Group, Adjuvant Colon Cancer Endpoints (ACCENT)

3618 Background: Current prognostic tools and staging systems in CC use relatively few patient (pt) characteristics; including additional covariates may improve prediction. Using the large ACCENT database, we constructed clinically based NGs for OS and TTR in stage III CC that better separate pts into risk groups compared to AJCC v7 staging. Methods: 15,936 stage III pts accrued to phase III clinical trials since 1989 were used to construct Cox models for TTR and OS. Variables included age, sex, race, BMI, performance status, tumor grade, tumor stage, ratio of positive lymph nodes to nodes examined, number/location of primary tumors (any multiple versus single left, right, or transverse), and treatment (5FU variations vs. 5FU with oxaliplatin or irinotecan). Missing data (<18%) were imputed, continuous variables modeled with splines, and clinically relevant pairwise interactions included if p < 0.001. Final models were internally validated via bootstrapping for corrected calibration and C-indices for survival data. NG-defined risk tertiles were compared to AJCC v7 stage III for observed 3-year (yr) TTR and 5-yr OS for a subset of 7400 pts with complete data. Results: All variables were statistically and clinically significant for OS; age and race did not predict TTR. No meaningful interactions existed. NGs for OS and TTR were well calibrated and associated with C-indices of 0.66 and 0.65, respectively, vs. 0.58 and 0.59 for AJCC. NG risk tertiles were better separated than AJCC groups, (3-yr TTR, 5-yr OS below), with fewer mid-risk NG pts. Removing treatment from NGs did not affect performance (C=0.66 for OS and 0.65 for TTR). Conclusions: The proposed ACCENT NGs are internally valid and have the potential to aid prognostication, decision-making, and patient/physician communication in pts with stage III CC. [Table: see text]