Developing and validation of a liver transplantation donation after cardiac death risk index using the UNOS database
Recommended Citation
Chau L, Delvecchio K, Mohamed A, Kitajima T, Lu M, Yedulla S, Collins K, Rizzari M, Yoshida A, Abouljoud M, and Nagai S. Developing and validation of a liver transplantation donation after cardiac death risk index using the UNOS database. Am J Transplant 2021; 21(SUPPL 1):21.
Document Type
Conference Proceeding
Publication Date
2-1-2021
Publication Title
Am J Transplant
Abstract
Introduction: Donation after cardiac death (DCD) liver transplantation is an increasing form of organ donation. Shlegal et. al. identified seven factors predicting 1-year DCD graft survival based on the UK transplantation population. This project aims to validate the existing predictive model and to develop a novel DCD graft failure prediction model based on the UNOS database.
Methods: We examined all adult DCD transplanted Jan 1 2014 to Mar 31 2020 in the UNOS registry. The population was divided into train (66%) and validation (34%) subsets. Variables of interest were selected from the train subset with backwards stepwise selection with criteria for entry P = 0.05 and exit P = 0.06. Logistic regression models were fitted based on selected variables to predict 1-year graft failure. Performance of the model was assessed in the validation population by computing the area under the receiver operating characteristic curve (AUROC) after 10-fold stratified cross-validation. The performance of the novel model was compared to the UK DCD prediction model.
Results: 2738 DCD transplants were included in this study with 1835 in the train and 903 in the validation subsets. The model identified 12 factors predictive for 1-year graft failure among DCD recipients. The model AUROC was 0.741 (95% CI: 0.686, 0.796). When validating the UK DCD model in the UNOS database, the model achieved AUROC of 0.628 (0.564, 0.691).
Conclusions: This model identified 12 predictive factors predictive of 1-year graft failure among DCD recipients from the UNOS database, which outperformed the existing model.
Volume
21
Issue
Suppl 1
First Page
21