Glycaemic control (GC) in the intensive care unit is clinically contentious. Hyperglycaemia, hypoglycaemia, and glycaemic variability are increased with many GC protocols, and, critically, all associated with increased morbidity and mortality. While some studies and physiological evidence suggests GC should benefit hyperglycaemic patients, others show no or negative effects and increased incidence of hypoglycaemia. Interpretation of results is made more difficult by differences in the measurement and reporting of glycaemic control, blood glucose (BG) levels and variability in patients. In addition, target ranges for glycaemic control are not universally accepted, and higher targets are often used out of fear of hypoglycaemia, rather than their relationship to a clinical outcome.

The issues of hyperglycaemia, hypoglycaemia, and glycaemic variability may be solved with the help of continuous glucose monitoring (CGM) devices. CGM sensors have a higher rate of BG measurement, and have been effective in managing diabetes, while offering potential benefits for use in the intensive care unit (ICU). Use of CGM devices in the ICU has been limited, primarily due to the higher point accuracy errors over currently used traditional intermittent blood glucose measures. General models of CGM errors, including drift and random errors, are lacking, but would enable better design of protocols to safely and effectively utilise these devices as integrated parts of GC protocols.

This research presents an auto-regressive (AR) based modelling method that separately characterises the drift and random noise of the GlySure CGM sensor (GlySure LLC, UK). Clinical sensor data (n=33) and reference measurements were used to generate 2 AR models to describe sensor drift and noise. These models were used to generate 100 Monte-Carlo simulations based on reference blood glucose measurements. These simulated CGM traces were then compared to the original CGM clinical data using mean absolute relative difference (MARD) and a Trend Compass. The point accuracy MARD was very similar between simulated and clinical data (9.6% vs 9.9%). A Trend Compass was used to assess trend accuracy, and found simulated and clinical sensor profiles were similar (simulated trend index 11.4° vs. clinical trend index 10.9°).

The model and method accurately represents cohort sensor behaviour over patients, providing a general modelling approach to any such sensor by separately characterising each type of error that can arise in the data. Overall, it enables better protocol design using validated virtual patients based on accurate expected CGM sensor behaviour, as well as enabling the analysis of what level of each type of sensor error would be necessary to obtain desired glycaemic control safety and performance with a given protocol.

The modelling of CGM sensors may then be used for in-silico virtual trials to replace intermittent measurements in GC. This research aims to delineate the trade-offs of performance, safety and workload that CGM sensors provide in GC protocols. Clinical data from 236 patients were used for clinically validated virtual trials. A CGM-enabled version of the STAR GC protocol was used to evaluate the use of guard rails and rolling windows. Safety was assessed through percentage of patients who had a severe hypoglycaemic episode (BG < 2.22 mmol/L) as well as percentage of resampled BG < 4.0 mmol/L. Performance was assessed as percentage of resampled measurements in the 4.4-7.0 mmol/L and the 4.4-8.0 mmol/L target bands. Workload was measured by number of manual BG measures per day.

CGM-enabled versions of STAR decreased the number of required blood draws by up to 74%, while maintaining performance (76.6% BG measurements in the 4.4-7.0 mmol/L range vs. 62.8% clinically, 87.9% in the 4.4-8.0 mmol/L range vs. 83.7% clinically) and maintaining patient safety (1.13% of patients experienced a severe hypoglycaemic event vs. 0.85% clinically, 1.37% of BG measurements were less than 4.0 mmol/L vs. 0.51% clinically). CGM sensor traces were simulated in virtual trials and used in place of intermittent BG measurements to guide GC interventions and decisions. Existing GC protocols, such as STAR, may only need to be adjusted slightly to gain the benefits of the increased temporal measurements of CGM sensors, through which workload may be significantly decreased while maintaining GC performance and safety.

This research also reviews differences in the reporting of BG level and its variability in literature, which is a growing issue with the emergence of CGM sensors and other high rate sensors. There are already a multitude of differences in the reporting of BG level and variability where only intermittent measurements are concerned. The rise of new high rate sensing technology can add more temporal factors that also need to be considered, but also adds to the differences of reporting of BG and variability. The research then proposes a vision for improved description of glycaemia and how it changes and evolves over time. This work then presents a continuous glucose monitoring sensor-based method to better quantify glycaemic level and variability, based on clinically defined metrics. A case study of this new method is presented using CGM sensor data from a study of 614 infants at risk of neonatal hypoglycaemia. The CGM sensor data is used to understand glycaemia and how it evolves over time in an infant cohort. Results show the new clinically defined method is able to describe changes in glycaemic level and variability in these patients and presents a flexible way forward for accurately describing state and variability from a clinically defined perspective. This method may provide better insight to patient glycaemia over time, and thus provide scope for improved control of glycaemia. The metric is then used to assess glycaemic State Changes in the infant cohort, and attempts to relate the number of State Changes per Day to neurodevelopmental impairment data from the original study from which this infant cohort data is provided (the CHYLD Study). State Changes per Day as a metric for variability alone was found to be a weak indicator of neurodevelopmental impairment, mostly due to the complex behaviour experienced by the infants recovering from hypoglycaemia identified as having impairment.

Overall, this thesis has approached the problem of using CGM sensors to understand glycaemia and its evolution over time, and to inform model-based, personalised GC. It develops and validates a novel and accurate CGM sensor modelling method. It uses this methodology in redesigning a successful model- based GC protocol for optimised use with CGM sensors. Finally, it presents and validates a method of evaluating trends and variability in high rate CGM sensor data that results from CGM enabled GC. Thus, the thesis has developed and validated a series of solutions to CGM enabled GC.