Artificial intelligence and statistical techniques to predict probability of injury survival

SALEH, Mohammed (2018). Artificial intelligence and statistical techniques to predict probability of injury survival. Doctoral, Sheffield Hallam University.

Preview

PDF Saleh_2019_PhD_ArtificialIntelligenceStatistical.pdf - Accepted Version Creative Commons Attribution Non-commercial No Derivatives. Download (5MB) | Preview

Abstract

The aim of this study is to design, develop and evaluate artificial intelligence and statistical techniques to predict the probability of survival in traumas using knowledge acquired from a database of confirmed traumas outcomes (survivors and not survivors). Trauma in this study refers to body injuries from accidents or other means. Quantifying the effects of traumas on individuals is challenging as they have many forms, affect different organs, differ in severity and their consequence could be related to the individual's physiological attributes (e.g. age, fragility, premedical condition etc). It is known that appropriate intervention improves survival and may reduce disabilities in traumas. Determining the probability of survival in traumas is important as it can inform triage, clinical research and audit. A number of methods have been reported for this purpose. These are based on a combination of physiological and anatomical examination scores. However, these methods have shortcomings as for example, combining the scores from injuries for different organs is complicated. A method for predicting probability of survival in traumas needs to be accurate, practical and accommodate broad cases. In this study Sheffield Hallam University, Sheffield Children's Hospital, Sheffield University and the Trauma Audit and Research Network (TARN) collaborated to develop improved means of predicting probability of survival in traumas. The data used in this study were trauma cases and their outcomes provided by the TARN. The data included 47568 adults (age: mean = 59.9 years, standard deviation = 24.7 years) with various injuries. In total, 93.3% of cases had survived and 6.7% of cases had not survived. The data were partitioned into calibration (2/3 of the data) and evaluation (1/3 of the data). The trauma parameters used in the study were: age, respiration rate (RR), systolic blood pressure (SBP), pulse (heart) rate (PR) and the values obtained from two trauma scoring systems called Abbreviated Injury Score (AIS) and Glasgow Coma Score (GCS). Intubation and Pre-exiting Medical Condition (PMC) data were also considered. Initially a detailed statistical exploration of the manner trauma these trauma parameters related to the probability of survival outcomes was carried out and the results were interpreted. The resulting information assisted the development of three methods to predict probability of survival. These were based on Bayesian statistical approach called predictive statistical diagnosis (PSD), a new method called Iterative Random Comparison classification (IRCC) and the third method combined the IRCC with the fuzzy inference system (FIS). The performance of these methods was compared with each other as well as the method of predicating survival used by the TARN called Ps14 (the name refers to probability of survival method reported in 2014). The study primarily focused on Trauma Brain Injury (TBI) as they represented the majority of the cases. For TBI, the developed IRCC performed best amongst all methods including Ps14. It predicted survivors and not survivors with 97.2% and 75.9% accuracies respectively. In comparison, the predication accuracy for Ps14 for survivors and not survivors were 97.4% and 40.2%. The study provided resulted in new findings that indicated the manner trauma parameters affect probability of survival and resulted in new artificial intelligence and statistical methods of determining probability survival in trauma.

Item Type:	Thesis (Doctoral)
Contributors:	Thesis advisor - Saatchi, Reza [0000-0002-2266-0187]
Additional Information:	Director of studies/Supervisor - Professor Reza Saatchi
Identification Number:	https://doi.org/10.7190/shu-thesis-00188
Depositing User:	Louise Beirne
Date Deposited:	28 Jun 2019 11:34
Last Modified:	03 May 2023 02:06
URI:	https://shura.shu.ac.uk/id/eprint/24775

Actions (login required)

View Item

Downloads

Publication Metrics