Predictive models for improved prognostication and selection of neoadjuvant and adjuvant systemic chemotherapy in upper tract urothelial cell carcinoma.

Abstract

456 Background: Chemotherapy is still underutilized in management of upper tract urothelial carcinoma (UTUC). We created pre and post-operative predictive tools combining independent prognostics to guide selection of patients for neoadjuvant and adjuvant chemotherapy. Methods: From the UTUC collaboration database (1,453 patients who underwent radical nephroureterectomy [RNU] at 13 academic institutions); a preoperative predictive model was created using 659 patients in whom all preoperative prognostic variables were available and a post-operative model was created using 586 patients with non-metastatic/ high-grade UTU. After multivariable survival analyses, a backward step-down selection process was applied to create a preoperative nomogram. Internal validation was performed using 200 bootstrap resamples. For the postoperative model, TALL score was created based on the sum of the independent prognostic variables. Results: Preoperative model: Grade, architecture and location of the tumor were independently associated with nonorgan confined disease. A nomogram including these 3 variables achieved 76.6% accuracy in predicting nonorgan confined upper tract urothelial cancer. Postoperative model: TALL score (1-7) was the sum of T ( ≤ T1 = 1, T2 = 2, T3 = 3 and T4 = 4), A (papillary = 0 and sessile = 1), LVI (absent = 0 and present = 1) and L (lymphadenectomy = 0 and no lymphadenectomy = 1). Five-year disease-free survival (DFS) and cancer-specific survival (CSS) were stratified into four risk categories according to the TALL score: low (TALL 0-2; 86 % DFS and 90 % CSS), intermediate (TALL = 3; 71 % DFS and 75 % CSS), high (TALL = 4; 57 % DFS and 58 % CSS) and very high risk (TALL ≥ 5; 34 % DFS and 38 % CSS) using Kaplan-Meier survival analyses. TALL score was externally validated in a single-center cohort of 85 UTUC patients. Conclusions: We developed validated multivariable prognostic tools for prediction of locally advanced UTUC and oncological outcomes after RNU for UTUC. These prediction models can be used for patient counseling, selection for neoadjuvant/adjuvant systemic therapies and design of clinical trials.