Continuous Glucose-error Grid Analysis Research Articles

Intelligent Ensemble Deep Learning System for Blood Glucose Prediction Using Genetic Algorithms

Forecasting blood glucose (BG) values for patients can help prevent hypoglycemia and hyperglycemia events in advance. To this end, this study proposes an intelligent ensemble deep learning system to predict BG values in 15, 30, and 60 min prediction horizons (PHs) based on historical BG values collected via continuous glucose monitoring devices as an endogenous factor and carbohydrate intake and insulin administration information (times) as exogenous factors. Although there are numerous deep learning algorithms available, this study applied five algorithms, namely, recurrent neural network (RNN), which is optimized for sequence data (e.g., time-series), and RNN-based algorithms (e.g., long short-term memory (LSTM), stacked LSTM, bidirectional LSTM, and gated recurrent unit). Then, a genetic algorithm (GA) was applied to the five prediction models to optimize their weights through ensemble techniques and to yield (output) the final predicted BG values. The performance of the proposed model was compared to that of the autoregressive integrated moving average (ARIMA) model as a baseline. The results show that the proposed model significantly outperforms the baseline in terms of the root mean square error (RMSE) and continuous glucose error grid analysis. For the valid 29 diabetic patients for the multivariate models, the RMSE was 11.08 ( ± 3.19), 19.25 ( ± 5.28), and 31.30 ( ± 8.81) mg/DL for 15, 30, and 60 min PH, respectively. When the same data were applied to univariate models, the RMSE was 11.28 ( ± 3.34), 19.99 ( ± 5.59), and 33.13 ( ± 9.27) mg/DL for 15, 30, and 60 min PH, respectively. Both the univariate and multivariate models showed a statistically significant difference compared with the baseline at a 5% statistical significance level. Instead of using a model with a single algorithm, applying a GA based on each output of a model with multiple algorithms was found to play a significant role in improving model performance.

Long-Term Prediction of Blood Glucose Levels in Type 1 Diabetes Using a CNN-LSTM-Based Deep Neural Network.

In this work, we leverage state-of-the-art deep learning-based algorithms for blood glucose (BG) forecasting in people with type 1 diabetes. We propose stacks of convolutional neural network and long short-term memory units to predict BG level for 30-, 60-, and 90-minute prediction horizon (PH), given historical glucose measurements, meal information, and insulin intakes. The evaluation was performed on two data sets, Replace-BG and DIAdvisor, representative of free-living conditions and in-hospital setting, respectively. For 90-minute PH, our model obtained mean absolute error of 17.30 ± 2.07 and 18.23 ± 2.97 mg/dL, root mean square error of 23.45 ± 3.18 and 25.12 ± 4.65 mg/dL, coefficient of determination of 84.13 ± 4.22% and 82.34 ± 4.54%, and in terms of the continuous glucose-error grid analysis 94.71 ± 3.89% and 91.71 ± 4.32% accurate predictions, 1.81 ± 1.06% and 2.51 ± 0.86% benign errors, and 3.47 ± 1.12% and 5.78 ± 1.72% erroneous predictions, for Replace-BG and DIAdvisor data sets, respectively. Our investigation demonstrated that our method achieved superior glucose forecasting compared with existing approaches in the literature, and thanks to its generalizability showed potential for real-life applications.

Comparative assessment of glucose prediction models for patients with type 1 diabetes mellitus applying sensors for glucose and physical activity monitoring.

The present work presents the comparative assessment of four glucose prediction models for patients with type 1 diabetes mellitus (T1DM) using data from sensors monitoring blood glucose concentration. The four models are based on a feedforward neural network (FNN), a self-organizing map (SOM), a neuro-fuzzy network with wavelets as activation functions (WFNN), and a linear regression model (LRM), respectively. For the development and evaluation of the models, data from 10 patients with T1DM for a 6-day observation period have been used. The models' predictive performance is evaluated considering a 30-, 60- and 120-min prediction horizon, using both mathematical and clinical criteria. Furthermore, the addition of input data from sensors monitoring physical activity is considered and its effect on the models' predictive performance is investigated. The continuous glucose-error grid analysis indicates that the models' predictive performance benefits mainly in the hypoglycemic range when additional information related to physical activity is fed into the models. The obtained results demonstrate the superiority of SOM over FNN, WFNN, and LRM with SOM leading to better predictive performance in terms of both mathematical and clinical evaluation criteria.

