Sage Journals: Discover world-class research

Abstract

Background:

Computer simulation has been shown over the past decade to be a powerful tool to study the impact of medical devices characteristics on clinical outcomes. Specifically, in type 1 diabetes (T1D), computer simulation platforms have all but replaced preclinical studies and are commonly used to study the impact of measurement errors on glycemia.

Method:

We use complex mathematical models to represent the characteristics of 3 continuous glucose monitoring systems using previously acquired data. Leveraging these models within the framework of the UVa/Padova T1D simulator, we study the impact of CGM errors in 6 simulation scenarios designed to generate a wide variety of glycemic conditions. Assessment of the simulated accuracy of each different CGM systems is performed using mean absolute relative deviation (MARD) and precision absolute relative deviation (PARD). We also quantify the capacity of each system to detect hypoglycemic events.

Results:

The simulated Roche CGM sensor prototype (RCGM) outperformed the 2 alternate systems (CGM-1 & CGM-2) in accuracy (MARD = 8% vs 11.4% vs 18%) and precision (PARD = 6.4% vs 9.4% vs 14.1%). These results held for all studied glucose and rate of change ranges. Moreover, it detected more than 90% of hypoglycemia, with a mean time lag less than 4 minutes (CGM-1: 86%/15 min, CGM-2: 57%/24 min).

Conclusion:

The RCGM system model led to strong performances in these simulation studies, with higher accuracy and precision than alternate systems. Its characteristics placed it firmly as a strong candidate for CGM based therapy, and should be confirmed in large clinical studies.

Keywords

continuous glucose monitors in-silico study simulation glucose measurements accuracy

Continuous glucose monitoring (CGM) has become a powerful tool to assist in the optimization of glycemic control in diabetes. Since the advent of CGM technology,^1-3 significant progress has been made toward versatile and reliable CGM devices that not only monitor the course of blood glucose (BG) fluctuations day and night, but also provide feedback such as alarms when BG reaches preset low or high thresholds. A number of studies have documented the benefits of CGM^4-7 and charted guidelines for its clinical use.^8-11

Most current CGM devices do not measure actual BG but form a BG representation based on the interstitial glucose (IG) concentration (the actual measurement matrix), sparse actual BG measurements (calibrations), and algorithmic tools (calibration algorithms, smoothers and short term prediction); all in an effort to account for the physiological process allowing transport from blood to the interstitium.^12-19 Because the time lag can greatly influence the accuracy of CGM, a number of studies were dedicated to its investigation, yielding various results.^16-22 For example, it was hypothesized that if a glucose fall is due to peripheral glucose consumption the physiological time lag would be negative, that is, fall in IG would precede fall in BG.^16,23 In most studies IG lagged behind BG by 4-10 minutes regardless of the direction of BG change. The formulation of the push-pull phenomenon offered reconciliation of these results and provided arguments for a more complex BG-IG relationship.²⁴

In addition, errors from calibration, transient loss of sensitivity, movement artifacts like pressure-induced sensor attenuation (PISA),^25,26 large errors (or large inaccuracies as defined in Leelarathna et al),²⁷ and random noise confound CGM data.^28,29 Nevertheless, the accuracy of CGM is increasing and may be approaching a physiological limit for subcutaneous glucose monitoring.^30-34

While the assessment of sensor accuracy during human clinical trials remains the main source for information about CGM performances and prediction of the utility of these systems,^35-38 computer simulation has also been shown to be a relevant means to infer the impact of different measuring systems on glucose control, based on sensor error modeling derived from said accuracy studies.^39-44 At the core of these simulations is the UVa/Padova simulation platform, a computer based software allowing for the simulation of glucose and insulin metabolism in adults, adolescents, and children with T1D, under a variety of treatment paradigms (eg, multiple daily injections, pump therapy, or closed loop systems). The simulator is equipped with 300 in-silico patients (100 in each age category), designed to span the T1D population and produce a robust assessment of the impact of the specified therapy on plasma glucose. While it is based on 30+ years of research, the simulator was validated in 2008 and is accepted by the Food and Drug Administration as a replacement for preclinical data.

