Death One Hour After Terminal Extubation (DONATE): A Multisite Cohort Study

Death One Hour After Terminal Extubation (DONATE): A Multisite Cohort Study

Winter M; Zhou A; Laksana E; Aczon M; Ledbetter D; Avesar M; Fainberg N; Hsu S; McCrory M; Morrow K; O'Brien C; Pringle C; Ross P; Shah S; Shpaner L; Siegel L; Tripathi S; Wetzel R

Critical Care Medicine

2024

child; Cohort Studies; adult; cohort analysis; female; human; major clinical study; male; retrospective study; aged; pediatric intensive care unit; machine learning; randomized controlled trial; clinical feature; family counseling; conference abstract; intensive care unit; drug therapy; vital sign; death; organ donor; predictive value; extubation; operating room

INTRODUCTION: Predicting time to death after terminal extubation is valuable to augment family counseling and identify suitable candidates for organ donation after circulatory death (DCD). Our objective was to train and validate a machine learning model that predicts time to death after terminal extubation. <br/>METHOD(S): The Death One Hour After Terminal Extubation (DONATE) Study is a multi-site retrospective cohort study using data from ten U.S. pediatric intensive care units (ICUs). Patients 0-21 years of age who died after terminal extubation in the ICU from 2009-2021 (n=957) were included. Demographic data, clinical features, vital signs, laboratory values, ventilator settings, medications, and procedures were collected. Using these input variables, machine learning models were trained to predict whether a patient would die within one hour after terminal extubation and then evaluated on a holdout set. Performance metrics included area under the receiver operator characteristic curve (AUROC), positive predictive value (PPV), and number needed to alert (NNA). <br/>RESULT(S): The final model was a parsimonious extremely randomized trees model with 21 input features, all of which can be obtained from the medical chart or bedside assessment. It was trained on the 2009-2018 data from eight sites (n=634) and evaluated on a holdout set composed of the 2019-2021 data of all ten sites (n=323), representing temporal and external validation. The AUROC was 0.84. At a sensitivity of 0.90, the PPV was 0.88 and the NNA was 1.14. Among patients who met United Network for Organ Sharing criteria for DCD, at the same sensitivity level, the PPV was 0.86 and the NNA was 1.17, implying that six of seven prepared operating rooms would be predicted to yield a viable organ. <br/>CONCLUSION(S): Our model trained and validated on multi-site data accurately predicted whether a child would die within one hour of terminal extubation. This has important applications to end-of-life counseling, especially for families who wish to attempt organ donation after circulatory death.

Article information provided for research and reference use only. PedPalASCNET does not hold any rights over the resource listed here. All rights are retained by the journal listed under publisher and/or the creator(s).

July List 2024