S499 Development and Internal Validation of a Simplified Clinical Prediction Model for Completion Colectomy in Patients With T1N0M0 Colorectal Cancer

Introduction: Completion colectomy is variably pursued following local excision for T1 colorectal cancer (CRC), often guided by histopathologic features. We aimed to develop a simplified model using Surveillance, Epidemiology, and End Results (SEER) database to predict the likelihood of completion colectomy in patients with T1N0M0 CRC on endoscopy. Methods: This was a retrospective analysis of SEER data from 2004–2015. Adults with histologically confirmed T1N0M0 CRC and known surgical treatment status were included. The primary outcome was receipt of completion colectomy following local excision. A multivariable logistic regression model was developed using Least Absolute Shrinkage and Selection Operator (LASSO) to identify predictive variables, followed by construction of a simplified logistic regression model. Final predictors included primary tumor site, race, sex, and tumor grade. Model discrimination was assessed using 10-fold cross-validation and bootstrap resampling. Calibration was evaluated using the GiViTI calibration belt. A manual scoring formula was derived using coefficient values from the final model. Results: Among 3,289 patients (55% men; 80% White), 1,399 (42.5%) underwent completion colectomy. The final model demonstrated excellent performance with a cross-validated area under the curve (AUC) of 0.84 (95% confidence interval [CI]: 0.81–0.85), bootstrapped AUC of 0.83 (95% CI: 0.81–0.84), and standard receiver operative characteristic (ROC) AUC of 0.84 (95% CI: 0.82–0.85). Calibration was strong (GiViTI P = 0.85), indicating no significant deviation from ideal prediction. Using a threshold probability ≥0.60, the model achieved 73.9% sensitivity, 83.9% specificity, 77.3% positive predictive value, 81.6% negative predictive value, and 78.6% overall accuracy. The full manual scoring equation was: logit (P) = 1.6807 0.5543 × Appendix 0.2014 × Ascending colon 0.0929 × Hepatic flexure − 0.9138 × Transverse colon − 0.9886 × Splenic flexure − 0.6383 × Descending colon − 2.8347 × Sigmoid colon − 0.3868 × Overlapping lesion − 1.8755 × Colon, NOS − 3.8004 × Rectosigmoid junction − 3.0419 × Rectum − 0.5823 × Asian or Pacific Islander − 0.1340 × Black − 1.1498 × Unknown race − 0.3480 × White − 0.1091 × Male 0.1875 × Poorly differentiated 0.3356 × Undifferentiated − 0.2220 × Unknown grade 0.0883 × Well differentiated. Conclusion: We developed a simplified model that accurately predicts completion colectomy after local excision for T1N0M0 CRC. A web-based calculator is available at https://tinyurl.com/Colectomy-Calculator. External validation is needed.