Continuous Measurement of Blood Glucose: Validation of a New Intravascular Sensor

• The value of tight glucose control by intensive insulin therapy in the perioperative period and in the intensive care unit is uncertain, perhaps because blood glucose concentrations fluctuate rapidly in response to treatment of intermittently monitored concentrations.

• A novel intravascular continuous glucose monitor performed with low mean bias and low 95% limits of agreement throughout clinically relevant blood glucose concentration ranges in a pig model with response times that may be sufficient for real time glucose control.

THE concept of tight blood glucose control has attracted great attention in perioperative and intensive care medicine. However, studies have shown contradictory effects on patient outcomes and have been inconclusive regarding appropriate blood glucose targets.1–5Contradictory results may be the result of insufficient monitoring of hypoglycemic events,6–8time outside blood glucose target ranges,9,10and fluctuations in blood glucose concentrations.11–13Several researchers have suggested the need for better measurement of the prime variable, blood glucose.14,15 

The standard methods for blood glucose monitoring are repetitive arterial, capillary, and venous blood samples analyzed with blood gas machines and point-of-care or laboratory analyzers. Laboratory analyzers use different techniques, commonly glucose oxidase (blood gas machines and point-of-care analyzers) and hexokinase (central laboratory analyzers) technologies.16Results based on analyses from blood gas machines are recommended as they are in agreement with laboratory analyzers.17There is, however, no consensus regarding how frequently glucose should be measured. Some studies report taking blood glucose measurements at intervals ranging from 1 to 4 h, whereas others do not specify measurement frequency.1–5 

Regardless of the type of sampling procedure and analyzer, repetitive measuring does not yield continuous glucose concentration readings. Because the optimal measurement interval is not known—and to minimize labor and discomfort associated with repetitive measuring—there has been growing interest in the use of continuous glucose monitors (CGM). Most commonly used CGMs are subcutaneous sensors, which measure interstitial glucose by the glucose-oxidase method. Other interstitial techniques, such as microdialysis, reverse iontophoresis, and impedance spectroscopy, have been used with limited clinical success. There are technical challenges with all CGMs, such as time lag between changed glucose values in the blood versus  interstitial fluid, fibrin clotting, local inflammation, and the regular need for CGM recalibration.18Although primarily intended for use in type I diabetes mellitus patients, subcutaneous CGMs have also been validated in intensive care settings.17,19,20One major problem is the rather unpredictable subcutaneous milieu of intensive care patients, caused by microcirculatory disturbances, fluid imbalance, and vasopressor use, complicating the time lag problem21and insulin dosing.22 

Theoretically, intravascular positioning of CGMs could yield instantaneous and accurate glucose readings. Modified subcutaneous CGMs (oxidase technology) have been used with promising results,23but commercial products have not yet reached the market. A different approach would be a system that automatically withdraws blood and analyzes blood glucose semicontinuously.24However, all intravascular devices are prone to clot formation—which inevitably impacts sensor performance and could lead to thromboembolism. Yet, many intensive care patients receive antithrombolic prophylaxis, which theoretically could reduce clotting.

This study describes the validation of a new intravascular CGM in an experimental pig model with hypoglycemic and hyperglycemic events. The sensor uses a new technological approach based on the shrinking and swelling of a boronic acid–incorporated hydrogel when exposed to various glucose concentrations.

The study was approved by the Norwegian State Commission for Animal Experimentation (Oslo). Seven pigs (22–28 kg) were acclimatized and treated in accordance with the European Convention for the Protection of Vertebrate Animals used for Experimental and Other Scientific Purposes.∥They were premedicated with 10 mg diazepam (Stesolid; Dumex-Alpharma, Copenhagen, Denmark) and 400 mg azaperone (Stresnil; Janssen-Cilag Pharma, Vienna, Austria) intramuscularly. Anesthesia was induced with intravenous 1 mg atropine (Nycomed Pharma AS, Oslo, Norway), 10 mg/kg ketamine hydrochloride (Parke-Davis, Solna, Sweden), and 5 mg/kg sodium thiopental (Pentothal Natrium; Abbott Scandinavia AB, Solna, Sweden). Animals received tracheotomy or were intubated and mechanically ventilated (Dameca, Copenhagen, Denmark). A fraction of inspired oxygen was kept at 0.3; tidal volume, 10 ml/kg; and minute ventilation was adjusted to maintain Paco²at 34–42 mmHg (4.5–5.5 kPa). Anesthesia was maintained by isoflurane, 0.5–1.0% (Forene; Abbott Scandinavia AB), and 100-μg bolus intravenous fentanyl (Pharmalink, Spnga, Sweden). Fluid balance was achieved using a continuous 10-ml · kg⁻¹· h⁻¹infusion of warmed (37°C) acetated Ringer's solution. Heparin (2,500 units) either in repetitive boluses or as a bolus followed by 1,500 units/h (Leo Orifarm, Oslo, Norway) was given throughout the experiment to prevent clot formation. Heart rate and oxygen saturation were monitored by pulse oximetry. Two animals were awakened after the procedures and were used for 3 consecutive days. They rested for 18 h between experiments.

