Belief rule-based inference for predicting trauma outcome

Abstract

A belief rule-based inference methodology using the evidential reasoning approach (RIMER) is employed in this study to construct a decision support tool that helps physicians predict in-hospital death and intensive care unit admission among trauma patients in emergency departments (EDs). This study contributes to the research community by developing and validating a RIMER-based decision tool for predicting trauma outcome. To compare the prediction performance of the RIMER model with those of models derived using commonly adopted methods, such as logistic regression analysis, support vector machine (SVM), and artificial neural network (ANN), several logistic regression models, SVM models, and ANN models are constructed using the same dataset. Five-fold cross-validation is employed to train and validate the prediction models constructed using four different methods. Results indicate that the RIMER model has the best prediction performance among the four models, and its performance can be improved after knowledge base training with historical data. The RIMER tool exhibits strong potential to help ED physicians to better triage trauma, optimally utilize hospital resources, and achieve better patient outcomes.