Search

The burden of chronic disease is growing at a fast pace, leading to poor quality of life and high healthcare expenditures in a large portion of the Australian population. Much of the burden is borne by hospitals, and therefore there is an ever-increasing interest in preventative interventions that can keep people out of hospitals and healthier for longer periods. There is a wide range of potential interventions that may be able to achieve this goal, and policy makers need to decide which one should be funded and implemented. This task is difficult for two reasons: first it is often not clear what is the short-term effectiveness of an intervention, and how it varies in specific sub-populations, and second it is also not clear what the long-term intended and unintended consequences might be. In this thesis I make contributions to address both these difficulties. On the short-term side I focus on the use of physical activity to prevent the development of chronic disease and to reduce hospital costs. Increasing physical activity has been long heralded as a way to achieve these goals but evidence of its effectiveness has been elusive. In this thesis I provide data driven evidence to justify policies that encourage higher levels of physical activity (PA) in middle age and older Australian population. I use data from the “45 and up” and the Social, Economic and Environmental Factors (SEEF) study, linked with the Admitted Patient Data Collection (APDC), to identify and study the cost and health trajectories of individuals with different levels of physical activity. The results show a clear statistically significant association between PA and lower hospitalisation cost, as well as between PA and reduced risk of heart disease, diabetes and stroke. On the long-term side of the analysis, I placed this thesis in the context of a larger program of work performed at Western Sydney University that aims to build a microsimulation model for the analysis of health policy interventions. In this framework I studied predictive models that use survey and/or administrative data to predict hospital costs and resource utilisation. I placed particular emphasis on the application of methods borrowed from Natural Language Processing to understand how to use the thousands of diagnosis and procedure codes found in administrative data as input to predictive models. The methods developed in this thesis go beyond the application to hospital data and can be used in any predictive model that relies on complex coding of healthcare information.