In this article we explore the characteristics of a new CGM prototype, the Roche CGM sensor prototype (RCGM), designed by Roche Diabetes Care. We first describe the overall accuracy metrics based on accuracy data provided by Roche, and decompose the error in its different components (gain, time lag, drift & colored noise) of the new prototype as well as the characteristics of 2 already commercially available CGM systems (CGM-1 and CGM-2) as a base for comparison. Based on this analysis we then leverage the UVa/Padova T1DM simulation platform to compare CGM performances during different scenarios designed to reproduce real-life use, and further describe the potential clinical benefits associated with CGM accuracy. Such setup allows for further testing of a wide variety of situations (see scenarios below), in a more diverse subject population.

Methods

Modeling Data

Data were provided by Roche Diabetes Care for modeling and analysis, including frequent glucose reference measurements and 3 different CGM systems as follows: 10 patients monitored by the RCGM, leading to 19 RCGM sensor traces;⁴⁵ 12 patients monitored by alternate system 1 (CGM-1), leading to 24 sensor traces;⁴⁶ 10 patients monitored by alternate system 2 (CGM-2), leading to 20 sensor traces.⁴⁷ All data were collected during a previous study and deidentified prior to analysis; details about subjects and protocols have been published.^45-47

The length of CGM recordings was on average 155 hours, with 163.7 ± 0.29 hours for RCGM, 142.43 ± 3.25 hours for CGM- 1, and 162.7 ± 1.42 hours for CGM-2. Frequent reference measurements (median time between reference 42 minutes [16 min-61 min]) were taken throughout; studies assessing the accuracy of RCGM and CGM-2 used Accu-Chek® Aviva, Roche Diagnostics GmbH, as reference measurements, while we had access to YSI 2300 for the modeling of CGM-1. Distribution of BG references and rate of change are shown in Figure 1.

Figure 1.

Distribution of reference values and rate of change in original data set.

Reference measurements and sensor signals between calibrations were used to model the sensors, leading to 3178, 4414, and 3346 data pairs for RCGM, CGM-1, and CGM-2 respectively.

Modeling Method

CGM signals have been shown to be a complex mixture of transformed BG signals and error processes. As discussed earlier, the transport of glucose from blood to the sensing matrix of a CGM (within the interstitium), associated with the calibration of the CGM signal using blood measurements (SMBG) leads to a complex relationship between CGM and BG traces. While the more exact mathematical model to represent this phenomenon has been described as part of the “push-pull” methodology, the calibration scheme, and its potential errors, do not allow for a precise estimation of the process parameters. Therefore, we have chosen to represent the transport/calibration “error” as a combination of a time dependent linear transform and time lag, associated with a multiplicative error term; namely:

C G M (θ, t_{i}) \underline{\underline{def}} ((α t_{i} + β) B G (t_{i} - τ) + γ) ε (t)

where $θ$ = [α,β,γ], α represents sensor drift in time, β corresponds to a gain of the CGM sensor, and γ is the bias of the sensor.

The error term, ϵ(τ), is not considered to be a white noise process: the multiplicative noise is considered to be following a Johnson distribution, and as such to be a transformation of a Gaussian process, then its dynamics are set to follow an autoregressive moving average (ARMA) scheme; leading to:

\begin{matrix} J o h n s o n d i s t r i b u t i o n : ε (t_{i}) = λ s i n h (\frac{z (t_{i}) - μ}{η}) + ξ \\ A R M A f o r m : z (t_{i}) = \sum_{j = 1}^{p} Θ_{j} z (t_{i - j}) + \sum_{j = 1}^{q} Φ_{j} ν (t_{i - j}) \end{matrix}

where $v$ is a standard white noise process. With such a form, the error terms are not independent in time (eg, white noise) but an error propagates in the future through the autoregressive terms $(Θ)$ and moving average term $(Φ) .$ Therefore, each error is correlated with previous errors.

Model parameters are estimated by least square methods, with a single τ per sensor, and $θ, Θ, Φ,$ being specific to a calibration window (between 2 calibrations of the sensor). For each sensor type (RCGM, and each of the alternate sensors) we collected the probability distributions of, and correlation between, each parameter.

