Search

Objective: Type-1 diabetes (T1D) is a metabolic disease, characterized by impaired blood glucose (BG) regulation, which forces patients to multiple daily therapeutic actions, the most critical of which is exogenous insulin administration. T1D management can considerably benefit of mathematical models enabling accurate BG predictions and effective/safe automated insulin delivery. In building these models, dealing with large inter- and intra-patient variability in glucose-insulin dynamics represents a major challenge. The aim of the present work is to assess linear black-box methods, including a novel non-parametric methodology, for learning individualized models of glucose response to insulin and meal, suitable for model-based prediction and control. Methods: We focus on data-driven techniques for linear model-learning and compare the state-of-art parametric pipeline, exploring all its degrees of freedom (including population vs. individualized parameter identification, model class chosen among ARX/ARMAX/ARIMAX/Box-Jenkins, model order selection criteria, etc.), with a novel non-parametric approach based on Gaussian regression and stable spline kernel. By using data collected in 11 T1D individuals, we evaluate effectiveness of the different models by measuring root mean squared error (RMSE), coefficient of determination (COD), and time gain of the associated BG predictors. Results: Among the tested approaches, the non-parametric technique results in the best prediction performance: median RMSE=29.8mg/dL, and median COD=57.4%, for a prediction horizon (PH) of 60 min. With respect to the state-of-the-art parametric techniques, the non-parametric approach grants a COD improvement of about 2%, 7%, 21%, and 41% for PH = 30, 60, 90, and 120 min (paired-sample t-test p 0.001, p=0.003, p=0.03, and p=0.07 respectively). Conclusion: Non-parametric linear model-learning grants statistically significant improvement with respect to the state-of-art parametric approach. The higher PH, the more pronounced the improvement. Significance: The use of a linear non-parametric model-learning approach for model-based prediction and control could bring to a more prompt, safe and effective T1D management.

Aims Time-in-ranges, such as time-in-range (TIR), and time-in-tight-range (TITR), time-below-range (TBR), time-above-range (TAR), are key metrics to evaluate glucose control. The duration of the CGM datastream used to compute them affects their reliability: short observation periods result in uncertain estimates, while long trials grant precise evaluation but are associated with higher costs and organizational difficulties. In this paper we propose a mathematical link between the precision/uncertainty of the estimates and the number of monitoring days, that could be useful for clinicians and researchers. Materials and methods Four formulas for the above-mentioned time-in-ranges were obtained by estimating the equation's parameters on a training set extracted from study A (226 subjects, ~180 days, 5-min Dexcom G4 Platinum sensor). The formulas were then validated on the remaining data. We also illustrate how to adjust the parameters for sensors with different sampling rates. Finally, we use Study B (45 subjects, ~365 days, 15-min Abbott Freestyle Libre sensor) to further validate our results. Results Our approach was effective in predicting the uncertainty when time-in-ranges are estimated using n days of CGM, matching the variability observed in the data. As an example, monitoring a population with TIR=70%, TITR=50%, TBR=5% and TAR=25% for n=30 days warrants a precision of ±3.50%, ±3.68%, ±1.33%, ±3.66%, respectively. Conclusions The presented approach can be used both to compute the uncertainty of time-in-ranges and to determine the minimal duration of a trial to achieve a desired precision. An online tool to facilitate its implementation is made freely available to the clinical investigator. This article is protected by copyright. All rights reserved.

Objective: This paper aims at proposing a new machine-learning based model to improve the calculation of mealtime insulin boluses (MIB) in type 1 diabetes (T1D) therapy using continuous glucose monitoring (CGM) data. Indeed, MIB is still often computed through the standard formula (SF), which does not account for glucose rate-of-change ( $\Delta$ G), causing critical hypo/hyperglycemic episodes. Methods: Four candidate models for MIB calculation, based on multiple linear regression (MLR) and least absolute shrinkage and selection operator (LASSO) are developed. The proposed models are assessed in silico , using the UVa/Padova T1D simulator, in different mealtime scenarios and compared to the SF and three $\Delta$ G-accounting variants proposed in the literature. An assessment on real data, by retrospectively analyzing 218 glycemic traces, is also performed. Results: All four tested models performed better than the existing techniques. LASSO regression with extended feature-set including quadratic terms (LASSO $_Q$ ) produced the best results. In silico, LASSO $_Q$ reduced the error in estimating the optimal bolus to only 0.86 U (1.45 U of SF and 1.36–1.44 U of literature methods), as well as hypoglycemia incidence (from 44.41% of SF and 44.60–45.01% of literature methods, to 35.93%). Results are confirmed by the retrospective application to real data. Conclusion: New models to improve MIB calculation accounting for CGM- $\Delta$ G and easy-to-measure features can be developed within a machine learning framework. Particularly, in this paper, a new LASSO $_Q$ model was developed, which ensures better glycemic control than SF and other literature methods. Significance: MIB dosage with the proposed LASSO $_Q$ model can potentially reduce the risk of adverse events in T1D therapy.

Type 1 diabetes (T1D) is a chronic life-threatening metabolic condition which needs to be accurately and continuously managed with care by multiple daily exogenous insulin injections, frequent blood glucose concentration monitoring, ad-hoc diet, and physical activity. In the last decades, new technologies, such as continuous glucose monitoring sensors, eased the burden for T1D patients and opened new therapy perspectives by fostering the development of decision support systems (DSS). A DSS for T1D should be able to provide patients with advice aimed at improving metabolic control and reducing the number of actions related to therapy handling. Major challenges are the vast intra-/inter-subject physiological variability and the many factors that impact glucose metabolism. The present work illustrates a new DSS for T1D management. The algorithmic core includes a module for optimal, personalized, insulin dose calculation and a module that triggers the assumption of rescue carbohydrates to avoid/mitigate impending hypoglycemic events. The algorithms are integrated within a prototype communication platform that comprises a mobile app, a real-time telemonitoring interface, and a cloud server to safely store patients' data. Tests made in silico show that the use of the new algorithms lead to metabolic control indices significantly better than those obtained by the standard care for T1D. The preliminary test of the prototype platform suggests that it is robust, performant, and well-accepted by both patients and clinicians. Future work will focus on the refinement of the communication platform and the design of a clinical trial to assess the system effectiveness in real-life conditions.Clinical Relevance- The presented DSS is a promising tool to facilitate T1D daily management and improve therapy efficacy.

Aims Reliable estimation of the time spent in different glycaemic ranges (time-in-ranges) requires sufficiently long continuous glucose monitoring. In a 2019 paper (Battelino et al., Clinical targets for continuous glucose monitoring data interpretation: recommendations from the international consensus on time in range. Diabetes Care. 2019;42:1593-1603), an international panel of experts suggested using a correlation-based approach to obtain the minimum number of days for reliable time-in-ranges estimates. More recently (in Camerlingo et al., Design of clinical trials to assess diabetes treatment: minimum duration of continuous glucose monitoring data to estimate time-in-ranges with the desired precision. Diabetes Obes Metab. 2021;23:2446-2454) we presented a mathematical equation linking the number of monitoring days to the uncertainty around time-in-ranges estimates. In this work, we compare these two approaches, mainly focusing on time spent in (70-180) mg/dL range (TIR). Methods The first 100 and 150days of data were extracted from study A (148subjects, ~180days), and the first 100, 150, 200, 250 and 300days of data from study B (45subjects, ~365days). For each of these data windows, the minimum monitoring duration was computed using correlation-based and equation-based approaches. The suggestions were compared for the windows of different durations extracted from the same study, and for the windows of equal duration extracted from different studies. Results When changing the dataset duration, the correlation-based approach produces inconsistent results, ranging from 23 to 64days, for TIR. The equation-based approach was found to be robust versus this issue, as it is affected only by the characteristics of the population being monitored. Indeed, to grant a confidence interval of 5% around TIR, it suggests 18days for windows from study A, and 17days for windows from study B. Similar considerations hold for other time-in-ranges. Conclusions The equation-based approach offers advantages for the design of clinical trials having time-in-ranges as final end points, with focus on trial duration.

