• View in gallery
    Figure 1—

    Box-and-whisker plots of interstitial glucose concentrations measured by use of CGMS sensors and blood glucose concentrations measured by use of a criterion-referenced (CR) method during a 72-hour period in each of 4 dogs (A through D, respectively). Each box represents the IQR, the horizontal line in each box represents the median, and the whiskers represent the range. In each dog, sensors 1 through 4 were implanted in the interstitial tissues over the left lumbar musculature, and sensors 5 through 8 were implanted in the interstitial tissues over the right lumbar musculature.

  • View in gallery
    Figure 2—

    Scatterplot of blood glucose concentration measured by use of a criterion-referenced method versus the difference between the interstitial glucose concentration measured by use of a CGMS (8 concurrently implanted sensors) and the blood glucose concentration measured by use of the criterion-referenced method for each of 4 dogs. Each type of symbol represents results for 1 dog. The horizontal dashed line represents the mean difference, and the dashed-and-dotted lines represent the 95% reference intervals (calculated as mean difference ± [1.96 × SD]).

  • View in gallery
    Figure 3—

    Box-and-whisker plots of the difference between the interstitial glucose concentration measured by use of a CGMS (sensors implanted at each of 4 locations) and the blood glucose concentration measured by use of the criterion-referenced method during a euglycemic steady state in 8 dogs. Each box represents the IQR, the horizontal line in each box represents the median, the whiskers represent the range, and circles represent outliers (> 1.5 times the IQR).

  • View in gallery
    Figure 4—

    Box-and-whisker plots of the difference between the interstitial glucose concentration measured by use of a CGMS (4 sensors implanted in the interstitial tissues over the left and 4 sensors implanted in the interstitial tissues over the right lumbar musculature) and the blood glucose concentration measured by use of the criterion-referenced method during a euglycemic steady state in each of 4 dogs (A through D). See Figure 3 for remainder of key.

  • View in gallery
    Figure 5—

    Box-and-whisker plots of the difference between the interstitial glucose concentration measured by use of a CGMS (sensors implanted at each of 4 locations) and the blood glucose concentration measured by use of the criterion-referenced method during a glycemic clamp procedure in 4 dogs. See Figure 3 for remainder of key.

  • View in gallery
    Figure 6—

    Box-and-whisker plots of the difference between the interstitial glucose concentration measured by use of a CGMS (sensors implanted at each of 4 locations) and the blood glucose concentration measured by use of the criterion-referenced method during various phases of a glycemic clamp procedure in 4 dogs. Phases included initiation of an initial hypoglycemic (50 mg/dL) or hyperglycemic (300 mg/dL) phase, hypoglycemic or hyperglycemic steady state maintained for approximately 1 hour, return to midrange (mid) glucose concentration (150 mg/dL), midrange steady state maintained for approximately 1 hour, initiation of opposite glycemic phase (ie, hyperglycemia or hypoglycemia), hyperglycemic or hypoglycemic steady state maintained for approximately 1 hour, and glucose infusions discontinued. For 2 dogs, the hyperglycemic phase was initiated first (Hyper 1), which was followed by the midrange concentration (Mid 1) and then the hypoglycemic phase (Hypo 2); for the other 2 dogs, the hypoglycemic phase was initiated first (Hypo 1), which was followed by the midrange phase (Mid 2) and then the hyperglycemic phase (Hyper 2). See Figure 3 for remainder of key.

  • View in gallery
    Figure 7—

    Clarke error grids of the interstitial glucose concentration measured by use of CGMS sensors implanted at each of 4 locations (A = dorsolateral aspect of the neck, B = abdomen, C = lumbar region, and D = lateral aspect of the thorax) and the blood glucose concentration measured by use of a criterion-referenced method in each of 4 dogs. Values were obtained during a euglycemic steady state (circles) or during glycemic clamp procedures (plus signs). Zones for Clarke error grid analysis are as follows: A = value deviates by < 20% from the criterion-referenced blood glucose concentration or both glucose concentrations < 70 mg/dL, B = value deviates by > 20% from the criterion-referenced blood glucose concentration but leads to no or benign treatment, C = value leads to overcorrection of an acceptable blood glucose concentration or misinterpretation of euglycemia as hyperglycemia or hypoglycemia, D = value leads to a dangerous failure to detect and treat hyperglycemia or hypoglycemia, and E = value leads to treatment that is contrary to that required (ie, treatment for hypoglycemia during a hyperglycemic state, and vice versa).

  • View in gallery
    Figure 8—

    Consensus error grids of the interstitial glucose concentration measured by use of CGMS sensors implanted at each of 4 locations (A = dorsolateral aspect of the neck, B = abdomen, C = lumbar region, and D = lateral aspect of the thorax) and the blood glucose concentration measured by use of a criterion-referenced method in each of 4 dogs. Zones for consensus error grid analysis are as follows: A = no effect on clinical action, B = altered clinical action unlikely to affect outcome, C = altered clinical action likely to affect clinical outcome, D = altered clinical action that could pose serious medical risk, and E = altered clinical action that could have dangerous consequences. See Figure 7 for remainder of key.

  • View in gallery
    Figure 9—

    Interstitial glucose concentrations (solid line) obtained for 2 representative sensors (A and B) during glycemic clamp procedures and corresponding blood glucose concentrations (circles) concurrently measured by use of a criterion-referenced method. Notice the lag between changes in blood glucose concentrations and changes in CGMS values.

  • 1. Metzger M, Leibowitz G, Wainstein J, et al. Reproducibility of glucose measurements using the glucose sensor. Diabetes Care 2002; 25: 11851191.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 2. Davison LJ, Slater LA, Herrtage ME, et al. Evaluation of a continuous glucose monitoring system in diabetic dogs. J Small Anim Pract 2003; 44: 435442.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 3. DeClue AE, Cohn LA, Kerl ME, et al. Use of continuous blood glucose monitoring for animals with diabetes mellitus. J Am Anim Hosp Assoc 2004; 40: 171173.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 4. Wiedmeyer CE, DeClue AE. Continuous glucose monitoring in dogs and cats. J Vet Intern Med 2008; 22: 28.

  • 5. Hafner M, Lutz TA, Reusch CE, et al. Evaluation of sensor sites for continuous glucose monitoring in cats with diabetes mellitus. J Feline Med Surg 2013; 15: 117123.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 6. Reineke EL, Fletcher DJ, King LG, et al. Accuracy of a continuous glucose monitoring system in dogs and cats with diabetic ketoacidosis. J Vet Emerg Crit Care (San Antonio) 2010; 20: 303312.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 7. Wiedmeyer CE, Johnson PJ, Cohn LA, et al. Evaluation of a continuous glucose monitoring system for use in dogs, cats, and horses. J Am Vet Med Assoc 2003; 223: 987992.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 8. Wiedmeyer CE, Johnson PJ, Cohn LA, et al. Evaluation of a continuous glucose monitoring system for use in veterinary medicine. Diabetes Technol Ther 2005; 7: 885895.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 9. Affenzeller N, Thalhammer JG, Willmann M. Home-based subcutaneous continuous glucose monitoring in 10 diabetic dogs. Vet Rec 2011; 169: 206210.

  • 10. Chico A, Vidal-Rios P, Subirà M, et al. A. The continuous glucose monitoring system is useful for detecting unrecognized hypoglycemias in patients with type 1 and type 2 diabetes but is not better than frequent capillary glucose measurements for improving metabolic control. Diabetes Care 2003; 26: 11531157.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 11. Summers LK, Clark ML, Humphreys SM, et al. The use of microdialysis to monitor rapid changes in glucose concentration. Horm Metab Res 1999; 31: 424428.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 12. National Research Council Committee for the Update of the Guide for the Care and Use of Laboratory Animals. Guide for the care and use of laboratory animals. 8th ed. Washington, DC: The National Academies Press, 2011.

    • Search Google Scholar
    • Export Citation
  • 13. Mazzaferro EM. Arterial catheterization. In: Silverstein D, Hopper K, eds. Small animal critical care medicine. 2nd ed. St Louis: Elsevier, 2015;10401043.

