Machine learning for predicting distant metastasis in nasopharyngeal carcinoma patients.

Background: Distant metastasis is the main cause of treatment failure and death in patients with nasopharyngeal carcinoma (NPC). The aim of this study was to explore the risk factors for distant metastasis in NPC patients using machine learning (ML) methods. Methods: We collected data from NPC patients who were treated at the Eye Ear Nose Throat Hospital of Fudan University between September 2017 and June 2024. Seven ML methods were employed to construct the predictive models. By comparing the predictive performance of different ML models, the best one was selected to establish a predictive model for distant metastasis of NPC. The SHapley Additive exPlanation (SHAP) method was utilized to ascertain the ranking of feature importance and to provide explanations for the predictive model. Results: A total of 1,845 NPC patients were included in this study. Among the seven models, Logistic Regression (LR) performed best in the test dataset (Area Under the ROC Curve [AUC] = 0.8499). SHAP analysis indicated that the most important variables for distant metastasis in NPC patients were targeted therapy, immunotherapy, N stage, Epstein-Barr virus (EBV), hypertension, T stage, lymphocyte count (LY) and lactate dehydrogenase (LDH) level. Conclusion: Targeted therapy, N stage, immunotherapy, EBV, hypertension, T stage, LY and LDH level are significantly associated with the risk of distant metastasis in NPC and could be used to identify high-risk populations for distant metastasis in NPC patients. For high-risk patients, early interventions such as targeted therapy and immunotherapy might be considered to reduce the risk of distant metastasis in NPC.