To determine whether any conventional clinical prognostic factors for metastasis from uveal melanoma retain prognostic significance in multivariate models incorporating gene expression profile (GEP) class of the tumor cells.
Prospective, interventional case series with a prognostic model.
Single-institution study of GEP testing and other conventional prognostic factors for metastasis and metastatic death in 299 patients with posterior uveal melanoma evaluated by fine-needle aspiration biopsy (FNAB) at the time of or shortly prior to initial treatment. Univariate prognostic significance of all evaluated potential prognostic variables (patient age, largest linear basal diameter of tumor [LBD], tumor thickness, intraocular location of tumor, melanoma cytomorphologic subtype, and GEP class) was performed by comparison of Kaplan-Meier event rate curves and univariate Cox proportional hazards modeling. Multivariate prognostic significance of combinations of significant prognostic factors identified by univariate analysis was performed using step-up and step-down Cox proportional hazards modeling.
GEP class was the strongest prognostic factor for metastatic death in this series. However, tumor LBD, tumor thickness, and intraocular tumor location also proved to be significant individual prognostic factors in this study. On multivariate analysis, a 2-term model that incorporated GEP class and largest basal diameter was associated with strong independent significance of each of the factors.
Although GEP test is the most robust prognostic indicator in uveal melanoma and early studies of mostly larger tumors found that no clinicopathologic factors had significant prognostic value independent of GEP, our single-center study, which included a substantial proportion of smaller tumors, showed that both GEP and LBD of the tumor are independent prognostic factors for metastasis and metastatic death in multivariate analysis.
In spite of extensive research over the past 50 years and many improvements in clinical diagnosis and treatment of posterior uveal melanoma, the mortality from metastasis has remained virtually unchanged for patients with tumors of a particular tumor size category during this period. While the cause of uveal melanoma remains elusive, multiple clinical, histopathologic, cytopathologic, chromosomal, and molecular biological prognostic factors for metastasis and metastatic death have been identified over the years. The most robust of these prognostic factors in patients with uveal melanoma currently appears to be a gene expression profile (GEP) test, available commercially at this time as the DecisionDx-UM test (Castle Biosciences, Friendswood, Texas, USA). The test categorizes tumors as either class 1 (low metastatic risk) or class 2 (relatively high metastatic risk within 3–5 years). Recently, class 1 tumors have been further distinguished into class 1A (lowest metastatic risk) and class 1B (moderate long-term metastatic risk). Although GEP class 2 assignment is strongly correlated with various chromosomal and clinical prognostic factors, previous publications have shown that no clinical features or chromosomal abnormalities modify the prognostic impact of GEP significantly in multivariate analyses.
Our group has used the GEP test developed and reported by Onken and associates in our practice since 2007. The principal purpose of our study was to determine whether any generally accepted clinical prognostic factors for metastasis retained independent significant prognostic value in multivariate analysis with GEP class in our series of cases. The principal reason for believing that we might find different results than those reported previously was the substantial proportion of patients with smaller tumors in our series of cases.
Our study was approved by the University of Cincinnati College of Medicine’s Institutional Review Board (IRB) to prospectively evaluate the clinical features, cytopathology, and gene expression profile of tumor cells sampled by fine-needle aspiration biopsy (FNAB) of patients with posterior uveal melanoma. Patients have signed a written consent to be enrolled in the GEP validation trial. Because FNAB for cytology had already been part of our service’s standard of care, patients are routinely consented for FNAB as a diagnostic procedure or for prognostic purposes prior to treatment of their posterior uveal tumor. The GEP test was incorporated into our routine testing of FNAB specimens after its validation in 2011. Subsequent to that, patients consented to this diagnostic/prognostic test as part of their care. Further, the same IRB has approved de-identified data collection and analysis of patients managed by the Ocular Oncology Service. The IRB monitors the ocular oncology database documents and HIPAA compliance yearly.
