JavaScript is disabled for your browser. Some features of this site may not work without it.
Please be advised that the site will be down for maintenance on Sunday, September 1, 2024, from 08:00 to 18:00, and again on Monday, September 2, 2024, from 08:00 to 09:00. We apologize for any inconvenience this may cause.
Model development to improve patient safety in the operating theatre
This project considers the need to use machine learning for supporting anaesthesiologists to predict
and detect patient issues. Anaesthesiologists play a vital role within medical care and, especially in
South Africa, are involved in nearly all medical care practices. This project is the first of its kind to look
holistically at the entire anaesthesiology process, where previous papers have aimed at controlling
and investigating only a small portion of the process. It was seen that in South Africa hospitals do not
capture live patient data electronically, but rather on paper format. After testing and considering live
patient data it was opted to construct an artificial data set as to take the sensitivity of the data into
account and aim for a higher model accuracy. An artificial patient data set was constructed using
interviews, medical knowledge available to the masses and intuition. This data set was described in
detail and the deep complex nature of interrelations of the different variables were highlighted. The
data set consisted of a 1000 patients, 500 male patients, 500 female patients, age distributions
between 20 and 80 years old, patient heights in metres, patient weights in kilograms, heart rates in
beats per minute and lastly, systolic and diastolic blood pressures in millimetres mercury. The data
set was analysed by a number of machine learning algorithms and it was found that: J48 decision tree
achieved a prediction accuracy of 98.9%, logistic regression 97.8%, k-nearest neighbour 98.3% and
lastly, neural networks obtained a 99.7% accuracy. Validation and verification was done via the J48’s
decision tree and the models were proven to be fit for use and accurate. From the data it could be
seen that future projects that would aim to use machine learning in the pre-,intra- and post-operative
care sections of anaesthesiology; that they would have to gather a large data set as to make the
models more accurate. Unlike other projects that aim to control the amount of anaesthesia or predict
the patient risk beforehand, this project has proven that it is possible to continuously predict patient’s
current health status while the operation is under way. The report concluded by stating that the neural
networks can be used as a second opinion to classify the patient’s current health status and will be
run live with current patient information.
Description:
Mini Dissertation (BEng)--University of Pretoria, 2016.