Files
Abstract
The demand for transplant organs surpasses supply, with xenotransplantation offering a potential solution to this shortage. Successful investigational transplants of genetically edited pig kidneys into brain-dead recipients and expanded access cases involving living human recipients suggest that the first human clinical trials are imminent. This dissertation focuses on patient selection for the initial trials. In Chapter 1, using the benchmark of 2-year survival of non-human primates in pre-clinical studies, we develop a tool that can identify individual wait-listed patients predicted to have a shorter life expectancy than with a xenotransplant, utilizing Random Survival Forest, DeepSurv and Cox Proportional-Hazards models. We find that it is hard to identify patients that reach clinical equipoise unless the expected xenograft survival exceeds two years. Few patients would benefit based on survival alone and potential beneficiaries are spread across more than 200 transplant centers. Several incentives could allow more patients to reach equipoise. Keeping patients inactive on the waitlist while they have a functioning xeno-kidney provides a modest incentive, while giving patients with failed xenografts the same or even more priority as prior living donors would represent a potent driver for participation in trials. We are able, however, to identify phenotypes that have high mortality and low transplant rates in the current allocation system that could serve as acceptable candidates. In Chapter 2, we extend this framework to jointly model decisions about who should receive a xenograft, when to proceed, and under which incentive scheme. We develop an individual-level optimal stopping model that captures the decision process of a forward-looking patient who evaluates whether to accept or decline allo- and xeno-offers over time. The model is structured as a two-stage problem: the first covers choices after initial waitlisting, and the second applies if the patient reenters the list following xenograft failure. Incentive mechanisms, such as awarding additional priority points after xenograft failure, are embedded in the second stage, influencing the value of rejoining the waitlist and shaping the initial acceptance decision. To capture heterogeneity, we represent patient health with high-dimensional covariates that integrate demographic, clinical, and socioeconomic factors, enabling individualized outcome predictions. For computation, we implement GPU-accelerated backward induction to efficiently evaluate large state spaces. We also employ classification analysis to identify patient characteristics most predictive of benefiting from xenografts. In Chapter 3, we examine the current eligibility criteria for initial xenotransplant clinical trials and apply them to patient-level data using Scientific Registry of Transplant Recipients decision aid and the Estimated Post-Transplant Survival score. We then compare this framework with our proposed approach.