basal diameter was taken from measurements and sketches recorded in the patient chart and the histopathology report. 

The paraffin blocks of primary tumors and tumors in recurrences and metastases were stained with hematoxylin and eosin, and the thickness of the primary tumor was measured with a calibrated ocular micrometer. ²³  

Assessment of Outcome

The date of diagnosis and date of local recurrence were taken to be the date of the first recorded observation of the tumor by a physician whose clinical diagnosis led to histopathologic confirmation of primary and recurrent malignant melanoma of the conjunctiva, respectively. The date of metastasis was the date on which dissemination was confirmed by biopsy, imaging, or clinical examination. Metastasis had been histopathologically confirmed in 14 (58%) of 24 patients. We were able to retrieve each of these specimens and to verify the diagnosis. In one case, the cause of death had been registered as primary hepatic carcinoma. Autopsy specimens, however, were consistent with metastatic conjunctival melanoma. 

Complete follow-up data until December 2000 were available for 84 of the 85 patients. One patient had received a diagnosis of conjunctival melanoma at the age of 82 in 1969. He was free of disease a year later, but was then lost to follow-up and censored from the analysis at that time. The Population Registry was unable to provide the date and cause of his death. The median follow-up time was 6.3 years (range, 0.3–33) for patients who were still alive. 

Population Data

Annual population data by age group were obtained from the Statistics Finland. During the study period, Finland had a stable population of 4.6 to 5.2 million inhabitants of white origin. Immigration from other countries was negligible, and the proportion of people of African and Asian origin was less than 1%. 

Statistical Analysis

Analyses were performed on computer (Stata, ver. 7.0; Stata Corp., College Station, TX). Contingency tables were analyzed with the Fisher exact test and the Pearson χ² test, and distributions of continuous variables were compared with the nonparametric Mann-Whitney test. Exact probabilities were then computed (StatXact-5; Cytel Software, Cambridge, MA). 

The crude, age-adjusted, and age-specific incidence rate of conjunctival melanoma were calculated, adjusting the incidence for the 2000 U.S. standard million population by the direct method. ²⁴ Annual incidence data were displayed in a scatterplot and summarized by line smoothing without weighting. ²⁵ For comparison, data on malignant cutaneous melanoma were obtained from the Finnish Cancer Registry and analyzed accordingly. The age-standardized incidence rate ratio of cutaneous to conjunctival melanoma was calculated and plotted. 

In addition to standard analysis of time to first local recurrence, ^{26 27} the full history of recurrences was analyzed using repeated events, multiple failure-time data. ^{28 29} In the latter analysis, patients were returned to the risk set after each recurrence. To analyze melanoma-specific survival, patients judged to have died from causes unrelated to conjunctival melanoma were censored at the time of death. 

Univariate analysis of survival time data were based on the Kaplan-Meier product-limit method. ^{26 27} Age at diagnosis was divided in tertiles, largest basal diameter was arbitrarily divided into three categories, and tumor thickness was divided into three categories based on previous studies that have suggested a thickness of 1 and 2 mm to indicate high risk of metastasis (<1 mm vs. 1 to 2 mm vs. >2 mm). ^{5 6 17 23 30} The effect of adjuvant treatment was analyzed by comparing outcomes in patients who received no adjuvant therapy with those who were treated either with adjuvant cryocoagulation ³¹ or topical mitomycin C. ^{14 16 32} Unordered and ordered survival curves were compared with the log-rank test and test for trend, respectively. ²⁶  

Multivariate analysis of time to local recurrence was based on an extension of the Cox proportional hazards model, which allows analysis of ordered multiple events. ^{28 29} The conditional risk set model of Prentice et al., ³³ which assumes that patients are not at risk for second recurrence until they have experienced the first recurrence, was fitted. The analysis was based on time from entry. Standard errors were calculated with the robust variance estimator of Lin and Wei ³⁴ and tied survival times were handled with the Efron approximation. ²⁷ Location of the primary tumor was dichotomized into limbal (including corneally displaced) and nonlimbal (bulbar, palpebral, caruncular, and diffuse), based on Kaplan-Meier analysis and previous publications. ^{11 18 22} Age at diagnosis and the thickness, largest basal diameter, and contact area of the primary tumor were modeled as continuous variables. 

