Incorporating Model for End-stage Liver Disease Trajectories into Liver Allocation

Abstract

Background: Liver transplant (LT) allocation prioritizes candidates by current Model for End-stage Liver Disease (MELD) score, without accounting for disease trajectory. However, the magnitude and direction of MELD change (ΔMELD) may convey additional prognostic information and could inform continuous distribution-based allocation.

Methods: We analyzed adult LT candidates in the Organ Procurement and Transplantation Network waitlist data set from January 2016 to June 2024, reformatted into a person-day structure. Discrete-time hazard models estimated associations between death or waitlist dropout at 1 and 90 d and current MELD, ΔMELD, and MELD velocity (ΔMELD/day). We derived dynamic MELD-based scores incorporating these measures (MELD-DY) and evaluated model performance using net reclassification improvement and integrated discrimination improvement.

Results: Among 90 622 candidates, ΔMELD was significantly associated with next-day waitlist mortality ( P < 0.001), with a strong interaction between current MELD and ΔMELD (interaction odds ratio, 1.004; 95% confidence interval, 1.003-1.005), indicating that the prognostic impact of MELD change varied by MELD severity. MELD velocity was independently associated with next-day mortality (odds ratio, 1.24; confidence interval, 1.22-1.27) without requiring an interaction term. Models incorporating ΔMELD/day (dynamic MELD score based on ΔMELD/day) improved prediction of death or dropout compared with static MELD (net reclassification improvement 0.28, P < 0.001; integrated discrimination improvement >0, P < 0.001).

Conclusions: MELD dynamics provide prognostic information beyond static MELD, supporting their use for real-time risk stratification and acuity-based LT allocation.