The optimal dynamic therapy: a Decision-Theoretic approach

Therapy planning is a very complex task, being the patient's therapeutic response aaected by several sources of uncertainty. In this paper we cope with this class of problems using the Decision-Theoretic Planning approach able to provide plans in presence of partial and qualitative information, while preserving a sound mathematical foundation. In particular, Markov Decision Processes (MDPs) represent a well-understood and ee-cient instrument to cope with time-dependent decision problems. They are suitable to describe and solve decision problems in which the optimal choice has to be revised periodically in accordance to the evolution of the patient's conditions. So, they represent a valuable solution to dynamic decision problems that are not handled properly by the formalisms traditionally used in medical decision making eld, useful only for managing static decision problems. The main problem in building a MDP is the knowledge elicitation from a speciic domain in order to derive the necessary transition probabilities. As interesting solution to these problems, we will exploit a novel graphical formalism for MDP knowledge acquisition called the Innuence View. This methodology has been applied in this paper to the choice of the prophylaxis in patients aaected by a mild Hereditary Spherocyt-osis. The results of the dynamic decision problem are compared with the ones of the static decision. 1. Introduction Therapy planning is a very complex task, being the pa-tient's therapeutic response aaected by several sources of uncertainty as inter-and intra-individual variability or wrong implementations of drug delivery protocol. Furthermore, the modellization of the patient's evolution is frequently hampered by the incompleteness of the medical knowledge; hence it is not often possible to derive a mathematical model that is able to take into account the characteristics of the uncertain environment. In that case, in medical practice, the deenition of a prognostic model for supporting therapeutic decisions usually relays on heuristic rules, that exploit a representation of the patient's state based on discrete and qualitative variables. An alternative way of coping with this class of problems is the Decision-Theoretic Planning approach, i.e. the formulations of policies on the basis of Decision Theory. Such approach is able to provide plans in presence of partial and qualitative information, while preserving a sound mathematical foundation. In particular , Markov Decision Processes (MDPs) represent a well-understood and eecient instrument to cope with time-dependent decision problems 1]. MDPs allow to manage a wide class of problems in medical decision making. They are suitable to …