A Simulation-Based Framework for Analyzing and Optimizing Organ Allocation Policies
Files
TR Number
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Organ allocation policies must balance medical urgency, expected benefit, fairness, logistical constraints, and organ utilization under severe supply scarcity. Because candidate conditions evolve over time and donor organs arrive unpredictably, allocation policy cannot be evaluated adequately as a static ranking problem. This dissertation develops a simulation-based framework for organ allocation policy analysis, with a primary application to the U.S. lung allocation system. The framework connects short-horizon prediction of dynamic candidate priority, empirically validated high-fidelity simulation, and evidence-based multi-objective simulation optimization. First, the dissertation studies short-horizon forecasting of dynamic lung allocation priority using U.S. lung allocation data from the Lung Allocation Score (LAS) era, prior to Continuous Distribution, as a stable empirical setting. Recurrent sequence models are compared with discrete-time Markov chain benchmarks over 7-, 14-, and 28-day horizons. The results show that near-term LAS movement can be forecast from longitudinal candidate information, that recurrent models outperform Markov chain benchmarks, and that component-wise forecasting of waiting-list urgency and post-transplant survival yields modest but consistent gains at longer horizons. Interpretable classification models further identify candidates at elevated risk of near-term priority increase. Second, the dissertation develops the Modular Lung Allocation Policy Simulation Model (MoLAPS), a discrete-event simulation framework for evaluating U.S. lung allocation policies. MoLAPS represents candidate arrivals, donor arrivals, dynamic patient-status updates, sequential organ offers, offer acceptance, non-transplant removals, and post-transplant outcomes. Calibration and validation using 2017--2019 Organ Procurement and Transplantation Network (OPTN) data show close alignment between simulated and empirical outcomes across aggregate, distributional, subgroup, and temporal checks. A modular C/C++ implementation with parallel computing substantially reduces runtime for multi-replication experiments relative to a prior lung-allocation simulator in the settings evaluated. Third, the dissertation develops an evidence-based multi-objective simulation-optimization framework for policy search under simulation uncertainty. Policy adoption is formulated as a baseline-relative decision problem. Efficiency is evaluated using replication-level log Bayes factors, while fairness is evaluated using Kullback--Leibler divergence in subgroup composition. The proposed trust-region framework, ASTRO-DF-MOO, supports finite-budget exploration of efficiency and fairness trade-offs. Benchmark experiments and a lung allocation case study illustrate how the approach traces evidence-supported efficiency--fairness trade-offs relative to the historical LAS-based policy. Within a one-dimensional LAS-type score family that varies the relative weight assigned to waiting-list urgency versus post-transplant survival, the approach identifies an evidence-supported menu of incumbent-adjacent policies rather than a single dominant replacement. These contributions define a connected methodological pipeline from candidate-level priority dynamics to validated counterfactual policy evaluation and then to evidence-based policy search under uncertainty.