Simulation Setup

Simulations were run on the 2008 version of the UVa/Padova T1DM simulator⁴⁰ using the adult population (100 in-silico subjects). Each in-silico subject was equipped with (1) a simulated BG meter with normally distributed measurement error, following the ISO15197:2003 standard, (2) a simulated CGM, with parameters randomly sampled from the distributions described above, and (3) a simulated error-free insulin pump. Each in-silico subject is associated with a set of CSII treatment parameters that was used to compute insulin doses during the simulation runs. Therapy was CGM based: the CGM data were used for both dosing insulin or additional carbohydrates. In-silico subjects were exposed to 6 testing scenarios designed to expose them to hyper and hypoglycemia and have them experience fast changing plasma glucose values triggered by mismatched therapy and physiological changes. A summary of these scenarios is provided below:

Optimal glucose control in which per-subject estimated carbohydrates and basal/bolus insulin doses are accurate and glucose is well-controlled. This is considered a baseline “control” protocol for comparison of the others demonstrating poor control.

Hyperglycemia in which carbohydrates and insulin doses are “underestimated” and BG levels rise above the normal range.

Hypoglycemia in which “overtreatment” of insulin occurs during the simulation and BG levels fall below the normal and safe range.

Exercise-induced hypoglycemia in which “insulin sensitivity” and “insulin utilization” is increased and BG levels fall below the normal and safe range.

Insulin stacking with mistimed dosing in which “insulin on board” is increased.

Rapid glycemic excursions in which BG fluctuations change ± 0.9 to 3.5 mg/dL/minute from high or normal glucose level into the hypoglycemic range.

The entire in-silico population was exposed to each protocol, wearing all 3 CGM systems, and repeated 50 times each, leading to approximately 44.5 million hours of simulated glucose signals.

Outcomes and Analysis

We compared the RCGM performance and clinical outcome measures to those of the alternate CGM models (CGM-1 and CGM-2) using standard performance metrics. The analyses included accuracy (MARD) of paired CGM measurements and plasma glucose values (SMBG) and precision (PARD) of paired CGM measurements across the range of glucose variability (see Obermaier et al⁴⁸ for precise definitions). BG ranges of interest included hyperglycemia of > 200 mg/dL and >240 mg/dL, euglycemia of 70-180 mg/dL, hypoglycemia at < 70 mg/dL and severe hypoglycemia at < 55 mg/dL; Rapidly changing glucose levels (rising and falling ≥ 1 mg/dL/ minute) were investigated as well. In addition, hypoglycemia event detection failure (false alarms or missed events) and delay to detection were examined. Differences in MARD and PARD were the primary accuracy measures with MAD being a secondary measure for the comparison of the RCGM to the other CGM devices. The median and variability of the metrics across all sensor models and the in-silico population are provided in the summary results tables below. No statistical analysis was performed due to the specific nature of simulation studies (the standard error of an estimate can be made artificially low by increasing the number of repetitions).

Results

Parameters were obtained for all 3 sensors and their distributions are reported in Figure 2. All systems showed similar drifts, but RCGM estimated time lag was approximately half as that of the other systems (4.5 vs 12.1 vs 11.4 minutes, P < .01). Both RCGM and CGM-2 showed a slightly elevated multiplicative Bias when compared to CGM-1 (0.97 vs 0.72 vs 0.99, P < .01), but CGM-1 displayed a much larger additive correction (−1.6 vs 32.7 vs −1.3, P < .01).

Figure 2.

Distribution (box plot) of the parameter estimates for each CGM system. Whiskers represent 95% of the data, the median is represented as a horizontal line, and the colored block represents the interquartile range. RCGM is shown in blue, CGM-1 in orange, and CGM-2 in yellow.

