Exploring the prognostic efficacy of machine learning models in predicting adenocarcinoma of the esophagogastric junction

Abstract

Abstract This study investigated the predictive performance of machine learning models for adenocarcinoma of esophagogastric union (AEG), based on 287 AEG patient data collected clinically. After grouping, Cox proportional hazards regression model (Cox-PH) and four machine learning models were constructed and internally validated. The AUC values of 3-year survival rate in validation set of Cox-PH, extreme gradient boosting (XGBoost), Random Forest (RF), support vector machines (SVM), and Multi-layer Perceptron (MLP) were 0.870, 0.901, 0.791, 0.832 and 0.725, respectively. The AUC values of 5-year survival rate in validation set of each model were 0.915, 0.916, 0.758, 0.905 and 0.737. The internal validation AUC values of the four machine learning models, XGBoost, RF, SVM and MLP were 0.818, 0.772, 0.804 and 0.745, respectively. In conclusion, compared with Cox-PH, machine learning models do not need to meet proportional assumption or linear regression model, and can include more influencing variables, which has good predictive performance for the 3-year and 5-year survival rate of AEG patients, among which XGBoost model is the most stable and has obvious superiority in prediction performance than other machine learning methods, practical and reliable.