Integration of Preoperative Neutrophil-to-Lymphocyte Ratio into Machine Learning Models for Predicting Lymph Node Metastasis in CRC

Background: Accurate preoperative prediction of lymph node metastasis (LNM) is essential for tailoring treatment strategies in T1 colorectal cancer (CRC). Although the neutrophil-to-lymphocyte ratio (NLR), an easily acquired inflammatory biomarker, correlates with tumor progression, its incremental value when incorporated into machine learning (ML) models for predicting LNM in T1 CRC remains unclear. Therefore, this study evaluated the independent predictive value of preoperative NLR for LNM and constructed interpretative ML models to stratify LNM risk in T1 CRC. Methods: We retrospectively enrolled 533 pathologically confirmed T1 CRC patients. NLR was calculated as the neutrophil–lymphocyte count ratio. Multivariable logistic regression identified NLR’s independent correlation with LNM. Features were screened via combined least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm. Six ML models, including Logistic Regression (LR), Random Forest (RF), XGBoost, LightGBM, CatBoost, and Logistic Regression with Splines (LR-Spline), were developed and validated. Model performance was evaluated in terms of discrimination, calibration, and clinical utility. Shapley additive explanations (SHAP) were applied for feature interpretability. Results: Elevated preoperative NLR was independently associated with an increased risk of LNM (adjusted OR = 2.83, 95% CI: 1.48–5.41, p = 0.002), exhibiting stability across all clinical and pathological subgroups. The optimized LightGBM outperformed other models with an area under the receiver operating characteristic curve (AUC) of 0.768, alongside favorable clinical utility. Conclusions: An elevated preoperative NLR is a robust, independent predictor of LNM in patients with T1 CRC. An optimized LightGBM model integrating NLR with routine clinicopathological indicators offers accurate risk stratification, potentially refining surgical decision-making and sparing low-risk T1 CRC patients from unnecessary radical resections.