Journal of clinical gastroenterology · 2024
Prediction of in-hospital Mortality of Intensive Care Unit Patients with Acute Pancreatitis Based on an Explainable Machine Learning Algorithm.
СтатьяНазвание 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.
Ren Wensen, Zou Kang, Huang Shu, Xu Huan, Zhang Wei, Shi Xiaomin, Shi Lei, Zhong Xiaolin, Peng Yan, Tang Xiaowei, Lü Muhan
Аннотация
BACKGROUND AND AIM: Acute pancreatitis (AP) is potentially fatal. Therefore, early identification of patients at a high mortality risk and timely intervention are essential. This study aimed to establish an explainable machine-learning model for predicting in-hospital mortality of intensive care unit (ICU) patients with AP.
METHODS: Data on patients with AP, including demographics, vital signs, laboratory tests, comorbidities, treatment, complication, and severity scores, were extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) and the eICU collaborative research database (eICU-CRD). Based on the data from MIMIC-IV, we used the least absolute shrinkage and selection operator algorithm to select variables and then established 9 machine-learning models and screened the optimal model. Data from the eICU-CRD were used for external validation. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, decision curve, and calibration plots were used to assess the models' efficacy. Shapley's additive explanation values were used to explain the model.
RESULTS: Gaussian naive Bayes (GNB) model had the best performance on the data from MIMIC-IV, with an AUC, accuracy, sensitivity, and specificity of 0.840, 0.787, 0.839, and 0.792, respectively. The GNB model also performed well on the data from the eICU-CRD, with an AUC, accuracy, sensitivity, and specificity of 0.862, 0.833, 0.848, and 0.763, respectively. According to Shapley's additive explanation values, the top 4 predictive factors were maximum red cell distribution width, minimum saturation of blood oxygen, maximum blood urea nitrogen, and the Sequential Organ Failure Assessment score.
CONCLUSION: The GNB model demonstrated excellent performance and generalizability in predicting mortality in ICU patients with AP. Therefore, it can identify patients at a high mortality risk.
METHODS: Data on patients with AP, including demographics, vital signs, laboratory tests, comorbidities, treatment, complication, and severity scores, were extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) and the eICU collaborative research database (eICU-CRD). Based on the data from MIMIC-IV, we used the least absolute shrinkage and selection operator algorithm to select variables and then established 9 machine-learning models and screened the optimal model. Data from the eICU-CRD were used for external validation. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, decision curve, and calibration plots were used to assess the models' efficacy. Shapley's additive explanation values were used to explain the model.
RESULTS: Gaussian naive Bayes (GNB) model had the best performance on the data from MIMIC-IV, with an AUC, accuracy, sensitivity, and specificity of 0.840, 0.787, 0.839, and 0.792, respectively. The GNB model also performed well on the data from the eICU-CRD, with an AUC, accuracy, sensitivity, and specificity of 0.862, 0.833, 0.848, and 0.763, respectively. According to Shapley's additive explanation values, the top 4 predictive factors were maximum red cell distribution width, minimum saturation of blood oxygen, maximum blood urea nitrogen, and the Sequential Organ Failure Assessment score.
CONCLUSION: The GNB model demonstrated excellent performance and generalizability in predicting mortality in ICU patients with AP. Therefore, it can identify patients at a high mortality risk.