A Markov Decision Process Framework for Optimal Cancer Chemotherapy Dose Selection

Abstract

Chemotherapy is arguably the most effective treatment for treating cancer. However, it triggers toxic effects on patients due to its interaction with normal cells. In this paper, we propose a finite-horizon Markov decision process framework for optimal chemotherapy dose selection during cancer treatment. In contrast to many other research works, our model accommodates all possible patient’s clinical health states that could be considered during treatment. We study the use of our model by applying it on Wilms tumor. We explore three different cases of varying transition probabilities and rewards. Our numerical results agree with clinical intuition and demonstrate the potential applicability of our model.