    • Search Google Scholar
    • Export Citation
  • 14. Wess G, Reusch C. Evaluation of five portable blood glucose meters for use in dogs. J Am Vet Med Assoc 2000; 216: 203209.

  • 15. Barnhart HX, Haber M, Song J. Overall concordance correlation coefficient for evaluating agreement among multiple observers. Biometrics 2002; 58: 10201027.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 16. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: 307310.

    • Search Google Scholar
    • Export Citation
  • 17. Clarke WL, Cox D, Gonder-Frederick LA, et al. Evaluating clinical accuracy of systems for self-monitoring of blood glucose. Diabetes Care 1987; 10: 622628.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 18. Parkes JL, Slatin SL, Pardo S, et al. A new consensus error grid to evaluate the clinical significance of inaccuracies in the measurement of blood glucose. Diabetes Care 2000; 23: 11431148.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 19. Hoenig M, Pach N, Thomaseth K, et al. Evaluation of long-term glucose homeostasis in lean and obese cats by use of continuous glucose monitoring. Am J Vet Res 2012; 73: 11001106.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 20. Derosa G, Salvadeo SA, Mereu R, et al. Continuous glucose monitoring system in free-living healthy subjects: results from a pilot study. Diabetes Technol Ther 2009; 11: 159169.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 21. Kerssen A, de Valk HW, Visser GH. Day-to-day glucose variability during pregnancy in women with type 1 diabetes mellitus: glucose profiles measured with the continuous glucose monitoring system. BJOG 2004; 111: 919924.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 22. Goldberg PA, Siegel MD, Russell RR, et al. Experience with the continuous glucose monitoring system in a medical intensive care unit. Diabetes Technol Ther 2004; 6: 339347.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 23. Wisniewski N, Moussy F, Reichert WM. Characterization of implantable biosensor membrane biofouling. Fresenius J Anal Chem 2000; 366: 611621.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 24. Li CM, Dong H, Cao X, et al. Implantable electrochemical sensors for biomedical and clinical applications: progress, problems, and future possibilities. Curr Med Chem 2007; 14: 937951.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 25. Chlup R, Jelenová D, Chlupová K, et al. Function and accuracy of glucose sensors beyond their stated expiry date. Diabetes Technol Ther 2006; 8: 495504.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 26. Enlite Glucose Sensor user's guide. MP6025676-2AF1/A. Northridge, Calif: Medtronic Inc, 2014.

  • 27. Vriesendorp TM, Devries JH, Holleman F, et al. The use of two continuous glucose sensors during and after surgery. Diabetes Technol Ther 2005; 7: 315322.

    • Crossref
    • Search Google Scholar
    • Export Citation

Advertisement

Effect of sensor location in dogs on performance of an interstitial glucose monitor

Amie KoenigDepartment of Small Animal Medicine and Surgery, College of Veterinary Medicine, University of Georgia, Athens, GA 30602.

Search for other papers by Amie Koenig in
Current site
Google Scholar
PubMed
Close
 DVM
,
Margarethe E. HoenigDepartment of Physiology and Pharmacology, College of Veterinary Medicine, University of Georgia, Athens, GA 30602.

Search for other papers by Margarethe E. Hoenig in
Current site
Google Scholar
PubMed
Close
 Dr Med Vet, PhD
, and
David A. JimenezDepartment of Veterinary Biosciences and Diagnostic Imaging, College of Veterinary Medicine, University of Georgia, Athens, GA 30602.

Search for other papers by David A. Jimenez in
Current site
Google Scholar
PubMed
Close
 DVM

Abstract

OBJECTIVE To identify variations in glucose values concurrently obtained by use of a continuous glucose monitoring system (CGMS) at the same site, reliability of results for each site, lag time for each site, and influence of site thickness on CGMS accuracy.

ANIMALS 8 random-source research dogs.

PROCEDURES In experiment 1, 8 CGMS sensors were implanted bilaterally at 1 site (4 sensors/side) in 4 dogs. In experiment 2, 2 CGMS sensors were implanted bilaterally at each of 4 sites (1 sensor/side) in 8 dogs; 4 of those 8 dogs then were subjected to a glycemic clamp technique. The CGMS results were compared among sensors and with criterion-referenced results during periods of euglycemia for all 8 dogs and during hyperglycemia and hypoglycemia for 4 dogs during the glycemic clamp procedure.

RESULTS Differences (median, −7 mg/dL; interquartile range [IQR], −18.75 to 3 mg/dL) between CGMS and criterion-referenced glucose concentrations differed significantly among dogs and sites; during euglycemia, they were not different from the expected normal variation between multiple sensors concurrently implanted at the same site. Differences (median, −35 mg/dL; IQR, −74 to −15 mg/dL) between CGMS and criterion-referenced concentrations were greater during changes in glucose concentrations. Thoracic sensors were most accurate but had the shortest mean functional life.

CONCLUSIONS AND CLINICAL RELEVANCE Significant differences were detected between CGMS and criterion-referenced glucose concentrations. Overall clinical utility of CGMS was acceptable at all sites, with most of the values from all sensors, sites, and dogs meeting guidelines for point-of-care glucometers.

Abstract

OBJECTIVE To identify variations in glucose values concurrently obtained by use of a continuous glucose monitoring system (CGMS) at the same site, reliability of results for each site, lag time for each site, and influence of site thickness on CGMS accuracy.

ANIMALS 8 random-source research dogs.

PROCEDURES In experiment 1, 8 CGMS sensors were implanted bilaterally at 1 site (4 sensors/side) in 4 dogs. In experiment 2, 2 CGMS sensors were implanted bilaterally at each of 4 sites (1 sensor/side) in 8 dogs; 4 of those 8 dogs then were subjected to a glycemic clamp technique. The CGMS results were compared among sensors and with criterion-referenced results during periods of euglycemia for all 8 dogs and during hyperglycemia and hypoglycemia for 4 dogs during the glycemic clamp procedure.

RESULTS Differences (median, −7 mg/dL; interquartile range [IQR], −18.75 to 3 mg/dL) between CGMS and criterion-referenced glucose concentrations differed significantly among dogs and sites; during euglycemia, they were not different from the expected normal variation between multiple sensors concurrently implanted at the same site. Differences (median, −35 mg/dL; IQR, −74 to −15 mg/dL) between CGMS and criterion-referenced concentrations were greater during changes in glucose concentrations. Thoracic sensors were most accurate but had the shortest mean functional life.

CONCLUSIONS AND CLINICAL RELEVANCE Significant differences were detected between CGMS and criterion-referenced glucose concentrations. Overall clinical utility of CGMS was acceptable at all sites, with most of the values from all sensors, sites, and dogs meeting guidelines for point-of-care glucometers.

Interstitial glucose concentrations are reportedly comparable to whole blood glucose concentrations in humans and other animals1–9 as a result of an equilibrium that develops between the glucose concentration in the blood and interstitium. Continuous interstitial glucose monitoring systems allow frequent monitoring of glucose concentrations with a minimum of venipuncture, and these devices have become a popular alternative method for evaluating glucose concentrations in animals.3–6,9 Furthermore, CGMSs also allow detection of hyperglycemic and hypoglycemic episodes that may otherwise be undetected.10

Several factors may influence the performance of interstitial sensors, including the proportion of interstitial fluid in a tissue, which is lower and more variable in subcutaneous tissues and dependent on size of adipocytes and subsequent adipocyte blood flow.11 In humans, it is recommended that sensors be implanted in tissues in the region of the hips, buttocks, ventral aspect of the abdomen, or thighs. In veterinary medicine, the neck and lateral aspect of the thorax are common sites for sensor implantation, perhaps because sensors in these sites are relatively easy to secure and less likely to be dislodged. Use of alternative sites for sensor implantation may be necessary in specific patients because of injury, infection, catheters, bandages, or feeding tubes that may interfere with sensor placement. Authors of 1 study5 suggested that the dorsal aspect of the neck was better than the lateral aspect of the thorax or lateral aspect of the stifle joint for sensor implantation in cats. Optimal implantation sites for dogs have not been identified.

