A personalised and adaptive insulin dosing decision support system for type 1 diabetes

• Main Author: Pesl, Peter
• Other Authors: Georgiou, Pantelis
• Published: Imperial College London 2016
• Subjects: 621.3
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.749127

People with type 1 diabetes (T1D) rely on exogenous insulin to maintain stable glucose levels. Despite the advent of diabetes technologies such as continuous glucose monitors and insulin infusion pumps, the majority of people with T1D do not manage to bring back glucose levels into a healthy target after meals. In addition to patient compliance, this is due to the complexity of the decision-making on how much insulin is required. Commercial insulin bolus calculators exist that help with the calculation of insulin for meals but these lack fine-tuning and adaptability. This thesis presents a novel insulin dosing decision support system for people with T1D that is able to provide individualised insulin dosing advice. The proposed research utilises Case-Based Reasoning (CBR), an artificial intelligence methodology, that is able to learn over time based on the behaviour of the patient and optimises the insulin therapy for various diabetes scenarios. The decision support system has been implemented into a user-friendly smartphone-based patient platform and communicates with a clinical platform for remote supervision. In-silico studies are presented demonstrating the overall performance of CBR as well as metrics used to adapt the insulin therapy. Safety and feasibility of the developed system have been assessed incrementally in clinical trials; initially during an eight-hour study in hospital settings followed by a six-week study in the home environment of the user. Human factors play an important role in the clinical adoption of technologies such as the one proposed. System usability and acceptability were evaluated during the second study phase based on feedback obtained from study participants. Results from in-silico tests show the potential of the proposed research to safely automate the process of optimising the insulin therapy for people with T1D. In the six-week study, the system demonstrated safety in maintaining glycemic control with a trend suggesting improvement in postprandial glucose outcomes. Feedback from participants showed favourable outcomes when assessing device satisfaction and usability. A six-month large-scale randomised controlled study to evaluate the efficacy of the system is currently ongoing.