Purpose
To validate a staging system for metastatic uveal melanoma that will facilitate planning, reporting, and interpreting the results of clinical trials.
Design
Reliability and validity study.
Methods
The performance index, the largest diameter of the largest metastasis and alkaline phosphatase level at the time of diagnosis of metastases, and overall survival of 249 patients from 7 ocular oncology centers who died of dissemination were analyzed. Predicted median survival time calculated according to the Helsinki University Hospital Working Formulation was used to assign patients to stages IVa, IVb, and IVc, which correspond to predicted survival times of ≥12, <12-6, and <6 months, respectively. The predictions were compared against observed survival.
Results
The 3 variables used to assign stage were independent predictors of survival in the validation dataset. Of the 249 patients, 110 (44%), 109 (44%), and 30 (12%) were classified to Working Formulation stages IVa, IVb, and IVc, respectively. Corresponding median observed survival times were 18.6, 10.7, and 4.6 months and worsened by increasing stage ( P < .001). Of 201 patients managed without surgical resection of metastases, 83 (41%), 89 (44%), and 29 (15%) were classified to stages IVa, IVb, and IVc, respectively, and their median observed survival times were 17.2, 10.0, and 4.6 months ( P < .001). Survival of 47 patients who underwent resection did not differ by working formulation stage ( P = .69).
Conclusions
This multicenter study confirms that the Working Formulation is a reliable and valid, repeatable system for dividing metastatic uveal melanoma into distinct prognostic subgroups, especially for stage-specific reporting of survival in prospective clinical trials.
Uveal melanoma is the second most common primary intraocular cancer after retinoblastoma, with an estimated annual incidence of 6679–7095 patients. Half of them develop metastases over 3 decades. Their life expectancy after detection of metastases is short, a median of 8 months for unselected patients, although median overall survival times ranging from 12 to 24 months are commonly reported in retrospective analyses of specific treatments. These patients typically participated in surveillance for early detection of asymptomatic metastases so that lead time bias likely contributed to their survival times. For example, patients reviewed annually with abdominal ultrasonography to detect hepatic metastases, the predominant type of metastasis from uveal melanoma, survived a median of 5 months longer than those who were not screened, but their survival times from diagnosis of the primary tumor to death did not differ.
Because uveal melanoma is rare, few randomized clinical trials are performed. Thus, there is a need for validated staging systems to adjust for differences in case mix in nonrandomized phase II studies that would aid in selecting patients for trials, in stratifying the analysis, and in interpreting the results. Recently, a nomogram-based system for this purpose was developed and externally validated.
A previous multivariate analysis of the overall survival of 91 consecutive patients who died of metastatic uveal melanoma identified 3 variables—performance index, reflecting general health of the patient; largest diameter of the largest metastasis (LDLM), reflecting measurable metastases; and serum alkaline phosphatase (AP) level, reflecting both liver function and, indirectly, unmeasurable miliary metastases—as being independently associated with survival. A prognostic Helsinki University Hospital Working Formulation based on this multivariate model divides newly diagnosed metastatic (stage IV ) uveal melanoma into 3 prognostic categories with a predicted median overall survival of either ≥12 months (IVa), <12-6 months (IVb), or <6 months (IVc). It categorizes patients who would normally be eligible for chemotherapy into 3 groups based on predicted median overall survival. Such an effort to categorize is warranted because the resulting prognostic estimate would be useful in clinical practice. Knowing the prognostic category of each patient would help the ophthalmologist in counseling the patient, in selecting patients for enrollment in clinical trials together with an oncologist, in stratifying them in trials, and in interpreting the results of clinical trials.
Until now, this prognostic model has not been validated. For this purpose, we performed the present collaborative European Ophthalmic Oncology Group (OOG) study.
Methods
Study Design and Inclusion Criteria
This reliability and validity study was approved by the institutional review boards of the 7 participating institutions and followed the tenets of the Declaration of Helsinki as well as all laws applicable in each participating country. Informed consent was not collected because of the study design, which was based on retrospective data and enrolled only deceased patients.
