Chao Liu1, Xiaoli Liu, Zhi Mao, Pan Hu, Xiaoming Li, Jie Hu, Quan Hong, Xiaodong Geng, Kun Chi, Feihu Zhou, Guangyan Cai, Xiangmei Chen, Xuefeng Sun. 1. Medical School of Chinese PLA, Beijing, China Department of Nephrology, Chinese PLA General Hospital, Chinese PLA Institute of Nephrology, State Key Laboratory of Kidney Diseases, National Clinical Research Center for Kidney Diseases, Beijing, China School of Biological Science and Medical Engineering, Beihang University, Beijing, China Department of Critical Care Medicine, Chinese PLA General Hospital, Beijing, China Department of Anesthesiology, The 920 Hospital of Joint Logistic Support Force of Chinese PLA, Kunming Yunnan, China.
Abstract
PURPOSE: Rhabdomyolysis (RM) is a complex set of clinical syndromes that involves the rapid dissolution of skeletal muscles. Mortality from rhabdomyolysis is approximately 10%. This study aimed to develop an interpretable and generalizable model for early mortality prediction in RM patients. METHOD: Retrospective analyses were performed on two electronic medical record databases: the eICU Collaborative Research Database (eICU-CRD) and the Medical Information Mart for Intensive Care III (MIMIC- III) database. We extracted data from the first 24h after patient ICU admission. Data from the two datasets were merged for further analysis. The merged datasets were randomly divided, with 70% used for training and 30% for validation. We used the machine learning model XGBoost (extreme gradient boosting (XGBoost) with the Shapley additive explanation method to conduct early and interpretable predictions of patient mortality. Five typical evaluation indexes were adopted to develop a generalizable model. RESULTS: In total, 938 patients with RM were eligible for this analysis. The AUC of the XGBoost model in predicting hospital mortality was 0.871, the sensitivity was 0.885, the specificity was 0.816, the accuracy was 0.915 and the F1 score was 0.624. The XGBoost model performance was superior to that of other models (logistic regression (AUC = 0.862), support vector machine (AUC = 0.843), random forest (AUC = 0.825) and naive Bayesian (AUC = 0.805) and clinical scores (SOFA (AUC = 0.747) and APS III (AUC = 0.721)). CONCLUSIONS: Although the XGBoost model is still not great from an absolute performance perspective, it provides better predictive performance than other models for estimating the mortality of patients with RM based on patient characteristics in the first 24h of admission to the ICU.
PURPOSE:Rhabdomyolysis (RM) is a complex set of clinical syndromes that involves the rapid dissolution of skeletal muscles. Mortality from rhabdomyolysis is approximately 10%. This study aimed to develop an interpretable and generalizable model for early mortality prediction in RM patients. METHOD: Retrospective analyses were performed on two electronic medical record databases: the eICU Collaborative Research Database (eICU-CRD) and the Medical Information Mart for Intensive Care III (MIMIC- III) database. We extracted data from the first 24h after patient ICU admission. Data from the two datasets were merged for further analysis. The merged datasets were randomly divided, with 70% used for training and 30% for validation. We used the machine learning model XGBoost (extreme gradient boosting (XGBoost) with the Shapley additive explanation method to conduct early and interpretable predictions of patientmortality. Five typical evaluation indexes were adopted to develop a generalizable model. RESULTS: In total, 938 patients with RM were eligible for this analysis. The AUC of the XGBoost model in predicting hospital mortality was 0.871, the sensitivity was 0.885, the specificity was 0.816, the accuracy was 0.915 and the F1 score was 0.624. The XGBoost model performance was superior to that of other models (logistic regression (AUC = 0.862), support vector machine (AUC = 0.843), random forest (AUC = 0.825) and naive Bayesian (AUC = 0.805) and clinical scores (SOFA (AUC = 0.747) and APS III (AUC = 0.721)). CONCLUSIONS: Although the XGBoost model is still not great from an absolute performance perspective, it provides better predictive performance than other models for estimating the mortality of patients with RM based on patient characteristics in the first 24h of admission to the ICU.
Authors: Kamyar M Hedayat; David Chalvet; Maël Yang; Shahrokh Golshan; Caroline Allix-Beguec; Serge Beneteaud; Thomas Schmit Journal: Front Med (Lausanne) Date: 2022-06-06