Objective To investigate the temporal changes in serum neurotrophin-3 NT-3 cadherin-3 CDH3 and pyruvate dehydrogenase E1α subunit PDHA1 after radical surgery in colorectal cancer CRC patients and to construct a dynamic multi-index prediction model for postoperative recurrence and metastasis. Method A prospective cohort of 108 CRC patients who underwent radical resection in the Department of Gastrointestinal Surgery from June 2021 to June 2022 was enrolled. Serum samples were collected preoperatively and at 6 12 24 and 36 months postoperatively. Follow-up ended in June 2025. Recurrence and metastasis events were recorded and prediction models were constructed. Result Dynamic monitoring of 108 postoperative CRC patients revealed that the recurrence group had higher rates of vascular invasion and perineural invasion P < 0. 05 a higher proportion of TNM stage Ⅲ and ≥ 3 lymph node metastases P = 0. 008 and elevated serum biomarker levels compared to the non-recurrence group NT-3 at 6 months 25. 30 ng / ml vs 21. 70 ng / ml P = 0. 018 and 12 months 30. 50 ng / ml vs 20. 80 ng / ml P = 0. 004 CDH3 at 12 months 22. 90 pg / ml vs 17. 50 pg / ml P = 0. 010 and PDHA1 at 12 months 11. 30 ng / L vs 8. 00 ng / L P = 0. 003 . Univariate analysis confirmed that postoperative NT - 3 CDH3 and PDHA1 levels and clinical stage were independent predictors of recurrence. The random survival forest model constructed based on these factors showed excellent predictive performance AUC = 0. 853 significantly outperforming the traditional TNM staging model AUC = 0. 671 and the Cox model AUC = 0. 789 with strong clinical utility. Conclusion The temporal dynamic changes in serum NT-3 CDH3 and PDHA1 can serve as effective predictive indicators for postoperative recurrence and metastasis in CRC. The RSF model incorporating multi - timepoint detection significantly improves predictive accuracy.