The model was further validated by testing the normality of the transformed noise process (Shapiro-Wilk test for normality > 0.75 for all sensors), the ARMA process chosen was a first order autoregressive model with irregular spacing, parameters were similar for all systems (no significant differences found), with partial autocorrelation of approximately 0.6 and spread of approximately 0.35. Noise generated by the models was not found to originate from a different distribution than the source data: Kolmogorov-Smirnov test with P > .7. Finally, the MARDs of 500 simulated sensors for each system were found to be indistinguishable from the empirical data.

Overall both accuracy and precision of the 3 CGM system were clearly different, with an order of decreasing accuracy: RCGM with an MARD of 8% (6.8%-10.2%), CGM-2 11.4% (9.3%-22.4%), and CGM-1 18% (12.8%-53.4%). These relative performances were confirmed across all testing conditions, with particular contrasts in hypoglycemia with RCGM performances remaining around 10% and the alternate sensors showing MARDs 2 to 4 times higher. Both RCGM and CGM-2 were much more precise than CGM-1 overall, with RCGM being the most precise (PARD: 6.8% vs 14.1% vs 7.4%). MARDs and PARDs are reported in Table 1.

Table 1.

Accuracy and Precision of the 3 CGM Systems Under Different Glycemic Conditions (Percentages).

	MARD			PARD
Condition	RCGM	CGM-1	CGM-2	RCGM	CGM-1	CGM-2
Overall	8.0	18.0	11.9	6.8	14.1	9.4
Protocol 1: euglycemia (70 mg/dL<BG<180 mg/dL)	6.8	12.8	9.3	6.7	14	7.4
Protocol 2: hyperglycemia (BG>200 mg/dL)	8.6	21.6	14.1	7.1	19.0	7.5
Protocol 2: hyperglycemia (BG>240 mg/dL)	8.8	25.7	16.3	7.2	20.2	7.9
Protocols 3-5: hypoglycemia (BG<70 mg/dL)	9.2	37.1	18.6	8.6	10.5	16.9
Protocols 3-5: hypoglycemia (BG<55 mg/dL)	10.2	53.4	22.4	11.7	11.7	21.2
Protocol 6: rapid changes	9.6	19.5	13.5	5.4	15.4	7.6

These differences were further exacerbated during episodes of fast changing BG, with RCGM accuracy (MARD) varying from 7.3% to 15.6%, CGM-1 from 16.6% to 26.6%, and CGM-2 from 10.1% to 22.9%. Accuracy was best for all systems in the −1 mg/dL/min to 1 mg/dL/min range, and worst during fast descent (< –1.5 mg/dL/min). See Figure 3 for further details.

Figure 3.

Analysis of the influence of blood glucose rate of change on CGM accuracy as measured by MARD, for RCGM (blue), CGM-1 (yellow), and CGM-2 (orange).

No hypoglycemia occurred during protocols 1 and 2 (nominal and hyperglycemia), so focusing only on the remaining 4 protocols we report the hypoglycemia detection rates and false alarm rates of the 3 CGM systems. In our simulations, the RCGM missed detection of fewer hypoglycemia events than the other 2 sensors (8.1% vs 43.4% vs 13.9%) and recorded lower rates of false detection of hypoglycemia (30%, 48%, and 29%). CGM-1 performed the worst and while failing to detect the presence of hypoglycemia and severe hypoglycemia as previously mentioned consequently had lower rates of false detection of hypoglycemia and severe hypoglycemia than the RCGM. CGM-2 failed to detect hypoglycemia at a rate nearly double that of the RCGM. Hypoglycemia detection characteristics are reported in Table 2.

Table 2.

Hypoglycemia Detection Performances.

	RCGM	CGM-1	CGM-2
Hypoglycemia <70 mg/dL
# of detected events per simulation	1.1	0.5	1.0
Detection rate (%)	92	57	86
# of false alarm per simulation	0.51	0.53	0.44
Rate of false alarms (%)	30	48	29
Delay to detection (min)	3.9	23.9	15.1
Hypoglycemia <55 mg/dL
# of detected events per simulation	0.52	0.31	0.48
Detection rate (%)	96	69	92
# of false alarm per simulation	0.36	0.44	0.46
Rate of false alarms (%)	36	43	43
Delay to detection (min)	4.6	26.5	16.4

Discussion

Based on mathematical modeling of the continuous glucose monitors characteristics, we showed that one can accurately simulate the behavior of CGM systems. This was performed by breaking the CGM signals into several components, from calibration to drift, time lag to colored noise. Borrowing and expanding from the literature,⁴¹ a new model of CGM noise is proposed and fitted to clinical trial data. And while each sensor is represented by a specific set of parameters, we showed significant differences in the distribution of these parameters across 3 different CGM systems.