The University of Virginia/Padova Type 1 Diabetes Simulator Matches the Glucose Traces of a Clinical Trial

In 2008, the Food and Drug Administration (FDA) accepted our type 1 diabetes mellitus (T1DM) simulator (S2008), equipped with 100 in silico adults, 100 adolescents, and 100 children, as a substitute for preclinical trials for certain insulin treatments, including closed-loop algorithms. Hypoglycemia was well described in the simulator, but recent closed-loop trials showed a much larger frequency of hypoglycemia events in patients compared with the in silico ones. In order to better describe the distribution of glucose concentration observed in clinical trials, the simulator has recently been updated, and modifications have been accepted by the FDA (S2013). The aim of this study is to assess the validity of the S2013 simulator against clinical data and compare its performance with that of the S2008. The database consists of 24 T1DM subjects who received dinner (70.7±3.3 g of carbohydrate) and breakfast (52.9±0.1 g of carbohydrate) in two occasions (open- and closed-loop), for a total of 96 postmeal glucose profiles. Measured plasma glucose profiles were compared with those simulated in 100 in silico adults, and the continuous glucose error grid analysis (CG-EGA) was used to assess the validity of the simulated traces. Moreover, the most common outcome metrics have been compared. The frequency of hypoglycemia episodes predicted by the S2013 well reproduces that observed during clinical trials as proven by the CG-EGA. In addition, the outcome metrics provided by the S2013 are similar to those observed in clinical trials in a set of T1DM subjects. We demonstrated that the virtual subjects of the S2013 are representative of the T1DM population observed in a clinical trial. We conclude that the S2013 is a valid tool usable to test the robustness of closed-loop control algorithms for artificial pancreas.

Performance Evaluations of Continuous Glucose Monitoring Systems: Precision Absolute Relative Deviation is Part of the Assessment

Even though a Clinical and Laboratory Standards Institute proposal exists on the design of studies and performance criteria for continuous glucose monitoring (CGM) systems, it has not yet led to a consistent evaluation of different systems, as no consensus has been reached on the reference method to evaluate them or on acceptance levels. As a consequence, performance assessment of CGM systems tends to be inconclusive, and a comparison of the outcome of different studies is difficult. Published information and available data (as presented in this issue of Journal of Diabetes Science and Technology by Freckmann and coauthors) are used to assess the suitability of several frequently used methods [International Organization for Standardization, continuous glucose error grid analysis, mean absolute relative deviation (MARD), precision absolute relative deviation (PARD)] when assessing performance of CGM systems in terms of accuracy and precision. The combined use of MARD and PARD seems to allow for better characterization of sensor performance. The use of different quantities for calibration and evaluation, e.g., capillary blood using a blood glucose (BG) meter versus venous blood using a laboratory measurement, introduces an additional error source. Using BG values measured in more or less large intervals as the only reference leads to a significant loss of information in comparison with the continuous sensor signal and possibly to an erroneous estimation of sensor performance during swings. Both can be improved using data from two identical CGM sensors worn by the same patient in parallel. Evaluation of CGM performance studies should follow an identical study design, including sufficient swings in glycemia. At least a part of the study participants should wear two identical CGM sensors in parallel. All data available should be used for evaluation, both by MARD and PARD, a good PARD value being a precondition to trust a good MARD value. Results should be analyzed and presented separately for clinically different categories, e.g., hypoglycemia, exercise, or night and day.