Data of consecutively registered patients who died of metastatic uveal melanoma (stage IV) were requested from members of the OOG; 90% of the patients had metastases diagnosed between 2000 and 2007. The diagnosis of metastases was based on autopsy, biopsy, or typical clinical course (progressive hepatic metastases in the absence of second cancer). The data requested were the date of diagnosis of metastases; the performance index, LDLM, and serum or plasma AP level when metastases were diagnosed; and the date of death. Patients without measurable metastases or recorded AP level were ineligible. No limitations regarding treatment for metastases were applied.
Patients and Clinical Data
Data on 249 patients were received from 7 medical and ocular oncology services: 90 patients from Institut Curie, Paris; 60 from Klinikum Benjamin Franklin, Berlin; 23 from University of Padua, Padua; 22 from Hopital Jules Gonin, Lausanne; 19 from Helsinki University Hospital, Helsinki; 18 from Croix-Rousse Hospital, Lyon; and 17 from Royal Liverpool University Hospital, Liverpool.
The Karnofsky index or the Eastern Cooperative Oncology Group (ECOG) score was taken from patient charts. The LDLM was the best estimate from ultrasonography (US), computed tomography (CT), or magnetic resonance imaging (MRI) reports. The serum or plasma AP level and the corresponding upper normal limit (UNL) were also taken from the charts.
Most patients participated in 6- to 12-monthly surveillance that included liver imaging and liver function tests (various combinations of US, CT, MRI, aspartate and alanine aminotransferase, AP, and lactate dehydrogenase), according to the preference of each center.
Staging of Metastatic Disease
We staged patients according to the Helsinki University Hospital Working Formulation by calculating their predicted median overall survival using the published final Cox proportional hazards regression model. AP and the LDLM were modeled as continuous variables. The performance index was divided into 3 categories: asymptomatic (Karnofsky 100-90; ECOG 0); symptomatic (Karnofsky 85-60; ECOG 1-2); and symptomatic, unfit for chemotherapy (Karnofsky <60; ECOG 3-4).
The original final model included time on chemotherapy as a confounder. This term helped in quantitating the contribution to prognosis of the other factors of interest when the model was built, irrespective of whether longer times on therapy were due to efficacy of the treatment or to less aggressive disease. When applying the model, the population mean of 5 months, taken from the building dataset, is used in all calculations as the likely time on chemotherapy or comparable medical treatment because it is not yet known how long such a therapy, if any, will continue.
Statistical Analysis
All analyses were performed with Stata (v10.0; Stata, College Station, Texas, USA). The stcurve command was used to plot the estimated survival function from which we read the corresponding median predicted survival. We assigned each patient correspondingly to stage IVa (≥12 months), IVb (<12-6 months), or IVc (<6 months).
We used Cox multivariate proportional hazards regression to confirm that the variables of the Working Formulation were independent predictors of survival. Regression coefficients and hazard ratios (HR) with 95% confidence intervals (CI) were calculated. Assumption of proportional hazards was assessed with complementary log plots.
We generated Kaplan-Meier survival curves for each stage and compared unordered and ordered categories with log-rank test and test for trend, respectively. All P values were 2-sided ( P < .05 was taken as significant). We adjusted P values in pairwise comparisons with Bonferroni correction. Primary analysis was based on all patients. Secondary analysis considered separately patients whose metastases had or had not been resected.
We compared predicted and observed survival visually using scatterplots and agreement between predicted and observed survival categories (≥12 months, <12-6 months, and <6 months) using weighted kappa (weighting matrix, 1 ∖ 0.50 1 ∖ 0 0.50 1).
Results
Patient Characteristics at Diagnosis of Metastases
Of the 249 eligible patients, most had an ECOG score 0 (62%) and 1-2 (37%) or an equivalent Karnofsky index ( Table 1 ). The distribution of performance indices was comparable in all participating centers except that ECOG 1-2 was the most common index in 1 center.
