Use this URL to cite or link to this record in EThOS:
Title: Identifying vital sign abnormality in acutely-ill patients
Author: Wong, D. C.
ISNI:       0000 0004 2745 2664
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
The Emergency Department (ED) provides the first line of care for anyone seeking treatment for an urgent problem caused by an accident or illness . Physiological observations in the ED are a required part of patient care, and are used to monitor a patient's condition. Manual observations are recorded regularly by nursing staff, using a Track and Trigger (T&T) system, in which higher scores indicate greater physiological abnormality. An observational study at the John Radcliffe Hospital, Oxford, was conducted to assess the effectiveness of T&T in the ED. Retrospective analysis showed that the effectivenessof T&T was limited by poor completion, and incorrect calculation of T&T scores. In response, we computed a retrospective, fully completed, scoring system which showedvery clear improvements in both sensitivity and specificity. In addition to nurse observations, higher acuity ED patients have their vital signs continuously monitored by bedside monitors. However, the alerts generated by the monitors are routinely ignored due to their high false alert rate. We investigated whether a baseline data fusion model and two alternative techniques, weighted Parzen windows and Support Vector Machines, could identify events relating to vital sign abnormality while keeping the number of false alerts to a minimum. The performance of each model was assessed by calculating its sensitivity and specificity. However, it was not possible to select an optimal model, due to the difficulty in assessing the relative importance of maximising true alertsand minimising false alerts. In the final part of this thesis, two limitations of the data fusion models are highlighted. Firstly, missing data is not handled coherently within the current models, and secondly the models do not make use of temporal information. One method of addressing both of these issues, Gaussian processes, was considered. Using this method, a novel framework was derived that allowed for alerts to be generated even when there is uncertainty in the vital sign values.
Supervisor: Tarassenko, L. Sponsor: Engineering and Physical Sciences Research Council
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Biomedical engineering ; Medical Engineering ; Data Fusion ; Machine Learning ; Patient ; Monitoring