Amperometric Glucose Sensors

8/6/2019 Amperometric Glucose Sensors

1/5

221

Amperometric Glucose Sensors:

Sources of Error and Potential Benet of Redundancy

Jessica R. Castle, M.D., and W. Kenneth Ward, M.D.

Author Afliation: Oregon Health Services University, Portland, Oregon

Abbreviations: (ARD) absolute relative difference, (NASA) National Aeronautics and Space Administration, (NIH) National Institutes of Health,(ZMAD) Z score with median absolute deviation

Keywords: accuracy, closed-loop system, multiple, redundancy, sensor

Corresponding Author: W. Kenneth Ward, M.D., Mail Code OP05, 3181 SW Sam Jackson Park Road, Portland, OR 97239; email address [email protected]

Journal of Diabetes Science and Technology

Volume 4, Issue 1, January 2010 Diabetes Technology Society

Abstract

Amperometric glucose sensors have advanced the care of patients with diabetes and are being studied tocontrol insulin delivery in the research setting. However, at times, currently available sensors demonstratesuboptimal accuracy, which can result from calibration error, sensor drift, or lag. Inaccuracy can be particularlyproblematic in a closed-loop glycemic control system. In such a system, the use of two sensors allows selection of

the more accurate sensor as the input to the controller. In our studies in subjects with type 1 diabetes, theaccuracy of the better of two sensors signicantly exceeded the accuracy of a single, randomly selected sensor.If an array with three or more sensors were available, it would likely allow even better accuracy with the useof voting.

J Diabetes Sci Technol 2010;4(1):221-225

Introduction

The commercialization of amperometric glucose sensorshas signicantly impacted the treatment of type 1 diabetes.Glucose sensors are now widely available as small,minimally invasive devices that measure interstitialglucose levels in subcutaneous fat. The ability tomeasure glucose levels every several minutes has notonly improved the treatment of type 1 diabetes,1,2 it hasalso launched research efforts into methods of automatedglycemic management. The accuracy of present-daysensors, while generally good, remains imperfect.For this reason, the U.S. Food and Drug Administration

does not allow the use of sensor data to make criticaldecisions regarding diabetes management withoutconrmation from standard capillary blood glucosereadings. The success of closed-loop systems, which utilizesensor values in the determination of insulin delivery

rates, will be improved by further enhancements ofsensor accuracy. This commentary discusses the knowncauses of sensor inaccuracy and how the use of sensorredundancy might improve accuracy.

Sources of Sensor Inaccuracy

There are three major brands of glucose sensors on theU.S. market today: DexCom SEVEN Plus, MedtronicGuardian , and FreeStyle Navigator. Although one studyfound substantial differences in accuracy among these

devices,3

later studies using updated versions of thesedevices found similar accuracy with each having meanabsolute relative differences (ARD) of approximately15%.4,5 The fact that median ARD values are usuallylower than mean ARD values underscores the nding

DEVELOPMENTS inContinuous Glucose Monitoring


2/5

222

Amperometric Glucose Sensors: Sources of Error and Potential Benet of Redundancy Castle

www.journalofdst.orgJ Diabetes Sci Technol Vol 4, Issue 1, January 2010

that a small number of sensors have very high ARDvalues, indicating substantial inaccuracy. Two situationsof particular clinical importance include the abilityof a sensor to (a) function well during hypoglycemiaand (b) avoid overestimating glucose, which could leadto excessive insulin delivery and the development ofhypoglycemia. Factors that may affect sensor accuracy

adversely are numerous and include calibration error,sensor delay, and sensor drift. Each is addressed here.

Calibration Error

Amperometric glucose sensors estimate interstitial glucosevalues by measuring an electrical current generated bythe reaction of glucose either with oxygen or with animmobilized redox mediator. Reference capillary bloodglucose is input into the system for the purpose ofcalibration, which includes quantication of the sensitivityof the sensor to glucose. After calibration, the sensor

current is used in conjunction with the sensitivity toestimate glucose values arithmetically. Not all of thecurrent generated by the glucose sensor is specic toglucose. Typically, there is a small background currentthat is unrelated to the glucose level. This backgroundcurrent must be subtracted from the total current duringcalibration and during the process of estimating eachunknown glucose level.

Glucose sensors require calibration at regular intervals,and accurate calibration is critical to sensor accuracy.The DexCom and Medtronic devices require calibration at

least every 12 hours. The FreeStyle Navigator has a morecomplex calibration scheme. More frequent calibration ofthese devices likely improves accuracy, although there isa paucity of published data on this topic. The accuracyof the reference glucose measurement method,6 the rateof glucose change during calibration, and the accuracyof the background current estimation all impact sensoraccuracy and are potential sources of calibration error.The presumption of linearity in the setting of a nonlinearsensor response to glucose also adds to calibration error.In such a case, if calibration occurs at a low glucose level, but measurement of glucose takes place at a high level,the glucose value will be underestimated. Overestimationwill result when calibration is carried out at high glucosevalues in the non-linear range and measurement takes placeat lower values. The magnitude of error is greatest whenthe glucose level changes markedly between the timeof calibration and the time of glucose measurement. Thiserror magnication during marked glucose level changeis also seen when the background current is estimatedincorrectly.

