The following is a summary of “Survival Machine Learning Model of T1 Colorectal Postoperative Recurrence after Endoscopic Resection and Surgical Operation: A Retrospective Cohort Study,” published in the February 2025 issue of the BMC Cancer by Li et al.

Accurately predicting postoperative recurrence in patients with T1 colorectal cancer following endoscopic resection and surgical intervention is essential for optimizing treatment strategies and improving patient outcomes. This study aimed to develop a robust recurrence prediction model using survival machine learning algorithms, leveraging clinical and pathological data from two tertiary-level affiliated hospitals. A cohort of 580 patients diagnosed with T1 colorectal cancer and treated with either endoscopic or surgical resection was retrospectively analyzed. Patient demographics, treatment modalities, and pathology-related variables were systematically extracted. To enhance predictive accuracy, Boruta’s algorithm was employed for feature selection, identifying key prognostic factors with significant contributions to recurrence risk. 

The dataset was subsequently divided into a training set (70%) and a test set (30%) to develop and validate five survival machine learning models: Randomized Survival Forest, Gradient Boosting, Survival Tree, Cox Proportional Hazards, and Coxnet. Model interpretability was assessed using the SHAP algorithm to elucidate the influence of individual predictors on recurrence risk. The results demonstrated that patients at high risk of lymph node metastasis exhibited significantly poorer prognoses, whereas different treatment approaches did not markedly affect recurrence outcomes. Among the models evaluated, RSF exhibited superior predictive performance, achieving a C-index of 0.848 and an Integrated Brier Score of 0.098 in the test set, with a time-dependent area under the curve of 0.918. SHAP-based interpretability analysis highlighted several key protective factors, including submucosal invasion depth <1000 µm, low-grade tumor budding (BD1), absence of lymphovascular and perineural invasion, well-differentiated cancer cells, and tumor size <20 mm. Conversely, negative gain characteristics contributed significantly to recurrence risk. 

These findings underscore the potential of survival machine learning models in facilitating precise, individualized recurrence risk assessment for patients with T1 colorectal cancer. The developed prognostic tool demonstrates strong predictive capabilities, offering clinicians valuable insights to guide post-treatment surveillance and therapeutic decision-making. Further prospective validation studies are warranted to refine the model and enhance its clinical applicability in personalized oncology care.

Source: bmccancer.biomedcentral.com/articles/10.1186/s12885-025-13663-6