e16524 Background: Renal cell carcinoma (RCC) patients suffering from local recurrence (LR) herald variable risk of second progression even after curative surgery for LR. Further risk stratification of these patients will facilitate selection of candidates for adjuvant immunotherapy. Here we evaluate the prognostic significance of tumor grade variation (GV) between the primary tumor (PT) and LR tumor in nonmetastatic RCC in the largest study to date. Methods: We reviewed Research of Multi-institution in East-China on Malignant and Benign Epithelial Renal Tumors (REMEMBER) database to identify ccRCC or pRCC patients who underwent surgery for LR. All pathology slides were centrally reviewed by three independent senior genitourinary pathologists and patients were assigned to groups based on GV as upgrading (WHO/ISUP Ⅰ/Ⅱ for PT grade and Ⅲ/Ⅳ for LR grade), downgrading (WHO/ISUP Ⅲ/Ⅳ for PT grade and Ⅰ/Ⅱ for LR grade), stable low-grade (WHO/ISUP Ⅰ/Ⅱ for both PT and LR grade) and stable high-grade groups (WHO/ISUP Ⅲ/Ⅳ for both PT and LR grade). To evaluate the predictive value of GV, four tumor grade measures (PT grade, LR grade, highest grade for the PT and LR, GV) were separately incorporated for model development. The primary end point was overall survival (OS), defined as the time from LR surgery to death from any cause or last follow-up. A multivariable cox regression analysis was performed to explore associations between timing and survival. Results: 345 patients were identified 114 stable low-grade (33.1%), 56 downgrading (16.2%), 97 upgrading (28.1%), and 78 stable high-grade (22.6%). Median OS were highest among patients with stable low-grade (114 stable low-grade vs. 86 Downgrading vs. 59 Upgrading vs. 33 Stable high-grade months). Univariate analysis indicates that PT, LR and GV were significant correlated with poor OS (all P values < 0.05). After separately constructing the PT grade model (#1) via LASSO cox regressions for predicting OS, LR grade model (#2), highest grade model (#3) and GV model (#4) were built by simply replacing the variable of PT grade, it turned out that GV model (#4) achieved the better discriminatory ability (AUC of model (#4) vs. model (#1), 0.812 vs. 0.743; model (#4) vs. model (#2), 0.812 vs. 0.788; model (#4) vs. model (#3), 0.812 vs. 0.762, all p values < 0.05). Conclusions: This is the first study showing that GV is independently associated with clinical outcomes in patients with LR. Routine reporting of LR upgrading or downgrading in relation to the PT and assessment of GV can facilitate more informed treatment decisions by tailoring strategies to an individual patient's risk of progression.