Machine learning-based model for predicting distant lymph node metastasis in esophageal cancer.

Background: Distant lymph node metastasis (LNM) from esophageal cancer is relatively difficult to detect and is associated with a poor prognosis. Our study aimed to develop machine learning models for predicting distant LNM in esophageal cancer, with the goal of enhancing diagnostic accuracy, optimizing treatment strategies, and identifying high-risk patients for personalized clinical management. Methods: The demographic and clinicopathological data of patients diagnosed with esophageal cancer were extracted from the Surveillance, Epidemiology, and End Results (SEER) database, 17 Registries, November 2023 (2000-2021). Univariate and multivariate logistic regression analyses were applied to select correlation variances related to distant LNM. Seven machine learning models were constructed, including random forest (RF), decision tree (DT), extreme gradient boosting (XGBoost), gradient boosting (GB), Naïve Bayes, artificial neural network (ANN), and adaptive boosting (ADB) models. The predictive performance of the models was evaluated in terms of accuracy, F1 score, recall rate, and area under the curve (AUC). Importance rank and Shapley additive explanation (SHAP) were performed to explain the machine learning models. We validated the machine learning model in 53 patients with esophageal cancer at our clinical center, Beijing Friendship Hospital. Results: Among the machine learning models, the GB model outperformed the other models, with an accuracy of 0.929, a recall of 0.427, an F1 score of 0.414, and an AUC of 0.912. The interpretable module revealed that tumor (T) stage, node (N) stage, age, liver metastasis, and lung metastasis were significant predictors. T stage was the most significant predictor of distant LNM in patients with esophageal cancer. The validation dataset revealed a high accuracy of 0.818, a recall of 0.857, and an AUC of 0.857. Conclusions: The GB model emerged as the best model for predicting LNM in the esophageal cancer model. The importance rankings provide a degree of explanation, and the model can assist physicians in the initial prediction of LNM, providing a clinical reference and guidance.