Patient-Specific Metabolic Variability and Precision Glycaemic Control in Critical Care

Critically ill patients often experience stress-induced hyperglycaemia. Elevated blood glucose levels are associated with increased morbidity and mortality. Glycaemic control demonstrated improved outcomes for these patients. However, other studies failed to replicate the results, primarily blaming the increased risk of hypoglycaemia and glycaemic variability, both associated with worse outcomes. These confounding outcomes have resulted in acceptance of hyperglycaemia and reduced outcomes, causing ongoing debate on glycaemic control.
The goal of the thesis is to define what makes glycaemic control hard to achieve safely, prove safe, effective control impacts patient outcome, and demonstrate it is possible to achieve safe, effective control for all patients, despite targeting lower glycaemic ranges.
Metabolic variability is the main factor making glycaemic control hard to achieve safely. More specifically, sudden changes in patient-specific response to insulin (intra-patient variability) can lead to severe hyper- and hypo- glycaemia. Novel analysis of model-based insulin sensitivity and its variability clearly showed while inter-patient variability can be significantly different across patients, intra-patient variability is equivalent. Therefore, no patient is harder nor easier to control, and thus all patients should be able to benefit from similar quality of control. In turn, conclusions on glycaemic control from studies failing to do so may be biased due to poor protocol design, rather than physiological factors related to severity and outcome.
Intra-patient variability is still very large, and it is not possible to discriminate more and less variable patients, reducing the quality of control deliverable in practical clinical scenarios. This research developed a novel 3D stochastic model to optimally segregate more and less variable patients based on prior behaviours. This approach enabled significantly improved, and tighter prediction of risks associated with a given insulin and/or nutrition intervention. Clinical trial results in NZ have shown improved control and safety using this new 3D stochastic model.
To demonstrate these outcomes, a clinical trial using STAR, a model-based, patient-specific glycaemic control framework, was designed and implemented at the University Hospital of Liège. Results showed STAR succeeded in providing safe, effective control to virtually all patients, despite targeting lower target bands associated with better outcomes. However, increased workload compared to the standard protocol was identified as a limitation.
Finally, this thesis develops a means to dramatically increase the STAR measurement interval from 1-3 hourly to 1-6 hourly without significantly degrading performance or safety. Virtual trials clearly defined the risk and reward trade-off between control performance, patient safety, workload, and nutrition. This result allows clinical staff to choose from a far wider range of options and approaches to provide safe, effective control, with clearly defined risk trade-offs.
Overall, a series of analyses and clinical trials have shown safe, effective control is necessary to improve outcomes, and can be achieved for all patients. These outcomes are possible using patient-specific, model-based glycaemic control protocols developed in this thesis, which directly account for both intra- and inter- patient variability and reduce workload.