Using these models within our T1D simulation platform, we looked at the reaction of each system under a broad range of conditions such as hypo and hyperglycemia, triggered by treatment errors (eg, over/under insulin dosing), behavior (eg, exercise), and physiology (eg, change in insulin sensitivity). These different scenarios allowed for the generation of glucose fluctuations akin to in-vivo data and therefore the expansion of the accuracy study of the CGM systems beyond the source accuracy obtained in the clinical trials.

We showed that the RCGM system characteristics led to strong accuracy and precision, in both hypo- and hyperglycemia, comparing favorably to the performance of 2 alternate CGM systems under identical glycemic conditions. In particular, RCGM maintained high accuracy (MARD<10%) and precision across all studied glucose ranges and rate of change; in contrast to the other systems, whose performance, though sometime comparable, fluctuated significantly (see table 1). CGM-1 in particular suffered from serious inaccuracy in the hypoglycemic range.

Focusing on hypoglycemia, we showed that the RCGM characteristics lead to a detection of mild (<70 mg/dL) hypoglycemia on average with less than 4 minutes time lag (compared to 15 min and 24 min for CGM-1 and CGM-2), with a detection rate of 92% (vs 57% and 86%) but with a nonnegligible rate of false alarm of approximately 30% (vs 48% and 29%). This performance (and how it compared to CGM-1 and CGM-2) was confirmed when focusing on more pronounced hypoglycemia events (<55 mg/dL).

Scenarios were specifically designed to reproduce glycemic events (eg, hypoglycemia or fast rate of change) regardless of the sensor in use, therefore, this analysis does not allow for the determination of potential improvement in glycemic control by using 1 of the 3 systems over the others. Nonetheless, Kovatchev et al published in 2015 a different in-silico study looking specifically at the CGM characteristics that would be necessary for an entirely CGM-based treatment of T1D.⁴⁹ They concluded that a combination of MARD below 10% and a rate of large errors (beyond 20% of the reference) below approximately 12% should be sufficient for use of CGM systems as replacement for BG measurements; this result was further depicted by a curve linking MARD and the rate of large errors, later used to show system improvement by optimal signal treatment.⁵⁰ In Figure 4, we have represented the published curve with published CGM accuracy data from several articles^38,50-52 as well as the data analyzed in this project. This figure clearly shows that the RCGM system’s predicted accuracy is well beneath the theoretical (obtained in simulation) threshold for nonadjunctive use in the treatment of T1D.

Figure 4.

Relationship between MARD and rate of large error, and how they relate to the threshold for sole use of CGM in T1D treatment. White stars represent accuracy data from the literature, and RCGM (blue), CGM-1 (yellow), and CGM-2 (orange) are placed on the curve.

Conclusion

Using computer simulations, we explored the clinical relevance of sensor error characteristics derived from human clinical trials. This exploration led to the determination that the RCGM system may be more accurate in a broad set of conditions that 2 alternate systems, and that the RCGM sensor was particularly stable in its performances across glucose ranges and rates of change. We further predicted that based on these characteristics, the RCGM sensor should detect hypoglycemia more often and with better precision than 2 alternate systems.

Finally, the simulated results obtained in this analysis seemed to indicate that the RCGM system would be a strong candidate for CGM-based insulin treatment, as determined by Kovatchev et al.⁴⁹

Major weaknesses of this analysis included the small size of the original dataset (10-12 patients per sensor model), and the reliance on a mathematical model to generate sensor errors. In addition, the results obtained are predictions of performances from computer simulations, and therefore inherently limited by the capacity of the simulation platform. Approaches closer to clinical data such as presented in Kovatchev et al,⁴⁹ though also simulations, could boost confidence in the presented results, but ultimately, actual clinical risk/benefits analysis would be needed to support nonadjunct claims.

