Accuracy of the Enlite 6-Day Glucose Sensor with Guardian and Veo Calibration Algorithms

Abstract

Objective:

This study investigates the accuracy of a newly developed, next-generation subcutaneous glucose sensor, evaluated for 6-day use.

Research Design and Methods:

Seventy-nine subjects (53 men, 26 women) with type 1 diabetes and 18 subjects (14 men, four women) with type 2 diabetes completed a three-center, prospective, sensor accuracy study. The mean age for the group was 42.2±15.0 years (mean±SD), ranging from 18 to 71 years, with a mean glycosylated hemoglobin level of 7.6±1.5%, ranging from 5.5% to 14%. Subjects wore Enlite^™ sensors (Medtronic Diabetes, Northridge, CA) in the abdominal and buttocks region for two separate 7-day periods and calibrated with a home-use blood glucose meter. Subjects participated in an in-clinic testing day where frequent sampled plasma glucose samples were acquired every 15 min for 10 h. Sensor data was retrospectively processed with Guardian^® REAL-Time (Medtronic) and Paradigm^® Veo^™ (Medtronic) calibration routines, and accuracy metrics were calculated for each algorithm and sensor location. Physiological time lag for each measurement site was calculated.

Results:

Based on 6,404 plasma–sensor glucose paired points, the Enlite sensor with Veo calibration algorithm produced a mean absolute relative difference of 13.86% with 97.3% of points within the A+B zones of the Clarke error grid. Threshold-only alarms detected 90.1% of hypoglycemia and 90% of hyperglycemia. Mean time lag measured at the abdominal region was 7.94±6.48 min compared with 11.70±6.71 min (P<0.0001) at the buttocks area.

Conclusions:

The Enlite sensor accurately measures glucose when compared with gold standard laboratory measurements over its 6-day use. Sensors placed in the buttocks region exhibited greater time lags than sensors placed in the abdomen.

Introduction

C ontinuous glucose monitoring (CGM) devices that are calibrated by self-monitoring of blood glucose (BG) provide trend information and enable more informative treatment decisions for patients with diabetes,¹ which can lead to improved glucose regulation.² Noted improvements in BG control have been demonstrated with the use of stand-alone CGM devices in several clinical studies.^3,4 In particular, adjunct therapy with integrated systems that combine CGM and continuous subcutaneous insulin infusion has been shown to significantly reduce glycosylated hemoglobin (A1C) levels compared with stand-alone therapy with continuous subcutaneous insulin infusion⁵ or multiple daily injections.^6

–11 The RealTrend⁶ and Eurythmics⁷ trials both demonstrated improvements in A1C of >1%, whereas the landmark STAR 3 Trial,⁹ the largest sensor-augmented pump therapy study to date (n=485), demonstrated an A1C reduction of approximately 1%, compared with multiple daily injections. A direct relationship was identified between sensor-augmented pump therapy efficacy and consistency of sensor wear.⁶

The accuracy and reliability of CGM devices are essential for effective hypo- and hyperglycemia detection, but also paramount in the realization of a closed-loop artificial pancreas (AP) system.¹⁰ In a recent closed-loop study,¹¹ the Cambridge AP group demonstrated tight glycemic control by using the Guardian^® REAL-Time (Medtronic Diabetes, Northridge, CA) and Freestyle Navigator^® (Abbott Diabetes, Alameda, CA) monitors for the CGM input to their control algorithm, showing sensor accuracies of 10% and 12%, respectively, when calibrated every 6 h. Similarly, the Oregon AP group, in a closed-loop study using the Guardian REAL-Time and DexCom (San Diego, CA) SEVEN^® Plus, showed a combined mean absolute relative difference (MARD) of 8.7%¹² with approximate 6-h calibrations. In all cases, monitors were calibrated with laboratory standard analyzer samples with the exception of the Navigator, which relied on capillary fingerstick measurement.

