Development of a prognostic tool using machine learning to identify high-risk mucoepidermoid carcinoma patients
1Department of Stomatology, Children's Hospital of Chongqing Medical University, National Clinical Research Center for Child Health and Disorders, Ministry of Education Key Laboratory of Child Development and Disorders, China International Science and Technology Cooperation Base of Child Development and Critical Disorders, Chongqing Key Laboratory of Structural Birth Defect and Reconstruction, No.136, Zhongshan 2 Road, Yuzhong District, Chongqing 400014, China.
Objective:
Mucoepidermoid carcinoma (MEC), typically arising in the salivary glands, has been documented in various anatomical locations. This study aimed to develop machine learning models for efficient prognosis prediction in patients with MEC across different anatomical sites and to create a novel tool for identifying high-risk patients in clinical settings.
Methods:
A retrospective cohort study involving 6280 patients with MEC from diverse anatomical sites was conducted. COX regression analysis identified prognostic variables. Five machine learning models-Support Vector Machine (SVM), Random Forest (RF), Generalized Additive Model (GAM), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM)-were constructed. Model performance was assessed using accuracy, recall, precision, and the area under the curve (AUC) of receiver operating characteristic (ROC) curves. However, the external validation was not performed. The best-performing model was developed into a prognostic tool.
Results:
The GAM outperformed the other models, and a web-based tool was developed for identifying patients with varying prognostic risks across MEC cases from different anatomical locations, supporting clinical decision-making, personalizing therapeutic strategies based on clinical characteristics, ensuring tailored treatment schedules and optimizing resource allocation. Extension, age, pathological grade, clinical stage, and N-stage were the most significant variables in the final model.
Conclusion:
This study established an online tool utilizing GAM to aid in identifying high-risk patients with MEC and enhancing prognostic accuracy in clinical practice. The anatomical site did not significantly influence survival outcomes, offering valuable insights into the shared pathogenesis of MEC.
