My PhD
Title:
Advanced Machine Learning to Improve Patient Care and Outcome using Real-time Hospitalisation Data
External Stakeholder:
Aneurin Bevan University Health Board
The Research:
Sepsis is characterised as a dysregulated host response to infection, which can lead to life threatening organ dysfunction and serious conditions such as septic shock. Sepsis is responsible for nearly 20% of global deaths, and is particularly prevalent in low/middle-income countries. Modern guidelines for defining sepsis utilise systems designed for ICU use, however, sepsis can develop in any hospitalised patient, therefore research outside of the ICU is particularly important. My PhD research aims to develop novel deep learning techniques for predicting early sepsis onset using ward patient data. If used in clinical practice, the aim of these models is to augment and improve the clinicians decision making process, not to replace them.
The main objective of the research is developing a sepsis prediction model using novel deep learning techniques. Furthermore, we intend to research explainable artificial intelligence (XAI) techniques to explore how the model can be fully utilised by a clinician. XAI algorithms describe why the model made a certain decision, which can build clinician and patient trust. In addition, the clinician needs to understand the context which the model works optimally in, developing an awareness of situations where the model may not perform well. Considering XAI could be beneficial in increasing adoption and sustaining use of machine learning in healthcare. While existing early warning systems, such as NEWS, explain what relevant measurements caused the patient to be identified as clinically deteriorating, machine learning techniques have an ability to capture relationships between key features, which can lead to more accurate results.