Nonetheless, simulations provide a unique environment to study the potential clinical impact of medical devices, allowing for precise control of the device characteristics and testing environment, as well as the capacity to explore a wide range of possible clinical scenario in a safe and cost-effective way. Technical performances of the RCGM were significantly better than the comparator systems in the original clinical data and these differences led to similar differences in simulated clinical outcomes. Such improvements would benefit from confirmation in long-term clinical studies.

Footnotes

Abbreviations

ARMA, autoregressive moving average; BG, blood glucose; CGM, continuous glucose monitor; CGM1, alternate continuous glucose monitor 1; CGM2, alternate continuous glucose monitor 2; CSII, continuous subcutaneous insulin infusion; IG, interstitial glucose; MAD, mean absolute deviation; MARD, mean absolute relative deviation; PARD, precision absolute relative deviation; PISA, pressure-induced sensor attenuation; RCGM, Roche continuous glucose monitor; T1D/T1DM, type 1 diabetes/type 1 diabetes mellitus.

Declaration of Conflicting Interests

MDB has received honorarium and consulting fees from Roche Diabetes. RH, MS, and GSR are Roche Diabetes Care employees. ECN and SR received funding from Roche for this work.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funding for this work was obtained from Roche Diabetes Care, Mannheim Germany. All simulations were run at The Epsilon Group.

References

Mastrototaro

JJ.

The MiniMed continuous glucose monitoring system. Diabetes Technol Ther. 2000;2(suppl 1):S13-S18.

Bode

BW.

Clinical utility of the continuous glucose monitoring system. Diabetes Technol Ther. 2000;2(suppl 1):S35-S42.

Feldman

Brazg

Schwartz

Weinstein

A continuous glucose sensor based on wired enzyme technology—results from a 3-day trial in patients with type 1 diabetes. Diabetes Technol Ther. 2003;5:769-778.

Deiss

Bolinder

Riveline

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

Garg

Zisser

Schwartz

. Improvement in glycemic excursions with a transcutaneous, real-time continuous glucose sensor. Diabetes Care. 2006;29:44-50.

Kovatchev

Clarke

WL.

Continuous glucose monitoring reduces risks for hypo- and hyperglycemia and glucose variability in diabetes. Diabetes. 2007;56(suppl 1):0086OR.

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.

Klonoff

DC.

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

Hirsch

Armstrong

Bergenstal

. Clinical application of emerging sensor technologies in diabetes management: consensus guidelines for continuous glucose monitoring. Diabetes Technol Ther. 2008;10:232-246.

10.

Hovorka

Continuous glucose monitoring and closed-loop systems. Diabet Med. 2006;23:1-12.

11.

Klonoff

DC.

The artificial pancreas: how sweet engineering will solve bitter problems. J Diabetes Sci Technol. 2007;1:72-81.

12.

Bequette

BW.

Continuous glucose monitoring: real-time algorithms for calibration, filtering, and alarms. J Diabetes Sci Technol. 2010;4(2):404-418.

13.

Rossetti

Bondia

Vehí

Fanelli

CG.

Estimating plasma glucose from interstitial glucose: the issue of calibration algorithms in commercial continuous glucose monitoring devices. Sensors. 2010;10(12):10936-10952.

14.

Sparacino

Facchinetti

Cobelli

“Smart” continuous glucose monitoring sensors: on-line signal processing issues. Sensors. 2010;10(7):6751-6772.

15.

Peyser

Nakamura

Price

Bohnett

Hirsch

Balo

Hypoglycemic accuracy and improved low glucose alerts of the latest Dexcom G4 Platinum continuous glucose monitoring system. Diabetes Technol Ther. 2015;17(8):548-554.

16.

Rebrin

Steil

van Antwerp

Mastrototaro

JJ.

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

17.

Rebrin

Steil

GM.

Can interstitial glucose assessment replace blood glucose measurements?

