A SEER based SHAP interpretable prognostic nomogram for overall survival in surgically treated patients with colorectal cancer and synchronous liver metastasis.

BACKGROUND: Colorectal cancer (CRC) with liver metastasis is associated with poor prognosis and substantial heterogeneity in survival outcomes. Accurate risk stratification is essential for individualized treatment planning, yet reliable and interpretable prognostic tools remain limited. Therefore, this study aimed to develop and validate a prognostic model for overall survival (OS) in a predominantly surgically treated cohort of patients with CRC and synchronous liver metastasis at diagnosis based on Surveillance, Epidemiology, and End Results (SEER) data, and to enhance model interpretability using SHapley Additive exPlanations (SHAP) analysis. METHODS: Data of patients with CRC with liver metastasis diagnosed between 2011 and 2018 were obtained from the SEER database. OS was used as the endpoint. Kaplan-Meier analysis and univariate and multivariable Cox regression were performed, and multivariable Cox regression was used to identify independent prognostic factors. A nomogram was constructed, and its performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). SHAP analysis was further applied to improve model interpretability. RESULTS: Kaplan-Meier analysis showed a poor overall prognosis, with a median OS of 25 months. Multivariable Cox regression identified age, tumor location, differentiation, metastatic stage, preoperative carcinoembryonic antigen (CEA), perineural invasion, bone metastases, metastasectomy, postoperative chemotherapy, and lymph node ratio (LNR) as independent prognostic factors for OS. Based on these variables, a nomogram was developed to predict 12-, 36-, and 60-month OS. The model showed good discrimination, calibration, and clinical utility in both the training and validation sets. SHAP analysis further improved model interpretability by showing the relative contribution of each predictor to survival estimation. CONCLUSIONS: We developed and internally validated a prognostic nomogram for OS in patients with CRC with liver metastasis based on SEER data. The model showed acceptable predictive performance and may provide supportive information for individualized prognosis evaluation in this population.