Logo image
Back
A Tool to Predict Donation After Circulatory Death Expiration at 30- and 120-Minutes Post Extubation
Abstract   Peer reviewed

A Tool to Predict Donation After Circulatory Death Expiration at 30- and 120-Minutes Post Extubation

J. White, J. Song, H. Zappacosta, P. Cho, S. McKay, A. Abramov, M. Daniel, T. Seto, S. West, D. Telesca, …
The Journal of heart and lung transplantation, v 45(5), pp 6-7
Jul 2026
DOI: https://doi.org/10.1016/j.healun.2026.02.019
Featured in Collection :   Drexel's Newest Publications

Abstract

Purpose: Donation after Circulatory Death (DCD) is a vital source of donor organs in the United States. However, not all potential donors progress to circulatory death rapidly enough to yield viable organs. Attempts to develop predictive algorithms have been unsuccessful due to sample size, insufficient number of variables and incomplete analysis. Prior models have been largely limited to regression analysis which is suboptimal due to significant variable correlation. Therefore, this study seeks to use machine learning techniques imposed on larger, more comprehensive data sets to predict which DCD donors will expire at 30- and 120-minutes post extubation. Methods: This is a retrospective cohort study with predictive modeling. Several thousand potential DCD donors from January 2014 to June 2025 will be analyzed across 71 variables through data sets provided directly from organ procurement organizations. The outcome of interest is a binary of expiration at the respective time interval. Univariate analysis conducted on categorical and continuous variables will be compared using the Chi-Squared and t-test, respectively with subsequent multivariable regression to create an initial model for later comparison. The data set will be inserted into XGboost and Random Forest for machine learning modeling with associated Receiver Operating Curves (ROC) and Precision-Recall (PR) curves. Endpoints The primary intent is to create a model that accurately discerns DCD expiration at 30- and 120-minute intervals, available via web-based application. This study is novel due to a sample size and variable list 10x larger than any previous DCD expiration study. Initial XGboost modeling with a limited sample resulted in PR AUCs of 0.84 and 0.90 at 30- and 120-minutes, respectively. In the 30-minute model a precision of 80% resulted in 65% recall. In the 120-minute model a precision of 80% resulted in 97% recall. This study will have tremendous impact on DCD heart and lung donation by radically improving procurement efficiency.

Metrics

1 Record Views

Details

Logo image