Optimizing Equitable Resource Allocation in Parallel Any-Scale Queues with Service Abandonment and its Application to Liver Transplant

TL;DR

This paper develops a finite approximation method for optimizing resource allocation in multiple queues with abandonment, applied to liver transplants, improving fairness and system performance over traditional fluid models.

Contribution

Introduces a finite approximation technique for multi-queue resource allocation with abandonment, tailored for small-scale queues, and demonstrates its application to liver transplant allocation.

Findings

Finite approximation improves allocation accuracy for small queues.

Allocating more resources to high-risk or small-scale queues enhances equity.

Method outperforms fluid models in system performance and fairness.

Abstract

We study the problem of equitably and efficiently allocating an arriving resource to multiple queues with customer abandonment. The problem is motivated by the cadaveric liver allocation system of the United States, which includes a large number of small-scale (in terms of yearly arrival intensities) patient waitlists with the possibility of patients abandoning (due to death) until the required service is completed (matched donor liver arrives). We model each waitlist as a GI/MI/1+GI queue, in which a virtual server receives a donor liver for the patient at the top of the waitlist, and patients may abandon while waiting or during service. To evaluate the performance of each queue, we develop a finite approximation technique as an alternative to fluid or diffusion approximations, which are inaccurate unless the queue's arrival intensity is large. This finite approximation for hundreds of queues is used within an optimization model to optimally allocate donor livers to each waitlist. A piecewise linear approximation of the optimization model is shown to provide the desired accuracy. Computational results show that solutions obtained in this way provide greater flexibility, and improve system performance when compared to solutions from the fluid models. Importantly, we find that appropriately increasing the proportion of livers allocated to waitlists with small scales or high mortality risks improves the allocation equity. This suggests a proportionately greater allocation of organs to smaller transplant centers and/or those with more vulnerable populations in an allocation policy. While our motivation is from liver allocation, the solution approach developed in this paper is applicable in other operational contexts with similar modeling frameworks.