The objective of the study reported here was to evaluate the accuracy and reliability of a CGMS. Specific objectives included determining the amount of variation between 8 CGMS sensors concurrently implanted at the same site in the same dog, evaluating reliability of the CGMS at 4 sites by correlating CGMS results with blood glucose measurements, determining the lag time for each site (time for interstitial glucose concentration to approximate blood glucose concentration after changes in glucose concentrations), and evaluating the effect of site thickness on accuracy of the CGMS and lag time. We hypothesized that duration of sensor viability or CGMS values would not differ among sites, differences among sites would be within the range of expected variation for multiple concurrently implanted sensors, there would be no difference in the lag time among sites, and site thickness would not affect CGMS accuracy.

Materials and Methods

Animals

Eight random-source research colony dogs with a body weight > 20 kg were used in the study. Dogs were deemed healthy on the basis of results of physical examination, a routine CBC, and biochemical analysis. Animals were cared for in accordance with established principles.12 The study protocol was approved by an institutional animal care and use committee. Dogs were housed separately in 1.2 × 2.4-m runs, and CGMS monitors were within 1.2 m of the dogs at all times. During periods of sensor implantation, dogs wore a plastic muzzle to prevent removal of the sensors or transmitters. The muzzles allowed the dogs to drink and pant freely, and the muzzles were removed at least 6 times daily during supervised periods to allow dogs to eat and drink unimpeded. Dogs were fed once daily. In addition, elastic tapea hobbles were placed on the hind limbs of the dogs during periods of sensor implantation to prevent abduction of the limbs and dislodgement of the sensors.

Study design

The study consisted of 2 experiments. Experiment 1 was designed to assess variability of concurrently implanted CGMS units. For this experiment, 4 dogs each were implanted with 4 sensors bilaterally at the same site (interstitial tissues over the left and right lumbar musculature). Expected normal variation was defined as the expected differences between CGMS values obtained at the same time from multiple sensors concurrently implanted at the same site during a euglycemic steady state. Experiment 2 was designed to assess differences among sensor sites. The 4 dogs used in experiment 1 were transitioned immediately into the second experiment, and 4 additional dogs were included. Four separate standardized sites were used for sensor implantation (dorsolateral aspect of the neck, lateral aspect of the thorax, lumbar region, and abdomen). For experiment 2, CGMS data were collected during a euglycemic steady state, and 4 of the 8 dogs were subjected to a glycemic clamp technique to create hyperglycemic and hypoglycemic conditions.

Data collection

The CGMSb consisted of a single-use sensor, wireless transmitter, remote data logger (monitor), and USB link with computer software. The sensor was an electroenzymatic 3-electrode cell housed in flexible tubing with a side window enclosed by a semipermeable polyurethane membrane that allowed the electrode to interact with interstitial fluid. Glucose traversed the membrane and was converted by glucose oxidase into gluconic acid and hydrogen peroxide, which in turn generated an electric current proportional to the interstitial glucose concentration.

Initial calibration of the system was performed approximately 2 hours after sensor implantation, then between 2 and 8 hours after implantation, and then approximately every 8 hours thereafter. Data were not collected by the system until it was recalibrated if more than 12 hours elapsed between calibrations. Additionally, calibration was performed when a dog's glucose concentration was relatively stable and within the sensor limits, which were a glucose concentration of 40 to 400 mg/dL. A smoothed average was recorded by the system at 5-minute intervals. The sensor was attached to a transmitter that relayed the glucose concentration to the monitor, and the data were downloaded for evaluation. The manufacturer intended that the sensors would be used for 72 hours, after which time the monitors would not provide data.

Experiment 1 (assessment of variability of concurrent CGMS units)

Hair was clipped from an area (approx 12 × 12 cm) over the right and left lumbar musculature; the area was then swabbed with isopropyl alcohol and allowed to air dry. Interstitial glucose sensorsc were prepared for implantation; new batteries were installed, and sensors were synchronized with regard to date and time. Four sensors were implanted as per the manufacturer's instructions on each side of each dog. Sensors were implanted by use of the provided insertion deviced and were located immediately adjacent to each other in a 4-leaf clover pattern. Sensors were secured with a drop of tissue gluee placed at each corner of the sensor. A transmitter was attached to each sensor and secured to the skin with a 2 × 2-cm piece of double-sided tape. All sensors were implanted and secured by the same investigator (AK) in both experiments.

Before data were collected, an 18-gauge, 20-cm single-lumen sampling catheterf was inserted into a jugular vein by use of a modified Seldinger technique and standard aseptic procedures. Catheter patency was maintained by flushing with 2 mL of heparinized saline (0.9% NaCl) solution (10 U of heparing/mL) at 8-hour intervals. A jugular vein blood sample (0.5 to 1 mL) was collected at various intervals ranging from 2 to 10 hours apart throughout a 72-hour period. All blood samples were acquired via the catheter by use of a 3-syringe technique with a presample of 3 mL.13 Glucose was measured immediately with a POC meter,h which was selected for use because it yielded a more accurate measurement of glucose concentrations, compared with results for other POC meters.14 The remainder of the blood sample was allowed to clot and then was centrifuged (within 20 minutes after collection) at 1,500 × g; serum was decanted and frozen at −20°C. Samples subsequently were batch analyzed for serum glucose concentration by use of a colorimetric glucose oxidase techniquei as the criterion-referenced method. All CGMSs were calibrated approximately every 8 hours (range, 6 to 10 hours) by use of a blood glucose concentration measured with the POC meter. All calibrations were entered into each CGMS at the same time, with ≤ 30 seconds between calibration of the first and last CGMS. Additional blood samples were collected at various intervals ranging from 2 to 10 hours apart for up to 72 hours; these samples were treated as described previously for POC samples and subsequently were batch analyzed by use of the criterion-referenced method. Sensor sites were evaluated at the time of each blood collection for evidence of complications such as redness, swelling, discharge, or sensor dislodgement. At the end of the 72-hour experimental period, sensors were removed and each dog was returned to its cage.

Experiment 2 (assessment among sensor sites)

One CGMS sensor was inserted at each of 4 sites on each side of the 8 dogs (8 sensors/dog). The neck implantation site was half the distance from the vertical ramus of the mandible to the thoracic inlet and half the distance from the jugular vein to the dorsal midline with the dog in a standing position. The thorax site was located between the fifth and sixth intercostal space at the level of the glenohumeral joint with the dog in lateral recumbency. The lumbar site was located half the distance between the wing of the ilium and the last rib and half the distance between the dorsal midline and ventral edge of the epaxial muscles with the dog in a standing position. The abdomen site was immediately ventral to the lumbar musculature and caudal to the last rib with the dog in lateral recumbency. Ultrasonography was used to measure depth of the skin and subcutaneous tissues at each of the 4 sites prior to sensor implantation. Four separate measurements were obtained at each site, and the ultrasonographer (DAJ) was not aware of the numeric measurements recorded during acquisition. The 4 measurements were used to calculate the mean value for a representative thickness at each site for each dog.

Sensors and transmitters were arbitrarily assigned to each of the sites in each of the dogs. Implantation, securing of the sensors and transmitters, sensor initialization, and calibration procedures were identical to those in experiment 1. Blood samples were collected at various intervals ranging from 2 to 10 hours apart during a 72-hour period. Implantation sites were evaluated for complications as described previously.