A progressive improvement in the accuracy of sensors measuring interstitial fluid glucose is evident, with significant performance advancements demonstrated by each of the three manufacturers' of percutaneous-based sensors and CGM systems.^13
–15 The performance of the Enlite^™ sensor, the latest in Medtronic Diabetes subcutaneous glucose sensing technology, is assessed in this study. The sensor will be used by both stand-alone CGM and sensor-augmented pump therapy platforms.

The primary end point for this study is to measure sensor accuracy across 7 days of continuous sensor wear. We performed an analysis of the dataset to recalibrate the study data with the Guardian and Paradigm^® Veo^™ (Medtronic) real-time calibration algorithms, similar to previous analyses,¹⁶ to demonstrate the accuracy of the sensor with both calibration routines.

Research Design and Methods

The study was approved by a central institutional review board for each of the three centers, and consent forms were signed by each participant. In total, 100 subjects were enrolled in this multicenter, prospective, single-sample study composed of 69 male and 31 female subjects with an age range of 18–71 years (42.2±15.0 years old). Two subjects withdrew, and 98 completed all study phases. Eighteen of the cohort were insulin-using type 2 diabetes subjects, and the average group A1C for the group was 7.6±1.5%.

The Enlite 6-day sensor

The Enlite glucose sensor is labeled for 6-day use and is based on enzymatic technology.¹⁷ It is an amperometric device consisting of three electrodes plated on a flexible substrate, where a constant voltage potential is maintained across working and reference electrodes to support an optimal electrochemical reaction. During this reaction process glucose is oxidized to produce hydrogen peroxide. The enzyme glucose oxidase is spread across the electrodes and acts as the catalyst that generates the reaction between oxygen and glucose. One electron is released per glucose and oxygen molecule, equalized by a glucose-limiting membrane, where the current produced is proportional to the interstitial fluid glucose concentration.¹⁸

The Enlite sensor components include sensor, introducer needle, and adhesive tape. The sensor is housed in tubing with minimal area to diminish any discomfort over long periods of sensor wear of at least 1 week. The introducer inserts the needle and sensor into the subcutaneous tissue, and the needle is automatically withdrawn with a retracting mechanism to ensure no accidental sticks and safe removal of the needle. The sensor connects to a MiniLink^® (Medtronic) transmitter with waterproof connectors.

Study design

Subjects wore two Enlite sensors—one in the abdomen and one in the buttocks. A CGMS^® iPro^™ (Medtronic) was used to record data from sensors placed in the buttocks region, and the Guardian REAL-Time recorded abdominal data. Subjects were required to test their BG levels at least four times daily with the provided study meter, the OneTouch^® UltraLink^™ (LifeScan Inc., Milpitas, CA). Fingerstick BG samples were transmitted over a radiofrequency link to the Guardian monitor for calibration. Participants wore both systems for two separate 7-day study periods and on one occasion were required to undergo an in-clinic testing phase. Although the study consisted of in-clinic and at-home phases, the primary end point is based on the in-clinic phase, which is solely presented herein.

Each subject participated in one 10-h, in-clinic frequent sampling day following enrollment occurring during either the first or second study period. The in-clinic visit occurred on one of days 1–7 of sensor life with an equal number of subjects tested on each day. Subjects fasted for 8 h prior to the morning of the in-clinic visit, at which time an intravenous line was inserted and frequent blood samples were taken every 15 min for 10 h. Each blood sample was centrifuged, and plasma glucose was measured using the YSI 2300 STAT Plus^™ glucose and lactate analyzer (YSI Inc., Yellow Springs, OH). Values were recorded using a computer connected to the YSI with specialized software developed for this purpose. Glucose and insulin challenges were performed in order to provide a sufficient glucose range for evaluation of sensor performance in the hypo- and hyperglycemic ranges. Glucose was allowed to drop to 70 mg/dL, at which point a meal was provided following a confirmatory YSI reading. The use of additional insulin at the discretion of the investigator was optional to assist in ensuring the low target was achieved. Similarly, during a meal response, glucose was allowed to reach 250 mg/dL. An example of one sensor profile for the 10-h frequent sampling period is shown in Figure 1.

FIG. 1.