Accurate blood glucose (BG) forecasting is key in diabetes management, as it allows preventive actions to mitigate harmful hypoglycemic/hyperglycemic episodes. Considering the encouraging results obtained by seasonal stochastic models in proof-of-concept studies, this work assesses the methodology in two datasets (open-loop and closed-loop) recorded in free-living conditions. First, similar postprandial glycemic profiles are grouped together with fuzzy C-means clustering. Then, a seasonal stochastic model is identified for each cluster. Finally, real-time BG forecasting is performed by weighting each model’s prediction. The proposed methodology (named C-SARIMA) is compared to other linear and nonlinear black-box methods: autoregressive integrated moving average (ARIMA), its variant with input (ARIMAX), a feed-forward neural network (NN), and its modified version (NN-X) fed by BG, insulin, and carbohydrates (timing and dosing) information for several prediction horizons (PHs). In the open-loop dataset, C-SARIMA grants a median root-mean-squared error (RMSE) of 20.13 mg/dL (PH = 30) and 27.23 mg/dL (PH = 45), not significantly different from ARIMA and NN. Over a longer PH, C-SARIMA achieves an RMSE = 31.96 mg/dL (PH = 60) and RMSE = 33.91 mg/dL (PH = 75), significantly outperforming the ARIMA and NN, without significant differences from the ARIMAX for PH ≥ 45 and the NN-X for PH ≥ 60. Similar results hold on the closed-loop dataset: for PH = 30 and 45 min, the C-SARIMA achieves an RMSE = 21.63 mg/dL and RMSE = 29.67 mg/dL, not significantly different from the ARIMA and NN. On longer PH, the C-SARIMA outperforms the ARIMA for PH > 45 and the NN for PH > 60 without significant differences from the ARIMAX for PH ≥ 45. Although using less input information, the C-SARIMA achieves similar performance to other prediction methods such as the ARIMAX and NN-X and outperforming the CGM-only approaches on PH > 45min.

In the management of type 1 diabetes (T1D), systematic and random errors in carb-counting can have an adverse effect on glycemic control. In this study, we performed an in silico trial aiming at quantifying the impact of different levels of carb-counting error on glycemic control.The T1D patient decision simulator was used to simulate 7-day glycemic profiles of 100 adults using open-loop therapy. The simulation was repeated for different values of systematic and random carb-counting errors, generated with Gaussian distribution varying the error mean from -10% to +10% and standard deviation (SD) from 0% to 50%. The effect of the error was evaluated by computing the difference of time inside (∆TIR), above (∆TAR) and below (∆TBR) the target glycemic range (70-180mg/dl) compared to the reference case, that is, absence of error. Finally, 3 linear regression models were developed to mathematically describe how error mean and SD variations result in ∆TIR, ∆TAR, and ∆TBR changes.Random errors globally deteriorate the glycemic control; systematic underestimations lead to, on average, up to 5.2% more TAR than the reference case, while systematic overestimation results in up to 0.8% more TBR. The different time in range metrics were linearly related with error mean and SD (iR/isup2/supgt;0.95), with slopes ofmml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"mml:mrowmml:msubmml:miβ/mml:mimml:mrowmml:miM/mml:mimml:miE/mml:mimml:miA/mml:mimml:miN/mml:mi/mml:mrow/mml:msubmml:mo=/mml:momml:mn0/mml:mnmml:mo./mml:momml:mn21/mml:mn/mml:mrow/mml:math,mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"mml:mrowmml:msubmml:miβ/mml:mimml:mrowmml:miS/mml:mimml:miD/mml:mi/mml:mrow/mml:msubmml:mo=/mml:momml:mo-/mml:momml:mn0/mml:mnmml:mo./mml:momml:mn07/mml:mn/mml:mrow/mml:mathfor ∆TIR,mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"mml:mrowmml:msubmml:miβ/mml:mimml:mrowmml:miM/mml:mimml:miE/mml:mimml:miA/mml:mimml:miN/mml:mi/mml:mrow/mml:msubmml:mo=/mml:momml:mo-/mml:momml:mn0/mml:mnmml:mo./mml:momml:mn25/mml:mn/mml:mrow/mml:math,mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"mml:mrowmml:msubmml:miβ/mml:mimml:mrowmml:miS/mml:mimml:miD/mml:mi/mml:mrow/mml:msubmml:mo=/mml:momml:mo+/mml:momml:mn0/mml:mnmml:mo./mml:momml:mn06/mml:mn/mml:mrow/mml:mathfor ∆TAR, andmml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"mml:mrowmml:msubmml:miβ/mml:mimml:mrowmml:miM/mml:mimml:miE/mml:mimml:miA/mml:mimml:miN/mml:mi/mml:mrow/mml:msubmml:mo=/mml:momml:mn0/mml:mnmml:mo./mml:momml:mn05/mml:mn/mml:mrow/mml:math,mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"mml:mrowmml:msubmml:miβ/mml:mimml:mrowmml:miS/mml:mimml:miD/mml:mi/mml:mrow/mml:msubmml:mo=/mml:momml:mo+/mml:momml:mn0/mml:mnmml:mo./mml:momml:mn01/mml:mn/mml:mrow/mml:mathfor ∆TBR.The quantification of carb-counting error impact performed in this work may be useful understanding causes of glycemic variability and the impact of possible therapy adjustments or behavior changes in different glucose metrics.

In type 1 diabetes management, the availability of algorithms capable of accurately forecasting future blood glucose (BG) concentrations and hypoglycemic episodes could enable proactive therapeutic actions, e.g., the consumption of carbohydrates to mitigate, or even avoid, an impending critical event. The only input of this kind of algorithm is often continuous glucose monitoring (CGM) sensor data, because other signals (such as injected insulin, ingested carbs, and physical activity) are frequently unavailable. Several predictive algorithms fed by CGM data only have been proposed in the literature, but they were assessed using datasets originated by different experimental protocols, making a comparison of their relative merits difficult. The aim of the present work was to perform a head-to-head comparison of thirty different linear and nonlinear predictive algorithms using the same dataset, given by 124 CGM traces collected over 10 days with the newest Dexcom G6 sensor available on the market and considering a 30-min prediction horizon. We considered the state-of-the art methods, investigating, in particular, linear black-box methods (autoregressive, autoregressive moving-average, and autoregressive integrated moving-average, ARIMA) and nonlinear machine-learning methods (support vector regression, SVR, regression random forest, feed-forward neural network, fNN, and long short-term memory neural network). For each method, the prediction accuracy and hypoglycemia detection capabilities were assessed using either population or individualized model parameters. As far as prediction accuracy is concerned, the results show that the best linear algorithm (individualized ARIMA) provides accuracy comparable to that of the best nonlinear algorithm (individualized fNN), with root mean square errors of 22.15 and 21.52 mg/dL, respectively. As far as hypoglycemia detection is concerned, the best linear algorithm (individualized ARIMA) provided precision = 64%, recall = 82%, and one false alarm/day, comparable to the best nonlinear technique (population SVR): precision = 63%, recall = 69%, and 0.5 false alarms/day. In general, the head-to-head comparison of the thirty algorithms fed by CGM data only made using a wide dataset shows that individualized linear models are more effective than population ones, while no significant advantages seem to emerge when employing nonlinear methodologies.