| Participating Center | ||||||||
|---|---|---|---|---|---|---|---|---|
| Berlin, Germany, N (%) | Helsinki, Finland, N (%) | Lausanne, Switzerland, N (%) | Liverpool, UK, N (%) | Lyon, France, N (%) | Padua, Italy, N (%) | Paris, France, N (%) | Total, N (%) | |
| Performance index | ||||||||
| ECOG 0 | 43 (72) | 13 (68) | 16 (73) | 11 (65) | 16 (89) | 16 (70) | 39 (43) | 154 (62) |
| ECOG 1-2 | 17 (28) | 6 (32) | 6 (27) | 5 (29) | 1 (6) | 6 (26) | 51 (57) | 92 (37) |
| ECOG 3-4 | – | – | – | 1 (6) | 1 (6) | 1 (4) | – | 3 (1) |
| LDLM (cm) | ||||||||
| ≤3.0 | 12 (20) | 14 (74) | 10 (45) | 6 (35) | 14 (78) | 10 (43) | 50 (56) | 116 (47) |
| 3.1–8.0 | 34 (57) | 5 (26) | 10 (45) | 9 (53) | 4 (22) | 13 (56) | 37 (41) | 112 (45) |
| >8.0 | 14 (23) | – | 2 (9) | 2 (12) | – | – | 3 (3) | 21 (8) |
| AP level, ×UNL | ||||||||
| 1.0 | 31 (52) | 16 (84) | 17 (77) | 8 (47) | 7 (39) | 17 (74) | 83 (92) | 179 (72) |
| 1.1–2.0 | 21 (35) | 1 (5) | 3 (14) | 5 (29) | 8 (44) | 5 (22) | 5 (6) | 48 (19) |
| >2.0 | 8 (13) | 2 (10) | 2 (9) | 4 (24) | 10 (56) | 1 (4) | 2 (2) | 22 (9) |
| Primary treatment | ||||||||
| Chemotherapy | ||||||||
| Dacarbazine | – | 16 (84) | – | 1 (6) | – | 35 (39) | 52 (21) | |
| Fotemustine | 9 (15) | – | 7 (32) | – | 6 (22) | 19 (21) | 41 (16) | |
| Treosulfan | 51 (85) | – | – | – | – | – | 51 (20) | |
| Other | – | – | – | 11 (65) | 5 (28) | 6 (22) | 2 (2) | 24 (10) |
| Chemoembolization | – | – | – | – | 4 (22) | – | – | 4 (2) |
| Interferon-tamoxifen | – | – | – | – | – | 6 (22) | – | 6 (2) |
| Tumor vaccine | – | – | – | – | – | – | 9 (10) | 9 (4) |
| Surgical resection | ||||||||
| No chemotherapy | – | – | – | – | 1 (6) | 1 (4) | 9 (10) | 11 (4) |
| Chemotherapy | – | 15 (64) | 1 (6) | 8 (44) | 1 (4) | 11 (12) | 36 (14) | |
| Best supportive care | 1 (5) | – | 4 (24) | – | 1 (4) | 5 (6) | 11 (4) | |
| Other | 2 (11) | – | – | – | 2 (9) | 4 (2) | ||
| Observed survival (y) | ||||||||
| ≤1.0 | 40 (67) | 7 (37) | 6 (27) | 14 (82) | 2 (11) | 10 (44) | 38 (42) | 117 (47) |
| 1.1–2.0 | 15 (25) | 5 (26) | 2 (9) | 2 (12) | 5 (28) | 9 (39) | 38 (42) | 76 (31) |
| 2.1–3.0 | 4 (7) | 5 (26) | 9 (41) | – | 2 (11) | 1 (4) | 9 (10) | 30 (12) |
| >3.0 | 1 (2) | 2 (10) | 5 (23) | 1 (6) | 9 (50) | 3 (13) | 5 (6) | 26 (10) |
| Working formulation | ||||||||
| All patients | ||||||||
| Stage IVa | 20 (33) | 13 (68) | 9 (41) | 6 (35) | 10 (56) | 12 (52) | 40 (44) | 110 (44) |
| Stage IVb | 23 (38) | 5 (26) | 12 (55) | 8 (48) | 5 (28) | 9 (39) | 47 (52) | 109 (44) |
| Stage IVc | 17 (28) | 1 (5) | 1 (5) | 3 (18) | 3 (17) | 2 (9) | 3 (3) | 30 (12) |
| Excluding surgically treated patients | ||||||||
| Stage IVa | 20 (33) | 13 (68) | 2 (29) | 5 (31) | 5 (56) | 11 (55) | 27 (39) | 83 (41) |
| Stage IVb | 23 (38) | 5 (26) | 4 (57) | 8 (50) | 2 (22) | 7 (35) | 40 (57) | 40 (57) |
| Stage IVc | 17 (28) | 1 (5) | 1 (14) | 3 (19) | 2 (22) | 2 (10) | 3 (4) | 29 (14) |
The median LDLM was 3.5 (range 0.8–19; interquartile range [IQR] 2.0–5.8) cm. The largest metastasis was small (≤3.0 cm) in 116 (47%) and large (>8.0 cm) in 21 patients (8%). Small metastases predominated in 3 centers and medium-sized ones in 3; the remaining center had both in equal proportions ( Table 1 ). The median AP level was 0.70× UNL (range 0.24–8.15; IQR 0.53–1.11). It was <1.0× UNL in 179 patients (72%) and >2.0× UNL in 22 (9%). Most patients had a normal level in all but 2 centers ( Table 1 ).
Treatment of Metastases and Observed Survival
The patients had been treated on an individual basis according to the prevailing practice in each participating center ( Table 1 ). Of the entire cohort, 11 patients (4%) received best supportive care; 165 (66%) were given single-agent or combination chemotherapy with or without immunotherapy or vaccine, usually a regimen based on dacarbazine, fotemustine, or treosulfan; 4 (2%) underwent chemoembolization; 4 (2%) received interferon, usually with tamoxifen; 9 (4%) were immunized with a tumor vaccine; 47 (18%) had metastases surgically resected with or without chemotherapy, immunotherapy, or vaccine; and 4 (2%) underwent other treatments.
The observed median survival was 13.5 months (range, 0.2–129 months; IQR 6.7–22.6 months; Figure 1 , Top left) for the entire cohort; 132 patients (53%) survived >1 year, 56 (22%) >2 years, 26 (10%) >3 years, and 8 (3%) >5 years ( Table 1 ).
The observed median survival was comparable for the 3 main types of chemotherapy, based on dacarbazine, fotemustine, or treosulfan ( P = .23, log-rank test). It was 10.9 months (95% CI 9.6–13.5) for the 202 patients managed without surgery, which was significantly shorter than 26.0 months (95% CI 17.4–32.1) for the 47 patients who underwent surgical resection of metastases as part of their treatment ( P < .001). The surgically treated group had more favorable performance indices ( P = .048, nonparametric test for trend) and more favorable largest diameters of the largest metastases ( P < .001, Mann-Whitney U test), but their AP levels were comparable ( P = .24).
Verification of the Prognostic Parameters
All 3 variables on which the Working Formulation is based were significantly and independently associated with overall survival, irrespective of whether AP level was treated as a continuous or a categorical variable relative to UNL (models 1 and 2; Table 2 ).
| Variable | Beta Coefficient (SE) | Wald χ 2 | P | Hazard Ratio (95% CI) |
|---|---|---|---|---|
| Model 1 (−2 log likelihood, 2216.1) | ||||
| Performance index a | 0.440 (0.134) | 10.8 | .001 | 1.55 (1.19–2.02) |
| Serum or plasma alkaline phosphatase level b | 0.195 (0.060) | 10.4 | .001 | 1.21 (1.08–1.37) |
| Largest diameter of the largest metastasis c | 0.087 (0.022) | 15.5 | <.001 | 1.09 (1.04–1.14) |
| Model 2 (–2 log likelihood, 2219.6) | ||||
| Performance index a | 0.445 (0.134) | 11.1 | .001 | 1.56 (1.20–2.03) |
| Serum or plasma alkaline phosphatase level d | 0.239 (0.111) | 4.6 | .032 | 1.27 (1.02–1.58) |
| Largest diameter of the largest metastasis c | 0.084 (0.224) | 14.1 | <.001 | 1.09 (1.04–1.14) |
| Model 3 (−2 log likelihood, 2214.4) | ||||
| Working formulation | ||||
| Stage IVa | Reference | |||
| Stage IVb | 0.394 (0.137) | 8.2 | .004 | 1.48 (1.13–1.94) |
| Stage IVc | 1.591 (0.216) | 54.3 | <.001 | 4.91 (3.22–7.50) |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