Example of a sensor tracing for a glucose and insulin challenge with calculated total time lag of 7 min—3.5 min for system group delay plus 3.5 min of physiological time lag.

Data processing and analysis

Performance metrics were calculated based on the Clinical and Laboratory Standards Institute's metric for continuous interstitial glucose monitoring¹⁹ to assess sensor accuracy for days 1–7 of each in-clinic frequent sample visit. The Clinical and Laboratory Standards Institute standards define an “event” as two YSI readings within a 30-min time window that exceed prespecified thresholds. We define hypoglycemic and hyperglycemic thresholds as less than 70 mg/dL and greater than 240 mg/dL, respectively. The criterion for sensing accuracy is based on the International Organization for Standardization guideline for home use BG meters,²⁰ which considers readings within 15 mg/dL of reference glucose values in the hypoglycemic range as accurate. Similarly, sensor readings within 20% of reference glucose values in the hyperglycemic range are also considered accurate.

Each sensor download was processed separately with Guardian and Veo real-time calibration algorithms. The unprocessed sensor waveforms are calibrated with each monitor routine to simulate the real-time respective device output. Comparisons are possible because all devices share the same data acquisition hardware and preprocessing techniques prior to calibration.²¹ Clinical accuracy is assessed through consensus,²² Clarke (CEG),²³ and continuous²⁴ error grid analyses. The delay for each site (abdomen and buttocks) was assessed by comparing plasma glucose values with raw sensor current using a correlation-based method.

Results

In total, 98 subjects completed the study, of which 83 sensor sets were used for analysis, which had a complete set of reference blood samples captured during the in-clinic phase. This produced 6,404 paired reference YSI and sensor evaluation points.

Performance evaluation

The performance of each calibration routine is presented in Table 1 for the entire dataset including both measurement sites. MARDs are calculated for 6,404 YSI plasma–sensor glucose paired points stratified by sensor day and glucose range. The aggregatedsensor errors for the entire glucose range and experiment duration are 13.9% and 13.8% for Veo and Guardian algorithms, respectively. The greatest distinction between algorithm performances can be seen in the low glucose range (40–80 mg/dL). Here, the Veo algorithm demonstrated superior performance with an aggregated difference in performance of 3.5% across the 7 days and of >7% on day 1. The Guardian algorithm produced MARDs of 13.1% and 14.6% for sensors placed in the buttocks and abdominal regions, respectively, and the Veo generated errors of 12.9% and 14.7%, correspondingly, for both sites. In all cases the differences in the results were statistically insignificant.

Table 1.

Mean Absolute Relative Difference Stratified by Day and Range

Glucose (mg/dL)	Total paired readings	Day 1 (n=941)	Day 2 (n=928)	Day 3 (n=1,074)	Day 4 (n=762)	Day 5 (n=1,055)	Day 6 (n=907)	Day 7 (n=737)	Overall (n=6,404)	Data ^a
40–80	728	20.0	17.8	14.3	18.4	19.6	23.9	16.1	18.4	V
		27.2	22.3	16.5	16.6	22.5	27.2	19.8	21.9	G
81–120	1,446	19.3	16.3	10.1	13.4	13.4	13.0	13.6	13.7	V
		21.0	16.5	9.8	11.9	11.5	14.0	12.0	13.3	G
121–240	3,086	13.0	15.2	9.2	12.2	14.9	11.8	14.3	12.9	V
		13.9	14.1	8.6	11.8	13.2	11.8	14.0	12.4	G
240–400	1,140	12.7	15.9	11.3	11.3	15.7	13.3	15.9	13.9	V
		11.9	14.9	11.7	11.2	14.6	12.1	14.6	13.1	G
Overall	6,404	14.9	15.8	10.3	13.0	15.2	13.4	14.8	13.9	V
		16.4	15.7	10.2	12.2	14.1	13.8	14.7	13.8	G

The upper value indicates Veo accuracy (V), and the lower value represents the Guardian Real-Time algorithm (G) for the full dataset including both sites.