Through surgical cut-down, a 20-cm triple lumen central venous catheter (Certofix-Trio S715; Braun, Melsungen, Germany) was introduced at the right internal jugular vein. The tip of the catheter was positioned in the superior caval vein under fluoroscopic guidance. The CGM (Invivosense Continuous Glucose Monitor [IvSCGM]; Invivosense ASA, Trondheim, Norway) was introduced through the distal lumen and secured with a Luer lock. A retractable sheet tube protected the CGM during insertion. When it was correctly positioned 5 mm distal to the tip of the catheter, the protecting sheet tube was retracted, exposing the sensor. An arterial line was placed either in the carotid or femoral artery for blood samples. A few glucose sensors were positioned in one of the carotid arteries. The carotid artery was reached from the tracheotomy incision, and a central venous catheter was advanced into the artery, although not more than a few centimeters to avoid ventricle irritation. The CGM was then inserted as described.

After the end of surgery, animals were stabilized for 30 min. To induce blood glucose fluctuations, animals were given 500 mg/ml 0.4–0.8 ml/kg glucose intravenously (Braun) and 0.2–0.4 unit/kg insulin (Actrapid Penfill; Novo Nordisk A/S, Bagsværd, Denmark) 60 min later several times during the experiment. A bolus of glucose followed by insulin is known as a glucose/insulin event . Blood glucose concentration was measured by repetitive arterial blood samples (ABL 725; Radiometer, Brønshøj, Denmark). Experiments lasted 5–7 h.

The IvSCGM sensor (Invivosense ASA) is a biosensor with a hydrogel matrix incorporated with 3-phenylboronic acid. The gel contracts at rising glucose concentrations as a consequence of glucose-induced cross-binding of phenylboronic molecules. The hydrogel is fabricated at the tip of an optical fiber, and the whole sensor is covered with a partly heparinized semipermeable coating. The diameter of the gel is measured by an interferometric technique. Changes in glucose concentration cause the volume of the gel to change, causing its diameter to change accordingly.25The sensor measures the relative change in glucose concentration and needs to be calibrated against another method of glucose measurement (in vivo  calibration) or in a solution with a known glucose concentration (in vitro  calibration).

To transform interferometric length measurement data into glucose concentration data, we used a nonlinear two-parameter calibration function. We used data from repetitive blood samples as calibration parameters retrospectively. In addition, to compensate for baseline drift, we used a fixed baseline drift rate. To compensate for pH interference, we used pH-corrected calibration parameters. When sensor responsiveness was grossly reduced (∼70%), it was regarded damaged, and no further data were collected from that CGM. For future clinical use, the calibration procedure has to be performed in vitro  in advance of probe insertion by a two-point calibration procedure, which exposes the sensors to two buffer solutions with two different glucose concentrations.

To calibrate and evaluate the IvSCGM, we used a blood gas analyzer (Radiometer). Glucose is transported across the outer membrane of the multilayer glucose electrode. Glucose oxidase, immobilized between the inner and outer membrane layers, converts glucose to hydrogen peroxide (glucose + oxygen → gluconic acid + H²O²), which crosses the inner membrane toward the electrode's anode. The oxidation of H²O²creates a proportional electric current that is also proportional to the amount of glucose. The analyzer uses similar techniques, but different electrodes, to measure Pao², Paco², pH, and lactate.26 

Arterial blood samples were collected at preset time intervals and analyzed instantaneously by a bedside blood gas analyzer (Radiometer). Blood sample times were marked electronically in the IvSCGM system. Some animals wore more than one sensor, yielding more than one paired value from each blood sample. A total of 20 probes were used in seven pigs, yielding 90 glucose/insulin events and 807 paired arterial blood sample (ABL) and IvSCGM measurements.