Nocturnal Continuous Glucose Monitoring: Accuracy and Reliability of Hypoglycemia Detection in Patients with Type 1 Diabetes at High Risk of Severe Hypoglycemia

A reliable method to detect biochemical nocturnal hypoglycemia is highly needed, especially in patients with recurrent severe hypoglycemia. We evaluated reliability of nocturnal continuous glucose monitoring (CGM) in patients with type 1 diabetes at high risk of severe hypoglycemia. Seventy-two type 1 diabetes patients with recurrent severe hypoglycemia (two or more events within the last year) participated for 4 nights in blinded CGM recordings (Guardian(®) REAL-Time CGMS and Sof-Sensor(®); Medtronic MiniMed, Northridge, CA). Blood was drawn hourly from 23:00 to 07:00 h for plasma glucose (PG) measurements (gold standard). Valid data were obtained in 217 nights. The sensitivity of CGM was 65% (95% confidence interval, 53-77%) below 4 mmol/L, 40% (24-56%) below 3 mmol/L, and 17% (0-47%) below 2.2 mmol/L. PG and CGM readings correlated in the total measurement range (Spearman's ρ=0.82; P<0.001). In the normo- and hyperglycemic ranges CGM underestimated PG by 1.1 mmol/L (0.9-1.2 mmol/L) (P<0.001); in contrast, in the hypoglycemic range (PG<4 mmol/L) CGM overestimated PG levels by 1.0 mmol/L (P<0.001). The mean absolute relative differences in the hypo- (≤3.9 mmol/L), normo- (4-9.9 mmol/L), and hyperglycemic (≥10 mmol/L) ranges were 45% (37-53%), 23% (22-25%), and 20% (19-21%), respectively. Continuous glucose error grid analysis indicated a clinical accuracy of 56%, 99%, and 93% in the hypo-, normo-, and hyperglycemic ranges, respectively. The accuracy in the hypoglycemic range of nocturnal CGM data using Sof-Sensor is suboptimal in type 1 diabetes patients at high risk of severe hypoglycemia. To ensure clinical useful sensitivity in detection of nocturnal hypoglycemic episodes, an alarm threshold should not be lower than 4 mmol/L.

305

Introduction: Stress hyperglycemia is a common result of critical illness, severe trauma, or major surgery, and is often directly correlated with increased morbidity and mortality. Control of blood glucose (BG) levels during critical illness results in improved outcomes but intermittent measurement of blood glucose (BG) may miss episodes of hyper- and hypoglycemia that a continuous monitor can quickly identify to facilitate improved glycemic control. The Symphony® Transdermal Continuous Glucose Monitor (tCGM) is a non-invasive system designed to address this need, using the Prelude® SkinPrep System to prepare the skin, providing interstitial fluid (ISF) access at the measurement site. Hypothesis: The Symphony tCGM system is able to continuously monitor glucose non-invasively on sites prepared with Prelude SkinPrep in the surgical intensive care unit. Methods: Fifteen adult patients undergoing cardiac surgery were evaluated. Prelude was used to prepare the skin and a tCGM sensor was applied to each of two test sites. Frequent venous reference blood samples were taken from IV lines at 30-60 minute intervals. The reference BG values were measured with a YSI 2300 Stat+ glucose analyzer. At the conclusion of the study, the skin at the test sites was inspected for redness or any other effects. Accuracy of the tCGM data relative to the reference BG values was determined using Continuous Glucose Error Grid Analysis (CG-EGA) and mean absolute relative difference (MARD). Results: Using over 600 Symphony tCGM glucose readings paired with reference BG measurements, CG-EGA showed that 99.6% of the readings were clinically accurate. The MARD was 12.3%. Values for BG measurements ranged from 73 to 251 mg/dL. No unanticipated adverse device effects were reported for the Prelude skin permeation or the Symphony tCGM. Conclusions: The Symphony tCGM system is able to continuously monitor glucose in cardiac surgery intensive care patients by transdermally measuring glucose in ISF. This non-invasive approach represents a simple, convenient, needle-free solution for continuous glucose monitoring in critically ill patients.

