5084 Background: Germ cell tumors (GCTs) are the most common solid malignancy in young adult men. Although GCTs are highly curable, when patients do succumb to disease, they have the greatest average number of life years lost of any adult cancer. Currently, patients are risk-stratified using clinical markers defined by the International Germ Cell Cancer Consensus Group (IGCCCG). Identification of molecular markers could improve outcome prediction. Methods: Expression profiling was performed on 74 non-seminomatous tumors (NSGCT) from patients who had received cisplatin based chemotherapy (training set) and 34 similarly treated NSGCT patients (test set). A modified version of Prediction Analysis for Microarrays (PAM) was used to develop a classifier for the binary endpoint of 5 year overall survival (5yOS). We also developed a predictive score for overall survival (OS) based on a weighted sum of the top genes derived using the univariate Cox model. Results: PAM identified a set of 170 transcripts that predicted 5yOS in the training set with a cross-validated classification rate of 60%. In the test set, the predictive gene set correctly classified 90% of the specimens. The predicted good and bad outcome groups showed significant separation in a Kaplan-Meier analysis. For the OS endpoint, we developed a 10 gene model that had a cross-validated predictive accuracy (concordance index, CI) of 0.66. Application of this model to the test set resulted in a CI of 0.83. Dichotomization of the samples based on the median score resulted in significant separation of the patients based on outcome in a Kaplan-Meier analysis. For both endpoints, the gene-based predictor was an independent prognostic factor in a multivariate model that included IGCCCG risk stratification (P<0.01 for both). Conclusions: We have developed and validated two gene-based models that predict outcome in GCT patients. These gene sets should have clinical utility for outcome prediction in GCT patients and should provide novel targets for therapeutic intervention. No significant financial relationships to disclose.
Read full abstract