Sensor Delay

Sensor delay is another important source of sensorinaccuracy and refers to the lag of the sensed interstitialglucose values behind blood glucose values. A small partof sensor delay is physiologic and accounts for the time ittakes for glucose in the intravascular compartment toequilibrate with the interstitial compartment. Physiologicdelay is now thought to be short and probably less than5 minutes.7 The delay inherent to the sensor itself andto data ltering and smoothing techniques, dependingon the particular system, can be of greater magnitudeThe time required for glucose to diffuse across outersensor membranes into the enzyme layer and for hydrogenperoxide to reach the indicating electrode, when an innermembrane is present, contribute to sensor delay. If a sensoris calibrated during a rapid change in glycemia, theeffect of sensor delay, if not accounted for, is magnied.

Sensor DriftAlthough sensor drift is, to some extent, universal, it ispoorly understood. In an attempt to better understandthe phenomenon of drift and oscillation, we studiedfully implantable subcutaneous sensor arrays, with eacharray having multiple indicating electrodes. During thesestudies in rabbits, half of the indicating electrodes werecoated with glucose oxidase and the other half wereuncoated. Over time, there was substantial drift of thecoated electrodes, which also demonstrated electricacurrent oscillations with several dominant frequenciesInterestingly, in the uncoated electrodes, the output

of which was only a background current, there was verylittle drift and very little oscillation.8 Our interpretationof the ndings was guided by the fact that theenzyme-coated electrodes, by design, respond to glucoseand oxygen. The uncoated electrodes do not respondto either compound, but do respond to extraneousnonphysiologic interference. Because oscillation anddrift were seen almost exclusively in the enzyme-coatedelectrodes, we believe that changes in blood ow andoxygen provision, not extraneous interference, werelikely to be a major source of the oscillations. It is knownthat there are regular cycles of vasoconstriction andvasodilation in the skin and subcutaneous tissue thatparticipate in thermoregulation.9 The role of changes invascular tone on sensor current was studied by Gilliganand colleagues.10 They found that administration of avasoconstrictor drug reduced sensor output, despite theabsence of a glucose change. Similarly, in unpublished datawe have observed that direct pressure over the site ofan implanted sensor leads to a transient loss of currentlikely due to a temporary reduction in blood ow.


3/5

223



impact of a malfunctioning unit can be minimizedThe occurrence of severe hypoglycemia in a person withdiabetes is analogous to failure of the flight controlsystem of a NASA spacecraft; both are life-threatening.

The use of two or more sensors in a closed-loop system isattractive for several reasons. From a practical standpoint

it provides a reserve device for cases of sensor ortelemetry failure. In addition, depending on the specicsensor, when a new sensor is inserted, it must be inplace for at least 2 to 10 hours before calibration canproceed and sensed glucose readings are provided. If theonly sensor in place fails, one must wait 2 to 10 hoursfor the new sensor to stabilize.

The concurrent use of two sensors in an individual alsoincreases the chance of at least one being highly accurateWe routinely use two sensors concurrently in ourclosed-loop studies and choose the more accurate one toprovide input to the controller. From recent closed-loopdata, we prepared two gures. These gures illustratesensor data in different individuals with type 1 diabeteseach of whom wore two sensors. In both gures, thesensor is calibrated only once, at the beginning of thestudy. Figure 1 represents a situation in which the twosensors track each other very closely over the 9-hourperiod. The situation in Figure 2 is quite different. In thiscase, sensor 2 (the higher tracing) tracked blood glucosequite well and had a mean ARD of 11.7%, whereassensor 1 functioned poorly with a mean ARD of 24.8%

In a closed-loop system, this degree of inaccuracy wouldaffect the rate of insulin delivery substantially. If the lessaccurate sensor were used in this case, it would haveled to inadequate delivery of insulin with subsequenthyperglycemia. Fortunately, sensor 2 was used for thevast majority of this study to control the insulin deliveryrate, and good glycemic control was achieved.

There are several potential ways of using data frommore than one sensor, and this topic has not been wellstudied. Using the average of two signals is not always theoptimal method because if one sensor is performing very

poorly, the average is also inaccurate to some extentAnother option is to compare the two sensor signalsand avoid using sensor data when the two signals arediscrepant beyond a specied criterion. This method wasexamined by Schmidtke and colleagues16 several yearsago. This technique signicantly improved the numberof glucose readings in regions A and B of the Clarkeerror grid from 92.4 to 98.8%. Of course, in a closed-loopsetting, the choice to avoid using either sensor deprivesthe controller of afferent input.

It is likely that the presence of a foreign body createsa state of heightened inammation. After sensor insertion,the measurement medium is not normal interstitial uid, but rather interstitial uid enriched with inammatorycells, cytokines, and mediators. During the entire dwelltime of a foreign body, macrophages are present andsecrete many cytokines and signaling compounds.