Pilot Study of a Prototype Minimally Invasive Intradermal Continuous Glucose Monitor

The purpose of this study was to assess point accuracy, rate-of-change accuracy, and safety of a prototype, minimally invasive continuous glucose monitoring (CGM) device over a 12 h in-clinic study. The CGM system consisted of a wireless electronics module with a disposable glucose sensor attached to the bottom. The electronics module was affixed to the abdomen using an adhesive pad on the bottom of the disposable sensor housing. Two CGM sensors were inserted into the abdominal tissue (left and right) of 15 adults aged 26-67 years, 5 with normoglycemia, 5 with type 1 diabetes, and 5 with type 2 diabetes. Over a 12 h period, each participant was fed three meals. Reference venous blood samples were drawn at periodic intervals (12.4 ± 5.3 min), and glucose was measured at the bedside using a laboratory reference method. For each participant, a single plasma equivalent glucose concentration was used for retrospective sensor calibration. A total of 1082 paired data points were obtained from 15 subjects and 25 of 30 sensors. Statistical analysis yielded a mean absolute relative difference of 12.6% and a mean absolute difference of 16.0 mg/dl. Continuous glucose error grid analysis showed the combined point and rate-of-change accuracy was 97.4% in zone A and 1.5% in zone B (98.9% A+B), with 1.1% erroneous readings. The prototype CGM system provided clinically accurate results 98.9% of the time and was well tolerated by participants, with little or no pain and no adverse events.

Determinants of the Accuracy of Continuous Glucose Monitoring in Non-Critically Ill Patients with Heart Failure or Severe Hyperglycemia

The accuracy of continuous glucose monitoring (CGM) in non-critically ill hospitalized patients with heart failure or severe hyperglycemia (SH) is unknown. Hospitalized patients with congestive heart failure (CHF) exacerbation (receiving IV or subcutaneous insulin) or SH requiring insulin infusion were compared to outpatients referred for retrospective CGM. Forty-three patients with CHF, 15 patients with SH, and 88 outpatients yielded 470, 164, and 2150 meter-sensor pairs, respectively. Admission glucose differed (188 versus 509 mg/dl in CHF and SH, p < .001) but not the first sensor glucose (p = .35). In continuous glucose error grid analysis, 67-78% of pairs during hypoglycemia were in zones A+B (p = .63), compared with 98-100% in euglycemia (p < .001) and 98%, 92%, and 99% (p = .001) during hyperglycemia for the CHF, SH, and outpatient groups, respectively. Mean absolute relative difference (MARD) was lower in the CHF versus the SH group in glucose strata above 100 mg/dl, but there was no difference between the CHF and outpatient groups. Linear regression models showed that CHF versus outpatient, SH versus CHF, and coefficient of variation were significant predictors of higher MARD. Among subjects with CHF, MARD was not associated with brain natriuretic peptide or change in plasma volume, but it was significantly higher in subjects randomized to IV insulin (p = .04). The results suggest that SH and glycemic variability are more important determinants of CGM accuracy than known CHF status alone in hospitalized patients.

Continuous glucose monitoring system: Is it really accurate, safe and clinically useful?

The continuous glucose monitoring system (CGM) has been used for constant checking of glucose level by measuring interstitial glucose concentrations, since the early days of the 21st century. It can potentially improve diabetes care if used carefully with the understanding of the characteristics of this system. Although there is a time lag of approximately 5-15 min between blood and interstitial glucose levels, the system is considered the most suitable device for meticulous glucose control and prevention of hypoglycemia. A large number of studies have examined its accuracy, safety and clinical effectiveness. The continuous glucose-error grid analysis (CG-EGA), designed by WL Clarke, evaluates the clinical accuracy of CGM. It examines 'temporal' characteristics of the data, analyzing pairs of reference and sensor readings as a process in time represented by a 'bidimensional' time series and taking into account inherent physiological time lags. Investment in CG-EGA is clearly meaningful, even though there are other methodologies for evaluation. The use of each method complementarily is the most effective way to prove the accuracy of the device. The device has improved gradually, and real-time CGM, which allows real-time monitoring of blood glucose level, is already available commercially. The use of real-time CGM could potentially lead to over- or undertreatment with insulin. Patient education through proper and effective handling of the new device is essential to improve diabetes care. (J Diabetes Invest, doi: 10.1111/j.2040-1124.2012.00197.x, 2012).

