BMC medical informatics and decision making · 2025
Development and validation of interpretable machine learning models to predict intensive care unit outcomes in patients on hemodialysis: a multicenter study.
СтатьяМногоцентровое исследованиеВалидационное исследованиеНазвание и аннотация: Перевод пока не подготовлен
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.
Chen Minjie, Li Pengan, Xu Yuanwen, Li Zhenghui, Xiong Yan, Wu Jianhua, Pandya Chintan, Wang Yunuo, Huang Guixin
Аннотация
BACKGROUND: Hemodialysis patients are at high risk for ICU admission due to elevated mortality, cardiovascular disease, and infection rates. Traditional ICU scoring systems (e.g., APACHE-II, SOFA) demonstrate limited accuracy in this population. This study aimed to identify key risk factors and develop interpretable machine learning (ML) models for predicting ICU outcomes to enable early intervention.
METHODS: This multicenter study analyzed data from three cohorts: The First Affiliated Hospital of Sun Yat-sen University (n = 248), MIMIC-IV (n = 769), and eICU-CRD (n = 1,878). Primary outcome was all-cause ICU mortality; secondary outcomes were cardiovascular and infection-related mortality. Thirteen ML algorithms and ensemble models were applied to 113 clinical variables collected within 24 h of ICU admission. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and benchmarked against existing ICU scoring systems. We employed SHapley Additive exPlanation (SHAP) analysis to enhance interpretability.
RESULTS: Key predictors numbered 6 (cardiovascular mortality), 11 (infection-related mortality), and 25 (all-cause mortality). Ensemble machine learning models, trained on the SYSU cohort, were initially screened by performance (8-fold cross-validation AUC ≥ 0.80) and evaluated in the eICU selection cohort, with the top-performing models subsequently validated in the external MIMIC-IV cohort. In the external validation, NeuralNetC achieved the highest AUC of 0.847 (95% confidence interval [CI] 0.806-0.885) for all-cause mortality among the ensemble models, outperforming ICU scoring systems. ExtraTreesA performed best for infection-related mortality (AUCs: 0.880; 95% CI 0.852-0.906), and NeuralNetD for cardiovascular mortality (AUCs: 0.790; 95% CI 0.733-0.844). An online predictive platform was developed to facilitate clinical application.
CONCLUSION: ML models provided high predictive accuracy for ICU mortality in hemodialysis patients, facilitating early identification of high-risk individuals and supporting targeted interventions. The online platform promotes clinical translation for intensive care decision-making.
METHODS: This multicenter study analyzed data from three cohorts: The First Affiliated Hospital of Sun Yat-sen University (n = 248), MIMIC-IV (n = 769), and eICU-CRD (n = 1,878). Primary outcome was all-cause ICU mortality; secondary outcomes were cardiovascular and infection-related mortality. Thirteen ML algorithms and ensemble models were applied to 113 clinical variables collected within 24 h of ICU admission. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and benchmarked against existing ICU scoring systems. We employed SHapley Additive exPlanation (SHAP) analysis to enhance interpretability.
RESULTS: Key predictors numbered 6 (cardiovascular mortality), 11 (infection-related mortality), and 25 (all-cause mortality). Ensemble machine learning models, trained on the SYSU cohort, were initially screened by performance (8-fold cross-validation AUC ≥ 0.80) and evaluated in the eICU selection cohort, with the top-performing models subsequently validated in the external MIMIC-IV cohort. In the external validation, NeuralNetC achieved the highest AUC of 0.847 (95% confidence interval [CI] 0.806-0.885) for all-cause mortality among the ensemble models, outperforming ICU scoring systems. ExtraTreesA performed best for infection-related mortality (AUCs: 0.880; 95% CI 0.852-0.906), and NeuralNetD for cardiovascular mortality (AUCs: 0.790; 95% CI 0.733-0.844). An online predictive platform was developed to facilitate clinical application.
CONCLUSION: ML models provided high predictive accuracy for ICU mortality in hemodialysis patients, facilitating early identification of high-risk individuals and supporting targeted interventions. The online platform promotes clinical translation for intensive care decision-making.