Using discrete event simulation to optimise donor lung allocation

Abstract

Background: Optimally allocating deceased donor lungs to candidates requiring a life saving lung transplant while balancing efficiency and equity is a difficult challenge, which is compounded by the limited availability of donor lungs in the UK, with less than 15% of offered lungs being utilised for transplantation. This thesis argues that the sequential, centre-based approach used by the existing UK lung allocation policy does not make optimal use of scarce donor lungs and leads to inequitable access to transplantation. This research identifies the need for a transparent, auditable, and equitable system that maximises the additional years of life recipients gain from transplant (‘net benefit’) by considering both clinical urgency and post-transplant outcomes. This research uses the concepts of the Lung Allocation Score (LAS) as a springboard to bridge the gap between the current sequential, centre-based UK lung allocation policy and a score-based national allocation policy. Methods: Lung transplant datasets were provided by NHS Blood and Transplant that included data on adult (aged 16+), first-time, lung-only lung transplant candidates (n = 4280) and recipients (n = 2131) listed or transplanted between 2002 and 2021. Custom Cox proportional hazards models were developed to simulate waiting list and post-transplant survival durations, and a novel lung allocation policy simulation engine was developed that used discrete event simulation to predict the impact of a number of potential national lung allocation policies. Five initial allocation policies were simulated, focusing on different priority-ratios between waiting list survival (WL) and post-transplant survival (PTX). Five additional policies were simulated that maximise the use of single-lung transplants (SLT) for recipients with interstitial lung disease (ILD). Additional scenarios were simulated to assess the impact of increased utilisation for each of the standard and SLT policies. The key performance metrics recorded for each policy were: annual waiting list deaths, mean net benefit per recipient, and post-transplant survival rates at 1 and 5 years. The analytic hierarchy process (AHP) was used to collect and evaluate stakeholder preferences (i.e., candidates, recipients, their family members (n = 100), and clinicians (n = 62)) to identify which simulated allocation policies aligned most closely with the goals and values of the lung transplant community. Results: The Cox models used for this work demonstrated reasonably strong predictive power for waiting list survival (C-statistic: training dataset = 0.73, validation dataset = 0.66), and moderate predictive power for post-transplant survival (C-statistic: training dataset = 0.60, validation dataset = 0.55). This demonstrates a significant improvement over the existing UK lung allocation policy, which had a C-statistic for waiting list survival of 0.51 (training dataset) and 0.56 (validation dataset), and for post-transplant survival: 0.54 (training dataset) and 0.52 (validation dataset). The five initial simulated policies revealed that a national score-based system would significantly decrease waiting list mortality relative to the existing policy (90 annual waiting list deaths) regardless of choice of priority-ratio: the WL policy (i.e., prioritising clinical urgency) resulted in 46 annual waiting list deaths (48.9% decrease), and the PTX policy (i.e., prioritising post-transplant outcomes) resulted in 77 deaths (14.4% decrease). The PTX policy yielded the highest net benefit (6.9 years, compared to 5.0 years with the existing policy), and post-transplant survival rates: 83.5% at 1 year and 59.5% at 5 years, compared to the existing policy with 80.2% and 53.3% respectively. The SLT policies further decreased waiting list mortality: when combined with the 1:2 WL:PTX priority-ratio, the SLT-1:2 policy reduced annual waiting list deaths to 31, a 65.6% decrease compared to the existing policy. Simulations showed a non-proportional relationship between increased utilisation rates and reduction in waiting list mortality: for the standard policies, a 5% increase in utilisation resulted in a 10.9% reduction in mortality, a 10% increase resulted in a 21.7% reduction, and a 25% increase resulted in a 45.7% reduction. This non-proportionality was also observed for the SLT policies. Survey responses highlighted a preference for policies that prioritise post-transplant survival, with the PTX policy aligning most closely with 52% of candidates, recipients and their family members, and 48.4% of clinicians (50.6% overall). Conclusion: This thesis demonstrates a comprehensive approach to evaluating and identifying improvements to the UK lung allocation policy, through a novel combination of methods from the fields of survival analysis, operations research, and computer science. Simulations demonstrated that a national score-based allocation policy could significantly decrease waiting list mortality, increase post-transplant survival, and ensure donor lungs are efficiently allocated to recipients that will benefit most from transplant. The results of this thesis calls into question the historical trend of decreasing use of SLT, and argues for a reversal of this trend by utilising SLT for candidates with ILD. Importantly, the methods developed and described in this thesis extend beyond lung transplantation, offering a framework that can be applied to other donor organs and other healthcare allocation challenges more generally. Furthermore, the proposed lung allocation scoring system would be the first in the world to integrate candidate and donor characteristics to ensure optimal matching between donor and recipient. Overall, this thesis contributes to the field of transplant data science by demonstrating the novel application of methods to balance benefit, urgency, and community values, and highlights the importance of data-driven, transparent, community-aligned research in allocation policy development. Future work should aim to refine predictive models, expand the target population, and explore the practical implications of implementing these recommendations, ensuring a careful, monitored transition to any new allocation system to mitigate unforeseen consequences.