Assessment of Blood Glucose Predictors: The Prediction-Error Grid Analysis

Prediction of the future blood glucose (BG) evolution from continuous glucose monitoring (CGM) data is a promising direction in diabetes therapy management, and several glucose predictors have recently been proposed. This raises the problem of their assessment. There were attempts to use for such assessment the continuous glucose-error grid analysis (CG-EGA), originally developed for CGM devices. However, in the CG-EGA the BG rate of change is estimated from past BG readings, whereas predictors provide BG estimation ahead of time. Therefore, the original CG-EGA should be modified to assess predictors. Here we propose a new version of the CG-EGA, the Prediction-Error Grid Analysis (PRED-EGA). The analysis is based both on simulated data and on data from clinical trials, performed in the European FP7-project "DIAdvisor." Simulated data are used to test the ability of the analyzed CG-EGA modifications to capture erroneous predictions in controlled situation. Real data are used to show the impact of the different CG-EGA versions in the evaluation of a predictor. Using the data of 10 virtual and 10 real subjects and analyzing two different predictors, we demonstrate that the straightforward application of the CG-EGA does not adequately classify the prediction performance. For example, we observed that up to 70% of 20 min ahead predictions in the hyperglycemia region that are classified by this application as erroneous are, in fact, accurate. Moreover, for predictions during hypoglycemia the assessments produced by the straightforward application of the CG-EGA are not only too pessimistic (in up to 60% of cases), but this version is not able to detect real erroneous predictions. In contrast, the proposed modification of the CG-EGA, where the rate of change is estimated on the predicted BG profile, is an adequate metric for the assessment of predictions. We propose a new CG-EGA, the PRED-EGA, for the assessment of glucose predictors. The presented analysis shows that, compared with the straightforward application of the CG-EGA, the PRED-EGA gives a significant reduction of the misclassification cases. A reduction by a factor of at least 4 was observed in the study. Moreover, the PRED-EGA is much more robust against uncertainty in the input and references.

A New Index to Optimally Design and Compare Continuous Glucose Monitoring Glucose Prediction Algorithms

Continuous glucose monitoring (CGM) data can be exploited to prevent hypo-/hyperglycemic events in real time by forecasting future glucose levels. In the last few years, several glucose prediction algorithms have been proposed, but how to compare them (e.g., methods based on polynomial rather than autoregressive time-series models) and even how to determine the optimal parameter set for a given method (e.g., prediction horizon and forgetting) are open problems. A new index, J, is proposed to optimally design a prediction algorithm by taking into account two key components: the regularity of the predicted profile and the time gained thanks to prediction. Effectiveness of J is compared with previously proposed criteria such as the root mean square error (RMSE) and continuous glucose-error grid analysis (CG-EGA) on 20 Menarini (Florence, Italy) Glucoday® CGM data sets. For a given prediction algorithm, the new index J is able to suggest a more consistent and better parameter set (e.g., prediction horizon and forgetting factor of choice) than RMSE and CG-EGA. In addition, the minimization of J can reliably be used as a selection criterion in comparing different prediction methods. The new index can be used to compare different prediction strategies and to optimally design their parameters.

