Recursive meta-Reinforcement Learning for Personalized Sequential Dynamic Treatment Policies

In recent years deep meta-reinforcement learning has extended the applicability of reinforcement learning algorithms: by integrating recurrent networks, trained models have the ability to quickly adapt to new unseen environments without the need for further backpropagation. We propose a novel recursive deep meta-reinforcement learning approach which enables the model of each decision of the sequential process to learn from and adapt to unseen circumstances by recursively integrating the feedback of the models of other decisions in the process. We apply this approach to a dataset of 3-step chemo-radiotherapeutic and surgical treatment of head and neck cancer patients, proving its ability to optimally handle previously unseen patient’s (and physician’s) preferences on survival and toxicity outcomes.