Consensus error grid²² and CEG²³ error grid analyses of paired evaluation and reference points were performed. CEG results are presented in Table 2 demonstrating the clinical accuracy and significance of the Enlite sensor with Veo and Guardian algorithms. The clinically accurate A+B zones for the CEG contained 97.3% and 96.2% of the evaluation points for the Veo and Guardian algorithms, respectively. The consensus grid A+B zones included 99.1% and 98.9% of all points for Veo and Guardian algorithms, respectively. The performance of each algorithm is comparable based on this analysis except in the low glucose range of the CEG, where the Veo algorithm showed a 10% increase in the number of points in the clinically accurate zones and 10% less in Zone D. Overall, 0.1% of all points reside in Zone E of the CEG for both algorithms, which could potentially lead to incorrect treatment decisions for hypo- and hyperglycemia. In all cases no points existed in Zone D or E of the consensus error grid.

Table 2.

Clarke Error Grid Analysis

		Zone
Glucose (mg/dL)	Total paired readings	A+B	A	B	C	D	E	Data ^a
40–80	721	90.4	73.8	16.6	0.0	9.6	0.0	V
		79.8	61.3	18.4	0.0	20.2	0.0	G
81–120	1,446	100.0	78.6	21.4	0.0	0.0	0.0	V
		99.8	79.4	20.4	0.2	0.0	0.0	G
121–240	3,086	99.3	80.5	18.7	0.5	0.0	0.3	V
		99.3	81.1	18.2	0.4	0.0	0.3	G
240–400	1,140	92.8	75.2	17.6	0.8	6.4	0.0	V
		93.6	78.4	15.2	0.7	5.7	0.0	G
Overall	6,404	97.3	78.4	18.9	0.4	2.2	0.1	V
		96.2	78.0	18.2	0.4	3.3	0.1	G

The upper value indicates Veo accuracy (V), and the lower value represents the Guardian Real-Time algorithm (G) for the full dataset including both sites.

A continuous glucose error grid analysis²⁴ for the comprehensive dataset processed using the Veo calibration algorithm is illustrated in Figure 2. The percentages of points occupying the clinically accurate A and benign B zones are 80.3% and 17.2%, respectively. Zone C has 0.3% and Zones D and E include 2% and 0.1% of the total points, respectively.

FIG. 2.

Continuous glucose error grid analysis with the Veo algorithm for both sites.

A sensitivity and specificity analysis based on the Clinical and Laboratory Standards Institute¹⁹ standards for CGM and International Organization for Standardization²⁰ standards for accuracy for self-monitoring of blood BG was performed for threshold alert settings of <70 mg/dL for hypoglycemia and >240 mg/dL for hyperglycemia. The analysis demonstrated detection rates of 90.1% and 79.2% (P=0.00012) for hypoglycemic events (n=80) and 90% and 90.1% (P=0.13) for hyperglycemic episodes (n=150) when applying the Veo and Guardian algorithms, respectively. The positive predictive value for each test was 83.8% (Veo) and 88.7% (Guardian) for hypoglycemia and 98.5% (Veo) and 97.3% (Guardian) for hyperglycemia.

Sensor reliability

The Enlite sensor is intended for 6 days (144 h) of continuous use; however, we evaluated sensor accuracy and reliability over 7 days. We define the sensor survival rate as the time duration the sensor exhibits full functionality from the first calibration to the end of sensor life. A total of 84.6% of sensors operated beyond the 6-day requirement with a mean functional lifetime of 152.7 h (median, 165.8 h) over the duration of the study. In this study 419 sensors were worn, where based on subject reporting from each clinical site 369 sensors remained correctly attached at the sensing location for 6 days of use. At the study end or after 6 days of sensor wear, 86.4% of sensors were functioning, with only 21 sensors having failed during initialization, which were identified at the time of the first calibration.