Multivariate analysis of melanoma-specific survival was based on an extension of the Cox proportional hazards model, which allows internal time-varying covariates. ²⁷ Three alternative strategies for modeling local recurrence were applied. In the first one, the corresponding variable was set to 0 until the first local recurrence and to 1 thereafter. This model suggests that subsequent recurrences do not additionally affect prognosis and the risk remains constant after the first recurrence. In the second, this variable was set to 0 and increased by 1 after each local recurrence, giving each recurrence an equal weight. In the third one, the variable was arbitrarily increased by 1, 0.8, 0.6, 0.4, and 0.2 after the first, second, third, fourth, and fifth local recurrence, respectively, giving each recurrence progressively less weight. The last model suggests that the first recurrence is of greatest significance—for example, because micrometastases spawned from the tumor may overgrow subsequent ones—but later recurrences still contribute to prognosis. 

The assumption of proportional hazards was verified by adding each covariate by log time interaction to the model and by assessing the significance of the product term using the partial likelihood ratio test, which was also used to compare alternative models. The number of variables in the final model was restricted to two, based on a rule that there had to be at least 15 to 20 events for each additional variable. ²⁶ The regression coefficients and HRs with 95% confidence intervals (CIs) were calculated. 

Results

Incidence

During the study period, 85 new patients with conjunctival melanoma were identified in the Finnish population (mean, 4.88 million inhabitants), corresponding to a mean of 2.5 patients per year (range, 0–8). The crude and age-standardized annual incidences were 0.51 and 0.54 per million, respectively, adjusted for the 2000 U.S. standard million population. The age-specific incidence was 0.06 for those under 30 years of age, and 0.48, 1.05, and 1.57 for the age groups of 30 to 49 years, 50 to 70 years, and more than 70 years, respectively. 

The smoothed, age-standardized incidence remained stable (mean, 0.4 per million) until 1975 and showed a steady increase thereafter to 0.8 per million (Fig. 1A) . The corresponding smoothed incidence of cutaneous melanoma increased at approximately the same rate from 1967 to 1998 from 41 to 101 per million (Fig. 1B) . The smoothed relative incidence of cutaneous compared with conjunctival melanoma remained stable at approximately 150:1 (Fig. 1C) . 

Gender and Age at Diagnosis

The male-to-female ratio did not differ statistically from unity (47:38, P = 0.39) binomial test; Table 1 ). The median age at diagnosis of the primary tumor was 60 years (range, 20–90; Table 1 ). 

Site of the Primary Tumor

The limbal conjunctiva was the most common primary site (59%; 95% CI, 48%–69%) followed by the bulbar conjunctiva, diffuse tumors, palpebral conjunctiva, corneally displaced tumors, and caruncle (Table 1) . Both eyes were equally affected (right-left, 41:44). None of the patients had bilateral melanoma. 

Predisposing Lesions

In 23 of 77 patients (30%; 95% CI, 20%–41%) on whom we had adequate clinical information, the melanoma probably had developed from a prior nevus, and 48 of 79 eyes (61%; 95% CI, 49%–72%) had had clinically detectable PAM (Table 1) . Six patients had a history of both nevus and PAM. 

In 7 (8%) additional patients, clinically detectable PAM appeared during follow-up, but had not been evident at the time of diagnosis. Limbal tumors tended to be less frequently associated with PAM than nonlimbal tumors (46% vs. 71%, P = 0.086 Pearson χ² test). 

Size of the Primary Tumor

The median thickness of 72 measurable primary tumors was 1.3 mm (range, 0.2–8.8). The largest basal diameter of 76 measured tumors ranged from 1 to 19 mm (median, 7). The contact area of 31 photographed tumors ranged from 4 to 132 mm² (median, 29). 

Limbal melanomas were thinner (median, 1.1 vs. 3.3 mm, P = 0.0003 Mann-Whitney test), and their diameter (median, 7 mm vs. 10 mm, P = 0.042) and contact area (median, 24 mm² vs. 63 mm², P = 0.042) were smaller than nonlimbal melanomas (Table 1) . 

Primary Treatment

Of the 85 primary tumors, 78 (92%) were removed by local excision, two of which were combined with lamellar resection of the underlying sclera. One corneally displaced melanoma was removed by penetrating keratoplasty (Table 1) . Four patients underwent primary exenteration (before 1981) and two enucleation (before 1979). In 13 patients, the primary resection was promptly followed by secondary surgery because the margins were involved in invasive melanoma. These included nine local resections, three exenterations and one enucleation (in 1983). 

As adjuvant treatment, the adjacent conjunctiva was cryocoagulated from beneath with a nitrous oxide probe in 16 eyes. Three eyes received two cycles of 0.04 mg/mL topical mitomycin C for 2 weeks at 1 month after surgery, and 2 eyes received both types of adjuvant treatment. 

Local Recurrence and Secondary Treatment

The tumor recurred at least once in 29 patients. The median number of recurrences was two (range, one–six). The 5- and 10-year cumulative proportions for first local recurrence were 0.36 (95% CI, 0.25–0.48) and 0.38 (95% CI, 0.28–0.52), respectively. 

By Kaplan-Meier analysis, the gender (P = 0.51 log-rank test, difference between curves), age at diagnosis of primary melanoma (Fig. 2B) , and history of previous conjunctival nevus (Fig. 2C) and PAM (Fig. 2D) were not significantly associated with recurrence. The site of the primary tumor was strongly associated with recurrence (Fig. 2E ; P = 0.006), mainly because palpebral location predicted frequent and rapid recurrence. Tumor thickness (Fig. 2F) , largest basal diameter (Fig. 2G) and contact area (P = 0.32), and adjuvant treatment (Fig. 2H) were not statistically associated with local recurrence. 

By univariate Cox regression (Table 2) , nonlimbal location had a tendency to be associated with time to first recurrence (HR 1.92, P = 0.096). This association strengthened when recurrence was analyzed as a repeated event (Wald χ², 5.08 vs. 2.78; HR 1.81, P = 0.024). None of the other variables was associated with local recurrence in either analysis. The fact that tumor thickness did not predict local recurrence was confirmed by bivariate analysis (Table 2) . 

After recurrence, four additional patients underwent secondary exenteration (in 1970, 1975, 1993, and 1997) and one painful eye was enucleated (in 1995). One patient refused exenteration for recurrence and was treated with radiotherapy followed by evisceration. The remaining 23 patients with recurrent melanoma were treated with local excision. 

Systemic Recurrence

Of the 85 patients, 24 had metastases and 23 of died during follow-up. One patient was alive 4 years after resection of a solitary submandibular lymph node metastasis. The median time from diagnosis of primary tumor to death from metastatic disease was 4.0 years (range, 1–19). The median time from diagnosis of metastasis to death was 1.3 years (range, 0.1–3.5) in the 19 patients for whom this interval was known. 

Melanoma-Specific Survival

During follow-up, 3 patients died of another tumor and 14 patients of causes unrelated to cancer. One patient was lost to follow-up and 44 patients were still alive. The cumulative 5- and 10-year melanoma-specific survival proportions were 0.80 (95% CI, 0.68–0.88) and 0.62 (95% CI, 0.47–0.74), respectively (Fig. 3A) . 

By Kaplan-Meier analysis, gender (Fig. 3B) , age at diagnosis of primary melanoma (Fig. 3C) , and history of previous conjunctival nevus (Fig. 3D) and PAM (Fig. 3E) did not predict melanoma-related death. Location (Fig. 3F ; P = 0.0002 log-rank test), thickness (Fig. 3G ; P = 0.009 log-rank test for trend), and largest basal diameter (Fig. 3H ; P = 0.003) of the primary tumor were associated with survival. The 5-year survival after limbal melanoma was 0.94, compared with 0.67 for bulbar and 0.38 for palpebral and diffuse melanomas. 