Software R was used for statistical analysis (version 2.10.1; The R Foundation for Statistical Computing, Vienna, Austria). Paired ABL-IvSCGM measurements were compared with mean bias (mean difference) and SD. We report the 95% limits of agreement, calculated as mean bias ± 1.96 × SD, the recommended statistic for assessing agreement in studies comparing two methods of measurement for physiologic variables.27–29We performed variance analyses to check for possible clustering at the animal and sensor level.30 

Because sensor performance at low values is important, the same test statistics (mean bias [SD] and 95% limits of agreement) were computed for all paired measurements with a mean below 2.2 mM.

The sensor was tested using a range of blood glucose values with the lowest value at 0.63 mM and the highest 15.75 mM (mean values of ABL and IvSCGM measures). The mean bias (SD) between IvSCGM-ABL measurements was 0.0131 (0.470) mM with 95% limits of agreement at −0.908 and 0.934 mM (fig. 1). Values are presented in a Bland–Altman plot with the differences of paired measures plotted against their mean. The mean difference and the 95% limits of agreement are marked with horizontal lines. As seen from the plot, the variance was equally spread throughout the range of measurements. There was no clustering of importance at the animal or sensor level.

There were 121 paired measurements in the hypoglycemic area below 2.2 mM with a mean bias (SD) bias of −0.0435 (0.260) mM and 95% limits of agreement at −0.553 and 0.466 mM.

Performance of the IvSCGM sensor was promising, with measurements that were in high agreement with the blood gas analyzer. The mean bias and the 95% limits of agreement were low. The sensor also performed well in the low glucose range.

Although response time was not a primary variable in this study, it is of great importance to the clinician. Although no systematic method to quantify response time was applied, we found the response to be almost instantaneous. Figure 2shows the rapid rise in blood glucose concentration after intravenous administration of bolus glucose. Sensor readings followed the rise in blood glucose; by the time of the next blood sample (i.e ., 5 min postadministration), sensor readings were falling.

To ensure rapid response times, intravascular CGM positioning is preferable to subcutaneous. Intravascular placement is especially important for intensive care patients, in whom peripheral circulation is rather unpredictable.31It is still unclear whether intra-arterial versus  intravenous placement is best, but this question is beyond the scope of the current investigation. Furthermore, there are obvious differences between central and peripheral venous positioning. All intravascular placements can lead to complications such as infection, thromboembolic incidents, and measurement errors caused by local circulatory factors as well as fluid and drug infusions. However, we still believe that this level of invasiveness is clinically acceptable because intensive care patients are already subjected to invasive treatment and monitoring.

Biocompatibility is an important issue for all in vivo  biosensors, with clotting being particularly devastating for intravascular biosensors.32Some IvSCGM sensors degraded over time. Under microscopic examination after removal, we found clotting on sensors surfaces that correlated with the degradation problem. Pigs hypercoagulate,33so we would expect clotting to be less severe in humans. Nevertheless, the issue of clotting needs to be addressed. Invivosense ASA is currently working with alternative surface materials to resolve the problem (Reinold Ellingsen, Ph.D., CEO, Invivosense ASA, Trondheim, Norway, oral communication, August 2010). In addition, interference with different endogenous and exogenous substances is a well-known problem in all biosensors. The IvSCGM was prone to interfere with pH, leading to drifting. Experiments are ongoing with different types of hydrogels to solve this important issue (Reinold Ellingsen, Ph.D., CEO, Invivosense ASA, Trondheim, Norway, oral communication, August 2010).

One strength of the present study was the use of an animal model, which allowed us to challenge the sensor using a broad range of blood glucose values. We were also able to compare sensor performance within tighter blood sampling intervals during periods of extreme variation. It was important to obtain multiple measurements in the hypoglycemic range, as hypoglycemia is the most dreaded complication of tight glycemic control. We are not aware of any other published studies on CGM performance using an animal model where the glycemic environment was so tightly controlled.

There is some divergence between authors in the statistical reporting of results from studies comparing CGM systems. This divergence is a problem when evaluating and comparing these studies. Surprisingly, the use of correlation coefficients and linear regression remains quite extensive. Analysis using Bland–Altman plots is a much better approach, and the use of 95% limits of agreement is a useful statistic.29The Clarke error grid, primary developed for comparing point-of-care blood sugar analyzers, is used extensively in studies comparing CGM systems.34We found the grid to be an inexact tool and one that is exceptionally bad when comparing measurements in the hypoglycemic range.35 

In conclusion, the current study presents a novel intravascular CGM. The performance of the sensor is in agreement with blood gas measurements over a wide range of glucose values including the clinically important hypoglycemic range. Response times seem sufficient for clinical use.