Duration of sensor life (ie, lifespan) was defined as the time from first calibration to the last interstitial measurement recorded by the monitor. The number and type of sensor alarms and reasons for premature removal (< 72 hours) were recorded. Alarms were defined as specific events that occurred after the initial calibration; these included alarms that indicated weak signal, lost sensor, calibration error, and change sensor. Weak signals were defined as the monitor did not receive data from the transmitter for 15 minutes (3 consecutive measurements) because the transmitter was too far away or because physical barriers impeded communication (eg, if a dog was lying with the sensor on the ground against a wall). Lost sensors were defined as 8 consecutive measurements (> 40 minutes) not detected by the monitor. Reestablishing communication between lost sensors and monitors was attempted, and the sensor was allowed to continue in the experiments if communication was restored. Calibration errors were considered as detection of a glucose concentration by the monitor that did not apparently correspond with the electrical current generated at the sensor. Potential causes of calibration errors included rapidly increasing or decreasing glucose concentrations or an incorrectly functioning sensor. Two consecutive attempts were made to recalibrate a sensor; if the error was corrected with the recalibration, the sensor was allowed to continue in the experiments. Change sensor alarms were sounded after 2 consecutive calibration errors, at which time the sensor was considered to have failed. Hyperglycemic and hypoglycemic monitor alarms were disregarded, and alerts requesting calibration of the glucose concentration at appropriate intervals were not recorded as alarms. Alarms that repeated prior to being addressed were counted as 1 alarm event.

Approximately 18 hours after the sensors were implanted, the 4 dogs previously included in experiment 1 were subjected to a glycemic clamp technique with multiple phases. The phases included baseline infusions of glucose during euglycemia, initiation of hypoglycemia (approx 50 mg/dL) or hyperglycemia (approx 300 mg/dL), hypoglycemic or hyperglycemic steady state maintained for approximately 1 hour, return to midrange glucose concentration (approx 150 mg/dL), midrange steady state maintained for approximately 1 hour, initiation of opposite glycemic phase (ie, hyperglycemia or hypoglycemia), hyperglycemic or hypoglycemic steady state maintained for approximately 1 hour, and glucose infusions discontinued. The hypoglycemic phase was initially induced in 2 dogs, which was followed by the hyperglycemic phase; these phases were reversed for the other 2 dogs.

At the start of the clamp technique, an 18-gauge, 3.8-cm catheter was placed in a cephalic vein for infusion of dextrose, somatostatin, and regular insulin. Somatostatinj was passed through a filter,k diluted in saline solution to a concentration of 100 μg/mL, and infused at a rate of 1.0 μg/kg/min during baseline. Regular insulinl was diluted in saline solution to a concentration of 100 mU/mL and infused at a rate of 0.15 mU/kg/min for the first 60 minutes. The insulin infusion then was increased to 1.1 mU/kg/min, and an IV infusion of 20% dextrosem was initiated and maintained at variable rates to achieve the various glycemic goals. The intent was to maintain the glucose concentration close to the desired value for approximately 1 hour, at which time the next phase of the clamp technique would be initiated.

The hypoglycemic phase began with an infusion of insulin (1.1mU/kg/min); an infusion of 20% dextrose (0.5 mL/kg/h) was initiated once the blood glucose concentration reached 60 mg/dL. Infusion rates during the hypoglycemic phase ranged from 0 to 1.1 mU/kg/min for insulin and 0 to 2 mL/kg/h for 20% dextrose. The midrange phase began with an infusion of insulin (1.1 mU/kg/min) and 20% dextrose (1.5 mL/kg/h); infusion rates during the midrange phase ranged from 0.06 to 6 mU/kg/min for insulin and 0.2 to 4.4 mL/kg/h for 20% dextrose. The hyperglycemic phase began with an infusion of insulin (1.1 mU/kg/min) and 20% dextrose (2.5 mL/kg/h); infusion rates during the hyperglycemic phase ranged from 0 to 1.1 mU/kg/min for insulin and 0.2 to 16 mL/kg/h for 20% dextrose. The blood glucose concentration was measured every 5 minutes during baseline and periods of rapid changes in glucose concentrations, then every 10 minutes once the desired glucose concentration was achieved. Blood samples collected at each time point were handled as described previously. At the end of the experimental period, infusions were discontinued; dogs then were fed, provided water, and returned to their runs. Approximately 3 hours after completion of the clamp technique, sensors were recalibrated and intermittent data collection resumed for the remainder of the experimental period. Data from the CGMS were downloaded to a computer database.n

Statistical analysis

Data analysis—Data were assessed for normality by evaluation of histograms or quantile-quantile plots. Significance was set at values of P < 0.05 for all tests.

Experiment 1 (assessment of variability of concurrent CGMS units)—Summary statistics (mean, median, minimum, and maximum) were evaluated for each sensor. The CGMS data were compared among the sensors via calculation of a correlation coefficient. The concordance correlation coefficient indicates agreement between 2 measurements; a value of 1 indicates perfect concordance. Overall concordance correlation coefficients and results of Bland-Altman analysis, whereby results for each sensor were compared with results for the criterion-referenced method, were determined for each dog.15,16

Experiment 2 (assessment among sensor sites)—Evaluation of multiple sensor sites included assessment of sensor lifespan and frequency of sensor alarms, which may have indicated problems with sensor function or communication with the monitor. A χ2 analysis was used to test for significant differences in the number of alarms among sites and dogs. Least squares means were used to test for differences between sites and among dogs.

Unexplainable rapid wide oscillations in CGMS values not supported by similar blood glucose concentrations were considered nonviable measurements. A change sensor alarm and failure to record CGMS values were considered sensor failures. The Freeman-Halton extension of the Fisher exact test was used to determine whether failure rate was independent of implantation site. Survival analysis was used to assess time to failure by site.

For each dog, data downloaded from each monitor was temporally matched with the data from the other monitors by use of starting time and calibration times as temporal anchor points. Because actual time and calibration times were synchronized across the monitors, time differences between monitor measurements were minimized. Data obtained from a sensor during rapid, unexplained oscillations in CGMS measurements prior to sensor failure were excluded from analysis; all other data obtained prior to sensor failure were included in analysis. Summary statistics of the differences between the CGMS and criterion-referenced glucose concentrations were calculated for each dog, site, side of implantation, and overall effect. Values outside 1.5 times the IQR were removed from additional analysis. The difference between the CGMS value and criterion-referenced glucose concentration was used as the response value for an ANOVA, which was used to examine the influence of dog, implantation site, site thickness, and the dog-by-implantation site interaction. Mean thickness of skin and subcutaneous tissues was calculated for each dog at each site, and these mean values were used to represent each dog in further calculations. Median values for thickness and range were calculated for each site across all dogs. Pairwise comparisons by use of least squares means were used to test for differences between any 2 locations. Bias also was calculated for sensors at each site. For each dog, raw CGMS data were examined by use of Kruskal-Wallis and Wilcoxon rank sum tests to detect significant differences among the monitor measurements at each site.

Duration of a clamp was defined as the period from initiation of baseline infusions until discontinuation of the infusions. All data obtained before and after a clamp represented a euglycemic steady state. Data obtained during the clamps were analyzed separately from data obtained during a euglycemic steady state. Differences between CGMS values and blood glucose concentrations measured with the criterion-referenced method were calculated for all monitors at all time points. Summary statistics, including mean, SD, minimum, first quartile, median, third quartile, maximum, and number of observations, were calculated. Kruskal-Wallis tests followed by Wilcoxon rank sum tests were used to compare CGMS values and criterion-referenced values by site and clamp phase. Differences between the CGMS and criterion-referenced glucose concentrations were median-adjusted by dog to remove the overall effect on differences that may have been contributed by a specific dog.

Clarke error grid and consensus error grid analyses were performed for euglycemic steady-state data, clamp data, and overall effects.17,18 The Clarke error grid contains 5 zones (A through E), which are as follows: A = value deviates by < 20% from the criterion-referenced blood glucose concentration or both glucose concentrations < 70 mg/dL, B = value deviates by > 20% from the criterion-referenced blood glucose concentration but leads to no or benign treatment, C = value leads to overcorrection of an acceptable blood glucose concentration or misinterpretation of euglycemia as hyperglycemia or hypoglycemia, D = value leads to a dangerous failure to detect and treat hyperglycemia or hypoglycemia, and E = value leads to treatment that is contrary to that required (ie, treatment for hypoglycemia during a hyperglycemic state, and vice versa). The consensus error grid also contains 5 zones (A through E), which are as follows: A = no effect on clinical action, B = altered clinical action unlikely to affect outcome, C = altered clinical action likely to affect clinical outcome, D = altered clinical action that could pose serious medical risk, and E = altered clinical action that could have dangerous consequences. A Fisher exact test was used to determine whether there were overall differences in error grid distributions among sites, and pairwise logistic regressions were then used to identify regions that differed.

