External Validation of the ARCAD Nomogram in a Real-World Cohort with Stage 4 Colorectal Cancer.

INTRODUCTION: The ARCAD-CRC nomogram predicts 1-year survival in clinical trial populations with stage 4 colorectal cancer (CRC). This study evaluated its performance in a real-world cohort. METHODS: Patients with stage 4 CRC were identified from the Flatiron database. A multivariable Cox proportional hazards model for overall survival was applied to fit the reduced model derived from the published full ARCAD nomogram, as some variables were unavailable in the database. Validation of the reduced model used C-index, while predictive performance was assessed using time-dependent ROC analysis and calibration in the Flatiron real-world cohort (FRWC). RESULTS: A total 9710 patients, including 5740 deceased, were analyzed. The FRWC was older (64 vs. 62 years) and had poorer ECOG-PS (15% vs. 4% with ECOG 2+) compared to the ARCAD cohort. The reduced nomogram's calibration was comparable to the full model (C-index 0.67 vs. 0.68). In the FRWC, the reduced nomogram showed good discrimination, with time-dependent AUROC at the 12-month time horizon of 0.741 (95% CI 0.729-0.753). Calibration plots indicated the reduced ARCAD-CRC nomogram performed quantitively well with a calibration slope of 1.010 (95% CI 0.966-1.054). The calibration intercept was -0.160 (95% CI -0.189, -0.131), indicating systematic underestimation of 1-year survival in the FRWC. The predicted and observed 1-year OS using the reduced ARCAD nomogram in the FRWC were 0.640 (95% CI 0.637, 0.644) and 0.694 (95% CI 0.684-0.703), respectively. CONCLUSIONS: Our findings show that the ARCAD-CRC nomogram is a useful tool for predicting 1-year survival in a real-world cohort. The observed underestimation of survival may be due to advances in CRC treatments, as the FRWC cohort (2013-2020) was collected later than the original ARCAD cohort (1997-2012).