Development and Validation of a Post-operative Risk Of Local Recurrence Model Integrating Serology and Evalution (PROMISE) in Local Advanced Rectal Cancer Receiving Neoadjuvant Therapy: a Multicenter Study

To address the gap in accessible and integrative risk prediction, this study aimed to develop and validate an interpretable machine learning model for the early prediction of postoperative local recurrence in LARC patients using a multicenter cohort. We employed SHapley Additive exPlanations (SHAP) analysis to elucidate feature importance and provide clear interpretations for individual predictions, with the ultimate goal of evaluating the model's clinical utility in guiding personalized patient management-particularly by identifying high-risk patients in clinical practice and informing tailored follow-up and treatment strategies to improve patient outcomes.

Inclusion criteria

• Diagnosed cT>=3 or cN+ stage rectal cancer by pelvic MRI or enhanced CT scan;
• Pathological proved primary rectal adenocarcinoma;
• Recieved full-cource radiation or more than 3 cycles of chemotherapy before surgery;
• Recieved radical tumor resection by TME surgery.

Exclusion criteria

• Diganosed with any distant metastasis disease or locally unresectable disease before neoadjuvant treatment or surgery;
• Recieved local resection surgery (R2 resection) or postoperative pathology confirmed positive surgical margins (R1 resection);
• Patients who died due to intra- or post-operative complications;
• Patient who die or relapse within 30 days after surgery;
• Patients who had a prior history of malignancy within 5 years before surgery;
• Patients who were loss of follow-up or had inadequate clinical data.