| This dissertation studies the effects of patient choice on the kidney allocation system. Most countries adopt centrally-planned allocation mechanisms and patients on the waiting list often respond strategically. They may: (i) refuse organ offers if they expect to receive a more compatible match in the future, and (ii) conceal or even misreport clinical information in order to improve their prospects of receiving a transplant. These incentive problems are studied using four separate models.; In the sequential stochastic assignment model, the problem is to make optimal dynamic allocation decisions, subject to patients' optimal stopping decisions (i.e. when to accept transplantation). We introduce a method based on large deviations to justify asymptotic optimality of fluid-limit solutions for a general class of sequential assignment problems, and use dynamic programming to characterize an incentive compatibility condition that ensures that patients accept all organ offers. We show that optimal allocation policies exhibit a partition structure: the kidney supply is partitioned into subsets, and each patient receives organs from exactly one of these subsets.; Next, we investigate how to use the queueing discipline to discourage patients from refusing organ offers. We develop a model based on stochastic games (multi-person Markov decision processes) to capture the dynamic strategic interactions among individuals in a queue. We show that First-Come-First-Served, although the most fair, is the least efficient priority rule, whereas Last-Come-First-Served, although the least fair, attains the first-best ideal.; In the mechanism design model, patient types are unobservable and kidneys are allocated based on reported types. This hidden-information problem is solved using the achievable regions/polymatroid optimization approach and Lagrangian duality techniques. We demonstrate that the allocation mechanism can be designed to induce truth-telling by ensuring that patients who wait longer also receive better kidneys.; Finally, we propose classifying kidneys into grades, and allowing patients to choose the range of kidney grades that they receive. We develop an agent-based (object-oriented) model that simulates the decisions of individual patients throughout the allocation process. This model closely mimics current procedures, and is based on realistic inputs obtained through survival analyses of historical transplant data. We show that our proposal achieves substantial improvements, with gains observed across all demographic groups. |
