MP39-09 A COMBINATION OF GENOMIC MARKERS AND CLINICAL VARIABLES PROVIDES SUPERIOR PROGNOSTIC PERFORMANCE IN HIGH-RISK BLADDER CANCER FOLLOWING CYSTECTOMY

Abstract

INTRODUCTION AND OBJECTIVES: Nearly half of muscleinvasive bladder cancer patients succumb to their disease following cystectomy. Selecting appropriate candidates for adjuvant therapy is currently based on clinical parameters with limited predictive power. Here we aimed to develop and validate a genomic-based signature that can better identify patients at high risk of recurrence following radical cystectomy. METHODS: A set of neoadjuvant chemotherapy-naive patients who underwent radical cystectomy for muscle-invasive urothelial carcinoma of the bladder from 1998 to 2004 was used as the study population. Whole transcriptomes of archival tumors were profiled using 1.4 million-feature oligonucleotide microarrays. A discovery cohort (n 1⁄4 133) was used to develop a genomic classifier (GC) for predicting recurrence. A multivariable classifier based on typical clinical covariates (CC) was also developed for the same endpoint. Finally, CC and GC were combined into a genomic-clinical classifier (G-CC). Performances of GC, CC and G-CC were compared to a post-cystectomy nomogram from the International Bladder Cancer Nomogram Consortium (IBCNC). Classifier performance was assessed by area under receiver-operating characteristic curves (AUCs) in the discovery cohort and an independent validation cohort (n 1⁄4 66). RESULTS: A 15-gene GC was developed on the discovery cohort (median follow-up, 9.3 years) with AUC 1⁄4 0.88 for predicting recurrence, which was higher than any individual clinical variable, and the IBCNC (AUC 1⁄4 0.73) and CC (AUC 1⁄4 0.81) nomograms. When genomic markers were combined with IBCNC (G-IBCNC) and CC (G-CC), AUCs of the models increased to 0.89 and 0.93, respectively. When the locked classifiers were blindly applied to the validation cohort (median follow-up, 10.8 years), GC retained its superior performance compared to individual clinical variables (AUC 1⁄4 0.77). Further, addition of GC to clinical nomograms improved their performance (IBCNC vs. G-IBCNC, 0.73 vs. 0.82; CC vs. G-CC, 0.78 vs. 0.86). G-CC high-risk patients had elevated recurrence probabilities (p < 0.001), with multivariable analysis indicating that the genomic component was the most significant predictor of recurrence (p 1⁄4 0.005). When compared to prior signatures, GC showed the best performance in validation. CONCLUSIONS: We present a validated G-CC that shows superior performance over clinical models for predicting post-cystectomy cancer recurrence. Such transcriptomic approaches can identify robust biological predictors of post-cystectomy outcomes.