We present a new mobile platform to be used in clinical trials aimed at both collecting data and assessing new technologies and treatments for diabetes care. The main components of the platform are a mobile app, that automatically collects data from continuous glucose monitoring sensors and activity trackers, and also allows users to manually log daily events; a cloud database for safe data storage; a web interface, which allows clinicians to monitor patients’ status in real-time. The platform is modular and highly customizable for a multitude of purposes in clinical research. Preliminary tests performed for daily-life data gathering by both clinicians and users are extremely encouraging.

Objective: The artificial pancreas (AP) is an innovative closed-loop system for type 1 diabetes therapy, in which insulin is infused by portable pumps and insulin dosage is modulated by a control algorithm on the basis of the measurements collected by continuous glucose monitoring (CGM) sensors. AP systems safety and effectiveness could be affected by several technological and user-related issues, among which insulin pump faults and missed meal announcements. This work proposes an algorithm to detect in real-time these two types of failure. Methods: The algorithm works as follows. First, a personalized autoregressive moving-average model with exogenous inputs is identified using historical data of the patient. Second, the algorithm is used in real time to predict future CGM values. Then, alarms are triggered when the difference between predicted vs measured CGM values is higher than opportunely set thresholds. In addition, by using two different set of parameters, the algorithm is able to distinguish the two types of failures. The algorithm was developed and assessed in silico using the latest version of the FDA-approved Padova/UVa T1D simulator. Results: The algorithm showed a sensitivity of $\sim$ 81.3% on average when detecting insulin pump faults with $\sim$ 0.15 false positives per day on average. Missed meal announcements were detected with a sensitivity of $\sim$ 86.8% and 0.15 FP/day. Conclusion: The presented method is able to detect insulin pump faults and missed meal announcements in silico , correctly distinguishing one from another. Significance: The method increases the safety of AP systems by providing prompt alarms to the diabetic subject and effectively discriminating pump malfunctioning from user errors.

Type 1 diabetes (T1D) therapy requires multiple daily insulin injections to compensate the lack of endogenous insulin production due to β-cells destruction. An empirical standard formula (SF) is commonly used for such a task. Unfortunately, SF does not include information on glucose dynamics, e.g. the glucose rate-of-change (ROC) provided by continuous glucose monitoring (CGM) sensor. Hence, SF can sometimes lead to under/overestimations that can cause critical hypo/hyperglycemic episodes during/after the meal. Recently, to overcome this limitation, we proposed new linear regression models, integrating ROC information and personalized features. Despite the first encouraging results, the nonlinear nature of the problem calls for the application of nonlinear models. In this work, random forest (RF) and gradient boosting tree (GBT), nonlinear machine learning methodologies, were investigated. A dataset of 100 virtual subjects, opportunely divided into training and testing sets, was used. For each individual, a single-meal scenario with different meal conditions (preprandial ROC, BG and meal amounts) was simulated. The assessment was performed both in terms of accuracy in estimating the optimal bolus and glycemic control. Results were compared to the best performing linear model previously developed. The two tree-based models proposed lead to a statistically significant improvement of glycemic control compared to the linear approach, reducing the time spent in hypoglycemia (from 32.49% to 27.57-25.20% for RF and GBT, respectively). These results represent a preliminary step to prove that nonlinear machine learning techniques can improve the estimation of insulin bolus in T1D therapy. Particularly, RF and GBT were shown to outperform the previously linear models proposed.Clinical Relevance— Insulin bolus estimation with nonlinear machine learning techniques reduces the risk of adverse events in T1D therapy.

Wearable continuous glucose monitoring (CGM) sensors are revolutionizing the treatment of type 1 diabetes (T1D). These sensors provide in real-time, every 1–5 min, the current blood glucose concentration and its rate-of-change, two key pieces of information for improving the determination of exogenous insulin administration and the prediction of forthcoming adverse events, such as hypo-/hyper-glycemia. The current research in diabetes technology is putting considerable effort into developing decision support systems for patient use, which automatically analyze the patient’s data collected by CGM sensors and other portable devices, as well as providing personalized recommendations about therapy adjustments to patients. Due to the large amount of data collected by patients with T1D and their variety, artificial intelligence (AI) techniques are increasingly being adopted in these decision support systems. In this paper, we review the state-of-the-art methodologies using AI and CGM sensors for decision support in advanced T1D management, including techniques for personalized insulin bolus calculation, adaptive tuning of bolus calculator parameters and glucose prediction.

Mathematical physiological models of type 1 diabetes (T1D) glucose-insulin dynamics have been of great help in designing and preliminary assessing new algorithm for glucose control. Derivation of models at the individual level is however difficult because of identifiability issues. Recently, fitting these models against data of real patients with T1D has been made possible by both the use of Bayesian estimation techniques and the availability of individual datasets including plasma glucose and insulin concentration samples gathered in clinical protocols. The aim of this work is to make a step further and develop a methodology able to estimate the parameters of T1D physiological models using easily accessible data only, i.e. continuous glucose monitoring (CGM) sensor, carbohydrate intakes (CHO), and exogenous insulin infusion (I) data. The methodology is tested on synthetic data of 100 patients generated by a composite model of glucose-insulin dynamics. To solve identifiability problems, a Bayesian approach numerically implemented by Markov Chain Monte Carlo (MCMC) has been used to obtain point estimates and confidence intervals of model unknown parameters exploiting a priori knowledge available from the literature. Results show goodness of model fit and acceptable precision of parameter estimates. The methodology is also successful in reconstructing of "non-accessible" glucose-insulin fluxes, i.e. glucose rate of appearance and plasma insulin. These preliminary results encourage further development of this framework and its assessment in more challenging setups.

The recent announcement of the production of new low-cost continuous glucose monitoring (CGM) sensors, the approval of marketed CGM sensors for making treatment decisions, and new reimbursement criteria have the potential to revolutionize CGM use. After briefly summarizing current CGM applications, we discuss how, in our opinion, these changes are expected to extend CGM utilization beyond diabetes patients, for example, to subjects with prediabetes or even healthy individuals. We also elaborate on how the integration of CGM data with other relevant information, for example, health records and other medical device/wearable sensor data, will contribute to creating a digital data ecosystem that will improve our understanding of the etiology and complications of diabetes and will facilitate the development of data analytics for personalized diabetes management and prevention.

