Lymph node metastasis in patients with hepatocellular carcinoma using machine learning: a population-based study

Abstract

Aim: This study aims to develo\p a population-adapted machine learning-based prediction model for hepatocellular carcinoma (HCC) lymph node metastasis (LNM) to identify high-risk patients requiring intensive surveillance. Methods: Data from 23511 HCC patients in the SEER database and 57 patients from our hospital were analyzed. Seven LNM risk indicators were selected. Four machine learning algorithms—decision tree (DT), logistic Regression (LR), multilayer perceptron (MLP), and extreme gradient boosting (XGBoost)—were employed to construct prediction models. Model performance was evaluated using area under the curve, accuracy, sensitivity, and specificity. Results: Among 23511 SEER patients, 1679 (7.14%) exhibited LNM. Race, Sequence number, Tumor size, T stage and AFP were identified as independent predictors of LNM. The LR model achieved optimal performance (area under the curve: 0.751; accuracy: 0.707; sensitivity: 0.711; specificity: 0.661). External validation with 57 patients from our hospital confirmed robust generalizability (area under the curve: 0.73; accuracy: 0.737; sensitivity: 0.829; specificity: 0.5), outperforming other models. Conclusions: The LR-based model demonstrates superior predictive capability for LNM in HCC, offering clinicians a valuable tool to guide personalized therapeutic strategies.