This is a prospective, interventional case series with a prognostic model. Our base group for this analysis consisted of all patients on whom we performed FNAB of an intraocular tumor believed to be a posterior uveal melanoma during the interval September 1, 2007 through December 31, 2012 (n = 330). The techniques we used for FNAB have been described. We excluded from this group (1) patients in whom cytomorphologic analysis and/or immunocytochemical staining of tumor aspirate was not consistent with a uveal melanoma (n = 6); (2) patients whose tumor was confined to the iris (n = 6), iris and ciliary body (n = 9), or ciliary body only (n = 1); (3) patients whose FNAB was performed more than 1 month prior to or 1 or more days after initial treatment of the intraocular tumor (n = 4); (4) patients whose tumor aspirates were submitted only for cytology and not GEP testing (n = 1); (5) patients whose tumor aspirates were submitted for GEP testing but not for cytology (n = 4); (6) patients whose GEP testing was unsuccessful for any reason (eg, failure of the technique to identify the control genes [n = 3], laboratory technician error [n = 4]); (7) patients for whom we were unable to measure the tumor (largest basal diameter or thickness) prior to biopsy and treatment (n = 1); and (8) any patients who were gross outliers from the remainder of the group (n = 1; a 4-year-old African-American child with uveal melanoma filling the eye plus extraocular extension to the orbit). Thirty-one patients were excluded because of 1 or more of these criteria. Consequently, our final analysis group consisted of 299 patients. All of these patients had a complete dataset of baseline clinical variables evaluated prospectively (age of patient at the time of biopsy, sex of the patient, largest linear basal diameter [LBD] of the patient’s tumor at the time of biopsy, maximal thickness of the patient’s tumor at the time of biopsy, intraocular location of the uveal tumor [exclusively choroidal, involving ciliary body], melanoma cytomorphologic subtype based on the tumor cells obtained by FNAB, and GEP classification of the tumor cells [class 1, class 2]). Because approximately one-third of our early cases preceded the distinction between class 1A and 1B, we kept all class 1 patients in the same category, understanding we would see a few metastases in this group related to the 1B subclass.
Finally, using the collected information on LBD of the primary intraocular tumor, its topographic location of the posterior uveal tumor (exclusively choroidal vs involving the ciliary body), and absence vs presence and extent of transscleral tumor extension, each case was assigned a prognostic stage according to the American Joint Commission on Cancer’s Tumor-Node-Metastasis (TNM) system (2010 edition).
We computed conventional descriptive statistics (mean, standard deviation, minimal, maximal, median) for all continuous numerical variables (age of patient, largest LBD of the tumor, maximal tumor thickness) and frequency distributions (number and percentage of cases) for all evaluated study variables. For intrinsically dichotomous variables (sex of the patient, intraocular tumor location, GEP class), we divided patients according to those discrete categories. For intrinsically numerical variables, we selected conventional values for those variables as cut points between trichotomous subgroups (for age: 50 and 70 years; for LBD: 10 and 15 mm; for maximal tumor thickness: 3.5 and 7 mm) and the mean or median value of the variable in our entire group as the cut point between dichotomous subgroups. Cytomorphology of melanoma cells was based on the multichotomous specification of cell type contained in the official pathology report of each case. These reports, based on the modified Calendar classification, mentioned the following original categories: insufficient specimen for classification, borderline melanocytic cells, spindle cell type, mixed cell type, epithelioid cell type, necrotic melanocytic cells, and melanoma of unspecified cell type. Although tumors may be composed of a mixture of different cells, we stipulated that the more aggressive feature present in at least 10% of the retrieved sample would determine the cytomorphologic classification for the purpose of this analysis. Considering that all of the insufficient aspirates in this series came from tumors sampled in at least 2 sites for cytology and prior studies have shown this finding to be associated with a low subsequent mortality rate from metastasis (similar to spindle cell melanoma), we combined these cases for data analysis into a subgroup termed spindle cell type . Similarly, because our prior studies of prognostic significance of melanoma cell type have shown patients with unspecified melanoma cell type to have similar mortality rates to those of patients having a uveal melanoma of mixed cell type, we combined these cases for data analysis into a subgroup termed mixed cell type . We also combined our cases of epithelioid and necrotic melanomas into a subgroup termed epithelioid cell type for data analysis. Further, before grouping the cell type categories, we also analyzed the outcome of the different cell types separately in this dataset and observed similar behavior among some of them. Such findings confirmed the proposed grouping used herein.
Cross-correlations Between Gene Expression Profile Class of Tumor Cells and Other Evaluated Variables
To evaluate the levels of correlation between GEP class of tumor cells and other evaluated variables in this study, we performed cross-table analysis with computation of χ 2 values as an expression of the homogeneity vs heterogeneity of the distributions of the evaluated variables in the GEP class subgroups. For this testing, we recoded the intrinsically multichotomous categorical variable cell type into a dichotomous variable with categories epithelioid cells absent and epithelioid cells present . For the continuous numerical variables LBD of tumor, tumor thickness, and patient age, we recoded the original values into trichotomous ordinal categories using 10 mm and 15 mm as the cutoff points for LBD, 3.5 mm and 7 mm as the cutoff points for thickness, and 50 and 70 years as the cutoff points for patient age. We computed the χ 2 statistic for each of these relationships. An alpha = 0.05 was assigned as the probability level for statistical significance.
Univariate Prognostic Factor Analysis
To assess the variable-by-variable prognostic significance of the evaluated variables in this study, we computed Kaplan-Meier survival curves for the dichotomous categories of each variable with computation of the log-rank test statistic for each comparison. We also computed individual prognostic significance of each of the variables using univariate Cox proportional hazards modeling. For this testing, we computed the beta coefficient for each variable, its log-rank test statistic, its significance level, its relative risk ratio, and the 95% confidence interval for the relative risk ratio.
