ELICITING PATIENT PREFERENCES AND PLACING EXPEDITED ORGANS

Liver transplantation plays a crucial role in saving lives when no other alternatives exist. Each year approximately 5,500 liver transplants are performed in the US. However, annually still 2,000 lives are lost due to lack of livers. Much effort has been spent on improving the organ allocation system. In this dissertation, we focus on patient preference elicitation which is an essential component of medical decision models and expedited organ placement which is relatively unexplored component of the organ allocation system.\ud When livers become available, they are offered to patients according to an order (match list) specified by a set of rules. Each patient can accept/reject the offer. Other researchers have considered this accept/decline decision. Estimating patient preferences over health states is an important component of these decision making models. Direct approaches, which involve asking patients abstract uestions, have significant drawbacks. We propose a new\ud approach that infers patient preferences based on observed decisions via inverse optimization techniques. We illustrate our method on the timing of a living-donor liver transplant.\ud If it appears that the standard allocation procedure will not result in a match before the organ becomes nonviable, the liver’s placement can be expedited, meaning that it is offered to a transplant center instead of an individual patient. We study the subsequent decision problem faced by a transplant center, namely which, if any, of its patients should receive the organ independent of their positions on the match list. We develop a simulation model and compare different policies for expedited liver placement. Our study indicates that a policy which gives higher priorities to patients whose likelihood of death is higher performs the best based on several metrics.\ud We also formulate the transplant center’s decision problems as an average reward Markov Decision Process (MDP). Due to the complexity of the model, traditional methods used to\ud solve MDP problems cannot be utilized for our model. Thus, we approximate the solution via Least Square Policy Iteration (LSPI) method. Despite the extensive search on basis functions, the LSPI method yields promising, yet not better outcomes than the policies found to be the best via simulation