By univariate Cox regression (Table 3) , nonlimbal location (HR 4.87), thickness (HR 1.39 for each millimeter increase) and largest basal diameter (HR 1.21 for each millimeter increase) were statistically significantly associated with an increased risk of melanoma-related death. The association with melanoma-related death and local recurrence strengthened (Wald χ², 25.2/18.8 vs. 2.47) when recurrence was analyzed as a repeated event (HR 1.61 and 1.98 for each recurrence, equal and decreasing weights, respectively). The model that gave each recurrence an equal weight showed the statistically strongest association with survival (Wald χ², 25.2 vs. 18.8; Table 3 ). 

Multivariate Cox regression models were fitted to asses the independent effect of location, tumor thickness, and repeated local recurrence, modeled by equal weights (Table 3) . All three variables retained statistical significance, but because of the small number of events the final model was restricted to two variables. Of the bivariate regressions, models that combined tumor thickness with location (likelihood ratio, 151.06 vs. 119.29, P < 0.001 χ² test, df = 1) and repeated local recurrences (likelihood ratio, 151.06 vs. 122.79, P < 0.001) showed statistically stronger association with survival than the model that included tumor location and repeated local recurrences (Table 3) . 

Discussion

Compared with the neighboring Nordic countries, the incidence of conjunctival melanoma in Finland was approximately double the crude annual incidence of 0.24 per million reported in Sweden from 1969 to 1991, ⁷ but comparable to an estimate of 0.45 per million in Denmark, derived from a study comprising the years 1960 to 1980 ⁸ by excluding patients who had PAM but never had invasive malignant melanoma. 

The rate of increase was comparable to that observed for cutaneous melanoma in Finland. No apparent change in registration procedures of the Finnish Cancer Registry and diagnostic criteria took place during the study period. The possibility remains that increased ultraviolet radiation exposure may contribute to the increase, as is suspected of skin melanoma. ³⁵ An equally high crude incidence of 0.8 per million, based on German Cancer Registry data from 1960 to 1988, is on record. ⁶ More population-based data are needed to determine whether the incidence of conjunctival melanoma is increasing globally. 

The population-based 5- and 10-year cumulative probabilities of having first local recurrence were comparable to corresponding probabilities of 0.39 and 0.43 reported in the Danish population. ⁸ A large, referral-based case series estimated the probabilities to be 0.26 and 0.51. ¹⁸ Throughout the study period, local resection of limbal and bulbar melanomas without lamellar scleral excision, followed by prompt re-excision if surgical margins were involved or a recurrence developed, was the treatment of choice in Finland. Nevertheless, 19% of patients eventually lost an eye. In Sweden, 38% of patients had an eye enucleated or exenterated during follow-up. ⁷ Change in attitudes has decreased, but not eliminated, the need for exenteration in this group of patients. ^{7 8 9 17}  

The melanoma-specific 5- and 10-year cumulative probabilities of survival resemble and fall within the confidence limits of Danish, German, and Swedish estimates that are population-based and range from 0.81 to 0.88 for 5-year survival and from 0.71 to 0.76 for 10-year survival. ^{6 7 8} In various case series, the 5-year cumulative probability of survival ranges from 0.77 to 0.93 and the 10-year survival from 0.70 to 0.76. ^{10 11 30} Because median follow-up time in all studies was less than 10 years, the 10-year estimates are not precise. 

Multiple local recurrences of conjunctival melanoma are frequent, especially if the tumor involves palpebral conjunctiva. Tumor recurrence can be modeled in several ways. Previously, time to the first recurrence and, in one study, ¹⁸ intervals between the first, second, and third recurrences have been analyzed. These methods disregard data on other recurrences. We applied more recently developed techniques and modeled local recurrence as a repeated event, which enables efficient use of all available information of the recurrence history of each patient. ²⁹ Furthermore, we modeled local recurrence as a time-varying covariate when analyzing survival. Recurrence is often inappropriately modeled as a baseline variable, although it can only be assessed at the time of recurrence. ³⁶ Of three strategies tested, giving an equal weight to each recurrence resulted in a model that best predicted melanoma-specific mortality. The clinical interpretation is that an effort must be made to eradicate the entire recurrence, however frequently the diagnosis is made in the patient. 

