Applied machine learning and the role of dynamic time-series data in optimising clinical management of acute febrile illnesses
File(s)
Author(s)
Ming, Damien Keng Yen
Type
Thesis
Abstract
Background
Accounting for time-series dependencies inherent in patient healthcare data is important but optimal approaches are not well understood. This thesis investigates the role, and implementation of machine learning models which utilise dynamic data to improve clinical decision-making in acute febrile illnesses.
Methods
Mixed methods involving data science, qualitative and bioengineering techniques were used. Real-world exemplars including datasets of bacterial bloodstream infections (n=4,198), of dengue (n=4,131) and undifferentiated fever (n=8,100) were analysed using data and machine learning approaches. Clinical implementation of dengue models developed in the thesis was investigated through use of process mapping and thematic analysis. The issues of enhancing patient data acquisition to support future model development were examined through minimally-invasive biosensing techniques.
Results
Machine learning models which account for time-series dependencies provided better predictive ability compared with time-invariant versions in a range of acute febrile illnesses. Applying such models to healthcare settings required understanding of decision-making processes and patient pathways. Continuous biosensing methods could offer an effective and acceptable approach to enhancing the patient healthcare data collection across different settings.
Conclusion
The thesis demonstrates the importance of time-series data in supporting clinical decision-making in acute febrile illnesses. Use of novel machine learning methods to enhance existing and routinely collected healthcare information can improve care and is likely to be cost-effective. It is crucial that factors relating to healthcare implementation and data acquisition are addressed to ensure research findings result in clinical impact.
Accounting for time-series dependencies inherent in patient healthcare data is important but optimal approaches are not well understood. This thesis investigates the role, and implementation of machine learning models which utilise dynamic data to improve clinical decision-making in acute febrile illnesses.
Methods
Mixed methods involving data science, qualitative and bioengineering techniques were used. Real-world exemplars including datasets of bacterial bloodstream infections (n=4,198), of dengue (n=4,131) and undifferentiated fever (n=8,100) were analysed using data and machine learning approaches. Clinical implementation of dengue models developed in the thesis was investigated through use of process mapping and thematic analysis. The issues of enhancing patient data acquisition to support future model development were examined through minimally-invasive biosensing techniques.
Results
Machine learning models which account for time-series dependencies provided better predictive ability compared with time-invariant versions in a range of acute febrile illnesses. Applying such models to healthcare settings required understanding of decision-making processes and patient pathways. Continuous biosensing methods could offer an effective and acceptable approach to enhancing the patient healthcare data collection across different settings.
Conclusion
The thesis demonstrates the importance of time-series data in supporting clinical decision-making in acute febrile illnesses. Use of novel machine learning methods to enhance existing and routinely collected healthcare information can improve care and is likely to be cost-effective. It is crucial that factors relating to healthcare implementation and data acquisition are addressed to ensure research findings result in clinical impact.
Version
Open Access
Date Issued
2023-03
Date Awarded
2023-09
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Holmes, Alison
Yacoub, Sophie
Georgiou, Pantelis
Publisher Department
Department of Medicine
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)