Multivariate Prognostic Factor Analysis
We assessed the independent prognostic significance of the various evaluated study variables using both stepwise step-up and stepwise step-down Cox proportional hazards modeling. In the step-down analysis, we started with all variables in the model and eliminated them one by one (the variable with the lowest independent significance) until a model was identified that contained only significant ( P < .05) prognostic variables. In the step-up analysis, we started with the most predictive variable and added variables in step-wise fashion provided that each of the variables in the multi-term model retained a significance level <.05.
Baseline Characteristics of Patients and Tumors
The characteristics of the 299 study patients and their posterior uveal melanomas at the time of FNAB are summarized in Table 1 . The patients ranged in age from 17.6 years to 99.2 years (mean age = 62.0 years [standard deviation = 14.6 years], median age = 62.3 years). The tumor largest LBD ranged from 4.0 mm to 22 mm (mean LBD = 11.9 mm [standard deviation = 3.5 mm], median LBD = 12.0 mm) and the maximal tumor thickness ranged from 1.4 mm to 16 mm (mean thickness = 5.6 mm [standard deviation = 3.0 mm], median thickness = 5.1 mm). The sex distribution was nearly equal (50.2% male, 49.8% female). Seventy-nine of the 299 tumors (26.4%) involved the ciliary body while 220 (73.6%) were exclusively choroidal. The TNM stage of the posterior uveal melanoma in our study patients was stage 1 in 84 (28.1%), stage 2 in 156 (52.2%), and stage 3 in 59 (19.7%).
|Age of patient (y)|
|>50 but ≤70||154||(51.5)|
|Sex of patient|
|Largest basal diameter of tumor (mm)|
|>10 to ≤15||137||(45.8)|
|Maximal thickness of tumor (mm)|
|>3.5 to ≤7||121||(40.5)|
|Intraocular tumor location|
|Involving ciliary body||79||(26.4)|
Classification of Melanoma Cells Obtained by Fine-Needle Aspiration Biopsy
FNAB of the intraocular tumor performed at the time of or shortly prior to initial treatment of the tumor yielded the results shown in Table 2 . The cells contained in the aspirates were classified cytomorphologically as epithelioid cells present in 136 cases (45.5%) and as epithelioid cells absent in 163 cases (54.5%). The cells were classified by GEP testing as class 1 in 211 cases (70.6%) and as class 2 in 88 cases (29.4%).
|Cytopathologic classification of melanoma cells obtained by FNAB|
|Spindle cell type||163||(54.5)|
|Mixed cell type||80||(26.8)|
|Epithelioid cell type||56||(18.7)|
|Gene expression profile class of melanoma cells|
Cross-correlations Between Gene Expression Profile Class of Tumor Cells and Other Prognostic Covariates
The cross-correlations between the evaluated clinical and cytopathologic variables in this study and the GEP class of the melanoma cells obtained by FNAB in this series are shown in Table 3 . Several of the associations were strong; more specifically, GEP class 2 was associated with older patient age, largest basal diameter of tumor, tumor thickness, ciliary body involvement, and presence of epithelioid cells.
|Variable||Gene Expression Profile Class||χ 2||df||P|
|Class 1||Class 2|
|Number (Row)||Number (Row %)|
|Age of patient (y)|
|≤50||47 (90.4)||5 (9.6)|
|>50 but ≤70||110 (71.4)||44 (28.6)||16.9||2||<.0001|
|>70||54 (58.1)||39 (41.9)|
|Sex of patient|
|Male||107 (71.3)||43 (28.7)||0.085||1||.44|
|Female||104 (69.8)||45 (30.2)|
|Largest basal diameter of tumor (mm)|
|≤10||97 (85.8)||16 (14.2)|
|>10 to ≤15||86 (62.8)||51 (37.2)||21.0||2||<.0001|
|>15||28 (57.1)||21 (42.9)|
|Maximal thickness of tumor (mm)|
|≤3.5||82 (85.4)||14 (14.6)|
|>3.5 to ≤7||81 (66.9)||40 (33.1)||16.7||2||<.0001|
|>7||48 (58.5)||34 (41.5)|
|Intraocular tumor location|
|Exclusively choroidal||169 (76.8)||51 (23.2)||15.7||1||<.0001|
|Involving ciliary body||42 (53.2)||37 (46.8)|
|Cytopathologic classification of tumor cells obtained by FNAB|
|Spindle cell type||131 (80.4)||32 (19.6)|
|Mixed cell type||45 (56.2)||35 (43.8)||17.19||2||<.0001|
|Epithelioid cell type||35 (62.5)||21 (37.5)|