TREATMENT TOXICITY MACHINE LEARNING PREDICTION OF RADIATION-INDUCED DYSPHAGIA AND XEROSTOMIA IN HEAD AND NECK CANCER PATIENTS

Abstract

Some of the most clinically relevant toxicity events seen after radiotherapy in head and neck cancer patients are radiation-induced dysphagia and xerostomia. These complications may have a negative impact on swallowing ability, nutritional intake, speech, oral health, treatment adherence and long-term quality of life. Hence, it is crucial to recognize and identify patients at high risk for developing these adverse effects, in order to plan for individual treatment and supportive care. This study suggests a machine learning based predictive model to estimate the risk of radiation induced dysphagia and xerostomia in HNC patients based on clinical, demographic, tumor-related and radiotherapy dosimetric variables. Various supervised learning models were suggested and validated for toxicity risk classification and to identify the most important predictors. Conventional evaluation metrics such as accuracy, precision, recall, f1 score and area under receiver operating characteristic curve were used to evaluate the performance of the model. The results indicate machine learning has significant potential to enhance the ability to stratify toxicity risks, by capturing the complex relationships between patient factors, tumor factors and dose-volume. Location of the tumor, type of therapy, doses of radiation to the swallowing structures, doses of radiation to the parotid gland and pre-treatment functional status, and concurrent chemotherapy are some of the most important predictive factors. The proposed approach could potentially help clinicians design more efficient radiotherapy plans, minimize unwanted side-effects, and enhance patient outcomes. Overall, the use of machine learning for personalized management of toxicity in head and neck oncology is a promising approach to this decision support.