Development of a Machine Learning-Powered Optimized Lung Allocation System for Maximum Benefits in Lung...

Area of Science:

• Transplantation research
• Medical informatics
• Biostatistics

Background:

• Current lung allocation systems aim to minimize waitlist deaths, improve transplant survival, and ensure fairness.
• Developing an optimal lung allocation score (LAS) is crucial for maximizing transplant benefits.

Purpose of the Study:

• To develop and validate a novel lung allocation score (LAS) system, termed MaxBenefit LAS.
• To maximize overall transplant benefits by balancing waitlist urgency and post-transplant survival.

Main Methods:

• Retrospective analysis of 1,599 lung transplant candidates from the Korean Network for Organ Sharing database (2009-2020).
• Development of the MaxBenefit LAS using elastic-net Cox regression, combining waitlist mortality and post-transplant survival models.
• Performance assessment using Area Under the Curve (AUC) and Uno's C-index, compared against the US LAS.

Main Results:

• The waitlist mortality model demonstrated strong predictive performance (AUC: 0.834 training, 0.818 validation).
• The post-transplant survival model showed good predictive ability (AUC: 0.708 training, 0.685 validation).
• MaxBenefit LAS effectively stratified risk, outperforming conventional LAS in predicting waitlist death and identifying candidates with greater transplant benefits.

Conclusions:

• The MaxBenefit LAS presents a promising strategy for optimizing lung allocation.
• This system balances candidate urgency with post-transplant survival likelihood.
• It has the potential to improve lung transplant recipient outcomes and reduce waitlist mortality through equitable organ allocation.