Predicting the Risk of Chronic Disease: A Framework Based on Graph Theory and Social Network Analysis
Year: 2016
Type: dissertation
Abstract: Chronic diseases progress slowly and the patients are often first diagnosed with the disease when admitted to the hospital, resulting in further complications and an increased burden on limited healthcare resources. This thesis focuses on to understand the chronic disease progression and then predict the risk of the disease, subsequently preventing hospital admissions. To do that, the thesis presents a framework that has two major parts: (i) to understand and represent the progression of a particular chronic disease, and (ii) to develop a model based on that understanding to predict chronic disease risk for non-chronic patients. The framework utilises graph theory and social network analysis for both of these parts. It uses hospital admission data which carries rich semantic information about the patients' overall health status and diagnosis. For the first part, it proposes the concept of 'baseline network' that can effectively model chronic disease comorbidities and their transition pattern, thereby representing the disease progression. For the second part, 'baseline network' is used to compare against the individual health trajectories of non-chronic patients. Graph theory and network-based methods are used to analyse parameters such as the prevalence of comorbidities, transition patterns and frequencies, disease cluster etc. Demographic and behavioural risk factor such as age, sex and smoking are also included in the prediction model. We implemented the overall framework on an administrative data drawn from the Australian healthcare context and chose type 2 diabetes as the chronic disease to predict. The dataset contained approximately 1.4 million hospital admission records from .75 million patients. The overall prediction accuracy was significantly higher than the trivial methods. This proves the effectiveness of the framework as well as network-based methods in chronic disease prediction which can be potential for different healthcare-related stakeholders.
Author: Mohammad Ariful Alam Khan
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2604018678