Lag time was defined as the amount of time required for the CGMS to reflect a dog's new steady state after a rapid increase or decrease in blood glucose concentration as controlled by the glucose clamp technique. For each phase of the clamp, lag time was estimated by calculating the duration of time between the point at which the blood concentration reflected a steady state and the CGMS value indicated a steady state. Steady state was identified as the time at which blood glucose concentrations oscillated minimally (± 10%) near the desired blood glucose concentration.

Results

Animals

All dogs successfully completed the experiments. Redness, swelling, hemorrhage, infection, or other complications were not detected at any sensor implantation site at any time during the experiments or after sensor removal.

Experiment 1 (assessment of variability of concurrent CGMS units)

Data for each dog were collected for 72 hours (Figure 1). There were intermittent short communication failures with the monitor because of lost sensors; thus, the number of measurements for each sensor ranged from 628 to 828 (mean, 768 measurements/sensor). Pairwise comparisons of the sensors at each time point yielded 18,618 to 22,934 comparisons/dog. Intersensor variability within each dog was large. Pairwise comparisons of sensors at any given time point yielded a median difference of 4 (range, 0 to 100), 8 (range, 0 to 194), 8 (range, 0 to 70), and 8 (range, 0 to 106) for the 4 dogs. The overall concordance coefficient correlation was 0.444, 0.243, 0.586, and 0.376 for the 4 dogs. Two dogs each had 1 sensor with a concordance coefficient correlation close to 0 for all pairs, which suggested that these sensors were not providing measurements similar to those for the other sensors. When these 2 sensors were eliminated from comparisons of concordance coefficient correlations, the overall concordance coefficient for these 2 dogs increased to 0.333 and 0.614, respectively. There were 13 to 18 measurements obtained by use of the criterion-referenced method available for each dog. Mean ± SD difference between blood glucose concentrations measured with the criterion-referenced method and CGMS values was −9.655 ± 18.396 mg/dL for all 4 dogs and −5.515 ± 15.264 mg/dL, −0.574 ± 15.012 mg/dL, −26.313 ± 16.780 mg/dL, and −6.382 ± 14.677 mg/dL for each of the individual dogs (Figure 2).

Figure 1—
Figure 1—

Box-and-whisker plots of interstitial glucose concentrations measured by use of CGMS sensors and blood glucose concentrations measured by use of a criterion-referenced (CR) method during a 72-hour period in each of 4 dogs (A through D, respectively). Each box represents the IQR, the horizontal line in each box represents the median, and the whiskers represent the range. In each dog, sensors 1 through 4 were implanted in the interstitial tissues over the left lumbar musculature, and sensors 5 through 8 were implanted in the interstitial tissues over the right lumbar musculature.

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

Figure 2—
Figure 2—

Scatterplot of blood glucose concentration measured by use of a criterion-referenced method versus the difference between the interstitial glucose concentration measured by use of a CGMS (8 concurrently implanted sensors) and the blood glucose concentration measured by use of the criterion-referenced method for each of 4 dogs. Each type of symbol represents results for 1 dog. The horizontal dashed line represents the mean difference, and the dashed-and-dotted lines represent the 95% reference intervals (calculated as mean difference ± [1.96 × SD]).

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

Experiment 2 (assessment among sensor sites)

A total of 16 of 64 sensors (abdomen, 3/16; lumbar, 2/16; neck, 4/16; and thorax, 7/16) did not function for the entire 72-hour experimental period. Three losses (1 each for the abdomen, lumbar, and thorax sensors) were classified as malfunctions (sensor terminated after error message). Thirteen sensors (2 abdomen, 1 lumbar, 4 neck, and 6 thorax) were prematurely removed because the entire sensor was absent from the skin or the sensor tip was dislocated from the subcutaneous tissues and had exited through the skin. Sensor lifespan ranged from 2.5 to 72 hours (mean, 60.8 hours; median, 72 hours). Lifespan for abdomen sensors ranged from 16.5 to 72 hours (mean, 64.4 hours; median, 72 hours), for lumbar sensors ranged from 44.2 to 72 hours (mean, 69.8 hours; median, 72 hours), for neck sensors ranged from 2.5 to 72 hours (mean, 59.4 hours; median, 72 hours), and for thorax sensors ranged from 6 to 72 hours (mean, 49.5 hours; median, 72 hours). Failure rate did not differ significantly (P = 0.28) on the basis of sensor location.

A total of 84 alarms were recorded (22 abdomen, 13 lumbar, 22 neck, and 27 thorax). There were 3 to 19 alarms/dog. The most common alarm message received was weak signal (35 alarms). There were 18 lost sensor alarms and 8 alarms for sensor progression from weak signal to lost sensor. Other alarms included calibration error (n = 11 alarms) and change sensor (5). There were no significant differences in the number of alarms among locations (P = 0.29) or dogs (P = 0.052).

Thickness of the skin and subcutaneous tissue for each site in the 8 dogs ranged from 3.1 to 5.7 mm (median, 3.6 mm) for the abdominal, 2.0 to 5.1 mm (median, 4.2 mm) for the lumbar, 2.2 to 3.9 mm (median, 3.4 mm) for the neck, and 2.6 to 6.4 mm (median, 3.1 mm) for the thoracic sites. Side of implantation (left vs right) and thickness at a site did not significantly (P = 0.063) affect differences in CGMS measurements among sites.

Examination of raw CGMS data revealed that for any specific dog, measurements obtained from at least 1 site differed significantly from measurements obtained at the other sites. In 5 of 8 dogs, there were significant differences among all 4 sites. In 1 dog, values for the neck sensor were significantly different from values for the other 3 sites. In 2 dogs, only one of the site pairs had values that did not differ (abdomen and lumbar in one dog; abdomen and thorax in the other dog).

A total of 842 pairs of data for CGMS values and criterion-referenced values were collected. Differences between 28 of the pairs were outside the limits (1.5 times the IQR) and were removed; thus, there were 814 pairs that remained for further analysis. There were significant differences between CGMS and criterion-referenced glucose concentrations for all dogs at all sites (range, −97 to 117 mg/dL; IQR, −18.75 to 3 mg/dL; mean ± SD, −8.36 ± 20.15 mg/dL; and median, −7 mg/dL). Differences between paired glucose concentrations differed significantly among dogs (P < 0.001) and between sites of implantation (P = 0.001; Figure 3). Specifically, differences in paired glucose values obtained at the thoracic site differed significantly from differences in values obtained at the lumbar (P = 0.002) and neck (P = 0.002) sites; differences in paired glucose concentrations obtained at the lumbar and neck sites did not differ significantly (P = 1.000). Paired differences at the abdominal site were not significantly different from paired differences at the lumbar (P = 0.47), neck (P = 0.46), or thoracic (P = 0.11) sites. There was a significant negative bias at all sites. The CGMS values were on average 7.2, 9.2, 9.4, and 3.6 mg/dL less than criterion-referenced values for the abdomen, lumbar, neck, and thoracic sensors, respectively.

Figure 3—
Figure 3—

Box-and-whisker plots of the difference between the interstitial glucose concentration measured by use of a CGMS (sensors implanted at each of 4 locations) and the blood glucose concentration measured by use of the criterion-referenced method during a euglycemic steady state in 8 dogs. Each box represents the IQR, the horizontal line in each box represents the median, the whiskers represent the range, and circles represent outliers (> 1.5 times the IQR).

