Date of Award

January 2025

Document Type

Thesis

Degree Name

Medical Doctor (MD)

Department

Medicine

First Advisor

Mathieu Bakhoum

Abstract

Purpose: Uveal melanoma (UM) is the most common primary intraocular cancer in adults, with a high propensity for metastasis. Despite advancements in treatment for the primary tumor, metastatic UM remains associated with poor survival outcomes. This study aims to validate previous findings on UM prognostication using a larger cohort and further investigate the role of tumor size in stratifying molecular risk groups.Methods: Using retrospective data from the Miguez et al. study (n=337) and prospective data from the Collaborative Ocular Oncology Group Study Number 2 (COOG2) dataset (n=1,577), we analyzed the prognostic utility of the Gene Expression Profile (GEP) test, Preferentially Expressed Antigen in Melanoma (PRAME) status, and tumor size in predicting metastatic risk. Statistical analyses included Cox proportional hazards regression, Kaplan-Meier survival analysis, and Receiver Operating Characteristic (ROC) curve analysis. Univariate and multivariate models were constructed to assess independent prognostic factors, with hazard ratios (HR) and 95% confidence intervals (CI) reported. Results: The study found no significant prognostic distinction between GEP Class 1A and 1B tumors. Univariate Cox regression demonstrated no significant difference in metastatic risk (HR: 1.15; 95% CI: 0.48-2.58; p=0.75 in the Initial Cohort, HR: 0.825; 95% CI: 0.481-1.37; p=0.47 in the COOG2 cohort). Log-rank tests confirmed the lack of significant difference in metastasis-free survival between Class 1A and 1B (p = 0.42). However, GEP Class 1 vs Class 2 categorization remains a significant predictor of metastasis. In a multivariate cox-proportional hazards regression analysis including GEP classification, PRAME positivity, tumor diameter, tumor size, ciliary body involvement and patient age, GEP Class 2 tumors were associated with a significantly increased risk of metastasis in both the Initial Cohort (HR: 3.07, 95% CI: 1.76-5.42, p<0.0001), and in the COOG2 Cohort (HR: 5.95, 95% CI: 4.45-8.04, p<0.0001). PRAME positivity in the COOG2 cohort was also significantly associated with metastatic risk (HR: 1.82, 95% CI: 1.43-2.34, p<0.0001). Moreover, stratifying tumors based on combined GEP/PRAME classification (GEP Class 1/PRAME Negative; GEP Class 1/PRAME Positive; GEP Class 2/PRAME Negative; GEP Class 1/PRAME Positive) yielded four distinct prognostic groups as demonstrated by Kaplan-Meier survival curves with a log-rank test (p < 0.0001). Tumor size also provided prognostic value, though whether tumor thickness or diameter was a more predictive of metastasis differed between cohorts. Stepwise Cox proportional hazards models showed that combining GEP, PRAME, and tumor size improved overall predictive performance, with C-statistics increasing from 0.770 (GEP alone) to 0.845 (GEP + PRAME + diameter) and 0.831 (GEP + PRAME + thickness). Therefore, when stratifying tumors based on GEP/PRAME classification, tumor size offered additional prognostic utility. Tumor size thresholds calculated by Youden Index within each molecular subgroup significantly differentiated metastatic risk. For example, GEP Class 1/PRAME Positive tumors were stratified by a diameter threshold of 14.05 mm (AUC: 0.748, sensitivity: 68.6%, specificity: 73.3%), while GEP Class 2/PRAME Negative tumors showed a thickness threshold of 8.05 mm (AUC: 0.691, sensitivity: 44.3%, specificity: 84.3%). Kaplan-Meier survival analyses confirmed that these thresholds significantly stratified metastatic risk (p<0.001). Moreover, when stratifying tumors based on size tertiles, GEP classification remained significant for all subgroups by cox proportional hazards regression analysis (p < 0.0001). In contrast, when combining tumor size and GEP categorization, PRAME’s prognostic significance varied by tumor size, showing stronger predictive value in larger tumors. In GEP Class 1 tumors, PRAME positivity was not significantly associated with metastasis in small tumors (HR: 1.33, 95% CI: 0.38-4.71, p=0.659) but was significant in large tumors (thickness HR: 6.21, 95% CI: 2.63-14.70, p<0.0001; diameter HR: 4.40, 95% CI: 1.98-9.77, p=0.0003). Similarly, in GEP Class 2 tumors, PRAME was more prognostic in larger tumors, particularly in the largest diameter tertile (HR: 2.24, 95% CI: 1.58-3.16, p<0.0001). Conclusion: These findings challenge the prognostic utility of GEP Class 1A vs 1B distinction and support a combined GEP/PRAME classification system refined by tumor size. Integrating tumor size into prognostic models enhances risk stratification which stands to improve patient management and clinical outcomes in UM.

Comments

This thesis is restricted to Yale network users only. This thesis is permanently embargoed from public release.

Download

Share

COinS