FreeStyle Navigator Continuous Glucose Monitoring System with TRUstart Algorithm, a 1-Hour Warm-up Time

The first-generation FreeStyle Navigator® Continuous Glucose Monitoring System (FreeStyle Navigator CGM) requires a 10 h warm-up period to avoid inaccurate glucose readings caused by sensor insertion trauma and wound-healing processes. The performance of a second-generation FreeStyle Navigator CGM that begins reporting glucose 1 h after sensor insertion is described. Second-generation FreeStyle Navigator CGM performance was evaluated in an in-clinic study using the YSI Model 2300 STATPlus Glucose Analyzer as reference with 47 subjects with type 1 diabetes. The reference readings were taken at 15 min intervals, and the study was designed to emphasize the first 10 h of use. The second-generation FreeStyle Navigator CGM exhibited continuous glucose error grid analysis ratings of 93.7% "clinically accurate," 3.6% "benign errors," and 2.8% "clinical errors" and a mean and median absolute relative difference of 14.5% and 10.7%, respectively. The second-generation algorithm detected signal instability in the first 10 h of use and suspended the reporting of 14.1% of first day continuous glucose readings. The clinical accuracy of the second-generation FreeStyle Navigator CGM was similar for the first 10 h versus subsequent hours, with 92.6% and 94.2% "clinically accurate" readings, respectively. The warm-up period for the second-generation FreeStyle Navigator CGM was reduced from 10 to 1 h, with minimal interruption of glucose reporting and without sacrificing clinical performance.

Evaluating the Clinical Accuracy of GlucoMen®Day: A Novel Microdialysis-Based Continuous Glucose Monitor

The objective of this work was to determine the clinical accuracy of GlucoMen®Day, a new microdialysis-based continuous glucose monitoring system (CGMS) from A. Menarini Diagnostics (Florence, Italy). Accuracy evaluation was performed using continuous glucose-error grid analysis (CG-EGA), as recommended by the Performance Metrics for Continuous Interstitial Glucose Monitoring; Approved Guideline (POCT05-A). Two independent clinical trials were carried out on patients with types 1 and 2 diabetes mellitus, the glycemic levels of whom were monitored in an in-home setting for 100-hour periods. A new multiparametric algorithm was developed and used to compensate in real-time the GlucoMen®Day signal. The time lag between continuous glucose monitoring (CGM) and reference data was first estimated using the Poincaré plot method. The entire set of CGM/reference data pairs was then evaluated following the CG-EGA criteria, which allowed an estimation of the combined point and rate accuracy stratified by glycemic ranges. With an estimated time lag of 11 minutes, the linear regression analysis of the CGM/reference glucose values yielded r = 0.92. The mean absolute error (MAE) was 11.4 mg/dl. The calculated mean absolute rate deviation (MARD) was 0.63 mg/dl/min. The data points falling within the A+B zones of CG-EGA were 100% in hypoglycemia, 95.7% in euglycemia, and 95.2% in hyperglycemia. The GlucoMen®Day system provided reliable, real-time measurement of subcutaneous glucose levels in patients with diabetes for up to 100 hours. The device showed the ability to follow rapid glycemic excursions and detect severe hypoglycemic events accurately. Its accuracy parameters fitted the criteria of the state-of-the-art consensus guideline for CGMS, with highly consistent results from two independent studies.

Comparison of the Clinical Information Provided by the FreeStyle Navigator Continuous Interstitial Glucose Monitor Versus Traditional Blood Glucose Readings