Nonlimbal location of the primary tumor was estimated to indicate approximately two times higher risk of local recurrence, analyzed as a repeated event, and almost five times higher risk of tumor-related death by univariate analysis. It was the only factor statistically related to local recurrence. The lower metastatic risk of limbal melanomas could, in part, be because the sclera and Bowman’s layer may act as natural barriers that delay invasion. ⁵  

Kaplan-Meier analysis suggested the risk of death to be significantly higher for tumors more than 2 mm in thickness. Comparison of survival when the tumor was less than 2 mm thick was inaccurate, because only six melanoma-related deaths occurred among the 47 patients who had tumors that were less than 2 mm thick, compared with 12 of 27 patients with tumors of 2 mm or thicker. Our data lend support to the theory that 2 mm may be close to the critical thickness that indicates a clinically significant higher risk of metastases. ^{5 6 23}  

Cox multivariate regression suggested that tumor location, thickness, and local recurrence independently contribute to melanoma-specific survival. Bivariate models that included tumor thickness predicted survival significantly better than the model that excluded it. The two models that combined thickness with tumor location and local recurrence, analyzed as a time-varying covariate, were equally effective in predicting survival. The latter model, however, is suitable only for retrospective assessment and not for clinical decision making. 

We could not establish any difference in survival between patients with and without clinically detectable PAM, and PAM was not associated with higher risk of recurrence in our data set. This is somewhat surprising, but it has been noted before. ³⁰ The nonassociation is unlikely to be due to misclassification, because conjunctival melanomas originating from completely amelanotic PAM are rare. ^{37 38} Identification of PAM from small histopathologic samples is similarly subject to error. However, the majority of conjunctival melanomas develop in association with PAM, ⁹ and patients who have it benefit from surveillance. 

Taken together with previous evidence, our data further support that nonlimbal tumor location, ^{5 11 23} thickness of the primary tumor, ^{30 37} and subsequent local recurrences ¹⁰ are statistically and clinically the most significant independent indicators of worse prognosis. Especially nonlimbal conjunctival pigmented lesions should be removed early to establish a diagnosis, before they grow to be thick. The data also underline the importance of efficient diagnosis and meticulous planning of the surgery for the primary tumor, especially in patients with nonlimbal tumors. These patients at high risk also are candidates for adjuvant treatment trials and review at short intervals during the first years after treatment. The beneficial effect of adjuvant treatment on recurrence, however, awaits proper prospective confirmation. 

 Supported by Grant TYH1217 from the Helsinki University Central Hospital, The Finnish Medical Foundation (Duodecim), The Eye Foundation, The Eye and Tissue Bank Foundation, The Friends of the Blind, and The Paulo Foundation, Finland.

 Submitted for publication February 8, 2002; revised May 15, 2002; accepted June 17, 2002.

 Commercial relationships policy: N.

 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. §1734 solely to indicate this fact.

 

Figure 1.

 

The annual crude and age-standardized incidence of conjunctival (A) and cutaneous (B) melanoma in Finland. The curve shown is that of the smoothed, age-standardized incidence. The smoothed incidences in (A) and (B) increase at approximately the same rate. The smoothed incidence of cutaneous melanoma relative to the incidence of conjunctival melanoma is stable (C).

Table 1.

 

