Return to search

Continuous Glucose Monitoring and Tight Glycaemic Control in Critically Ill Patients

Critically ill patients often exhibit abnormal glycaemia that can lead to severe complications and potentially death. In critically ill adults, hyperglycaemia is a common problem that has been associated with increased morbidity and mortality. In contrast, critically ill infants often suffer from hypoglycaemia, which may cause seizures and permanent brain injury. Further complicating the matter, both of these conditions are diagnosed by blood glucose (BG) measurements, often taken several hours apart, and, as a result, these conditions can remain poorly managed or go completely undetected. Emerging ‘continuous’ glucose monitoring (CGM) devices with 1-5 minute measurement intervals have the potential to resolve many issues associated with conventional intermittent BG monitoring. The objective of this research was to investigate and develop methods and models to optimise the clinical use of CGM devices in critically ill patients.

For critically ill adults, an in-silico study was conducted to quantify the potential benefits of introducing CGM devices into the intensive care unit (ICU). Mathematical models of CGM error characteristics were implemented with existing, clinically validated, models of the insulin-glucose regulatory system, to simulate the behaviour of CGM devices in critically ill patients. An alarm algorithm was also incorporated to provide a warning at the onset of predicted hypoglycaemia, allowing a virtual dextrose intervention to be administered as a preventative measure. The results of the in-silico study showed a potential reduction in nurse workload of approximately 75% and a significant reduction in hypoglycaemia, while also providing insight into the optimal rescue dose size and resulting dynamics of glucose recovery.

During 2012, ten patients were recruited into a pilot clinical trial of CGM devices in critical care with a primary goal of assessing the reliability of CGM devices in this environment, with a specific interest in the effects of CGM device type and sensor site on sensor glucose (SG) data. Results showed the mean absolute relative difference of SG data across the cohort was between 12-24% and CGM devices were capable of monitoring some patients with a high degree of accuracy. However, certain illnesses, drugs and therapies can potentially affect sensor performance, and one particular set of results suggested severe oedema may have affected sensor performance. A novel and first of its kind metric, the Trend Compass was developed and used to assesses trend accuracy of SG in a mathematically precise fashion without approximation, and, importantly, does so independent of glucose level or sensor bias, unlike any other such metrics. In this analysis, the trend accuracy between CGM devices was typically good.

A recent hypothesis suggesting that glucose complexity is associated with mortality was also investigated using the clinical CGM data. The results showed that complexity results from detrended fluctuation analysis (DFA) were influenced far more by CGM device type than patient outcome. In addition, the location of CGM sensors had no significant effect on complexity results in this data set. Thus, while this emerging analytical method has shown positive results in the literature, this analysis indicates that those results may be misleading given the impact of technology outweighing that of physiology. This particular result helps to further delineate the range of potential applications and insight that CGM devices might offer in this clinical scenario.

In critically ill infants, CGM devices were used to investigate hypoglycaemia during the first 48 hours after birth. More than 50 CGM data sets were obtained from several studies of CGM in infants at risk of hypoglycaemia at the Waikato hospital neonatal ICU (NICU). In light of concerns regarding CGM accuracy, particularly during the first few hours of monitoring and/or at low BG levels, an alternative, novel calibration scheme was developed to increase the reliability of SG data. The recalibration algorithm maximised the value of very accurate calibration BG measurements from a blood gas analyser (BGA), by forcing SG data to pass through these calibration BG measurements.

Recalibration increased all metrics of hypoglycaemia (number, duration, severity and hypoglycaemic index) as the factory CGM calibration was found to be reporting higher values at low BG levels due to its least squares calibration approach based on the assumption of a less accurate calibration glucose meter. Thus, this research defined new calibration methods to directly optimise the use of CGM devices in this clinical environment, where accurate reference BG measurements are available. Furthermore, this work showed that metrics such as duration or area under curve were far more robust to error than the typically used counted-incidence metrics, indicating how clinical assessment may have to change when using these devices.

The impact of errors in calibration measurements on metrics used to classify hypoglycaemia was also assessed. Across the cohort, measurement error, particularly measurement bias, had a larger effect on hypoglycaemia metrics than delays in entering calibration measurements. However, for patients with highly variable glycaemia, timing error can have a significantly larger impact on output SG data than measurement error. Unusual episodes of hypoglycaemia could be successfully identified using a stochastic model, based on kernel density estimation, providing another level of information to aid decision making when assessing hypoglycaemia.

Using the developed algorithms/tools, with CGM data from 161 infants, the incidence of hypoglycaemia was assessed and compared to results determined using BG measurements alone. Results from BG measurements showed that ~17% of BG measurements identified hypoglycaemia and over 80% of episodes occurred in the first day after birth. However, with concurrent BG and SG data available, the SG data consistently identified hypoglycaemia at a higher rate suggesting the BG measurements were not capturing some episodes. Duration of hypoglycaemia in SG data varied from 0-10+%, but was typically in the range 4-6%. Hypoglycaemia occurred most frequently on the first day after birth and an optimal measurement protocol for at risk infants would likely involve CGM for the first week after birth with frequent intermittent BG measurements for the first day.

Overall, CGM devices have the potential to increase the understanding of certain glycaemic abnormalities and aid in the diagnosis/treatment of other conditions in critically ill patients. This research has used a range of prospective and retrospective clinical studies to develop methods to further optimise the use of CGM devices within the critically ill clinical environment, as well as delineating where they are less useful or less robust. These latter results clearly define areas where clinical practice needs to adapt when using these devices, as well as areas where device makers could target technological improvements for best effect. Although further investigations are required before these devices are regularly implemented in day-to-day clinical practice, as an observational tool they are capable of providing useful information that is not currently available with conventional intermittent BG monitoring.

Identiferoai:union.ndltd.org:canterbury.ac.nz/oai:ir.canterbury.ac.nz:10092/8458
Date January 2013
CreatorsSignal, Matthew Kent
PublisherUniversity of Canterbury. Department of Mechanical Engineering
Source SetsUniversity of Canterbury
LanguageEnglish
Detected LanguageEnglish
TypeElectronic thesis or dissertation, Text
RightsCopyright Matthew Kent Signal, http://library.canterbury.ac.nz/thesis/etheses_copyright.shtml
RelationNZCU

Page generated in 0.0033 seconds