Pediatric critical care medicine : a journal of the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies · 2025
Death One Hour After Terminal Extubation in Children: Validation of a Machine Learning Model to Predict Cardiac Death After Withdrawal of Life-Sustaining Treatment in a Multicenter Cohort, 2009-2021.
СтатьяМногоцентровое исследованиеВалидационное исследованиеНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Winter Meredith C, Zhou Alice X, Laksana Eugene, Aczon Melissa D, Ledbetter David R, Avesar Michael, Burkiewicz Kimberly, Chandnani Harsha K, Fainberg Nina, Hsu Stephanie, McCrory Michael C, Morrow Katie R, Noguchi Anna, O'Brien Caitlin E, Ojha Apoorva, Pringle Charlene, Ross Patrick A, Shah Jui, Shah Sareen, Shpaner Leonid, Siegel Linda B, Tripathi Sandeep, Wetzel Randall C, Pediatric Data Science and Analytics (PEDAL) subgroup of the Pediatric Acute Lung Injury and Sepsis Investigators (PALISI) Network
Аннотация
OBJECTIVES: In the PICU, predicting death within 1 hour after terminal extubation (TE) is valuable in augmenting family counseling and in identifying suitable candidates for organ donation after circulatory determination of death (DCDD). The objective of this study was to train and validate a machine learning model to predict death within 1 hour after TE.
DESIGN: The Death One Hour After Terminal Extubation (DONATE) database was generated using multicenter retrospective data from 2009 to 2021. Data covering demographics, clinical features, vital signs, laboratory values, ventilator settings, medications, and procedures were collected. Machine learning models were trained to predict whether a pediatric patient would die within 1 hour after TE and evaluated on a holdout set.
SETTING: Ten U.S. PICUs.
PATIENTS: Children and adolescents, 0-21 years old, who died after TE ( n = 957).
INTERVENTIONS: None.
MEASUREMENTS AND MAIN RESULTS: The final model was a parsimonious extra-trees model with 21 input features. It was trained on the 2009-2018 data from eight sites ( n = 634) and evaluated on a holdout set comprised of the 2019-2021 data of all ten sites ( n = 323), representing temporal and external validation. The area under the receiver operating characteristic curve and 95% CI was 0.84 (95% CI, 0.81-0.87). At a sensitivity of 90%, the positive predictive value (PPV) was 88%, the negative predictive value (NPV) was 70%, and the number needed to alert (NNA) was 1.14. Among potential organ donors, at the same sensitivity level, the PPV was 86%, the NPV was 74%, and the NNA was 1.17.
CONCLUSIONS: Our model, trained and validated on multisite data, predicted whether a child will die within 1 hour of TE with high discrimination and a low false alarm rate. This finding has important applications to end-of-life counseling and institutional resource utilization when families wish to attempt DCDD.
DESIGN: The Death One Hour After Terminal Extubation (DONATE) database was generated using multicenter retrospective data from 2009 to 2021. Data covering demographics, clinical features, vital signs, laboratory values, ventilator settings, medications, and procedures were collected. Machine learning models were trained to predict whether a pediatric patient would die within 1 hour after TE and evaluated on a holdout set.
SETTING: Ten U.S. PICUs.
PATIENTS: Children and adolescents, 0-21 years old, who died after TE ( n = 957).
INTERVENTIONS: None.
MEASUREMENTS AND MAIN RESULTS: The final model was a parsimonious extra-trees model with 21 input features. It was trained on the 2009-2018 data from eight sites ( n = 634) and evaluated on a holdout set comprised of the 2019-2021 data of all ten sites ( n = 323), representing temporal and external validation. The area under the receiver operating characteristic curve and 95% CI was 0.84 (95% CI, 0.81-0.87). At a sensitivity of 90%, the positive predictive value (PPV) was 88%, the negative predictive value (NPV) was 70%, and the number needed to alert (NNA) was 1.14. Among potential organ donors, at the same sensitivity level, the PPV was 86%, the NPV was 74%, and the NNA was 1.17.
CONCLUSIONS: Our model, trained and validated on multisite data, predicted whether a child will die within 1 hour of TE with high discrimination and a low false alarm rate. This finding has important applications to end-of-life counseling and institutional resource utilization when families wish to attempt DCDD.