Predicting Radiation Replanning and Toxicity Outcomes to Support Personalized Treatment in Oropharyngeal Cancer
Abstract
Patients with oropharyngeal cancer (OPC) treated with chemoradiation can experience weight loss and tumor shrinkage, altering the prescribed radiation treatment. In addition, nearly all patients with OPC will suffer treatment-related toxicities. Treatment replanning ensures patients do not receive excessive doses to the normal tissue. There remains an unmet clinical need for robust, predictive models to assist in identifying patients who could benefit from a replan, allowing for better patient management and reduced toxicities. Similarly, given the high survival rates in OPC, clinicians are focused on methods of treatment de-escalation to reduce toxicity. However, there is limited data correlating dosimetry and patient-reported outcomes (PROs) in the de-escalated setting. There remains a need to identify dosimetric predictors to inform current treatment planning. Computed tomography (CT) images are typically taken during the routine radiotherapy workflow. Radiomics aims to extract quantitative imaging features which can be used to develop predictive models. These features can be used in combination with clinical and dosimetric information. The objective of this work was to develop and evaluate tools to assist clinicians in providing personalized treatment options to help reduce treatment-related toxicities and improve quality of life for patients with OPC. Radiomic features were extracted from tumor volumes and parotid glands in a retrospective dataset of patients treated with chemoradiation. Multiple machine learning classifiers were built using the top radiomics features to predict the need for a replan. The best-performing classifier significantly outperformed the baseline clinical model. Next, the model was evaluated in a prospective dataset. The model successfully validated, in addition to achieving a higher accuracy compared to clinicians who were asked to predict which patients they think will require a replan. Finally, regression analyses were performed to evaluate correlations between normal tissue dose metrics and PROs from a randomized control trial investigating de-escalation. Multiple predictors of toxicity were identified, across swallowing, saliva, speech, and general domains. This work has the potential to aid clinicians in identifying patients at a higher risk of the need for a replan, and establish toxicity predictors in patients receiving radiation, to help reduce treatment-related toxicities and improve quality of life.