Prediction of traumatic hemorrhagic shock using a Multi-scale exogenous variable model (MS-TimeXer-MoE).

Area of Science:

• Medical Informatics
• Trauma Care
• Predictive Modeling

Background:

• Hemorrhagic shock is a critical condition in trauma patients requiring timely prediction.
• Existing predictive models often lack accuracy or fail to incorporate diverse patient data.

Purpose of the Study:

• To develop and validate a novel multi-scale exogenous variable model, MS-TimeXer-MoE, for predicting hemorrhagic shock in trauma patients.
• To improve the accuracy and specificity of hemorrhagic shock prediction using a mixed-expert approach.

Main Methods:

• Utilized the TimeXer method incorporating endogenous (vital signs, labs) and exogenous (trauma cause, demographics, injury site) variables.
• Developed the MS-TimeXer-MoE model integrating multi-scale feature learning and a multi-expert mechanism.
• Analyzed data from 4,870 trauma patients from the MIMIC IV database, split into training, validation, and testing sets.

Main Results:

• The MS-TimeXer-MoE model achieved an AUC of 0.8995 and a recall rate of 0.8607 for hemorrhagic shock prediction.
• Regression analysis showed high accuracy with R² of 87.34%, MAE of 3.4397, and MSE of 8.2735.
• Five-fold cross-validation demonstrated low error fluctuation (MAE variance 0.031, MSE variance 0.174), ensuring robust predictions.

Conclusions:

• The MS-TimeXer-MoE model significantly outperforms existing methods in predicting hemorrhagic shock occurrence in trauma patients.
• The model's high accuracy and reliability support its potential for clinical application in early shock detection.