Abstract / Summary
Abstract Purpose To develop and internally validate a nomogram combining preoperative serum CA19-9, vascular endothelial growth factor (VEGF), and clinicopathological factors for predicting lymph node metastasis (LNM) in colorectal cancer (CRC), a clinical assessment for which imaging often misses small metastatic nodes. Methods We retrospectively enrolled 1,024 CRC patients who underwent curative resection between June 2019 and June 2022 and randomly stratified them 7:3 into training (n = 717) and validation (n = 307) cohorts. ROC analysis determined CA19-9 and VEGF cutoffs. LASSO regression and multivariable logistic regression identified predictors and generated a nomogram. Discrimination, calibration, and decision curve analysis assessed performance. Results Of 1,024 patients, 392 (38.3%) had LNM. Independent predictors were CA19-9 > 37.0 U/mL (OR = 2.12), VEGF > 186.8 pg/mL (OR = 1.89), T stage (T3: OR = 2.54; T4: OR = 4.73), poor/undifferentiated histology (OR = 2.31), and vascular invasion (OR = 3.15; all P < 0.001). The nomogram achieved a C-index of 0.847 in the training cohort and an AUC of 0.839 in the validation cohort, exceeding CA19-9 (AUC = 0.692) or VEGF (AUC = 0.674) alone. Calibration MAE was 0.031 and 0.028, respectively, and decision curve analysis showed net clinical benefit across thresholds of 10%–85%. Conclusion A nomogram integrating CA19-9, VEGF, and three clinicopathological factors showed good discrimination, calibration, and potential clinical utility for LNM risk stratification in CRC. Multicenter prospective external validation is needed to confirm generalizability.