City Research Online

Abstract

The Diabetes Control and Complications Trial (DCCT) and the United Kingdom Prospective Diabetes Study provide motivation for the intensive insulin therapy (IIT) in type 1 and also type 2 diabetes mellitus. Decision support systems (DSS) for the adjustment of insulin dosage in insulin-treated subjects with type 2 diabetes have the potential to attain significant importance in clinical practice by offering the means to improve glycaemic control.

The aims of the research were to extend the existing Diabetes Insulin Advisory System (DIAS) model of carbohydrate metabolism with a component representing the insulin secretion present in subjects with type 2 diabetes and to develop a decision support system, DIAS-NIDDM, to assist in clinical practice with the adjustment of insulin dosage in insulin-treated subjects with type 2 diabetes. In relation to assessing the long-term complications of diabetes, the aim was to incorporate a model that predicts steady state HbA)c concentrations in response to changes in diet and insulin therapy. Furthermore, the research evaluated DIAS-NIDDM with regard to performance, clinical utility, and safety of its advice on retrospective data to justify prospective clinical testing of DIAS-NIDDM.

The new model, implemented using causal probabilistic networks (CPN), represents the insulin secretion present in subjects with type 2 diabetes and assumes a linear relationship between plasma glucose concentration and endogenous insulin secretion.

DIAS-NIDDM was used to predict patient-specific plasma glucose profiles and advise on insulin doses during a pilot study in eight subjects with type 2 diabetes, with five subjects treated by insulin. Case studies showed that the advice is plausible and safe. However, a systematic error has been identified when the system predicted BG values in the hyperglycaemic BG range. Another evaluation step was performed as a pilot peer assessment of the insulin dose advice generated by DIAS-NIDDM. The peer blind assessment provided valuable information about the competence of DIAS-NIDDM compared to diabetes specialists, confirming the clinical utility and the safety of the advice. However, in the peer review study, DIAS-NIDDM recommendations performed less well than advice from a clinical diabetologist, as the system does not use the subject demoghraphic data to assess the feasibility of the recommended therapy in aged and insulin resistant subjects.

A new model using stochastic differential equations has been built to describe the relationship between DIAS-NIDDM predicted plasma glucose levels and glycated haemoglobin (HbAic) at steady state conditions. The model uses two physiological compartments to represent the glycated and unglycated haemoglobin where the glycation process is controlled by the plasma glucose concentration. The HbAic assay has been modelled, in a novel approach, as a separate process in order to deal with the glycation-induced heterogeneity and the mix of cell life spans in the sample. A retrospective pilot study has been performed in order to carry out a preliminary validation of the glycation model. The accuracy of the model was excellent when predicting thirty-two HbA)C retrospective measurements.

In conclusion, the results confirm that DIAS-NIDDM can generate advice that is similar in performance to the advice recommended by diabetes specialists and that the advice is safe, plausible, and of clinical utility. The system can predict steady state HbAic in response to changes in diet and insulin therapy. Despite possible further optimisations of the system, prospective clinical testing of DIAS-NIDDM is justified. The thesis, nevertheless, identifies DIAS-NIDDM as a decision support system with valuable potential.

Publication Type:	Thesis (Doctoral)
Subjects:	R Medicine > RZ Other systems of medicine
Departments:	School of Health & Psychological Sciences School of Health & Psychological Sciences > School of Health & Psychological Sciences Doctoral Theses Doctoral Theses

Export

Downloads