The purpose of the analysis was to compare the clinical utility of data from traditional self-monitoring of blood glucose (SMBG) to that of continuous glucose monitoring (CGM). A clinical study of the clinical accuracy of the FreeStyle Navigator CGM System (Abbott Diabetes Care, Alameda, CA), which includes SMBG capabilities, was conducted by comparison to the YSI blood glucose analyzer (YSI Inc., Yellow Springs, OH) using 58 subjects with type 1 diabetes. The Continuous Glucose-Error Grid Analysis (CG-EGA) was used as the analytical tool. Using CG-EGA, the "clinically accurate," "benign errors," and "clinical errors" were 86.8%, 8.7%, and 4.5% for SMBG and 92.7%, 3.7%, and 3.6% for CGM, respectively. If blood glucose is viewed as a process in time, SMBG would provide accurate information about this process 86.8% of the time, whereas CGM would provide accurate information about this process 92.7% of the time (P < 0.0001). In the hypoglycemic range, however, SMBG is more accurate as the "clinically accurate," "benign errors," and "clinical errors" were 83.5%, 6.4%, and 10.1% for SMBG and 57.1%, 8.4%, and 34.5% (P < 0.0001) for CGM, respectively. While SMBG produces more accurate instantaneous glucose values than CGM, control of blood glucose involves a system in flux, and CGM provides more detailed insight into the dynamics of that system. In the normal and elevated glucose ranges, the additional information about the direction and rate of glucose change provided by the FreeStyle Navigator CGM System increases the ability to make correct clinical decisions when compared to episodic SMBG tests.

Evaluating Clinical Accuracy of Continuous Glucose Monitoring Systems:Continuous Glucose – Error Grid Analysis (CG-EGA)

Continuous Glucose Sensors (CGS) generate rich and informative continuous data streams which have the potential to improve the glycemic condition of the patient with diabetes. Such data are critical to the development of closed loop systems for automated glycemic control. Thus the numerical and clinical accuracy of such must be assured. Although numerical point accuracy of these systems has been described using traditional statistics, there are no requirements, as of yet, for determining and reporting the rate (trend) accuracy of the data generated. In addition, little attention has been paid to the clinical accuracy. of these systems. Continuous Glucose-Error Grid Analysis (CG-EGA) is the only method currently available for assessing the clinical accuracy of such data and reporting this accuracy for each of the relevant glycemic ranges, - hypoglycemia, euglycemia, hyperglycemia. This manuscript reviews the development of the original Error Grid Analysis (EGA) and describes its inadequacies when used to determine point accuracy of CGS systems. The development of CG-EGA as a logical extension of EGA for use with CGS is described in detail and examples of how it can be used to describe the clinical accuracy of several CGS are shown. Information is presented on how to obtain assistance with the use of CG-EGA.

Comparison of the Numerical and Clinical Accuracy of Four Continuous Glucose Monitors

The purpose of this study was to compare the numerical and clinical accuracy of four continuous glucose monitors (CGMs): Guardian, DexCom, Navigator, and Glucoday. Accuracy data for the four CGMs were collected in two studies: Study 1 enrolled 14 adults with type 1 diabetes at the University of Virginia (UVA), Charlottesville, Virginia; study 2 enrolled 20 adults with type 1 diabetes at the Profil Institute for Metabolic Research, Neuss, Germany. All participants underwent hyperinsulinemic clamps including 1.5-2 h of maintained euglycemia at 5.6 mmol/l followed by descent into hypoglycemia, sustained hypoglycemia at 2.5 mmol/l for 30 min, and recovery. Reference blood glucose sampling was performed every 5 min. The UVA study tested Guardian, DexCom, and Navigator simultaneously; the Profil study tested Glucoday. Regarding numerical accuracy, during euglycemia, the mean absolute relative differences (MARDs) of Guardian, DexCom, Navigator, and Glucoday were 15.2, 21.2, 15.3, and 15.6%, respectively. During hypoglycemia, the MARDs were 16.1, 21.5, 10.3, and 17.5%, respectively. Regarding clinical accuracy, continuous glucose-error grid analysis (CG-EGA) revealed 98.9, 98.3, 98.6, and 95.5% zones A + B hits in euglycemia. During hypoglycemia, zones A + B hits were 84.4, 97.0, and 96.2% for Guardian, Navigator, and Glucoday, respectively. Because of frequent loss of sensitivity, there were insufficient hypoglycemic DexCom data to perform CG-EGA. The numerical accuracy of Guardian, Navigator, and Glucoday was comparable, with an advantage to the Navigator in hypoglycemia; the numerical errors of the DexCom were approximately 30% larger. The clinical accuracy of the four sensors was similar in euglycemia and was higher for the Navigator and Glucoday in hypoglycemia.

