The development and validation of risk assessment tools for non-diabetic hyperglycaemia or undiagnosed diabetes
thesisposted on 14.06.2017, 15:35 authored by Shaun Richard Barber
Risk assessment tools quantify the risk of an outcome using multiple covariates (risk factors). Risk assessment tools are recommended by diabetes prevention guidelines to allow blood tests to be targeted at individuals with an increased risk of currently having non-diabetic hyperglycaemia or undiagnosed diabetes. This thesis presents work on the identification, development and validation of such risk assessment tools. Key Findings: - A systematic review of risk assessment tools for prevalent non-diabetic hyperglycaemia was undertaken. This is the first systematic review to focus on risk assessment tools for prevalent non-diabetic hyperglycaemia. Eighteen risk assessment tools for prevalent non-diabetic hyperglycaemia, and prevalent non-diabetic hyperglycaemia or undiagnosed Type 2 diabetes are summarised. - An empirical comparison of logistic regression, decision trees, support vector machines and the novel application of chain event graphs for developing risk assessment tools found logistic regression and linear support machine vectors had the best external performance. This is the first empirical comparison for a binary medical outcome in cross-sectional data to include an external validation. - Risk groups for the Leicester Practice risk score were established, allowing consistent advice to be given across general practices when utilising the tool. - The Leicester Self-Assessment and Leicester Practice risk scores were externally validated using a nationally representative longitudinal dataset. Both gave comparable performance for identifying prevalent non-diabetic hyperglycaemia or undiagnosed diabetes to the dataset on which they were developed. Furthermore, both identified a small proportion of the population with a substantially increased risk of developing diabetes when utilised in the recommended two-stage screening programme and thus are advocated for use across England. This thesis aids those wishing to use a risk assessment tool for non-diabetic hyperglycaemia in their selection or development of an appropriate tool, as well as addressing some of the previous limitations of the Leicester Self-Assessment and Leicester Practice risk scores.