Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.772534
Title: Bayesian Gaussian processes for identifying the deteriorating patient
Author: Colopy, Glen Wright
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
Patients discharged from the ICU will commonly be placed in intermediary care, such as the step-down ward, where the nurse-to-patient ratio is reduced (compared to that of the ICU). Although most of these patients will continue to recover and stabilise, a significant portion will suffer cardiac arrest and/or other clinical emergencies, and readmission into intensive care. Upon readmission, the risk of mortality is significantly higher than that of the general ICU population. Evidence suggests that early detection of deterioration may prevent or alleviate the severity of clinical emergencies. Notable shortcomings of current practices are that they (i) involve manual calculation of risk scores, (ii) depend on heuristic decision criteria, (iii) ignore time-series dynamics of physiological measurements, and (iv) lack patient-specificity. Gaussian process regression (GPR) models are proposed as a principled, probabilistic method to address the clinical need to continuously monitor patient vital-sign time-series with the flexibility to address the aforementioned weaknesses of current methods. The proposed GPR models focus on the robust forecasting of patient vital-sign time-series and early detection of patient deterioration. The primary contributions of this thesis describe how: 1. Probabilistic models may be used to identify artefactual measurements from continuously-acquired vital-sign monitoring devices. 2. GP covariance functions may be constructed and regularised for robust modelling, suitable for both patient-cohorts and personalised care. 3. GPR-based methods may quantify erratic physiological time-series and provide useful advanced warning of deterioration events. Each of the above contributions use the time-series correlation of vital-sign measurements for advantageous clinical inference.
Supervisor: Not available Sponsor: Clarendon Fund
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.772534  DOI: Not available
Keywords: Precision Medicine ; Bayesian Statistics ; Personalised Medicine ; Machine learning ; Gaussian Processes ; Timeseries ; Healthcare ; Medical Statistics ; Artificial Intelligence ; Bayesian Timeseries
Share: