MARKOV DECISION PROCESS MODELS FOR IMPROVING EQUITY IN LIVER ALLOCATION

In the United States, end-stage liver disease (ESLD) patients are prioritized primarily by their Model for End-stage Liver Disease score (MELD) to receive organ offers. Therefore, patients are required to update their MELD score at predefined frequencies that depend on the patient's last reported score. One aim of this dissertation is to mitigate inequities that stem from patients' flexibility regarding MELD score updates. We develop a Markov decision process (MDP) model to examine the degree to which an individual patient can benefit from the updating flexibility, and provide a menu of updating requirements that balance inequity and data processing more efficiently than the current updating requirements. We also derive sufficient conditions under which a structured optimal updating policy exists.\ud \ud As the coordinator of the harvesting Organ Procurement Organization (OPO) extends offers according to MELD score prioritization, the organ becomes less desirable. To avoid\ud not placing the organ, the OPO coordinator can initiate an expedited placement, i.e., offer the organ to a transplant center, which can then allocate it to any of its patients. A second aim of this dissertation is to mitigate inequities induced by the OPO coordinator's premature departure from the prioritized list of patients via an expedited placement.\ud \ud As a preliminary step to studying the inequity induced by expedited liver placement, we conduct an extensive analysis of the current expedited liver placement practice based on\ud recent data. We investigate different aspects of extending offers, e.g., the number of offers extended concurrently, and patients' response characteristics. Several of the results from this analysis serve as inputs for a second MDP model that examines how many concurrent offers the OPO coordinator should extend and when the coordinator should initiate an expedited placement. Numerical experimentation reveals a structured optimal policy, and we test the sensitivity of the model outcomes with respect to changes in model inputs. Lastly, we examine how our model outputs compare to the analogous measures observed in current practice and how they can be used to improve current practice