Clinical Characteristics of the 85 Patients with Primary Conjunctival Melanoma

Variable	Total	Tumor Location			P
				Limbal		Nonlimbal
Gender, n (%)
Female	38 (45)	24 (44)	14 (45)	1.0*
Male	47 (56)	30 (56)	17 (55)
Median age, y (range)	60 (20–90)	60 (20–89)	60 (29–90)	0.50, †
Primary site, n (%)
Limbal conjunctiva		50 (59)		N/A
Corneally displaced tumors		4 (5)
Bulbar conjunctiva			11 (13)
Caruncle			3 (4)
Palpebral conjunctiva			8 (9)
Diffuse tumors			9 (11)
Previous nevus, n (%)
Present	23 (27)	17 (31)	6 (19)	0.35, ‡
Absent	54 (64)	31 (57)	23 (74)
Unknown	8 (9)	6 (11)	2 (6)
Previous PAM, n (%)
Present	47 (55)	25 (46)	22 (71)	0.086, ‡
Absent	32 (38)	24 (44)	8 (26)
Unknown	6 (7)	5 (9)	1 (3)
Median thickness, mm (range)	1.3 (0.2–8.8)	1.1 (0.3–3.9)	3.3 (0.2–8.8)	0.0003, †
Median largest basal diameter, mm (range)	7 (1–19)	7 (1–17)	10 (4–19)	0.042, †
Area, mm2 (range)	29 (4–132)	24 (4–56)	63 (5–132)	0.042, †
Primary treatment, n (%)
Local excision	76 (89)	49 (91)	27 (87)	0.41, ‡
Lamellar excision	2 (2)	2 (4)	0
Penetrating keratoplasty	1 (1)	1 (2)	0
Exenteration	4 (5)	1 (2)	3 (10)
Enucleation	2 (2)	1 (2)	1 (3)
Adjuvant treatment
Cryocoagulation	18 (21)	9 (18)	9 (29)
Topical mitomycin C	5 (6)	2 (4)	3 (10)

 PAM, primary acquired melanosis; N/A, not applicable.

*  Fisher exact test, two-tailed.

†  Mann-Whitney test, exact probability distribution, two-tailed.

‡  Pearson’s χ² test, exact probability distribution, two-tailed.

Figure 2.

 

Kaplan-Meier analysis of time to first local recurrence in the 85 patients with primary conjunctival melanoma (A). Also shown is a survival curve plotted for local recurrence as a repeated event. Rather than the proportion of patients with local recurrence, the latter curve is a summary of the time frame for any recurrence among the 85 patients. Analysis of time to the first local recurrence according to age at diagnosis (B); the presence of prior conjunctival nevus (C) and PAM (D); the location (E), thickness (F), and largest basal diameter (G) of the primary tumor; and adjuvant treatment (H) indicated the location of the primary tumor to be a statistically significant prognostic indicator. Numbers below each graph represent patients remaining in follow-up. The unknown groups in (C) and (D) are displayed but not used in the statistical analysis.

Table 2.

 

Cox Proportional Hazards Regression Analysis of First and Repeated Local Recurrence in 85 Patients with Primary Conjunctival Melanoma

Variable	Coefficient (SE)	Wald χ2	P	Hazard Ratio (95% CI)
Univariate analysis, first event
Gender*	0.248 (0.371)	0.45	0.50	1.28 (0.62–2.65)
Age, †	0.016 (0.011)	2.25	0.13	1.02 (1.00–1.04)
Location, ‡	0.651 (0.391)	2.78	0.096	1.92 (0.89–4.12)
Previous nevus, §	0.097 (0.429)	0.0004	0.98	1.01 (0.44–2.34)
Previous PAM, §	0.193 (0.394)	0.24	0.62	1.21 (0.56–2.62)
Thickness, †	0.083 (0.120)	0.48	0.49	1.09 (0.86–1.38)
Largest basal diameter, †	0.021 (0.055)	0.14	0.70	1.02 (0.92–1.14)
Adjuvant treatment, ∥	−0.356 (0.553)	0.41	0.52	0.70 (0.24–2.07)
Univariate analysis, repeated event
Gender*	0.030 (0.270)	0.01	0.91	1.03 (0.61–1.75)
Age, †	0.015 (0.009)	2.36	0.12	1.01 (1.00–1.03)
Location, ‡	0.592 (0.262)	5.08	0.024	1.81 (1.08–3.02)
Previous nevus, §	−0.046 (0.327)	0.02	0.89	0.95 (0.50–1.81)
Previous PAM, §	0.185 (0.273)	0.46	0.50	1.20 (0.70–2.05)
Thickness, †	0.084 (0.087)	0.92	0.34	1.09 (0.92–1.29)
Largest basal diameter, †	0.029 (0.041)	0.51	0.47	1.03 (0.95–1.12)
Adjuvant treatment, ∥	−0.420 (0.348)	1.46	0.23	0.66 (0.33–1.30)
Bivariate analysis, repeated event
Location, ‡	0.782 (0.334)	5.48	0.019	2.19 (1.14–4.20)
Thickness, †	−0.031 (0.097)	0.10	0.75	0.97 (0.80–1.17)

 Efron approximation for ties, robust standard error, conditional risk set analysis for repeated events.

