A Dynamic Stratification Strategy for Assessing Postoperative Recurrence Risk in Colorectal Cancer

{'en': 'Objective: To investigate the value of a dynamic stratification strategy based on continuous updating at follow-up visits for predicting postoperative recurrence risk in colorectal cancer. Methods: A retrospective cohort of 596 patients with stage II–III colorectal cancer who underwent R0 curative resection between January 2016 and June 2021 was enrolled, with the first local recurrence or distant metastasis as the primary outcome. Baseline pathological features were combined with longitudinal information including CEA, ctDNA, and imaging findings. A time-aware and missing-aware multimodal fusion model was used to update risk at follow-up nodes of 3, 6, 9, 12, 18, and 24 months after surgery, outputting the current risk, the 6-month and 12-month recurrence probabilities, and the direction of risk change. TNM staging, conventional Cox model, clinicopathological-plus-single-CEA, and clinicopathological-plus-CEA-trajectory models served as comparators. Dynamic AUC, time-dependent C-index, Brier Score, calibration, and decision curve analysis were used for evaluation, and ablation experiments were performed. Results: With a median follow-up of 41 months, recurrence occurred in 134 patients (22.5%). The full dynamic model achieved dynamic AUCs of 0.87, 0.85, and 0.82 at 12, 24, and 36 months, respectively, with a time-dependent C-index of 0.82, all superior to the comparator models (P < 0.05). The model classified patients as high risk a median of 4.6 months before imaging-confirmed recurrence; among patients with low or intermediate baseline risk who eventually recurred, 72.9% were upgraded to high risk in advance by the model. In ablation experiments, replacing the CEA trajectory with single CEA, removing the ctDNA module, removing the imaging module, and removing the missing-aware fusion decreased the 24-month AUC to 0.80, 0.83, 0.82, and 0.79, respectively. Conclusions: The dynamic stratification strategy can continuously update individual risk during follow-up and provides better discrimination and earlier-warning ability than static assessment, offering a reference for individualized postoperative follow-up management of colorectal cancer.'}