Model-Based Therapeutics for Type 1 Diabetes Mellitus

The incidence of Type 1 diabetes is growing yearly. Worryingly, the aetiology of the disease is inconclusive. What is known is that the total number of affected individuals, as well as the severity and number of associated complications are growing for this chronic disease. With increasing complications due to severity, length of exposure, and poor control, the disease is beginning to consume an increasingly major portion of healthcare costs to the extent that it poses major economic risks in several nations. Research has shown that intensive insulin therapy aimed at certain minimum glycosylated haemoglobin threshold levels reduces the incidence of complications by up to 76% compared to conventional insulin therapy. Moreover, the effects of such intensive therapy regimes over a 6.5y duration persists for at least 10y after, a so called metabolic memory. Thus, early intervention can slow the momentum of complications far more easily than later intervention. Early, safe, intensive therapy protocols offer potential solutions to the growing social and economic effects of diabetes. Since the 1970s, the artificial endocrine pancreas has been heralded as just this type of solution. However, no commercial product currently exists, and ongoing limitations in sensors and pumps have resulted in, at best, modest clinical advantages over conventional methods of insulin administration or multiple daily injection. With high upfront costs, high costs of consumables, significant complexity, and the extensive infrastructure and support required, these systems and devices are only used by 2-15% of individuals with Type 1 diabetes. Clearly, there is an urgent need to address the large majority of the Type 1 diabetes population using conventional glucose measurement and insulin administration. For these individuals, current conventional or intensive therapies are failing to deliver recommended levels of glycaemic control. This research develops an understanding of clinical glycaemic control using conventional insulin administration and glucose measurement techniques in Type 1 diabetes based on a clinically validated in silico virtual patient simulation. Based on this understanding, a control protocol for Type 1 diabetes that is relatively simple and clinically practical is developed. The protocol design incorporates physiological modelling and engineering techniques to adapt to individual patient clinical requirements. By doing so, it produces accurate, patient-specific recommendations for insulin interventions. Initially, a simple, physiological compartmental model for the pharmacokinetics of subcutaneously injected insulin is developed. While the absorption process itself is subject to significant potential variability, such models enable a real-time estimation of plasma insulin concentration. This information would otherwise be lacking in the clinical environment of outpatient Type 1 diabetes treatment due to the inconvenience, cost, and laboratory turnaround for plasma insulin measurements. Hence, this validated model offers significant opportunity to optimise therapy selection. An in silico virtual patient simulation tool is also developed. A virtual patient cohort is developed on patient data from a representative cohort of the broad diabetes population. The simulation tool is used to develop a robust, adaptive protocol for prandial insulin dosing against a conventional intensive insulin therapy, as well as a controls group representative of the general diabetes population. The effect on glycaemic control of suboptimal and optimal, prandial and basal insulin therapies is also investigated, with results matching clinical expectations. To gauge the robustness of the developed adaptive protocol, a Monte Carlo analysis is performed, incorporating realistic and physiological errors and variability. Due to the relatively infrequent glucose measurement in outpatient Type 1 diabetes, a method for identifying the diurnal cycle in effective insulin sensitivity and modelling it in retrospective patient data is also presented. The method consists of identifying deterministic and stochastic components in the patient effective insulin sensitivity profile. Circadian rhythmicity and sleep-wake phases have profound effects on effective insulin sensitivity. Identification and prediction of this rhythm is of utmost clinical relevance, with the potential for safer and more effective glycaemic control, with less frequent measurement. It is thus a means of further enhancing any robust protocol and making it more clinically practical to implement. Finally, this research presents an entire framework for the realistic, and rapid development and testing of clinical glycaemic control protocols for outpatient Type 1 diabetes. The models and methods developed within this framework allow rapid and physiological identification of time-variant, patient-specific, effective insulin sensitivity profiles. These profiles form the responses of the virtual patient and can be used to develop and robustly test clinical glycaemic control protocols in a broad range of patients. These effective insulin sensitivity profiles are also rich in dynamics, specifically those circadian in nature which can be identified, and used to provide more accurate glycaemic prediction with the potential for safer and more effective control