Time lag

The system time lag for each site was calculated by means of cross-correlation for 69 (n=138 sensors) paired sensor downloads. The mean and interquartile range of Pearson product-moment correlation coefficients was r=0.98±0.019 and r=0.98±0.021 with mean and SEs of 3.60±2.30 and 5±2.66 for abdominal and buttocks regions, respectively. Sensor current is processed by a 7^th-order finite impulse response low-pass filter²⁴ that introduces a group delay of 3.5 min, which is correlated with YSI plasma glucose samples using biased estimates. A mean time lag of 7.94±6.48 min (95% confidence interval, 6.38, 9.50 min) was estimated for the abdominal site compared with a 11.70±6.71 min (95% confidence interval, 10.08, 13.31 min) time lag estimated for sensors placed in the buttocks region. The sensor time lags for the abdomen and the buttocks were significantly different (P<0.0001).

Discussion

The most noteworthy aspect of this accuracy study is the high degree of correlation between the Enlite sensor and gold standard laboratory YSI reference samples, achieving extremely low SE measurements. The Enlite sensor is robust and accurate over 6 days of continuous sensor use and effective throughout the 40–400 mg/dL glucose range. The sensor is least accurate during the first day of operation, likely because of sensor initialization and run-in effects common at startup with enzyme-based in vivo glucose sensing. Overall, the sensors are very accurate with a mean error of <14%. Although the aggregate errorwas approximately the same for both calibration methods, the Veo algorithm demonstrated superior performance in the low glucose range. The additional accuracy observed at low glucose levels is not always reflected in the total error because of the disproportion in the distribution of evaluation pointsat lower ranges. However, this increased accuracy in the lower glucose range is critical for detecting hypoglycemic events and central to effective diabetes care through CGM.

Each error grid analysis resulted in highly populated A+B zones, demonstrating the clinical accuracy and utility of the Enlite sensor for CGM-based adjunct therapy. In particular, the accuracy of the Enlite sensor in combination with the Veo algorithm demonstrated increased precision with 97.3% and 99.1% of paired points in the clinically accurate and benign zones of the CEG and consensus error grid, respectively. It is interesting that although the CEG zones were comparable for both algorithms, the Veo algorithm achieved an additional 12.5% paired points in Zone A for the 40–80 mg/dL range and 10.6% less points in Zone D. Reliability was significantly improved as points in Zone D represent a potential failure to correctly identify hypo- or hyperglycemic events. The difference between the algorithms is more apparent in the hypoglycemic range because of the increased sensitivity of the Veo algorithm at low glucose values, demonstrating a statistically significant improvement of >10% in detecting hypoglycemia. Continuous glucose error grid analysis revealed similar accuracies with 97.5% of points in Zones A+B, signifying effective dynamics in tracking BG, with only 0.3% of points occupying Zone C, which suggests a potential for slight overcorrection in treatment. The 2% of points in Zone D indicate a lack of sensor responsiveness to fast BG divergence, which could be related to time lag. The erroneous Zone E accounted for only 0.1% of points that could potentially lead to incorrect treatment. Therefore 97.5% of points reside in Zones A+B zone, which would result in effective or no treatment. The 2% of points in Zone D signifies a failure to detect rapid rates of change, which could suggest a loss of sensor sensitivity or responsiveness to changes in glucose concentration. Only 0.1% of points make up Zone E, indicating erroneous behavior with the sensor exhibiting reverse dynamics with sensor glucose digressing contrary to plasma glucose.

Clearly the Enlite sensor is extremely sensitive to hypoglycemia. The positive predictive value rate, also known as the precision rate, provides a measure of reliability where higher rates will generally lead to increased patient satisfaction. Often a trade-off exists between increased sensitivity and the incident of false alerts. The high positive predictive values using either the Guardian Real-Time or Veo calibrations reflect an accurate detection rate with correspondingly few false-positives.