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

Median difference between CGMS and criterion-referenced values for all 4 dogs at all sites during all phases of the clamp technique (n = 1,055 pairs) was −35 mg/dL (IQR, −74 to −15 mg/dL). Similar to results for the nonclamp portions of the experiment, differences between CGMS and criterion-referenced glucose concentrations were significantly different among dogs (Figure 4). The median difference was −36 mg/dL for the abdominal, −32 mg/dL for the lumbar, −40 mg/dL for the neck, and −29 mg/dL for the thoracic sites (Figure 5). The difference between CGMS and criterion-referenced values was smallest for the baseline (n = 162 pairs) and hypoglycemic (165) phases, intermediate for the midrange phase (321), and largest for hyperglycemic phases (407; Figure 6). Similar to results obtained for experiment 1, bias was significantly different for values obtained at baseline for the neck and thorax (P = 0.003) and for the neck and abdomen (P = 0.005). Similar to data obtained during the nonclamp portions of the experiment, bias was least evident for sensors located in the thorax. Evidence for differences in bias among sites was less substantial during other phases of the clamp technique. There was a significant difference between thorax and lumbar values (P = 0.007) and thorax and neck values (P = 0.039) during hyperglycemia. There were no significant (P = 0.087 to P = 0.99) differences between sites during periods of hypoglycemia.

Figure 4—
Figure 4—

Box-and-whisker plots of the difference between the interstitial glucose concentration measured by use of a CGMS (4 sensors implanted in the interstitial tissues over the left and 4 sensors implanted in the interstitial tissues over the right lumbar musculature) and the blood glucose concentration measured by use of the criterion-referenced method during a euglycemic steady state in each of 4 dogs (A through D). See Figure 3 for remainder of key.

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

Figure 5—
Figure 5—

Box-and-whisker plots of the difference between the interstitial glucose concentration measured by use of a CGMS (sensors implanted at each of 4 locations) and the blood glucose concentration measured by use of the criterion-referenced method during a glycemic clamp procedure in 4 dogs. See Figure 3 for remainder of key.

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

Figure 6—
Figure 6—

Box-and-whisker plots of the difference between the interstitial glucose concentration measured by use of a CGMS (sensors implanted at each of 4 locations) and the blood glucose concentration measured by use of the criterion-referenced method during various phases of a glycemic clamp procedure in 4 dogs. Phases included initiation of an initial hypoglycemic (50 mg/dL) or hyperglycemic (300 mg/dL) phase, hypoglycemic or hyperglycemic steady state maintained for approximately 1 hour, return to midrange (mid) glucose concentration (150 mg/dL), midrange steady state maintained for approximately 1 hour, initiation of opposite glycemic phase (ie, hyperglycemia or hypoglycemia), hyperglycemic or hypoglycemic steady state maintained for approximately 1 hour, and glucose infusions discontinued. For 2 dogs, the hyperglycemic phase was initiated first (Hyper 1), which was followed by the midrange concentration (Mid 1) and then the hypoglycemic phase (Hypo 2); for the other 2 dogs, the hypoglycemic phase was initiated first (Hypo 1), which was followed by the midrange phase (Mid 2) and then the hyperglycemic phase (Hyper 2). See Figure 3 for remainder of key.

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

During nonclamp euglycemic steady state, there were no significant differences between the abdomen, lumbar, thorax, and neck sites for the percentage of measurements within zone A (82.2%, 76.8%, 82.2%, and 76.9%, respectively; P = 0.32) or zones A or B (99.5%, 100%, 100%, and 99.5%, respectively; P = 0.70) of Clarke error grids (Figure 7). Similarly, there were no significant differences between abdomen, lumbar, thorax, and neck locations for the percentage of measurements within zone A (83.1%, 79.5%, 84.7%, and 79.9%, respectively; P = 0.49) or zones A or B (99.1%, 100%, 100%, and 99.5%, respectively; P = 0.33) of consensus error grids (Figure 8).

Figure 7—
Figure 7—

Clarke error grids of the interstitial glucose concentration measured by use of CGMS sensors implanted at each of 4 locations (A = dorsolateral aspect of the neck, B = abdomen, C = lumbar region, and D = lateral aspect of the thorax) and the blood glucose concentration measured by use of a criterion-referenced method in each of 4 dogs. Values were obtained during a euglycemic steady state (circles) or during glycemic clamp procedures (plus signs). Zones for Clarke error grid analysis are as follows: A = value deviates by < 20% from the criterion-referenced blood glucose concentration or both glucose concentrations < 70 mg/dL, B = value deviates by > 20% from the criterion-referenced blood glucose concentration but leads to no or benign treatment, C = value leads to overcorrection of an acceptable blood glucose concentration or misinterpretation of euglycemia as hyperglycemia or hypoglycemia, D = value leads to a dangerous failure to detect and treat hyperglycemia or hypoglycemia, and E = value leads to treatment that is contrary to that required (ie, treatment for hypoglycemia during a hyperglycemic state, and vice versa).

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

Figure 8—
Figure 8—

Consensus error grids of the interstitial glucose concentration measured by use of CGMS sensors implanted at each of 4 locations (A = dorsolateral aspect of the neck, B = abdomen, C = lumbar region, and D = lateral aspect of the thorax) and the blood glucose concentration measured by use of a criterion-referenced method in each of 4 dogs. Zones for consensus error grid analysis are as follows: A = no effect on clinical action, B = altered clinical action unlikely to affect outcome, C = altered clinical action likely to affect clinical outcome, D = altered clinical action that could pose serious medical risk, and E = altered clinical action that could have dangerous consequences. See Figure 7 for remainder of key.

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

Clarke error grid analysis revealed that the percentage of values in zone A for the lumbar sensors during the glycemic clamp procedures (39.4%) was significantly (P = 0.024) lower than that for the thoracic sensors (48.5%; Figure 7). There were a significantly greater percentage of values that were not in zones A or B for lumbar sensors (14.9%) than for abdomen sensors (6.4%; P < 0.001), neck sensors (9.0%; P = 0.031), and thorax sensors (8.1%; P = 0.006). Consensus grid analysis revealed that there was a significantly greater percentage of values in zone A for the thorax sensors (49.8%) than for the lumbar sensors (38.2%; P = 0.002) and neck sensors (38.8%; P = 0.005; Figure 8). There was a significantly greater percentage of values that were not in zones A or B for lumbar sensors (7.8%) than for abdomen sensors (1.1%; P < 0.001), neck sensors (3.4%; P = 0.022), and thorax sensors (1.5%; P < 0.001).

On the basis of 158 transition points during the glycemic clamp procedures, minimum and maximum lag times for all sites were 0 and 50 minutes, respectively; median lag time for the lumbar site was 7.5 minutes, whereas median lag time for each of the other 3 sites was 10 minutes (Figure 9). Statistical analyses were not performed to compare lag times among locations because the large bias for the monitors coupled with the 5-minute intervals between CGMS measurements provided no evidence that lag times differed among locations.

Figure 9—
Figure 9—

Interstitial glucose concentrations (solid line) obtained for 2 representative sensors (A and B) during glycemic clamp procedures and corresponding blood glucose concentrations (circles) concurrently measured by use of a criterion-referenced method. Notice the lag between changes in blood glucose concentrations and changes in CGMS values.

Citation: American Journal of Veterinary Research 77, 8; 10.2460/ajvr.77.8.805

Discussion

The main objective of the study reported here was to evaluate the feasibility of multiple sensor sites for use when monitoring interstitial glucose concentrations. Identification of the best site for implantation should take into account accuracy for the site, duration of sensor life at each site, and lag time. This study found that results for a CGMS were dependent on sensor location and the specific dog in which the sensor was implanted. Values for the CGMS at the thorax site had the best correlation with blood glucose concentrations, but the sensors had the shortest functional duration. The CGMS values for the lumbar site were the least accurately correlated with blood glucose concentrations, but the sensors had the longest mean functional duration. Overall, despite differences in CGMS and criterion-referenced glucose concentrations, the general clinical utility of CGMS appeared to be acceptable, with most of the results for all sensors, sites, and dogs meeting guidelines established by the International Organization for Standardization for POC glucometerso or in zones A or B of error grid analysis and thus unlikely to alter clinical decision making or outcome.