Background: In type 1 diabetes (T1D) therapy, the calculation of the meal insulin bolus is performed according to a standard formula (SF) exploiting carbohydrate intake, carbohydrate-to-insulin ratio, correction factor, insulin on board, and target glucose. Recently, some approaches were proposed to account for preprandial glucose rate of change (ROC) in the SF, including those by Scheiner and by Pettus and Edelman. Here, the aim is to develop a new approach, based on neural networks (NN), to optimize and personalize the bolus calculation using continuous glucose monitoring information and some easily accessible patient parameters. Method: The UVa/Padova T1D Simulator was used to simulate data of 100 virtual adults in a single-meal noise-free scenario with different conditions in terms of meal amount and preprandial blood glucose and ROC values. An NN was trained to learn the optimal insulin dose using the SF parameters, ROC, body weight, insulin pump basal infusion rate and insulin sensitivity as features. The performance of the NN for meal bolus calculation was assessed by blood glucose risk index (BGRI) and compared to the methods by Scheiner and by Pettus and Edelman. Results: The NN approach brings to a small but statistically significant ( P < .001) reduction of BGRI value, equal to 0.37, 0.23, and 0.20 versus SF, Scheiner, and Pettus and Edelman, respectively. Conclusion: This preliminary study showed the potentiality of using NNs for the personalization and optimization of the meal insulin bolus calculation. Future work will deal with more realistic scenarios including technological and physiological/behavioral sources of variability.

Factory-calibrated continuous glucose monitoring (FC-CGM) sensors are new devices used in type 1 diabetes (T1D) therapy to measure the glucose concentration almost continuously for 10&ndash, 14 days without requiring any in vivo calibration. Understanding and modelling CGM errors is important when designing new tools for T1D therapy. Available literature CGM error models are not suitable to describe the FC-CGM sensor error, since their domain of validity is limited to 12-h time windows, i.e., the time between two consecutive in vivo calibrations. The aim of this paper is to develop a model of the error of FC-CGM sensors. The dataset used contains 79 FC-CGM traces collected by the Dexcom G6 sensor. The model is designed to dissect the error into its three main components: effect of plasma-interstitium kinetics, calibration error, and random measurement noise. The main novelties are the model extension to cover the entire sensor lifetime and the use of a new single-step identification procedure. The final error model, which combines a first-order linear dynamic model to describe plasma-interstitium kinetics, a second-order polynomial model to describe calibration error, and an autoregressive model to describe measurement noise, proved to be suitable to describe FC-CGM sensor errors, in particular improving the estimation of the physiological time-delay.

In Type 1 diabetes (T1D) standard treatment, the mitigation of hypoglycemia is achieved by the assumption of small amounts of carbohydrates (CHO), called hypotreatments (HTs), as soon as hypoglycemia is revealed. However, since CHO takes time to reach the blood stream, hypoglycemia cannot be totally avoided. Our purpose is to evaluate in-silico the effectiveness of preventive HTs and to propose a new real-time algorithm for the mitigation/avoidance of hypoglycemia, based on continuous glucose monitoring (CGM) sensor data. To such a purpose, the algorithm exploits the "dynamic risk" non linear-function that, by combining CGM value and trend, allows predicting the forthcoming hypoglycemic event. The algorithm is tested in an ideal noise-free environment on 100 virtual subjects (VSs) generated by the UVA/Padova T1D simulator and undergoing a single-meal experiment, with induced post-meal hypoglycemia. Compared to a reference HT rule, which suggest to assume HTs when hypoglycemia is detected, the algorithm reduces, on median [25th – 75th percentiles], both the time spent in hypoglycemia (from 36 [29 – 43] min to 10 [0 – 20] min) and the post-treatment rebound (from 136 [121 – 148] mg/dl to 114 [98 – 130] mg/dl). In conclusion, the proposed real-time algorithm efficiently generates preventive HTs that allow to almost totally avoid hypoglycemia. Future work will concern to modify the algorithm for detecting in advance the severity of the hypoglycemic episode –since performance are influenced on the hypoglycemic episode aggressiveness level- and to assess algorithm in a more challenging environment, including CGM measurement error.

Background: Many glycemic variability (GV) indices exist in the literature. In previous works, we demonstrated that a set of GV indices, extracted from continuous glucose monitoring (CGM) data, can distinguish between stages of diabetes progression. We showed that 25 indices driving a logistic regression classifier can differentiate between healthy and nonhealthy individuals; whereas 37 GV indices and four individual parameters, feeding a polynomial-kernel support vector machine (SVM), can further distinguish between impaired glucose tolerance (IGT) and type 2 diabetes (T2D). The latter approach has some limitations to interpretability (complex model, extensive index pool). In this article, we try to obtain the same performance with a simpler classifier and a parsimonious subset of indices. Methods: We analyzed the data of 62 subjects with IGT or T2D. We selected 17 interpretable GV indices and four parameters (age, sex, BMI, waist circumference). We trained a SVM on the data of a baseline visit and tested it on the follow-up visit, comparing the results with the state-of-art methods. Results: The linear SVM fed by a reduced subset of 17 GV indices and four basic parameters achieved 82.3% accuracy, only marginally worse than the reference 87.1% (41-features polynomial-kernel SVM). Cross-validation accuracies were comparable (69.6% vs 72.5%). Conclusion: The proposed SVM fed by 17 GV indices and four parameters can differentiate between IGT and T2D. Using a simpler model and a parsimonious set of indices caused only a slight accuracy deterioration, with significant advantages in terms of interpretability.

Even if still at an early stage of development, non-invasive continuous glucose monitoring (NI-CGM) sensors represent a promising technology for optimizing diabetes therapy. Recent studies showed that the Multisensor provides useful information about glucose dynamics with a mean absolute relative difference (MARD) of 35.4% in a fully prospective setting. Here we propose a method that, exploiting the same Multisensor measurements, but in a retrospective setting, achieves a much better accuracy. Data acquired by the Multisensor during a long-term study are retrospectively processed following a two-step procedure. First, the raw data are transformed to a blood glucose (BG) estimate by a multiple linear regression model. Then, an enhancing module is applied in cascade to the regression model to improve the accuracy of the glucose estimation by retrofitting available BG references through a time-varying linear model. MARD between the retrospectively reconstructed BG time-series and reference values is 20%. Here, 94% of values fall in zone A or B of the Clarke Error Grid. The proposed algorithm achieved a level of accuracy that could make this device a potential complementary tool for diabetes management and also for guiding prediabetic or nondiabetic users through life-style changes.

In the DexCom G4 Platinum (DG4P) continuous glucose monitoring (CGM) sensor, the raw current signal generated by glucose-oxidase is transformed to glucose concentration by a calibration function whose parameters are periodically updated by matching self-monitoring of blood glucose references, usually twice a day, to compensate for sensor variability in time. The aim of this work is to reduce DG4P calibration frequency to once a day by a recently proposed Bayesian calibration algorithm, which employs a time-varying calibration function and suitable day-specific priors.The database consists of 57 CGM signals that are collected by the DG4P for 7 days. The Bayesian calibration algorithm is used to calibrate the raw current signal following two different schedules, that is, two and one calibration per day. Calibrated glycemic profiles are compared with those originally acquired by the manufacturer, on days 1, 4, and 7, where frequent blood glucose references were available, by using standard metrics, that is, mean absolute relative difference (MARD), percentage of accurate glucose estimates, and percentage of data in the A-zone of Clarke Error Grid.The one per day Bayesian calibration algorithm has accuracy similar to that of two per day (11.8% vs. 11.7% MARD, respectively), and it is statistically better (P-value of 0.0411) than the manufacturer calibration algorithm, which requires two calibrations per day (13.1% MARD).A Bayesian calibration algorithm employing a time-varying calibration function and suitable priors enables a reduction of the calibrations of DG4P sensor from two to one per day.