We used a cross-correlation approach to estimate time lag as the sensor and plasma glucose time series were very highly correlated. We believe that the correlation approach yielded an accurate representation of true time lag encompassing filter delay, sensor diffusion, and physiological time lag. This analysis excluded capillary fingerstick measurements whose inherent BG meter bias and variability can lead to ambiguity with error-based methods. This method is less sensitive to sensor offset, where offset can appear as an additive time lag. The difference in mean physiological time lag between the two sites is just under 4 min (approximately 8 min for the abdomen and 11 min for the buttocks). It is interesting that sensors placed in the buttocks region outperformed the abdominal sensors by an average of 1.45% and 1.76% when calibrated with the Guardian and Veo algorithms, respectively. However, based on the correlation analysis, both locations had approximately the same correlation and SE for the subset used for the analysis. It is possible that the difference in mean time lag contributed to performance differences based on a fixed time delay in the pairing of calibration BG and sensor samples during the calibration process.

Current CGM devices are calibrated by capillary fingerstick measurements with BG meters. It is likely that most CGM errors are incurred through the calibration process (i.e., meter error, human error, fixed time lag parameters, and offsets from background current). This implies that, with additional calibration enhancements tailored to the new sensor dynamics, increased performance is achievable. In the future, factory calibrations will be possible that will challenge fingerstick accuracy and potentially BG meters, thereby removing the common occurrence of human error.

Footnotes

Acknowledgments

The authors would like to thank Pratik Agrawal (Medtronic) for analysis of sensor data, in addition to Bob Janowski (Medtronic) and the Clinical Research Department at Medtronic Diabetes.

Author Disclosure Statement

D.B.K., J.J.M., K.A.C., G.R., S.W.L., J.Y. and R.V.S. are employees of Medtronic Diabetes. H.Z. is an employee of Samsung Diabetes Research Center and received funding from Medtronic. T.B. is an employee of the University of California San Diego School of Medicine and has received research grants from Medtronic, Abbott, DexCom, and Bayer. R.L.B. is an employee of Rainer Clinical Research Center, Inc. with no conflicts.

References

Klonoff

. Continuous glucose monitoring: roadmap for 21^st century diabetes therapy. Diabetes Care, 2005; 28:1231–1239.

Juvenile Diabetes Research Foundation Continuous Glucose Monitoring Study Group. Continuous glucose monitoring and intensive treatment of type 1 diabetes. N Engl J Med, 2008; 359:1464–1476.

Deiss

, Bolinder

, Riveline

, Battelino

, Bosi

, Tubiana-Rufi

, Kerr

, Phillip

. Improved glycemic control in poorly controlled patients with type 1 diabetes using real time continuous glucose monitoring. Diabetes Care, 2006; 29:2730–2732.

Garg

, Kelly

, Voelmle

, Ritchie

, Gottlieb

, McFann

, Ellis

. Continuous home monitoring of glucose: improved glycemic control with real-life use of continuous glucose sensors in adult subjects with type 1 diabetes. Diabetes Care, 2007; 30:3023–3025.

Hirsch

, Abelseth

, Bode

, Fischer

, Kaufman

, Mastrototaro

, Parkin

, Wolpert

, Buckingham

. Sensor-augmented insulin pump therapy: results from the first randomized treat-to-target study. Diabetes Technol Ther, 2008; 10:377–383.

Raccah

, Sulmont

, Reznik

, Guerci

, Renard

, Hanaire

, Jeandidier

, Nicolino

. Incremental value of continuous glucose monitoring when starting pump therapy in patients with poorly controlled type 1 diabetes: the RealTrend study. Diabetes Care, 2009; 32:2245–2250.

Hermanides

, Nørgaard

, Bruttomesso

, Mathieu

, Frid

, Dayan

, Diem

, Fermon

, Wentholt

, Hoekstra

, DeVries

. Sensor-augmented pump therapy lowers HbA_1c in suboptimally controlled Type 1 diabetes; a randomized controlled trial. Diabet Med, 2011; 28:1158–1167.

Kordonouri

, Pankowska

, Rami

, Kapellen

, Coutant

, Hartmann

, Lange

, Knip

, Danne

. Sensor-augmented pump therapy from the diagnosis of childhood type 1 diabetes: results of the Paediatric Onset Study (ONSET) after 12 months of treatment. Diabetologia, 2010; 53:2487–2495.