There was considerable variability in CGMS values among multiple sensors implanted concurrently at the same site in an individual dog. Sensors were implanted as close as physically possible to each other, although not in exactly the same spot. It seemed unlikely that the short distance between sensors would have contributed substantially to variability. Sensors in experiment 1 were implanted in the lumbar location, which was later found to have more variability than at other locations, particularly the thorax. We elected to use the lumbar location for experiment 1 because of the authors' experiences, whereby sensors in the lumbar region were easily affixed and maintained for long periods. Although there was significant variability among sensors, it likely would not have been clinically relevant. Guidelineso for licensing of POC glucometers allow a difference of ± 20 mg/dL between the POC device and criterion-referenced method for blood glucose concentrations < 100 mg/dL and a difference of ± 20% for glucose concentrations > 100 mg/dL on the basis that these differences would be unlikely to alter clinical decision making. Most of the differences in the present study were within these limits.

When sensors were implanted at 4 locations in experiment 2, there was significant variation between CGMS data obtained among sites of each dog. During a euglycemic steady state, the mean CGMS value at each site was typically 3.6 to 9.4 mg/dL lower than the concentration for the criterion-referenced method, which would be acceptable on the basis of the guidelines for POC glucometerso and results for error grid analysis. Although the difference between blood glucose concentrations and CGMS values during changes in glucose concentrations was substantially greater and often outside limits indicated in the guidelines,o the difference generally was within the range of acceptability on the basis of results for error grid analysis. There was a negative bias between CGMS and criterion-referenced values at all locations, and magnitude of the bias increased with increasing blood glucose concentrations, which was similar to findings in another study2 of the use of a CGMS in diabetic dogs. During euglycemia, this variability was no more than that expected if the sensors were all at the same site. During and immediately after periods of changes in glucose concentrations as well as when the blood glucose concentrations were extremely high, the magnitude of the difference increased. Although the thorax site had the least bias at baseline, evidence for differences in bias among sites was less substantial during other phases; this may have indicated that once blood glucose concentrations began to fluctuate, problems with bias were evenly distributed among the various locations. A lower magnitude of bias would be expected during a hypoglycemic phase because the glucose concentration cannot decrease too low without clinical consequences. Oscillations in blood glucose concentrations around the target value determined for the glycemic clamp technique (± 10%) would also yield a smaller range of values for hypoglycemia than for hyperglycemia.

Several possible explanations exist for the differences between blood and interstitial glucose concentrations. First, calibrations were performed when the glucose concentrations were within the euglycemic reference range. Correlations with blood glucose values may have improved during periods of hyperglycemia if the monitor had been calibrated during hyperglycemia. Calibration during hyperglycemia was not possible in this study because the dogs were euglycemic and the monitors could not be calibrated during periods of rapid change in blood glucose concentrations because of the lag time between blood and interstitial glucose concentrations. Although the goal of the glucose clamp procedures was to maintain the blood glucose concentration within a relatively small range during a steady state for various glucose concentrations, the blood glucose concentration oscillated ± 10%, and interstitial values required time to equilibrate with changes in blood concentrations. Thickness of the dermis at each site did not appear to influence results. Most sites and dogs had similar skin thickness, and all dogs were in good body condition, so the influence of various amounts of adiposity could not be ascertained. It was possible that CGMS units would perform differently in emaciated or obese animals.

The authors are not aware of any studies conducted on humans or other animals to specifically evaluate the influence of adiposity or body condition score on CGMS accuracy, although it has been included as a variable in several broader studies. In 1 study,6 there were no differences in CGMS accuracy related to body condition score for dogs and cats with diabetic ketoacidosis; however, the spectrum of body condition scores evaluated was not reported, so its potential influence is unclear. In another study,19 lean cats required sensor replacement far more frequently than did obese cats. This may suggest that the lean cats were more active, and thus more likely to remove sensors, or that the thicker adipose layer provided a more stable fixation site. In humans, results regarding influence of adiposity on CGMS results have been mixed.20,21 In 1 study22 conducted in an intensive care unit, 3 of 3 sensors in an extremely thin elderly man failed. This accounted for half the sensor failures for the entire study, although the authors did not identify the reason that so many sensors failed in that 1 patient.

It is worthy of mention that the magnitude of variability between CGMS values obtained with multiple concurrently implanted sensors and differences between CGMS and criterion-referenced values was dependent on the specific dog. This information has implications for interpretation of the present study and other studies involving CGMS units as well as for use of CGMS units in a clinical setting. There was no obvious reason for differences among the dogs of the study reported here, but differences could have included composition of the skin attributable to effects of age or physical condition, activity, sensor implantation or fixation techniques, or hydration status. The fact that the same investigator implanted and secured all sensors in the study decreased the likelihood that these factors influenced sensor performance or functional duration.

Inflammation at the site, bioincompatibility, or biofouling (adhesion of proteins, cells, or microbes [biofilm] to a sensor) also could have contributed to sensor failure.23,24 Biofouling begins immediately after insertion of devices and is part of the reason that the manufacturers originally designated a maximum functional duration for each sensor.23,24 In this study, however, there were no clinical signs of inflammation, hemorrhage, discharge, or other abnormalities at any sensor implantation site to suggest substantial amounts of biofouling, although microscopic evaluation of the tissues was not performed. Furthermore, in 1 study25 conducted specifically to evaluate viability of 20 sensors 3 to 18 months after the stated expiration date and for longer functional durations, investigators identified that sensors remained viable for at least 7 days of use. Newer sensors are designated with longer functional lifespans.26

Investigators of other veterinary studies have suggested that there is patient-dependent variability. In 1 study2 of diabetic dogs, investigators identified that 3 of the animals had significant differences between the blood glucose concentration and CGMS values, although the authors did not speculate on the reasons for those differences. In a study6 of CGMS units in dogs and cats with diabetic ketoacidosis, the absolute percentage difference between CGMS values and blood glucose concentrations differed significantly in some patients; however, the animals were implanted with only 1 sensor at a time, and the number of animals receiving multiple (consecutive) sensors was not indicated, which makes it difficult to ascribe differences to a specific patient rather than to an individual sensor. There was apparent dysfunction of 2 sensors in experiment 1 of the present study as indicated by the fact that concordance coefficient correlations were approximately 0 for all sensor pairs.

The thorax sensors appeared to have the best correlation with results for the criterion-referenced method in all phases. This differed from results for a study5 of cats in which sensors located in the dorsal aspect of the neck performed better than those in the lateral aspect of the thorax or lateral fold of the stifle joint but was similar to results for a study27 of human surgical patients in which sensors in the shoulder region performed better than those in the proximal portion of the pelvic limb. The reason for the improved correlation between criterion-referenced and thoracic-sensor results in the dogs of the present study is unclear. We hypothesize that the blood supply overlying the muscles of respiration was more robust than the blood supply in the other regions or that increased movements at the thorax site improved regional circulation.

Lumbar sensors had the longest mean functional duration; only 1 lumbar sensor was lost, whereas half of the thorax sensors were lost before completion of the study. Mean functional duration for thorax sensors was approximately 20 hours less than that for lumbar sensors, which would be clinically important. We suspect the shortened thoracic sensor lifespan was related to movement at the implantation site as a result of typical activities, rising from a recumbent position, or rubbing of the sensors on a cage. The method to secure the sensors was used in an attempt to reduce sensor dislodgement and was ascertained to be the best available, as determined on the basis of the authors' experiences with various combinations of sensor adhesive, tape, suture, tissue glue, and bandage material. Although the dogs of the present study were confined, they were healthy and active. Functional duration of thoracic sensors may be increased in clinical patients with less activity or that are confined to a small space. Healthy diabetic animals, however, may be extremely active, and CGMS units are used in these animals in their home environments where they are free roaming, which perhaps renders the thoracic site less optimal. Because there were differences between sites, it may be desirable to use the same general implantation site each time in animals that are being monitored by use of a CGMS.

Lag time did not differ among sites. The 5-minute interval between CGMS measurements coupled with variability at each site and failure of the CGMS to achieve the same magnitude of hyperglycemia as for the criterion-referenced method after changes in glucose concentration prevented the investigators from identifying exact differences in lag time after changes in glucose concentrations. Obtaining CGMS measurements at more frequent intervals might allow calculation of lag times for the various sites.

