Diabetes is a world-wide problem with epidemically increasing prevalence. By 2040, more than 10% of adults world-wide will have diabetes and it is estimated to cost US$2.1 trillion per year by 2030. Type 2 diabetes mellitus (T2DM) is a complex condition causing high blood sugar levels, or hyperglycaemia, and is the result of dysregulation of insulin and glucose dynamics in the body. Uncontrolled hyperglycaemia damages tissues, leading over longer terms to an increasing range of complications, organ failure, and early death.

The Intensive Control Insulin-Nutrition-Glucose (ICING) model describes glucose-insulin dynamics in the human body and can be used to model dysregulated metabolic dynamics and conditions, including diabetes. The use of modelling can increase physiological understanding of any physiological dysfunction, including diabetes, and, in turn, help to determine, personalise, and optimise potential treatment options. This thesis analyses and optimises key model elements and dynamics related to diabetes control and outcomes in the ICING model, and then uses the model to evaluate changes in people with Type 2 Diabetes Mellitus (T2DM) after an intensive lifestyle intervention, as a final validation of the overall research.

The first model element considered was endogenous glucose production (EGP), a physiological important contribution to overall glycaemia, which is difficult to model and measure. In an analysis of intensive care patients, model-based insulin sensitivity (SI) was constrained in 2.3% of patient hours on model-based glycaemic control. This constrained SI value most likely represents model error in the assumed EGP parameter value in these cases, as most of the constraints occurred around low or no nutritional input periods. It also reflects the stress response, as the majority of constrained SI values occurred in the first 12 hours of insulin therapy, when the patient is most acute. Minimum EGP values calculated in the case of constrained SI varied from 1.16 to 2.72 mmol/min, which are within reported ranges for this cohort and thus physiologically realistic, further validating the model and analysis. Overall, the current population value of 1.16 mmol/min used in the ICING model is justified given 2.3% of hours are constrained, representing a small subset of patients and hours physiology not well-captured in the current model. These results translate well towards diabetes patients as both ICU and diabetes patients suffer from similarly dysregulated glycaemic control within the body, but at different levels. Thus, in both cases, EGP is poorly restrained when it is unnecessary, and thus models using “typical” values may cause significant inaccuracy.

EGP has typically been estimated using isotope tracers, arteriovenous difference, and magnetic resonance spectroscopy, all of which can have significant errors, are not continuous, and are clinically intensive. Using literature and study data, blood lactate concentration curves were generated as a function of heart rate (HR) during intense exercise and during recovery. EGP was then estimated from literature using HR and blood lactate concentration during exercise. With the generated phenomenological model of lactate and HR, EGP could then be estimated using only a HR measurement during one bout of exercise and recovery. This method provides a novel and non- invasive way of estimating EGP during intense exercise and recovery, and helps translate the model to a level necessary for outpatient management and consideration of exercise-based lifestyle interventions.

Insulin sensitivity is a parameter highly subject to physiological stress response in the body. This value was identified for recreational athletes who participated in a ramped exercise protocol to exhaustion, which also creates a significant physiological stress response similar in milieu to the consistent stress response seen in diabetes. Two oral glucose boluses were consumed during exercise. Insulin sensitivity increased during exercise, and then decreased during the recovery period, unlike other lower intensity exercise literature studies. The changes broadly matched

expectations, but are limited in resolution because of the several varied pathways by which EGP can increase or decrease, including ingestion of a glucose bolus, all of which impacts the identified insulin sensitivity value. Strenuous and intense exercise can cause hyperinsulinemia and hyperglycaemia when a glucose bolus is consumed at the end of intense exercise with significantly reduced insulin sensitivity, in effect, a type of temporary diabetes. This effect is also similar to the stress response that occurs during critical illness.

Insulin secretion is a further, difficult to measure parameter directly impacting glycaemic outcomes. Parameterization from the well-known and widely used Van Cauter model was used to calculate estimated insulin secretion from C-peptide measurements and a Monte Carlo analysis was used to determine the expected range of estimated insulin secretion. It was shown plasma volume (Vp) has the largest effect on modelled insulin secretion estimates and subjects with a body mass index (BMI) > 30 have larger variation in insulin secretion estimates than subjects with BMI < 30. Overall, the modelled insulin secretion estimates can vary by approximately +/- 15%. In particular, subjects with a BMI > 30, where the interquartile range of 34.3% compared to subjects with a BMI < 30 with 24.7%, a significant change in variability. This is the first time this variation has been quantified, and it should be considered when applying this insulin secretion estimation method to clinical diagnostic thresholds and interpretation of model-based analyses, such as insulin sensitivity.

Glucose clearance can vary via insulin-mediated and non-insulin mediated pathways, and these effects are not typically able to be separated in human studies. A modified Dynamic Insulin Secretion and Sensitivity Test (DISST) protocol with 7 samples over 60 minutes was created to identify non- insulin mediated glucose uptake (NIMGU) and SI in people with impaired glucose tolerance or T2DM before and after an intervention. Changes can be identified as small as 1x10-4 L/mU/min for SI and 0.002 min-1 for NIMGU. The results also quantify more precisely the trade-off of these parameters and the likelihood of different values along this trade-off line. As a result, the estimated contributions of each can be delineated within these ranges, providing further, heretofore unavailable insight.

Remission of T2DM is possible and has been extensively reported in literature studies, often with limited patient numbers. The Diabetes Reset through Intermittent Fasting Trial (DRIFT) study was conducted to determine exactly which metabolic changes occur during fasting in people with insulin dependent T2DM. In particular, an alternate day fasting protocol was employed as a short-term, intensive lifestyle intervention, and changes in insulin sensitivity and insulin secretion were assessed using the model-based DISST and the prior elements of this research. To date, two participants completed the DRIFT fasting protocol. Both showed a significant increase in endogenous insulin secretion and were able to significantly reduce or stop exogenous insulin use. More participants are needed in this ongoing clinical study to determine if the results are robust. However, these first results indicate how fasting actually impacts individuals with diabetes, and can cause remission. While the study protocol is not sustainable for long-term use, it shows the impact of fasting and further studies may thus follow onward towards a sustainable approach.

The prevalence of T2DM continues to increase. Diabetes medications help to reduce the risk of complications from high blood glucose concentration, but have side effects, significant cost, and sometimes poor adherence. Thus, lifestyle changes still need to be made to reverse progression of the disease. The use of fasting as an intervention can be a safe and effective option for many people with T2DM, and model-based methods developed in this thesis can provide the scientific and computational tools to assess, guide, personalise, and optimise these interventions going forward.