{"title":"Prediction of traumatic hemorrhagic shock using a Multi-scale exogenous variable model (MS-TimeXer-MoE).","authors":"Wenxin Wang, Bing Chen, Qiuyi Wang, Jian Rong","doi":"10.1007/s00068-025-02878-8","DOIUrl":null,"url":null,"abstract":"<p><strong>Objective: </strong>To predict the likelihood of hemorrhagic shock in trauma patients using a multi-scale exogenous variable model, MS-TimeXer-MoE.</p><p><strong>Methods: </strong>This study is the first to use the TimeXer method incorporating exogenous variables to establish a predictive model for traumatic hemorrhagic shock, achieving notable prediction results. Data from trauma patients were extracted from the MIMIC IV database according to inclusion and exclusion criteria. After data processing, the most relevant indicators were selected, including endogenous variables (e.g., vital signs, laboratory indicators) and exogenous variables (e.g., cause of trauma, gender, age, injury site). By integrating exogenous variables and multi-scale feature learning, and innovatively combining the multi-expert mechanism with the TimeXer method, a multi-expert mixed model, MS-TimeXer-MoE, was developed to predict hemorrhagic shock occurrence with higher accuracy and specificity. A total of 4,870 patients were included, divided into an experimental group of 2,432 cases and a control group of 2,438 cases based on the occurrence of hemorrhagic shock post-admission. The dataset was split into training, validation, and testing sets in a 60%:20%:20% ratio RESULTS: The AUC value of the MS-TimeXer-MoE model in predicting hemorrhagic shock in trauma patients was 0.8995, with a recall rate of 0.8607, demonstrating high efficiency in shock identification, indicating that the model can distinguish positive and negative samples with high accuracy and recall rate. In regression tasks, the MS-TimeXer-MoE model's mean absolute error (MAE) was 3.4397, mean squared error (MSE) was 8.2735, mean absolute percentage error (MAPE) was 5.9933%, and the coefficient of determination (R<sup>2</sup>) reached 87.3436%, showing good fitting and accuracy in time-series data prediction. In the five-fold cross-validation experiment, the variances of MAE and MSE were 0.031 and 0.174, respectively, further reflecting the model's low error fluctuation across different folds, ensuring accurate prediction of shock occurrence time.</p><p><strong>Conclusion: </strong>Compared to existing mainstream modeling methods for predicting shock occurrence, the MS-TimeXer-MoE model can more accurately predict the occurrence of hemorrhagic shock in trauma patients.</p>","PeriodicalId":520620,"journal":{"name":"European journal of trauma and emergency surgery : official publication of the European Trauma Society","volume":"51 1","pages":"222"},"PeriodicalIF":2.2000,"publicationDate":"2025-06-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"European journal of trauma and emergency surgery : official publication of the European Trauma Society","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1007/s00068-025-02878-8","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Objective: To predict the likelihood of hemorrhagic shock in trauma patients using a multi-scale exogenous variable model, MS-TimeXer-MoE.
Methods: This study is the first to use the TimeXer method incorporating exogenous variables to establish a predictive model for traumatic hemorrhagic shock, achieving notable prediction results. Data from trauma patients were extracted from the MIMIC IV database according to inclusion and exclusion criteria. After data processing, the most relevant indicators were selected, including endogenous variables (e.g., vital signs, laboratory indicators) and exogenous variables (e.g., cause of trauma, gender, age, injury site). By integrating exogenous variables and multi-scale feature learning, and innovatively combining the multi-expert mechanism with the TimeXer method, a multi-expert mixed model, MS-TimeXer-MoE, was developed to predict hemorrhagic shock occurrence with higher accuracy and specificity. A total of 4,870 patients were included, divided into an experimental group of 2,432 cases and a control group of 2,438 cases based on the occurrence of hemorrhagic shock post-admission. The dataset was split into training, validation, and testing sets in a 60%:20%:20% ratio RESULTS: The AUC value of the MS-TimeXer-MoE model in predicting hemorrhagic shock in trauma patients was 0.8995, with a recall rate of 0.8607, demonstrating high efficiency in shock identification, indicating that the model can distinguish positive and negative samples with high accuracy and recall rate. In regression tasks, the MS-TimeXer-MoE model's mean absolute error (MAE) was 3.4397, mean squared error (MSE) was 8.2735, mean absolute percentage error (MAPE) was 5.9933%, and the coefficient of determination (R2) reached 87.3436%, showing good fitting and accuracy in time-series data prediction. In the five-fold cross-validation experiment, the variances of MAE and MSE were 0.031 and 0.174, respectively, further reflecting the model's low error fluctuation across different folds, ensuring accurate prediction of shock occurrence time.
Conclusion: Compared to existing mainstream modeling methods for predicting shock occurrence, the MS-TimeXer-MoE model can more accurately predict the occurrence of hemorrhagic shock in trauma patients.