Machine Learning-Based Prediction of Lymphovascular Invasion in Superficial Esophageal Carcinoma: Model Development and Risk Factor Analysis

Background Lymphovascular invasion (LVI) represents a critical prognostic determinant in superficial esophageal carcinoma (SEC), significantly influencing therapeutic decision-making and clinical outcomes. Despite its clinical importance, reliable predictive tools for early LVI detection remain unavailable. The current study was designed to develop and validate a machine learning-based predictive model for accurate LVI risk stratification in SEC patients. Methods Predictive factor selection was conducted using least absolute shrinkage and selection operator (LASSO) regression followed by multivariable logistic regression analysis. Multiple machine learning algorithms were systematically evaluated, with model performance quantified through receiver operating characteristic (ROC) curve analysis. Model interpretability was enhanced through implementation of Shapley Additive Explanations (SHAP) methodology. Results Eight independent predictors of LVI were identified: neutrophil-to-lymphocyte ratio (NLR), esophageal wall thickness on computed tomography (CT), endoscopic ultrasound or magnifying endoscopy (EOM) findings, tumor diameter, multiple lesions, circumferential involvement proportion (CIP), consumption of pickled food and preoperative biopsy results. The logistic regression model demonstrated superior predictive performance, with area under the curve (AUC) values of 0.871 (training cohort), 0.852 (validation cohort), and 0.902 (test cohort). Conclusion The developed SHAP-interpretable logistic regression model provides an effective tool for early LVI detection in SEC, enabling personalized risk assessment and optimized clinical management strategies. This approach may significantly improve treatment decision-making for SEC patients.