Background: In type 1 diabetes (T1D) management, short-term glucose prediction can allow to anticipate therapeutic decisions when hypo/hyperglycemia is imminent. Literature prediction methods mainly use past continuous glucose monitoring (CGM) readings. Sophisticated algorithms can use information on insulin delivered and meal carbohydrate (CHO) content. The quantification of how much insulin and CHO information improves glucose prediction is missing in the literature and is investigated, in an open-loop setting, in this proof-of-concept study. Methods: We adopted a versatile literature prediction methodology able to utilize a variety of inputs. We compared predictors that use (1) CGM; (2) CGM and insulin; (3) CGM and CHO; and (4) CGM, insulin, and CHO. Data of 15 T1D subjects in open-loop setup were used. Prediction was evaluated via absolute error and temporal gain focusing on meal/night periods. The relative importance of each individual input of the predictor was evaluated with a sensitivity analysis. Results: For a prediction horizon (PH) ≥ 30 minutes, insulin and CHO information improves prediction accuracy of 10% and double the temporal gain during the 2 hours following the meal. During the night the 4 methods did not give statistically different results. When PH ≥ 45 minutes, the influence of CHO information on prediction is 5-fold that of insulin. Conclusions: In an open-loop setting, with PH ≥ 30 minutes, information on CHO and insulin improves short-term glucose prediction in the 2-hour time window following a meal, but not during the night. CHO information improves prediction significantly more than insulin.

The most important objective of any diabetes therapy is to maintain the blood glucose concentration within the euglycemic range, avoiding or at least mitigating critical hypo/hyperglycemic episodes. Modern continuous glucose monitoring (CGM) devices bear the promise of providing the patients with an increased and timely awareness of glycemic conditions as these get dangerously near to hypo/hyperglycemia. The challenge is to detect, with reasonable advance, the patterns leading to risky situations, allowing the patient to make therapeutic decisions on the basis of future (predicted) glucose concentration levels. We underline that a technically sound performance comparison of the approaches proposed in recent years has yet to be done, thus it is unclear which one is preferred. The aim of this study is to fill this gap by carrying out a comparative analysis among the most common methods for glucose event prediction. Both regression and classification algorithms have been implemented and analyzed, including static and dynamic training approaches. The dataset consists of 89 CGM time series measured in diabetic subjects for 7 subsequent days. Performance metrics, specifically defined to assess and compare the event-prediction capabilities of the methods, have been introduced and analyzed. Our numerical results show that a static training approach exhibits better performance, in particular when regression methods are considered. However, classifiers show some improvement when trained for a specific event category, such as hyperglycemia, achieving performance comparable to the regressors, with the advantage of predicting the events sooner.

Type 1 diabetes (TID) therapy is based on multiple daily injections of exogenous insulin. The so-called insulin bolus calculators facilitate insulin dose calculation to the patients by implementing a standard formula SF which, besides some patient-related parameters, also considers the current value of blood glucose concentration (BG), normally measured by the patient through a fingerprick device. The recent approval by the U.S. Food and Drug Administration to use the measurements collected by wearable continuous glucose monitoring (CGM) sensors for insulin dosing of fers new perspectives. Indeed, CGM sensors provide real-time information on both glucose concentration and rate of change, currently not considered in the SF. The purpose of this work is to preliminary investigate the possibility of using neural networks (NN)s for the calculation of meal insulin bolus dose exploiting CGM-based information. Using the UVa/Padova TID Simulator, we generated data of 100 subjects in 9-h, single-meal, noise-free scenarios. In particular, for each subject we analyzed different meal conditions in terms of carbohydrate intakes, preprandial BG and glucose rate-of -change. Then, a fully-connected feedforward NN was trained, with the aim of estimating the insulin bolus needed to obtain the best glycemic outcomes according to the blood glucose risk index (BGRI). Preliminary results show that by using the NN to calculate insulin doses lower BGRI values are obtained, on average, compared to the SF. These results encourage further development of the approach and its assessment in more challenging scenarios.

Background: The standard formula (SF) used in bolus calculators (BCs) determines meal insulin bolus using “static” measurement of blood glucose concentration (BG) obtained by self-monitoring of blood glucose (SMBG) fingerprick device. Some methods have been proposed to improve efficacy of SF using “dynamic” information provided by continuous glucose monitoring (CGM), and, in particular, glucose rate of change (ROC). This article compares, in silico and in an ideal framework limiting the exposition to possibly confounding factors (such as CGM noise), the performance of three popular techniques devised for such a scope, that is, the methods of Buckingham et al (BU), Scheiner (SC), and Pettus and Edelman (PE). Method: Using the UVa/Padova Type 1 diabetes simulator we generated data of 100 virtual subjects in noise-free, single-meal scenarios having different preprandial BG and ROC values. Meal insulin bolus was computed using SF, BU, SC, and PE. Performance was assessed with the blood glucose risk index (BGRI) on the 9 hours after meal. Results: On average, BU, SC, and PE improve BGRI compared to SF. When BG is rapidly decreasing, PE obtains the best performance. In the other ROC scenarios, none of the considered methods prevails in all the preprandial BG conditions tested. Conclusion: Our study showed that, at least in the considered ideal framework, none of the methods to correct SF according to ROC is globally better than the others. Critical analysis of the results also suggests that further investigations are needed to develop more effective formulas to account for ROC information in BCs.

Minimally invasive continuous glucose monitoring (CGM) sensors are wearable medical devices that provide real-time measurement of subcutaneous glucose concentration. This can be of great help in the daily management of diabetes. Most of the commercially available CGM devices have a wire-based sensor, usually placed in the subcutaneous tissue, which measures a “raw” current signal via a glucose-oxidase electrochemical reaction. This electrical signal needs to be translated in real-time to glucose concentration through a calibration process. For such a scope, the first commercialized CGM sensors implemented simple linear regression techniques to fit reference glucose concentration measurements periodically collected by fingerprick. On the one hand, these simple linear techniques required several calibrations per day, with the consequent patient’s discomfort. On the other, only a limited accuracy was achieved. This stimulated researchers to propose, over the last decade, more sophisticated algorithms to calibrate CGM sensors, resorting to suitable signal processing, modelling, and machine-learning techniques. This review paper will first contextualize and describe the calibration problem and its implementation in the first generation of CGM sensors, and then present the most recently-proposed calibration algorithms, with a perspective on how these new techniques can influence future CGM products in terms of accuracy improvement and calibration reduction.