Glucose Prediction Algorithms from Continuous Monitoring Data: Assessment of Accuracy via Continuous Glucose Error-Grid Analysis

The aim of this article was to use continuous glucose error-grid analysis (CG-EGA) to assess the accuracy of two time-series modeling methodologies recently developed to predict glucose levels ahead of time using continuous glucose monitoring (CGM) data. We considered subcutaneous time series of glucose concentration monitored every 3 minutes for 48 hours by the minimally invasive CGM sensor Glucoday® (Menarini Diagnostics, Florence, Italy) in 28 type 1 diabetic volunteers. Two prediction algorithms, based on first-order polynomial and autoregressive (AR) models, respectively, were considered with prediction horizons of 30 and 45 minutes and forgetting factors (ff) of 0.2, 0.5, and 0.8. CG-EGA was used on the predicted profiles to assess their point and dynamic accuracies using original CGM profiles as reference. Continuous glucose error-grid analysis showed that the accuracy of both prediction algorithms is overall very good and that their performance is similar from a clinical point of view. However, the AR model seems preferable for hypoglycemia prevention. CG-EGA also suggests that, irrespective of the time-series model, the use of ff = 0.8 yields the highest accurate readings in all glucose ranges. For the first time, CG-EGA is proposed as a tool to assess clinically relevant performance of a prediction method separately at hypoglycemia, euglycemia, and hyperglycemia. In particular, we have shown that CG-EGA can be helpful in comparing different prediction algorithms, as well as in optimizing their parameters.

Continuous Glucose Sensors: Continuing Questions about Clinical Accuracy

Continuous glucose sensors (CGS) offer the potential to greatly change the lives of people with diabetes. Even though two of these systems (Guardian RT, Medtronic, Northridge, CA, and DexCom STS, DexCom, San Diego, CA) have been approved by the Food and Drug Administration for use as adjuncts to self-blood glucose monitoring (SBGM), questions remain concerning the accuracy of these devices. When considering accuracy, two distinct approaches should be emphasized: (1) numerical and (2) clinical. Because CGS data are a process in time, each of these two approaches includes two subtypes of accuracy: point and rate. Conventional statistics such as correlation coefficients, mean and median relative absolute differences, and International Standards Organization criteria are measures of numerical point accuracy. A new measure, the R deviation, is introduced to quantify numerical rate accuracy. Error-grid analysis (Clarke EGA) measures clinical point accuracy. The only measure of both clinical point accuracy and rate accuracy is continuous glucose error-grid analysis. This analysis is a combination of two components, P-EGA measuring point accuracy and R-EGA measuring rate accuracy, which are designed to assess the information that distinguishes continuous glucose measurements from intermittent SBGM determinations. Further, a better understanding of the source of the error associated with time lag and its effect on CGS readings may improve sensor output. Finally, the reliability of the CGS sensors, in terms of initial calibration and long-term application, needs to be assessed carefully if current CGS systems are to be used as hypoglycemia monitors or incorporated in the future design of closed loop (artificial pancreas) systems.

A Critical Appraisal of the Continuous Glucose–Error Grid Analysis

In a recent publication, Wentholt et al. (1) stated that their aim was to critically explore the continuous glucose–error grid analysis (CG-EGA) (2) and to compare it with traditional techniques using data previously reported from two sensors. As developers of the CG-EGA, we hoped that our method might stimulate a discussion on the important problem of the accuracy of continuous monitoring sensors (CGS); therefore, we read this critique with interest. The methods used by Wentholt et al. (1) unfortunately failed to take into account the basic structure of CGS data, which represent time series (i.e., sequential readings that are ordered in time) (3). This structure leads to two fundamental requirements in their analysis. First, consecutive sensor readings taken from the same subject within a relatively short time are highly interdependent. Therefore, standard statistical analyses such as t tests, while appropriate …