e15045 Background: Patients with stage ≥T2 transitional cell carcinoma (TCC) of the bladder are typically treated with radical cystectomy (RC) + pelvic node dissection (PND) +/- chemotherapy. Adjuvant radiation therapy (RT) has no clearly defined role, although local-regional failure (LF) following RC remains a significant problem. Adjuvant RT could potentially reduce LF, but the morbidity reported in the past for such RT discouraged its use. Modern RT techniques with improved normal tissue sparing have rekindled interest in adjuvant RT. Identifying subgroups with differing LF risk could facilitate patient selection for adjuvant RT protocols. Methods: From 1990 - 2008, 442 patients with TCC of the bladder were followed after RC + PND +/- chemotherapy at the Hospital of the University of Pennsylvania. 132 (30%) received neoadjuvant or adjuvant chemotherapy. LF was scored as any failure within the pelvis before or within 3 months of distant failure. Univariate, multivariate Cox regression, and recursive partitioning analyses (RPA) were performed to identify subgroups with differing LF risk. Results: On univariate analysis, stage ≥pT3, nodes removed (<10 vs. ≥10), positive margins, positive nodes (pN+), node density, hydronephrosis, lymphovascular invasion, mixed histology disease and type of surgical diversion were significant predictors of LF, while chemotherapy, gender, race and BMI were not. Age at surgery was a marginal predictor of LF. On multivariate analysis, stage ≥pT3, nodes removed (<10 vs. ≥10), positive margins, pN+ and age were significant independent predictors of LF with hazard ratios of 3.36, 2.98, 2.08, 1.84, 1.04(per yr), respectively (p <0.05 for all variables). Different cut points and definitions were explored for pT stage, pN status, number of nodes removed and chemotherapy status, but none improved the regression models. RPA identified 3 patient subgroups with significantly different LF risk (logrank/p<0.01): low (<pT3), intermediate (≥pT3 + >10 nodes), and high (≥pT3 + <10 nodes) with 5 yr LF rates of 11%, 31% and 73%. Conclusions: The risk of LF after RC varies significantly among different subgroups of TCC patients. This risk stratification model could facilitate patient selection in future trials.