Use this URL to cite or link to this record in EThOS:
Title: Developing clinical prediction models for diabetes classification and progression
Author: Lynam, A.
ISNI:       0000 0004 8510 7420
Awarding Body: University of Exeter
Current Institution: University of Exeter
Date of Award: 2020
Availability of Full Text:
Access from EThOS:
Access from Institution:
Patients with type 1 and type 2 diabetes have very different treatment and care requirements. Overlapping phenotypes and lack of clear classification guidelines make it difficult for clinicians to differentiate between type 1 and type 2 diabetes at diagnosis. The rate of glycaemic deterioration is highly variable in patients with type 2 diabetes but there is no single test to accurately identify which patients will progress rapidly to requiring insulin therapy. Incorrect treatment and care decisions in diabetes can have life-threatening consequences. The aim of this thesis is to develop clinical prediction models that can be incorporated into routine clinical practice to assist clinicians with the classification and care of patient diagnosed with diabetes. We addressed the problem first by integrating features previously associated with classification of type 1 and type 2 diabetes to develop a diagnostic model using logistic regression to identify, at diagnosis, patients with type 1 diabetes. The high performance achieved by this model was comparable to that of machine learning algorithms. In patients diagnosed with type 2 diabetes, we found that patients who were GADA positive and had genetic susceptibility to type 1 diabetes progressed more rapidly to requiring insulin therapy. We built upon this finding to develop a prognostic model integrating predictive features of glycaemic deterioration to predict early insulin requirement in adults diagnosed with type 2 diabetes. The three main findings of this thesis have the potential to change the way that patients with diabetes are managed in clinical practice. Use of the diagnostic model developed to identify patients with type 1 diabetes has the potential to reduce misclassification. Classifying patients according to the model has the benefit of being more akin to the treatment needs of the patient rather than the aetiopathological definitions used in current clinical guidelines. The design of the model lends itself to implementing a triage-based approach to diabetes subtype diagnosis. Our second main finding alters the clinical implications of a positive GADA test in patients diagnosed with type 2 diabetes. For identifying patients likely to progress rapidly to insulin, genetic testing is only beneficial in patients who test positive for GADA. In clinical practice, a two-step screening process could be implemented - only patients who test positive for GADA in the first step would go on for genetic testing. The prognostic model can be used in clinical practice to predict a patient’s rate of glycaemic deterioration leading to a requirement for insulin. The availability of this data will enable clinical practices to more effectively manage their patient lists, prioritising more intensive follow up for those patients who are at high risk of rapid progression. Patients are likely to benefit from tailored treatment. Another key clinical use of the prognostic model is the identification of patients who would benefit most from GADA testing saving both inconvenience to the patient and a cost-benefit to the health service.
Supervisor: Jones, A. ; Shields, B. ; Hattersley, A. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Diabetes ; Logistic regression ; Survival analysis ; Royston-Palmer flexible parametric ; Prediction model ; R-shiny ; Machine Learning