Creation of Machine Learning Based Models Predicting Donation After Circulatory Death Liver Transplantation Survival and Length of Stay

Document Type

Conference Proceeding

Publication Date

6-1-2023

Publication Title

Am J Transplant

Abstract

Purpose: Machine learning techniques allow for complex modeling of associations with iterative cloud-based model tuning and deployment. We aim to apply these methods to estimate DCD liver transplantation outcomes. Methods: All adult DCD liver transplant recipients in the UNOS STAR file from Feb 1, 2003, to Sept 14, 2021 were included. Pediatric and simultaneous transplantation were excluded. Study cohort was divided into 80% training, 10% validation, and 10% testing sub-groups with more recent transplantations weighted for testing. Python and TensorFlow based feed-forward neural networks and gradient boosting decision tree models hosted on Google Cloud platform were deployed to predict 1-year patient survival, 1-year liver graft survival, and length of index hospitalization (LOS). Performance of the models was assessed using area under the receiver operating curve (AUC-ROC) and root mean square error (RMSE) with confidence intervals constructed using bootstrapping with 2000 resamples. Results: 8074 DCD liver transplant recipients were included in the study with 791 recipients in the testing cohort. 142 donor, recipient, and waitlist characteristics available at time of transplantation were used for model construction. The model was predictive of 1-year patient survival with AUC-ROC 0.794 (95% C.I. 0.659- 0.929), 1-year liver graft survival with AUC-ROC 0.825 (95% C.I. 0.716-0.934), and index LOS with RMSE 16.9. Conclusions: We demonstrate a cloud-based machine learning model that robustly predicts DCD liver transplantation index hospitalization length of stay, as well as 1-year patient and graft survival. Further hyperparameter tuning with future iterations of registry data will allow for continuous improvement. CITATION INFORMATION: Chau L., Lu Z., Miyake K., Kitajima T., Wickramaratne N., Rizzari M., Yoshida A., Abouljoud M., Nagai S. Creation of Machine Learning Based Models Predicting Donation After Circulatory Death Liver Transplantation Survival and Length of Stay AJT, Volume 23, Issue 6, Supplement 1. DISCLOSURES: L. Chau: None. [Figure presented]

Volume

23

Issue

6

First Page

S856

Last Page

S857

Find It @ Sladen

Share

COinS