MELD 3.i: A Bayesian framework for updating the model for end-stage liver disease score

Background & Aims: The model for end-stage liver disease (MELD) score has been central to liver transplant (LT) allocation since 2002, with iterative updates culminating in MELD 3.0. However, given temporal changes and variations in liver disease epidemiology across allocation systems that utilize MELD, continuous refinements are essential to ensure its optimal performance across diverse patient populations and transplant frameworks. Methods: This retrospective cohort study included all US adult LT candidates listed between July 13, 2023 to June 30, 2024. Candidates from the first two-thirds of listing dates formed the training set, while the last third comprised the validation set. We applied a Bayesian proportional hazards model, using MELD 3.0 as informative priors to generate posterior coefficient distributions. The resulting model, MELD 3.1, represents the first iteration within the MELD 3.i framework. Model performance was assessed using concordance (C-) statistics, and reclassification analyses evaluated patient redistribution and mortality risk across MELD tiers. Results: The cohort included 13,764 candidates (41.1% female). MELD 3.1 assigned a higher coefficient for female sex and a lower coefficient for creatinine, and showed improved C-statistics for 90-day waitlist mortality in the validation set (0.7195 vs. 0.7152 for MELD 3.0, p = 0.036). MELD 3.1 led to a net 3.1% up-categorization of patients who died or dropped out while on the waitlist, with the net gain entirely accounted for by female candidates. MELD 3.1 also showed net gains across age groups and liver disease etiologies. Conclusion: Our findings provide proof-of-concept for the MELD updating framework (MELD 3.i) as a sustainable and adaptive approach for periodically refining MELD using contemporary data and Bayesian methods. This iterative process enhances predictive accuracy, ensuring MELD remains responsive to evolving demographics and clinical practices in the US and other allocation systems. Impact and implications: We present MELD 3.i, a Bayesian framework for ongoing refinement of the MELD score, with MELD 3.1 as its first iteration. The findings demonstrate improved predictive accuracy for 90-day waitlist mortality compared to the original MELD 3.0, particularly for women. These results are important for clinicians, transplant centers, and policymakers aiming to optimize organ allocation while addressing persistent sex disparities. By enabling adaptive refinements within an existing model structure, this approach has the potential to offer a practical method to align liver transplant prioritization with evolving patient demographics and varying practices across nations and regions.