Predicting extensive metastasis in postoperative oligometastatic colorectal cancer.

Abstract

Purpose: Oligometastatic colorectal cancer (OMCRC) patients can achieve long-term disease control with multidisciplinary treatment. However, the development of extensive metastasis worsens prognosis and restricts treatment options. This study aims to develop a predictive model for extensive metastasis in OMCRC to assist in clinical decision-making.

Methods: Clinical and pathological data for OMCRC patients were collected from the Second Affiliated Hospital of Soochow University. Patients were randomly divided into training and testing cohorts. Risk factors for extensive metastasis were identified through LASSO regression analysis and COX regression analysis. Three predictive models were developed in the training cohort and validated in the testing cohort: COX regression analysis, Extreme Gradient Boosting (XGBoost), and Survival Support Vector Machine (SurvSVM). Finally, the optimal model was visualized with the nomogram.

Results: A total of 214 patients with OMCRC were enrolled in the study. Four independent risk factors were identified: whether surgery has been undertaken following oligometastasis (WST), histological type (HT), carcinoembryonic antigen at the last follow-up (CAE at last-FU), and preoperative albumin to globulin ratio (Preop-AGR). In the testing cohort, the COX model (1-year AUC = 0.82, 3-year AUC = 0.72, 5-year AUC = 0.85, mean AUC = 0.80) performed best. Decision curve analysis (DCA) confirmed the net benefit of the Cox model, and the nomogram provided accurate predictions of metastasis risk.

Conclusion: CAE at last-FU, Preop-AGR, HT, and WST are independent risk factors for extensive metastasis in OMCRC. The nomogram model incorporating risk factors can assist clinicians in developing optimal treatment for OMCRC patients.