SCH: Personalized Rescheduling of Adaptive Radiation Therapy for Head & Neck Cancer

点击翻译按钮获取中文摘要

Head and neck cancers (HNCs) account for nearly 3% of all cancers in the U.S. and most commonly affect aging individuals. While chemo-radiotherapy is the standard treatment approach for HNC, the method is known to cause substantial side-effects. In particular, anatomical changes occurring during the treatment may result in under-coverage of the clinical target volume or over-dosage of organs at risk. The project will develop novel optimization models for Adaptive Radiation Therapy (ART) – a customized treatment planning approach for individual patients designed by evaluating the systematic and random variations in tumor response. The models will use imaging data of tumor volume and normal tissue complication probabilities to determine the optimal number and timing of treatment replans. The models will provide personalized optimization through sequential decision-making based on response to treatment, as well as optimization and evaluation of simple threshold replanning policies often used by doctors. Optimal behavior in the face of conflicting payer/provider perspectives for emerging technologies will also be analyzed using mechanism design techniques. ART requires information about patient-state as well as transition probabilities describing the tumor's evolution over time. Since the process inherently calls for sequential decision-making under uncertainty, the proposed optimization models use a Markov Decision Process (MDP). The resulting optimal policies may be difficult to implement in practice, especially in centers lacking state-of-the-art equipment. Therefore, the proposed work will further evaluate simple threshold replanning policies using a bilevel programming framework. In particular, the bilevel program will find threshold values for various patient classes by minimizing the deviation from the MDP-prescribed policy. The proposed framework offers multiple avenues for methodological contributions. The novel MDP design framework is an incredibly powerful tool that can be used to model many interesting questions. We will explore ways of discretizing the continuous state spaces and estimating transition probabilities based on patient imaging data. We will explore algorithms for solving bilevel programs, especially utilizing the structure and properties of the lower-level MDP models. Finally, we will study applications of the principal-agent framework from economics in modeling payer/provider interactions for emerging clinical therapies. RELEVANCE (See instructions): This study uses novel models to develop and validate an integrated approach to reduce cancer radiotherapy side effects while maintaining or improving cancer control. It will maximize efficiency for patients and providers. Its findings will inform decisions about individual radiation planning, optimize risk- stratified treatment, and healthcare policy implementation of effective new technology interventions.