*  Categories: male, 0; female, 1.

†  Continuous variable.

‡  Categories: Limbal, 0; nonlimbal, 1 (see Table 1 ).

§  Categories: absent, 0; present, 1

∥  Categories: No, 0; cryocoagulation, topical mitomycin C, or both treatments, 1.

Figure 3.

 

Kaplan-Meier analysis of survival after diagnosis in the 85 patients with conjunctival melanoma (A). Analysis of survival according to gender (B); age at diagnosis (C); the presence of prior conjunctival nevus (D) and PAM (E); and the location (F), thickness (G), and largest basal diameter (H) of the primary tumor indicated location and both measures of primary tumor size to be statistically significant prognostic indicators. Numbers below each graph represent patients remaining in follow-up. The unknown groups in (D) and (E) are displayed but not used in the statistical analysis.

Table 3.

 

Cox Proportional Hazards Regression Analysis of Survival in 85 Patients with Primary Conjunctival Melanoma

Variable	Coefficient (SE)	Wald χ2	P	Hazard Ratio (95% CI)
Univariate analysis
Gender*	0.006 (0.416)	0.00	0.99	1.10 (0.45–2.27)
Age, †	0.005 (0.128)	0.15	0.70	1.00 (0.98–1.03)
Location, ‡	1.582 (0.435)	13.2	<0.001	4.87 (2.07–11.4)
Previous nevus, §	−0.220 (0.529)	0.32	0.57	0.74 (0.27–2.07)
Previous PAM, §	−0.451 (0.422)	1.14	0.29	0.64 (0.26–1.47)
Thickness, †	0.332 (0.076)	18.93	<0.001	1.39 (1.20–1.62)
Largest basal diameter, †	0.193 (0.056)	11.67	0.001	1.21 (1.09–1.35)
Adjuvant treatment, ∥	0.816 (0.525)	2.41	0.12	2.26 (0.81–6.34)
Local recurrence, ¶	0.654 (0.416)	2.47	0.12	1.92 (0.85–4.34)
Local recurrence, #	0.475 (0.095)	25.2	<0.001	1.61 (1.34–1.93)
Local recurrence, **	0.684 (0.158)	18.8	<0.001	1.98 (1.45–2.70)
Multivariate analysis Model 1 (−2 log likelihood, 116.66)
Location, ‡	1.406 (0.619)	5.15	0.023	4.08 (1.21–13.7)
Thickness, †	0.172 (0.093)	3.46	0.063	1.19 (0.99–1.42)
Local recurrence, #	0.300 (0.121)	6.00	0.014	1.39 (1.09–1.76)
Model 2 (−2 log likelihood, 119.29)—Final Model
Location, ‡	1.561 (0.591)	6.97	0.008	4.76 (1.50–15.2)
Thickness, †	0.167 (0.095)	3.13	0.077	1.18 (0.98–1.42)
Model 3 (−2 log likelihood, 151.06)
Location, ‡	1.462 (0.413)	12.5	<0.001	4.31 (1.92–9.69)
Local recurrence, #	0.456 (0.096)	22.4	<0.001	1.58 (1.31–1.91)
Model 4 (−2 log likelihood, 122.79)—Alternative Final Model
Thickness, †	0.317 (0.074)	18.5	<0.001	1.37 (1.19–1.59)
Local recurrence, #	0.376 (0.1118)	10.1	0.001	1.46 (1.16–1.84)

 Efron approximation for ties, robust standard error.

*  Categories: male, 0; female, 1.

†  Continuous variable.

‡  Categories: limbal, 0; nonlimbal, 1 (see Table 1 ).

§  Categories: absent, 0; present, 1.

∥  Categories: no, 0; cryocoagulation, topical mitomycin C, or both treatments, 1.

¶  Time-varying covariate, before first recurrence, 0; after first recurrence, 1.

#  Time-varying covariate, before first recurrence, 0; equal weights after nth recurrence.

**  Time-varying covariate, before first recurrence, 0; decreasing weights after nth recurrence.