For the rst few days, up to one week, there are alsoacute inammatory cells such as neutrophils andeosinophils. Although not well studied, it is likelythat some of these cells consume oxygen and glucoseand produce compounds such as hydrogen peroxide.For these reasons, a foreign implant can perturb theapparent local glucose concentration in interstitial uid.In the rst several hours after sensor insertion, thesignal is often unstable and more likely to be inaccurate.At later time points, collagen and other proteins aresecreted into the extracellular matrix, due in part to theeffect of transforming growth factor 1 and connectivetissue growth factor,11 which lead to a reduced diffusionof glucose into the sensor.1214

Nitric oxide may also play a key role. For example, Giffordand associates15 observed that changes in nitric oxideappear to affect sensor output and affect the magnitudeof current drift. Although not yet peer reviewed, theKreutzer group at the University of Connecticut reporteda very interesting nding regarding mast cells at the2009 American Diabetes Association scientic session.They presented data showing that mast cells, and released

histamine, appeared to have a direct effect of reducingsensor current over time. When mast cells were stabilizedwith cromolyn, the loss of current was decreased.This nding may contribute to the phenomenon of biofouling, the deposition of material on the sensorsurface, which tends to reduce sensor output over time.

Use of Multiple Sensing Units to ImproveAccuracy

The causes and mitigations of suboptimal accuracy arecomplex and require further study. We address here

the question of whether the use of two or more sensorsyields better accuracy than one.

The concept of redundancy is not new. It has long beenused in settings in which failure could be catastrophic.For example, many computer systems on NationalAeronautics and Space Administration (NASA) spacecraftsare redundant and employ the concept of voting.Voting algorithms process multiple data streams andreject discrepant or outlying data. In this manner, the


4/5

224



In order to address possible advantages of sensorredundancy, we examined the last year of data from ourhuman closed-loop study. Given this study has not yet been completed and data have not yet been submitted forpeer review, our ndings are presented in general form.During each study, subjects with type 1 diabetes woretwo sensors for 28 hours. Early in the course of the study,

the more accurate of the two sensors was selected for thecontrol of insulin and glucagon delivery. The accuracyof both sensors was followed for the duration of the studyand was compared to venous glucose performed every10 minutes on the highly accurate HemoCue analyzer.Each sensor was calibrated at the start of the studyand again 4 hours later. The selected sensor had amean ARD of 14%, which was signicantly lower whencompared to the unselected sensor, with a mean ARD of19%. In approximately 65% of subjects, the mean ARD ofthe two sensors were within four mean ARD percentagepoints of each other and close to 75% were withinseven points of each other. Thus, in approximately 25%of cases, there was a large discrepancy in accuracythat exceeded seven mean ARD percentage points. Theaccuracy difference in this group of individuals wouldhave led to substantial differences in the amounts ofinsulin and glucagon given by the algorithm, dependingon which sensor was used. These ndings support theuse of two sensors in settings where accuracy is critical.We also found that in over 80% of cases, the sensor thatwas selected early continued to be the more accurate ofthe two throughout the remainder of the 28-hour study.

In the small number of cases in which this was not thecase, the overall difference in accuracy between selectedand unselected sensors was less than 1.5 mean ARDpercentage points.

Although wearing three or more sensors is impracticalwith current sensor technology, sensor arrays withmultiple sensing units may be available in the future.The advantage of three or more sensors lies in the abilityto vote out data from one or more sensors when thereading is discrepant from the others. This techniqueis based on the fact that sensor signals that are quite

similar to others in the array are usually almost alwaysmore accurate than outliers. In animals, we tested sucha technique using a statistical technique termed theZ score with median absolute deviation (ZMAD), whichis based on median statistics. The subcutaneous sensorarrays contained four sensing units, each with its ownplatinum indicating electrode and its own telemetricchannel. The Ag/AgCl reference electrode was sharedamong the four sensing units. A Z score for each sensingunit was calculated every several minutes. This score

Figure 1. Two subcutaneous sensors in a person with type 1 diabetesundergoing closed-loop control. Blue symbols indicate data fromsensor 1, and red symbols indicate data from sensor 2. Note thatthe two sensors track each other very well; the tracings are nearlysuperimposable. Sensors were calibrated once at the onset of thestudy.

Figure 2. Two subcutaneous sensors in a person with type 1 diabetesundergoing closed-loop control. Blue symbols indicate data fromsensor 1, and red symbols indicate data from sensor 2. Note that thetwo sensors track each other poorly. Sensor 1 registered low glucosevalues at times that sensor 2 and reference glucose were normalSensors were calibrated once at the onset of the study.

compared how much the sensed glucose from one unitdeviated from the median sensed glucose of all unitsIf the Z score rose above a predetermined thresholdthen data from that unit were considered an outlier andwere therefore automatically excluded. This technique

can be employed in real time. During long-termimplantations, sensor array accuracy using the ZMADtechnique was signicantly and substantially better thanusing an average of the sensed glucose values.17

The use of redundant sensors addresses sensor drift, butdoes not address sensor delay. Delay, which is inuenced by physiologic elements, inherent sensor properties, andata ltering, is expected to affect all sensors similarlyIn a similar fashion, if all sensors in an array are similarly


5/5

Amperometric Glucose Sensors

Documents

Transcript of Amperometric Glucose Sensors