Modelling outcome prediction for trauma patients : an artificial intelligence approach

Ali, NA

Abstract

Trauma, a term used in medicine to describe a physical injury, is believed to be one
of the major causes of death and disability in modern societies. The development of
the Trauma and Injury Severity Score (TRISS) method by Boyd et al. (1987) can be
considered as a high-impact initiative in order to improve the trauma patient's care.
This method was used to compare the expected and the observed outcomes in
relation to mortality. Thus, the rate of unexpected deaths or survivals can be
examined and any related problems such as improper trauma patient's care can be
identified. In general, the model with this particular task can be referred to as a
prediction or classification model. In our study, the Trauma Audit and Research
Network (TARN) has applied the TRISS method to assist them in a comparative
audit among the participating hospitals since 1989. Despite the fact that the TRISS
model is simple and easy to use, there is some limitation in the logistic regression
technique which the TRISS model is based upon. In fact, some preliminary results
from other researchers have indicated that prediction accuracy may be improved by
using alternative modelling approaches, such as the artificial intelligence (AI) based
methods. Therefore, attempts are made in this study with the aim of developing new
outcome prediction models using the AI methods namely; artificial neural networks,
support vector machines, A>nearest neighbour and naive Bayesian, and then the
results will be compared to the TRISS-FP model (for comparison purposes, we refer
to the outcome prediction model based on the TRISS method developed by TARN as
the TRISS-FP model throughout this thesis). The model's predictive performances
are evaluated using performance measures, including sensitivity, specificity, area
under the receiving operating characteristic curve (AUC) and geometric mean (Gmean).
The data for this research is drawn from the TARN database. The empirical
result has shown that the new Al-based models developed in this study obtained
better predictive performances compared to the TRISS-FP model. The ANN model
has emerged as the best Al-based model. Thus, this model will be recommended to
the TARN for future consideration.

Citation

Ali, N. Modelling outcome prediction for trauma patients : an artificial intelligence approach. (Thesis). University of Salford

• Thesis Type: Thesis
• Deposit Date: Aug 13, 2021
• Additional Information: Funders : Universiti Teknologi Malaysia;Ministry of Higher Education, Malaysia
• Award Date: Apr 1, 2011