Session Name: Liver: Living Donor Liver Transplant and Partial Grafts
Session Date & Time: None. Available on demand.
*Purpose: To maximize non-directed living liver donor graft utility, we developed analytic models predicting 10-year graft survival post liver transplant (LT).
*Methods: We analyzed OPTN living liver donor and recipient data (1/2000-12/2019), with follow up to 3/2020. Data included: liver graft type (partial right, partial left, partial left lateral [segments 2&3]), 7 donor & 21 recipient factors, CIT, and donor-to-recipient body surface area (BSA) ratio. To accommodate data complexity with many variables and interactions, we constructed random forest survival models predicting 10-year graft survival for each of the 3 graft types, and variable selection to optimize the c-statistic. Data for each graft type was split into training and test sets. Survivals for predicted groups were calculated by Kaplan-Meier analyses.
*Results: The data included 6328 live liver donors (4621 partial right, 644 partial left, 1063 partial left lateral). Of these, 21% (n=1328) were adult to child, 79% (n=5000) were adult to adult. Most partial left lateral grafts (95%) went to children, and the majority of partial right grafts (98%) went to adults. Donor-to-recipient BSA ratio was an important survival predictor in all 3 graft type models (degree of importance in the models was: partial left lateral>partial left>partial right). Variables common to all 3 graft types were: malignant diagnosis, medical location at the time of LT (inpatient/ICU), and moderate ascites. Biliary atresia diagnosis was of high importance in partial left and left lateral graft models, but of no importance in the partial right graft model. Re-transplant status was only important in the partial right graft model. The best c-statistic was 0.70 for the partial left lateral model, and 0.63 and 0.61 for partial left and partial right models, respectively.(Fig. 1) To compare model predictions to actual survival, the 10-year graft survival predictions were stratified into upper quartile versus the combined lower quartiles. The upper quartile group in each model had significantly better 10-yr graft survival than the lower quartile groups, (p<0.005).
*Conclusions: Our utility-based models identify and stratify potential recipients for non-directed living donor livers based on the predicted 10-year graft survivals, while accounting for complex donor-recipient interactions. There is an unmet need for evidence to help guide living liver donor programs to more objectively allocate non-directed living donor livers, and these analyses set the stage for further investigation into this important area.
To cite this abstract in AMA style:Bambha K, Perkins J, Sturdevant M, Biggins SW, Bakthavatsalam R, Healey P, Dick A, Reyes J. Machine Learning Informs Utility-Based Non-Directed Living Liver Donor Allocation [abstract]. Am J Transplant. 2021; 21 (suppl 3). https://atcmeetingabstracts.com/abstract/machine-learning-informs-utility-based-non-directed-living-liver-donor-allocation/. Accessed June 18, 2021.
« Back to 2021 American Transplant Congress