Continuous glucose monitoring (CGM) sensors need to be calibrated twice/day by using self-monitoring of blood glucose (SMBG) samples. Recently, to reduce the calibration frequency, we developed an online calibration algorithm based on a multiple-day model of sensor time variability and Bayesian parameter estimation. When applied to Dexcom G4 Platinum (DG4P) sensor data, the algorithm allowed the frequency of calibrations to be reduced to one-every-four-days without significant worsening of sensor accuracy. The aim of this study is to assess the same methodology on raw CGM data acquired by a next-generation Dexcom CGM sensor prototype and compare the results with that obtained on DG4P.The database consists of 55 diabetic subjects monitored for 10 days by a next-generation Dexcom CGM sensor prototype. The new calibration algorithm is assessed, retrospectively, by simulating an online procedure using progressively fewer SMBG samples until zero. Accuracy is evaluated with mean absolute relative differences (MARD) between blood glucose versus CGM values.The one-per-day and one-every-two-days calibration scenarios in the next-generation CGM data have an accuracy of 8.5% MARD (vs. 11.59% of DG4P) and 8.4% MARD (vs. 11.63% of DG4P), respectively. Accuracy slightly worsens to 9.2% (vs. 11.62% of DG4P) when calibrations are reduced to one-every-four-days. The calibration-free scenario results in 9.3% MARD (vs. 12.97% of DG4P).In next-generation Dexcom CGM sensor data, the use of an online calibration algorithm based on a multiple-day model of sensor time variability and Bayesian parameter estimation aids in the shift toward a calibration-free scenario with even better results than those obtained in present-generation sensors.

Background and aims This study compared the accuracy of the FreeStyle Libre (Abbott, Alameda, CA) and Dexcom G4 Platinum (DG4P, Dexcom, San Diego, CA) CGM sensors. Methods and results Twenty-two adults with type 1 diabetes wore the two sensors simultaneously for 2 weeks. Libre was used according to manufacturer-specified lifetime (MSL); DG4P was used 7 days beyond MSL. At a clinical research center (CRC), subjects were randomized to receive the same breakfast with standard insulin bolus (standard) or a delayed and increased (delayed & increased) bolus to induce large glucose swings during weeks 1 and 2; venous glucose was checked every 5–15 min for 6 h. Subjects performed ≥4 reference fingersticks/day at home. Accuracy was assessed by differences in mean absolute relative difference (%MARD) in glucose levels compared with fingerstick test (home use) and YSI reference (CRC). During home-stay the Libre MARD was 13.7 ± 3.6% and the DG4P MARD 12.9 ± 2.5% (difference not significant [NS]). With both systems MARD increased during hypoglycaemia and decreased during hyperglycaemia, without significant difference between sensors. In the euglycaemic range MARD was smaller with DG4P [12.0 ± 2.4% vs 14.0 ± 3.6%, p = 0.026]. MARD increased in both sensors following delayed & increased vs. standard bolus (Libre: 14.9 ± 5.5% vs. 10.9 ± 4.1%, p = 0.008; DG4P: 18.1 ± 8.1% vs. 13.1 ± 4.6%, p = 0.026); between-sensor differences were not significant (p = 0.062). Libre was more accurate during moderate and rapid glucose changes. Conclusions DG4P and Libre performed similarly up to 7 days beyond DG4P MSL. Both sensors performed less well during hypoglycaemia but Libre was more accurate during glucose swings. Trial registration The study was registered in ClinicalTrials.gov ( NCT02734745 ) April 12, 2016.

Objective: In most continuous glucose monitoring (CGM) devices used for diabetes management, the electrical signal measured by the sensor is transformed to glucose concentration by a calibration function whose parameters are estimated using self-monitoring of blood glucose (SMBG) samples. The calibration function is usually a linear model approximating the nonlinear relationship between electrical signal and glucose concentration in certain time intervals. Thus, CGM devices require frequent calibrations, usually twice a day. The aim here is to develop a new method able to reduce the frequency of calibrations. Methods : The algorithm is based on a multiple-day model of sensor time-variability with second-order statistical priors on its unknown parameters. In an online setting, these parameters are numerically determined by the Bayesian estimation exploiting SMBG sparsely collected by the patient. The method is assessed retrospectively on 108 CGM signals acquired for 7 days by the Dexcom G4 Platinum sensor, testing progressively less-calibration scenarios. Results : Despite the reduction of calibration frequency (on average from 2/day to 0.25/day), the method shows a statistically significant accuracy improvement compared to manufacturer calibration, e.g., mean absolute relative difference when compared to a laboratory reference decreases from 12.83% to 11.62% ( p -value of 0.006). Conclusion : The methodology maintains (sometimes improves) CGM sensor accuracy compared to that of the original manufacturer, while reducing the frequency of calibrations. Significance : Reducing the need of calibrations facilitates the adoption of CGM technology both in terms of ease of use and cost, an obvious prerequisite for its use as replacement of traditional SMBG devices.

Many glycaemic variability (GV) indices extracted from continuous glucose monitoring systems data have been proposed for the characterisation of various aspects of glucose concentration profile dynamics in both healthy and non-healthy individuals. However, the inter-index correlations have made it difficult to reach a consensus regarding the best applications or a subset of indices for clinical scenarios, such as distinguishing subjects according to diabetes progression stage. Recently, a logistic regression-based method was used to address the basic problem of differentiating between healthy subjects and those affected by impaired glucose tolerance (IGT) or type 2 diabetes (T2D) in a pool of 25 GV-based indices. Whereas healthy subjects were classified accurately, the distinction between patients with IGT and T2D remained critical. In the present work, by using a dataset of CGM time-series collected in 62 subjects, we developed a polynomial-kernel support vector machine-based approach and demonstrated the ability to distinguish between subjects affected by IGT and T2D based on a pool of 37 GV indices complemented by four basic parameters—age, sex, BMI, and waist circumference—with an accuracy of 87.1%.

Minimally-invasive continuous glucose monitoring (CGM) sensors have revolutionized perspectives in the treatment of type 1 diabetes (T1D). Their accuracy relies on an internal calibration function that transforms the raw, physically measured, electrical data into blood glucose concentration values. Usually, a unique, pre-determined, calibration functional is adopted, with parameters periodically updated in individual patients by using “gold standard” references suitably collected by finger prick devices. However, retrospective analysis of CGM data suggests that variability of sensor-subject characteristics is often inefficiently coped with. In the present study, we propose a conceptual Bayesian model- selection framework aimed at guaranteeing wide margins of flexibility for both the determination of the most appropriate calibration functional and the numerical values of its unknown parameters. The calibration model is determined among a finite specified set of candidates, each one depending on a set of unknown model parameters, for which a priori statistical expectations are available. Model selection is based on predictive distributions carrying out asymptotic calculations through Monte Carlo integration methods. Performance of the proposed approach is assessed on synthetic data generated by a well-established T1D simulation model.

Background: Abnormal glucose variability (GV) is a risk factor for diabetes complications, and tens of indices for its quantification from continuous glucose monitoring (CGM) time series have been proposed. However, the information carried by these indices is redundant, and a parsimonious description of GV can be obtained through sparse principal component analysis (SPCA). We have recently shown that a set of 10 metrics selected by SPCA is able to describe more than 60% of the variance of 25 GV indicators in type 1 diabetes (T1D). Here, we want to extend the application of SPCA to type 2 diabetes (T2D). Methods: A data set of CGM time series collected in 13 T2D subjects was considered. The 25 GV indices considered for T1D were evaluated. SPCA was used to select a subset of indices able to describe the majority of the original variance. Results: A subset of 10 indicators was selected and allowed to describe 83% of the variance of the original pool of 25 indices. Four metrics sufficient to describe 67% of the original variance turned out to be shared by the parsimonious sets of indices in T1D and T2D. Conclusions: Starting from a pool of 25 indices assessed from CGM time series in T2D subjects, reduced subsets of metrics virtually providing the same information content can be determined by SPCA. The fact that these indices also appear in the parsimonious description of GV in T1D may indicate that they could be particularly informative of GV in diabetes, regardless of the specific type of disease.