Diabetes Technol Ther. 2000;2:461-472.

18.

Aussedat

Dupire-Angel

Gifford

Klein

Wilson

Reach

Interstitial glucose concentration and glycemia: implications for continuous subcutaneous glucose monitoring. Am J Physiol Endocrinol Metab. 2000;278:E716-E728.

19.

Steil

Rebrin

Hariri

. Interstitial fluid glucose dynamics during insulin-induced hypoglycaemia. Diabetologia. 2005;48:1833-1840.

20.

Boyne

Silver

Kaplan

Saudek

Timing of changes in interstitial and venous blood glucose measured with a continuous subcutaneous glucose sensor. Diabetes. 2003;52:2790-2794.

21.

Kulcu

Tamada

Reach

Potts

Lesho

MJ.

Physiological differences between interstitial glucose and blood glucose measured in human subjects. Diabetes Care. 2003;26:2405-2409.

22.

Stout

Racchini

Hilgers

ME.

A novel approach to mitigating the physiological lag between blood and interstitial fluid glucose measurements. Diabetes Technol Ther. 2004;6:635-644.

23.

Wientjes

Schoonen

AJ.

Determination of time delay between blood and interstitial adipose tissue glucose concentration change by microdialysis in healthy volunteers. Int J Artif Organs. 2001;24:884-889.

24.

Wentholt

IME

Hart

AAM

Hoekstra

JBL

DeVries

. Relationship between interstitial and blood glucose in type 1 diabetes patients: delay and the push-pull phenomenon revisited. Diabetes Technol Ther. 2004;9:169-175.

25.

Baysal

Cameron

Buckingham

. A novel method to detect pressure-induced sensor attenuations (PISA) in an artificial pancreas. J Diabetes Sci Technol. 2014;8(6):1091-1096.

26.

Helton

Ratner

Wisniewski

NA.

Biomechanics of the sensor-tissue interface—effects of motion, pressure, and design on sensor performance and the foreign body response—part II: examples and application. J Diabetes Sci Technol. 2011;5(3):647-656.

27.

Leelarathna

Nodale

Allen

. Evaluating the accuracy and large inaccuracy of two continuous glucose monitoring systems. Diabetes Technol Ther. 2013;15(2):143-149.

28.

Kovatchev

Clarke

WL.

Peculiarities of the continuous glucose monitoring data stream and their impact on developing closed-loop control technology. J Diabetes Sci Technol. 2008;2:158-163.

29.

Clarke

Kovatchev

BP.

Continuous glucose sensors—continuing questions about clinical accuracy. J Diabetes Sci Technol. 2007;1:164-170.

30.

Kovatchev

Anderson

Heinemann

Clarke

WL.

Comparison of the numerical and clinical accuracy of four continuous glucose monitors. Diabetes Care. 2008;31:1160-1164.

31.

Diabetes Research in Children Network (DirecNet) Study Group. The accuracy of the Guardian® RT continuous glucose monitor in children with type 1 diabetes. Diabetes Technol Ther. 2008;10:266-272.

32.

Garg

Smith

Beatson

Lopez-Baca

Voelmle

Gottlieb

PA.

Comparison of accuracy and safety of the SEVEN and the Navigator continuous glucose monitoring systems. Diabetes Technol Ther. 2009;11:65-72.

33.

Calhoun

Lum

Beck

Kollman

Performance comparison of the Medtronic Sof-Sensor and Enlite glucose sensors in inpatient studies of individuals with type 1 diabetes. Diabetes Technol Ther. 2013;15:758-761.

34.

Christiansen

Bailey

Watkins

. A new-generation continuous glucose monitoring system: improved accuracy and reliability compared with a previous-generation system. Diabetes Technol Ther. 2013;15:881-888.

35.

Kovatchev

Anderson

Heinemann

Clarke

Comparison of the numerical and clinical accuracy of four continuous glucose monitors. Diabetes Care. 2008;31(6):1160-1164.

36.

Weinstein

Schwartz

Brazg

Bugler

Peyser

McGarraugh

GV.