Study limitations included the use of clinically normal dogs, which may not have fully represented interstitial conditions in diabetic patients. The rapid shifts between pharmacologically generated hyperglycemic and hypoglycemic states may have stressed the CGMS sensors more than if they had been used under typical conditions, and this may have influenced sensor performance. Also, all the dogs were of similar body condition, which precluded assessment of the influence of adiposity on sensor performance.

Results obtained for the study reported here by use of a CGMS were dependent on sensor location and the specific dog in which the sensor was implanted. Values for the CGMS at the thorax site had the best correlation with blood glucose concentrations, but thorax sensors had the shortest functional lifespan. The CGMS results for lumbar sensors had the least accurate correlation with blood glucose concentrations, but lumbar sensors had the longest functional duration. Overall, despite significant differences in the CGMS and criterion-referenced glucose concentrations, clinical utility of the CGMS appeared to be acceptable. Most of the values for all sensors, sites, and dogs met guidelines for POC glucometers and were located within zones A or B for error grid analysis; thus, they were unlikely to alter clinical decision making or outcome.

Acknowledgments

Supported by a grant from the American College of Veterinary Internal Medicine Foundation.

The first author has conducted a separate study funded by Medtronic.

The authors thank Dr. Dan Fletcher for assistance with error grid analysis and Christie Hamilton, Amanda Torres, and Dr. Steve Budsberg for technical assistance.

ABBREVIATIONS

CGMS

Continuous glucose monitoring system

IQR

Interquartile range

POC

Point of care

Footnotes

a.

Elastikon, Johnson and Johnson, Skillman, NJ.

b.

Medtronic Guardian Real Time, Minneapolis, Minn.

c.

Soft Sensor, Medtronic Inc, Minneapolis, Minn.

d.

Sensertor, Medtronic Inc, Minneapolis, Minn.

e.

Gluture, Abbott Laboratories, Abbott Park, Ill.

f.

Arrow International Inc, Reading, Pa.

g.

Heparin, Hospira, Lake Forest, Ill.

h.

i-STAT, Abaxis Inc, Union City, Calif.

i.

Glucose (Trinder) assay, Genzyme Diagnostics, Charlottetown, PE, Canada.

j.

Somatostatin, Sigma-Aldrich Corp, St Louis, Mo.

k.

Supor sterile syringe filter, Pall Life Sciences, Ann Arbor, Mich.

l.

Novolin R, Novo Nordisk, Plainsboro, NJ.

m.

Dextrose, Hospira, Lake Forest, Ill.

n.

Carelink, Medtronic Inc, Minneapolis, Minn.

o.

ISO 15197, International Organization for Standardization, Geneva, Switzerland.

References

  • 1. Metzger M, Leibowitz G, Wainstein J, et al. Reproducibility of glucose measurements using the glucose sensor. Diabetes Care 2002; 25: 11851191.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 2. Davison LJ, Slater LA, Herrtage ME, et al. Evaluation of a continuous glucose monitoring system in diabetic dogs. J Small Anim Pract 2003; 44: 435442.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 3. DeClue AE, Cohn LA, Kerl ME, et al. Use of continuous blood glucose monitoring for animals with diabetes mellitus. J Am Anim Hosp Assoc 2004; 40: 171173.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 4. Wiedmeyer CE, DeClue AE. Continuous glucose monitoring in dogs and cats. J Vet Intern Med 2008; 22: 28.

  • 5. Hafner M, Lutz TA, Reusch CE, et al. Evaluation of sensor sites for continuous glucose monitoring in cats with diabetes mellitus. J Feline Med Surg 2013; 15: 117123.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 6. Reineke EL, Fletcher DJ, King LG, et al. Accuracy of a continuous glucose monitoring system in dogs and cats with diabetic ketoacidosis. J Vet Emerg Crit Care (San Antonio) 2010; 20: 303312.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 7. Wiedmeyer CE, Johnson PJ, Cohn LA, et al. Evaluation of a continuous glucose monitoring system for use in dogs, cats, and horses. J Am Vet Med Assoc 2003; 223: 987992.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 8. Wiedmeyer CE, Johnson PJ, Cohn LA, et al. Evaluation of a continuous glucose monitoring system for use in veterinary medicine. Diabetes Technol Ther 2005; 7: 885895.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 9. Affenzeller N, Thalhammer JG, Willmann M. Home-based subcutaneous continuous glucose monitoring in 10 diabetic dogs. Vet Rec 2011; 169: 206210.

  • 10. Chico A, Vidal-Rios P, Subirà M, et al. A. The continuous glucose monitoring system is useful for detecting unrecognized hypoglycemias in patients with type 1 and type 2 diabetes but is not better than frequent capillary glucose measurements for improving metabolic control. Diabetes Care 2003; 26: 11531157.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 11. Summers LK, Clark ML, Humphreys SM, et al. The use of microdialysis to monitor rapid changes in glucose concentration. Horm Metab Res 1999; 31: 424428.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 12. National Research Council Committee for the Update of the Guide for the Care and Use of Laboratory Animals. Guide for the care and use of laboratory animals. 8th ed. Washington, DC: The National Academies Press, 2011.

    • Search Google Scholar
    • Export Citation
  • 13. Mazzaferro EM. Arterial catheterization. In: Silverstein D, Hopper K, eds. Small animal critical care medicine. 2nd ed. St Louis: Elsevier, 2015;10401043.

    • Search Google Scholar
    • Export Citation
  • 14. Wess G, Reusch C. Evaluation of five portable blood glucose meters for use in dogs. J Am Vet Med Assoc 2000; 216: 203209.

  • 15. Barnhart HX, Haber M, Song J. Overall concordance correlation coefficient for evaluating agreement among multiple observers. Biometrics 2002; 58: 10201027.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 16. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: 307310.

    • Search Google Scholar
    • Export Citation
  • 17. Clarke WL, Cox D, Gonder-Frederick LA, et al. Evaluating clinical accuracy of systems for self-monitoring of blood glucose. Diabetes Care 1987; 10: 622628.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 18. Parkes JL, Slatin SL, Pardo S, et al. A new consensus error grid to evaluate the clinical significance of inaccuracies in the measurement of blood glucose. Diabetes Care 2000; 23: 11431148.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 19. Hoenig M, Pach N, Thomaseth K, et al. Evaluation of long-term glucose homeostasis in lean and obese cats by use of continuous glucose monitoring. Am J Vet Res 2012; 73: 11001106.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 20. Derosa G, Salvadeo SA, Mereu R, et al. Continuous glucose monitoring system in free-living healthy subjects: results from a pilot study. Diabetes Technol Ther 2009; 11: 159169.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 21. Kerssen A, de Valk HW, Visser GH. Day-to-day glucose variability during pregnancy in women with type 1 diabetes mellitus: glucose profiles measured with the continuous glucose monitoring system. BJOG 2004; 111: 919924.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 22. Goldberg PA, Siegel MD, Russell RR, et al. Experience with the continuous glucose monitoring system in a medical intensive care unit. Diabetes Technol Ther 2004; 6: 339347.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 23. Wisniewski N, Moussy F, Reichert WM. Characterization of implantable biosensor membrane biofouling. Fresenius J Anal Chem 2000; 366: 611621.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 24. Li CM, Dong H, Cao X, et al. Implantable electrochemical sensors for biomedical and clinical applications: progress, problems, and future possibilities. Curr Med Chem 2007; 14: 937951.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 25. Chlup R, Jelenová D, Chlupová K, et al. Function and accuracy of glucose sensors beyond their stated expiry date. Diabetes Technol Ther 2006; 8: 495504.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • 26. Enlite Glucose Sensor user's guide. MP6025676-2AF1/A. Northridge, Calif: Medtronic Inc, 2014.

  • 27. Vriesendorp TM, Devries JH, Holleman F, et al. The use of two continuous glucose sensors during and after surgery. Diabetes Technol Ther 2005; 7: 315322.

    • Crossref
    • Search Google Scholar
    • Export Citation

Contributor Notes

Dr. Hoenig's present address is Department of Clinical Veterinary Medicine, College of Veterinary Medicine, University of Illinois, Urbana, IL 61802.

Address correspondence to Dr. Koenig (akoenig@uga.edu).