ObjectiveType 2 diabetes arises from the interaction of physiological and lifestyle risk factors. Our objective was to develop a model for predicting the risk of T2D, which could use various amounts of background information.Research design and methodsWe trained a survival analysis model on 8483 people from three large Finnish and Spanish data sets, to predict the time until incident T2D. All studies included anthropometric data, fasting laboratory values, an oral glucose tolerance test (OGTT) and information on co-morbidities and lifestyle habits. The variables were grouped into three sets reflecting different degrees of information availability. Scenario 1 included background and anthropometric information; Scenario 2 added routine laboratory tests; Scenario 3 also added results from an OGTT. Predictive performance of these models was compared with FINDRISC and Framingham risk scores.ResultsThe three models predicted T2D risk with an average integrated area under the ROC curve equal to 0.83, 0.87 and 0.90, respectively, compared with 0.80 and 0.75 obtained using the FINDRISC and Framingham risk scores. The results were validated on two independent cohorts. Glucose values and particularly 2-h glucose during OGTT (2h-PG) had highest predictive value. Smoking, marital and professional status, waist circumference, blood pressure, age and gender were also predictive.ConclusionsOur models provide an estimation of patient’s risk over time and outweigh FINDRISC and Framingham traditional scores for prediction of T2D risk. Of note, the models developed in Scenarios 1 and 2, only exploited variables easily available at general patient visits.

Objective: Type-1 diabetes (T1D) treatment requires exogenous insulin administrations finely tuned based on glucose monitoring to avoid hyper/hypoglycemia. The safety and effectiveness of insulin treatments is commonly assessed in clinical trials, which are time demanding and expensive. These limitations can be overtaken by in silico clinical trials (ISCT) that require realistic patient and treatment models. The aim is to develop a T1D patient decision simulator usable to perform reliable ISCT. Methods: The T1D patient decision simulator was developed by connecting the UVA/Padova T1D model, which describes glucose, insulin, and glucagon kinetics, with modules describing glucose monitoring devices, like self-monitoring of blood glucose (SMBG) and continuous glucose monitoring (CGM), the patient's behavior in making treatment decisions, and insulin administration. The reliability of the simulator was assessed by comparing its predictions with data collected in 44 T1D subjects using the Dexcom G5 Mobile CGM sensor as an adjunct to the Bayer Contour Next USB SMBG device. Results: Metrics like time spent in eu/hypo/hyperglycemia of simulated data well match those observed in subjects. In particular, mean time in euglycemia, mean time in hyperglycemia, and median time in hypoglycemia are 61.75% versus 63.60% ( p -value = 0.4825), 33.38% versus 33.40% ( p -value = 0.9950), and 3.17% versus 2.14% (p-value = 0.1134), respectively, in real versus simulated data. Conclusion: The proposed simulator can be used to perform credible ISCT in realistic insulin treatment scenarios. Significance: The T1D patient decision simulator can be used to reliably assess novel insulin treatments, e.g., based on use of CGM only, in a realistic multiple-day scenario.

Continuous glucose monitors (CGMs) are prone to inaccuracy due to time lags, sensor drift, calibration errors, and measurement noise. The aim of this study is to derive the model of the error of the second generation Medtronic Paradigm Veo Enlite (ENL) sensor and compare it with the Dexcom SEVEN PLUS (7P), G4 PLATINUM (G4P), and advanced G4 for Artificial Pancreas studies (G4AP) systems. An enhanced methodology to a previously employed technique was utilized to dissect the sensor error into several components. The dataset used included 37 inpatient sessions in 10 subjects with type 1 diabetes (T1D), in which CGMs were worn in parallel and blood glucose (BG) samples were analyzed every 15 ± 5 min Calibration error and sensor drift of the ENL sensor was best described by a linear relationship related to the gain and offset. The mean time lag estimated by the model is 9.4 ± 6.5 min. The overall average mean absolute relative difference (MARD) of the ENL sensor was 11.68 ± 5.07% Calibration error had the highest contribution to total error in the ENL sensor. This was also reported in the 7P, G4P, and G4AP. The model of the ENL sensor error will be useful to test the in silico performance of CGM-based applications, i.e., the artificial pancreas, employing this kind of sensor This project has been partially supported by the Spanish Government through Grants DPI-2013-46982-C2-2-R and DPI-2016-78831-C2-2-R, the National Council of Technological and Scientific Development, CNPq Brazil through Grant 202050/2015-7, the University of Girona through Grant BR2014/51, and the Agency for Management of University and Research Grants of the Government of Catalonia, Spain (Beatriu de Pinós [BP-DGR 2013])

Tens of glycemic variability (GV) indices are available in the literature to characterize the dynamic properties of glucose concentration profiles from continuous glucose monitoring (CGM) sensors. However, how to exploit the plethora of GV indices for classifying subjects is still controversial. For instance, the basic problem of using GV indices to automatically determine if the subject is healthy rather than affected by impaired glucose tolerance (IGT) or type 2 diabetes (T2D), is still unaddressed. Here, we analyzed the feasibility of using CGM-based GV indices to distinguish healthy from IGTT2D and IGT from T2D subjects by means of a machine-learning approach.The data set consists of 102 subjects belonging to three different classes: 34 healthy, 39 IGT, and 29 T2D subjects. Each subject was monitored for a few days by a CGM sensor that produced a glucose profile from which we extracted 25 GV indices. We used a two-step binary logistic regression model to classify subjects. The first step distinguishes healthy subjects from IGTT2D, the second step classifies subjects into either IGT or T2D.Healthy subjects are distinguished from subjects with diabetes (IGTT2D) with 91.4% accuracy. Subjects are further subdivided into IGT or T2D classes with 79.5% accuracy. Globally, the classification into the three classes shows 86.6% accuracy.Even with a basic classification strategy, CGM-based GV indices show good accuracy in classifying healthy and subjects with diabetes. The classification into IGT or T2D seems, not surprisingly, more critical, but results encourage further investigation of the present research.

BACKGROUND AND OBJECTIVES: Impaired glucose control in very preterm infants is associated with increased morbidity, mortality, and poor neurologic outcome. Strategies based on insulin titration have been unsuccessful in achieving euglycemia in absence of an increase in hypoglycemia and mortality. We sought to assess whether glucose administration guided by continuous glucose monitoring (CGM) is more effective than standard of care blood glucose monitoring in maintaining euglycemia in very preterm infants. METHODS: Fifty newborns ≤32 weeks’ gestation or with birth weight ≤1500 g were randomly assigned (1:1) within 48-hours from birth to receive computer-guided glucose infusion rate (GIR) with or without CGM. In the unblinded CGM group, the GIR adjustments were driven by CGM and rate of glucose change, whereas in the blinded CGM group the GIR was adjusted by using standard of care glucometer on the basis of blood glucose determinations. Primary outcome was percentage of time spent in euglycemic range (72–144 mg/dL). Secondary outcomes were percentage of time spent in mild (47–71 mg/dL) and severe (180 mg/dL) hyperglycemia; and glucose variability. RESULTS: Neonates in the unblinded CGM group had a greater percentage of time spent in euglycemic range (median, 84% vs 68%, P < .001) and decreased time spent in mild (P = .04) and severe (P = .007) hypoglycemia and in severe hyperglycemia (P = .04) compared with the blinded CGM group. Use of CGM also decreased glycemic variability (SD: 21.6 ± 5.4 mg/dL vs 27 ± 7.2 mg/dL, P = .01; coefficient of variation: 22.8% ± 4.2% vs 27.9% ± 5.0%; P < .001). CONCLUSIONS: CGM-guided glucose titration can successfully increase the time spent in euglycemic range, reduce hypoglycemia, and minimize glycemic variability in preterm infants during the first week of life.