Bergenstal

, Tamborlane

, Ahmann

, Buse

, Dailey

, Davis

, Joyce

, Peoples

, Perkins

, Welsh

, Willi

, Wood

. STAR 3 Study Group. Effectiveness of sensor-augmented insulin-pump therapy in type 1 diabetes. N Engl J Med, 2010; 363:311–320.

10.

Keenan

, Cartaya

, Mastrototaro

. The pathway to the closed-loop artificial pancreas: research and commercial perspectives. Pediatr Endocrinol Rev, 2010; 7,Suppl 3:445–451.

11.

Hovorka

, Allen

, Elleri

, Chassin

, Harris

, Xing

, Kollman

, Hovorka

, Larsen

, Nodale

, De Palma

, Wilinska

, Acerini

, Dunger

. Manual closed-loop insulin delivery in children and adolescents with type 1 diabetes: a phase 2 randomised crossover trial. Lancet, 2010; 375:743–751.

12.

Castle

, Engle

, El Youssef

, Massoud

, Yuen

, Kagan

, Ward

. Novel use of glucagon in a closed-loop system for prevention of hypoglycemia in type 1 diabetes. Diabetes Care, 2010; 33:1282–1287.

13.

Weinstein

, Schwartz

, Brazg

, Bugler

, Peyser

, McGarraugh

. Accuracy of the 5-day FreeStyle Navigator Continuous Glucose Monitoring System: comparison with frequent laboratory reference measurements. Diabetes Care, 2007; 30:1125–1130.

14.

Zisser

, Bailey

, Schwartz

, Ratner

, Wise

. Accuracy of the SEVEN continuous glucose monitoring system: comparison with frequently sampled venous glucose measurements. J Diabetes Sci Technol, 2009; 3:1146–1154.

15.

Mastrototaro

, Shin

, Marcus

, Sulur

. STAR 1 Clinical Trial Investigators. The accuracy and efficacy of real-time continuous glucose monitoring sensor in patients with type 1 diabetes. Diabetes Technol Ther, 2008; 10:385–390.

16.

Keenan

, Cartaya

, Mastrototaro

. Accuracy of a new real-time continuous glucose monitoring algorithm. J Diabetes Sci Technol, 2010; 4:111–118.

17.

Mastrototaro

, Johnson

, Morff

, Lipson

, Andrew

, Allen

. An electroenzymatic glucose sensor fabricated on a flexible substrate. Sensors Actuators, 1991; 5:139–144.

18.

Rebrin

, Steil

, van Antwerp

, Mastrototaro

. Subcutaneous glucose predicts plasma glucose independent of insulin: implications for continuous monitoring. Am J Physiol, 1999; 277:E561–E571.

19.

Clinical and Laboratory Standards Institute. Performance Metrics for Continuous Interstitial Glucose Monitoring; Approved Guideline. CLSI document POCT05-A. Wayne, PA: Clinical and Laboratory Standards Institute, 2008.

20.

International Organization for Standardization: ISO 15197. In Vitro Diagnostic Test Systems—Requirements for Blood Glucose Monitoring Systems for Self Testing in Managing Diabetes Mellitus. Geneva: International Organization for Standardization, 2003.

21.

Keenan

, Voskanyan

, Mastrototaro

, Steil

. Delays in minimally invasive CGM devices: a review of the current technology. J Diabetes Sci Technol, 2009; 3:1207–1214.

22.

Parkes

, Slatin

, Pardo

, Ginsberg

. A new consensus error grid to evaluate the clinical significance of inaccuracies in the measurement of blood glucose. Diabetes Care, 2000; 23:1143–1148.

23.

Clarke

, Cox

, Gonder-Frederick

, Carter

, Pohl

. Evaluating clinical accuracy of systems for self-monitoring of blood glucose. Diabetes Care, 1987; 10:622–628.

24.

Kovatchev

, Gonder-Frederick

, Cox

, Clarke

. Evaluating the accuracy of continuous glucose-monitoring sensors: continuous glucose-error grid analysis illustrated by TheraSense Freestyle Navigator data. Diabetes Care, 2004; 27:1922–1928.