Predicting Lymph Node Metastasis in T1 Colorectal Cancer Patients Using Interpretable Machine Learning Models: A Multicenter Retrospective Study

Whether lymph node metastasis (LNM) is present is crucial for treatment decisions in T1 colorectal cancer (T1 CRC).This study developed predictive models using data from 1,205 patients across seven Chinese medical centers.We evaluated 29 machine learning algorithms and identified CatBoost as the top performer (AUC: 86%, accuracy: 96%).SHAP analysis revealed key predictors of LNM risk, including lymphovascular invasion, age, tumor size, invasion depth, and total lymph node count.Less influential features included perineural invasion and tumor location.The study highlights the importance of retrieving more lymph nodes during surgery to improve staging accuracy.A user-friendly online tool was developed to support clinical decision-making.