Gene expression profile of primary cutaneous melanomas to distinguish between low and high risk of metastasis.

Abstract

9022 Background: Wide variability exists in metastatic rates of primary cutaneous melanoma (CM), even within TNM stage groupings. We developed an RT-PCR based gene expression profile (GEP) test to predict distant metastases in early stage CM. Methods: RNA was isolated from formalin-fixed, paraffin embedded biopsies or wide excisions of primary CM from patients with stage 1-4 CM (87% stage 1-2) converted to cDNA and analyzed using RT-PCR. All analyses were done using JMP Genomics and WinSTAT. Radial Basis Machine (RBM) modeling was used to predict metastasis-free survival (MFS) for the training set of 149 samples. Independent validation was performed on an additional 107 CM samples. RBM reports a binary Class 1 (low risk) and Class 2 (high risk). Results: Analysis of the training set resulted in a model with 85% receiver operating characteristic (ROC) and sensitivity of 82%. Prediction of metastatic risk for the validation set resulted in a 90% ROC and sensitivity of 89%. 30 of 33 stage 1-3 cases with a known metastatic event were accurately called Class 2. Kaplan-Meier analysis showed 5-year MFS rates of 88% and 25% for predicted Class 1 and Class 2, respectively, in the training set (p<0.0001, overall MFS=60%). Similarly, MFS was 95% and 26% for Class 1 and Class 2, respectively, in the independent validation set (p<0.0001, overall MFS=67%). Univariate Cox regression analysis of the validation set revealed that GEP, AJCC stage, Breslow thickness, and ulceration were each predictors of metastatic risk (HR=27.2, 11.4, 3.0, and 11.6, respectively, p<0.002 for each). Multivariate analysis showed GEP and AJCC stage were independent predictors of risk (HR=8.4 and 6.4 respectively, p<0.007 for both). Conclusions: This GEP signature provides an accurate stratification of metastatic risk in the training and validation samples independent of all other histologic factors. This test may serve as a prognostic tool for outcomes in patients with melanoma and for stratifying patients for clinical trials.