Statistical Modelling and Analysis of Structured Data
Student: Nick Tierney
When: Wednesday, 12th April 2017 2:00 PM- 3:00 PM
Where: GP-C Block, Level 4, Room 405
- D/Prof Kerrie Mengersen (Principal)
- Prof Antonietta Mira
- Prof Angelo Aurichhio
- D/Prof Kerrie Mengersen (Chair)
- Dr Paul Wu
- Prof Adrian Barnett
The overall aim of this thesis is to develop approaches for modelling and analysis of structured data, where the data are partially or fully observed. We consider two case studies in the health domain: onsite medical checkups of employees, and spatial instances of cardiac arrest in urban and rural settings. The first case study data contain multiple and often different measurements of an individual’s health with varying numbers of visits and duration between visits, and where individuals are grouped according to workplace exposure groups. The second case study data contain GPS co-ordinates of out-of-hospital cardiac arrest (OHCA) events and current and potential locations of automated external defibrillators (AEDs).
These datasets possess a number of challenging features for analysis. Four canonical features pertaining to these case study examples described above are missing data, irregular time periods in a workplace structure, optimum allocation of facility resources, and measuring and evaluating geospatial accessibility and resource allocation. These structures for the motivation for the research presenting in this thesis. The overal aim of the research is described as four objectives.
The first objective is to develop approaches to eliciting missing data structure. This is achieved by extending decision tree methods. The second objective is to identify group and individual risks profiles, and individual health trajectories. This is achieved by developing and applying Bayesian hierarchical model that accommodates the complex characteristics of the data. Models developed in objective one and two will be used to assist in the development of Occupational Health Risk (OHR) profiles of individual employees over time.
The third objective is to develop methods to optimally locate and relocate facilities to maximise coverage. This is achieved through the development of a maximal covering location problem formulation, and extending this methodology to incorporate relocation of facilities. The fourth objective is to measure health resource demand and access, and the impact on access as resources are allocated. This is achieved through developing Bayesian geospatial models of OHCA risk, measuring facility access with the floating catchment methodology, and measuring subsequent access change as models from objective three are applied. Together the models developed in objectives three and four will be used to help obtain ideal AED locations to to maximise coverage of historical OHCA events. Additionally, these objectives will inform spatial risk of OHCA events, and evaluate the impact of AED placement on spatial accessibility.
Overall this thesis makes two substantive contributions. The first is with respect to statistical methodology and computation resources, by extending existing models to account for data structure and providing corresponding new software tools. The second contribution is to the field of occupational Health Surveillance and health resource usage, in the hope of improving the health of workers and citizens.