519 SPARC (SURVIVAL PREDICTION AFTER RADICAL CYSTECTOMY): A MULTIFACTORIAL OUTCOME PREDICTION MODEL FOR PATIENTS UNDERGOING RADICAL CYSTECTOMY FOR BLADDER CANCER

INTRODUCTION AND OBJECTIVES: While multiple independent clinicopathologic variables associated with outcome following radical cystectomy (RC) for bladder cancer (BC) have been identified, limited prediction tools exist to facilitate an individualized risk assessment based on the relative weight of these various features. Herein, we developed an outcome prediction model based on clinical and pathologic characteristics found to be significantly associated with death from BC after RC. METHODS: We evaluated 1829 patients who underwent RC without neoadjuvant therapy at our institution between 1980–2008. Of these, 1,751 with non-metastatic, non-small cell histology were identified. Univariate analysis evaluating variables associated with BC specific mortality was conducted. A multivariate model was obtained using a stepwise selection process from variables significant at the 0.05 level. A scoring system based on the -coefficients of this model was created, assigning 1 point for each parameter estimate value of 0.20. RESULTS: Median follow-up after RC was 10.7 years (interquartile range 7.7, 15.7), during which time 652 patients experienced death due to BC. On multivariate analysis (Table), race, Charlson comorbidity index, ECOG status, current smoking status, preoperative hydronephrosis, pathologic stage, multifocality, lymphovascular invasion, and concurrent CIS were significantly associated with BC death. Cumulative scores from these variables were used to stratify patients into quintiles according to risk score of BC death as 4, 5-8, 9-12, 13-16, and 17. 5-year cancer specific survival (CSS) from the lowest to highest risk quintile was 95%, 82%, 58%, 35%, and 17%, respectively (p 0.0001). The c-index for this model was 0.76. CONCLUSIONS: We present a model for individualizing estimation of CSS following radical cystectomy. Pending external validation, these data may be used for patient counseling, specifically with regard to recommendations for adjuvant therapy and surveillance frequency, as well as in clinical trial development.