Accuracy of the 5-day freestyle navigator continuous glucose monitoring system comparison with frequent laboratory reference measurements. Diabetes Care. 2007;30(5):1125-1130.

37.

Mazze

Strock

Borgman

Wesley

Stout

Racchini

Evaluating the accuracy, reliability, and clinical applicability of continuous glucose monitoring (CGM): Is CGM ready for real time?

Diabetes Technol Ther. 2009;11(1):11-18.

38.

Luijf

Mader

Doll

. Accuracy and reliability of continuous glucose monitoring systems: a head-to-head comparison. Diabetes Technol Ther. 2013;15(8):721-726.

39.

Kovatchev

Breton

Dalla Man

Cobelli

In silico preclinical trials: a proof of concept in closed-loop control of type 1 diabetes. J Diabetes Sci Technol. 2009;3(1):44-55.

40.

Dalla Man

Micheletto

Breton

Kovatchev

Cobelli

. The UVa/Padova type 1 diabetes simulator new features. J Diabetes Sci Technol. 2014;8(1):26-34.

41.

Breton

Kovatchev

Analysis, modeling, and simulation of the accuracy of continuous glucose sensors. J Diabetes Sci Technol. 2008;2(5):853-862.

42.

Facchinetti

Sparacino

Cobelli

Modeling the error of continuous glucose monitoring sensor data: critical aspects discussed through simulation studies. J Diabetes Sci Technol. 2010;4(1):4-14.

43.

Breton

Kovatchev

BP.

Impact of blood glucose self-monitoring errors on glucose variability, risk for hypoglycemia, and average glucose control in type 1 diabetes: an in silico study. J Diabetes Sci Technol. 2010;4(3):562-570.

44.

McQueen

Breton

Ott

Koa

Beamer

Campbell

. Economic value of improved accuracy for self-monitoring of blood glucose devices for type 1 diabetes in Canada. J Diabetes Sci Technol. 2015:1932296815599551.

45.

Zschornack

Schmid

Pleus

. Evaluation of the performance of a novel system for continuous glucose monitoring. J Diabetes Sci Technol. 2013;7(4):815-823.

46.

Freckmann

Pleus

Link

Zschornack

Klötzer

Haug

Performance evaluation of three continuous glucose monitoring systems: comparison of six sensors per subject in parallel. J Diabetes Sci Technol. 2013;7(4):842-853.

47.

Pleus

Schmid

Link

. Performance evaluation of a continuous glucose monitoring system under conditions similar to daily life. J Diabetes Sci Technol. 2013;7(4):833-841.

48.

Obermaier

Schmelzeisen-Redeker

Schoemaker

. Performance evaluations of continuous glucose monitoring systems: precision absolute relative deviation is part of the assessment. J Diabetes Sci Technol. 2013;7(4):824-832.

49.

Kovatchev

Patek

Ortiz

Breton

MD.

Assessing sensor accuracy for non-adjunct use of continuous glucose monitoring. Diabetes Technol Ther. 2015;17(3):177-186.

50.

Kovatchev

BP.

Hypoglycemia reduction and accuracy of continuous glucose monitoring. Diabetes Technol Ther. 2015;17(8):530-533.

51.

Kropff

Bruttomesso

Doll

. Accuracy of two continuous glucose monitoring systems: a head-to-head comparison under clinical research centre and daily life conditions. Diabetes Obes Metab. 2015;17(4):343-349.

52.

Thabit

Leelarathna

Wilinska

. Accuracy of continuous glucose monitoring during three closed-loop home studies under free-living conditions. Diabetes Technol Ther. 2015;17(11):801-807.

Analysis of the Accuracy and Performance of a Continuous Glucose Monitoring Sensor Prototype: An In-Silico Study Using the UVA/PADOVA Type 1 Diabetes Simulator

Abstract

Background:

Method:

Results:

Conclusion:

Keywords

Methods

Modeling Data

Modeling Method

Simulation Setup

Outcomes and Analysis

Results

Discussion

Conclusion

Footnotes

Abbreviations

Declaration of Conflicting Interests

Funding

References