It is clinically well-established that minimally invasive subcutaneous continuous glucose monitoring (CGM) sensors can significantly improve diabetes treatment. However, CGM readings are still not as reliable as those provided by standard fingerprick blood glucose (BG) meters. In addition to unavoidable random measurement noise, other components of sensor error are distortions due to the blood-to-interstitial glucose kinetics and systematic under-/overestimations associated with the sensor calibration process. A quantitative assessment of these components, and the ability to simulate them with precision, is of paramount importance in the design of CGM-based applications, e.g., the artificial pancreas (AP), and in their in silico testing. In the present paper, we identify and assess a model of sensor error of for two sensors, i.e., the G4 Platinum (G4P) and the advanced G4 for artificial pancreas studies (G4AP), both belonging to the recently presented "fourth" generation of Dexcom CGM sensors but different in their data processing. Results are also compared with those obtained by a sensor belonging to the previous, "third," generation by the same manufacturer, the SEVEN Plus (7P). For each sensor, the error model is derived from 12-h CGM recordings of two sensors used simultaneously and BG samples collected in parallel every 15 ± 5 min. Thanks to technological innovations, G4P outperforms 7P, with average mean absolute relative difference (MARD) of 11.1 versus 14.2%, respectively, and lowering of about 30% the error of each component. Thanks to the more sophisticated data processing algorithms, G4AP resulted more reliable than G4P, with a MARD of 10.0%, and a further decrease to 20% of the error due to blood-to-interstitial glucose kinetics.

By providing blood glucose (BG) concentration measurements in an almost continuous-time fashion for several consecutive days, wearable minimally-invasive continuous glucose monitoring (CGM) sensors are revolutionizing diabetes management, and are becoming an increasingly adopted technology especially for diabetic individuals requiring insulin administrations. Indeed, by providing glucose real-time insights of BG dynamics and trend, and being equipped with visual and acoustic alarms for hypo- and hyperglycemia, CGM devices have been proved to improve safety and effectiveness of diabetes therapy, reduce hypoglycemia incidence and duration, and decrease glycemic variability. Furthermore, the real-time availability of BG values has been stimulating the realization of new tools to provide patients with decision support to improve insulin dosage tuning and infusion. The aim of this paper is to offer an overview of current literature and future possible developments regarding CGM technologies and applications. In particular, first, we outline the technological evolution of CGM devices through the last 20 years. Then, we discuss about the current use of CGM sensors from patients affected by diabetes, and, we report some works proving the beneficial impact provided by the adoption of CGM. Finally, we review some recent advanced applications for diabetes treatment based on CGM sensors.

Input from continuous glucose monitors (CGMs) is a critical component of artificial pancreas (AP) systems, but CGM performance issues continue to limit progress in AP research. While G4 PLATINUM has been integrated into AP systems around the world and used in many successful AP controller feasibility studies, this system was designed to address the needs of ambulatory CGM users as an adjunctive use system. Dexcom and the University of Padova have developed an advanced CGM, called G4AP, to specifically address the heightened performance requirements for future AP studies. The G4AP employs the same sensor and transmitter as the G4 PLATINUM but contains updated denoising and calibration algorithms for improved accuracy and reliability. These algorithms were applied to raw data from an existing G4 PLATINUM clinical study using a simulated prospective procedure. The results show that mean absolute relative difference (MARD) compared with venous plasma glucose was improved from 13.2% with the G4 PLATINUM to 11.7% with the G4AP. Accuracy improvements were seen over all days of sensor wear and across the plasma glucose range (40–400 mg/dl). The greatest improvements occurred in the low glucose range (40–80 mg/dl), in euglycemia (80–120 mg/dl), and on the first day of sensor use. The percentage of sensors with a MARD

In diabetes research, non-invasive continuous glucose monitoring (NI-CGM) devices represent a new and appealing frontier. In the last years, some multi-sensor devices for NI-CGM have been proposed, which exploit several sensors measuring phenomena of different nature, not only for measuring glucose related signals, but also signals reflecting some possible perturbing processes (temperature, blood perfusion). Estimation of glucose levels is then obtained combining these signals through a mathematical model which requires an initial calibration step exploiting one reference blood glucose (RBG) sample. Even if promising results have been obtained, especially in hospitalized volunteers, at present the temporal accuracy of NI-CGM sensors may suffer because of environmental and physiological interferences. The aim of this work is to develop a general methodology, based on Monte Carlo (MC) simulation, to assess the robustness of the calibration step used by NI-CGM devices against these disturbances. The proposed methodology is illustrated considering two examples: the first concerns the possible detrimental influence of sweat events, while the second deals with calibration scheduling. For implementing both examples, 45 datasets collected by the Solianis Multisensor system are considered. In the first example, the MC methodology suggests that no further calibration adjustments are needed after the occurrence of sweat events, because the “Multisensor+model” system is able to deal with the disturbance. The second case study shows how to identify the best time interval to update the model’s calibration for improving the accuracy of the estimated glucose. The methodology proposed in this work is of general applicability and can be helpful in making those incremental steps in NI-CGM devices development needed to further improve their performance.

OBJECTIVE Reliability of continuous glucose monitoring (CGM) sensors is key in several applications. In this work we demonstrate that real-time algorithms can render CGM sensors smarter by reducing their uncertainty and inaccuracy and improving their ability to alert for hypo- and hyperglycemic events. RESEARCH DESIGN AND METHODS The smart CGM (sCGM) sensor concept consists of a commercial CGM sensor whose output enters three software modules, able to work in real time, for denoising, enhancement, and prediction. These three software modules were recently presented in the CGM literature, and here we apply them to the Dexcom SEVEN Plus continuous glucose monitor. We assessed the performance of the sCGM on data collected in two trials, each containing 12 patients with type 1 diabetes. RESULTS The denoising module improves the smoothness of the CGM time series by an average of ∼57%, the enhancement module reduces the mean absolute relative difference from 15.1 to 10.3%, increases by 12.6% the pairs of values falling in the A-zone of the Clarke error grid, and finally, the prediction module forecasts hypo- and hyperglycemic events an average of 14 min ahead of time. CONCLUSIONS We have introduced and implemented the sCGM sensor concept. Analysis of data from 24 patients demonstrates that incorporation of suitable real-time signal processing algorithms for denoising, enhancement, and prediction can significantly improve the performance of CGM applications. This can be of great clinical impact for hypo- and hyperglycemic alert generation as well in artificial pancreas devices.

Degludec is an ultralong-acting insulin analogue with a flat and reproducible pharmacodynamic profile. As some patients with type 1 diabetes (T1D) fail to achieve 24-h coverage with glargine or detemir despite twice-daily injections, we studied the effect of switching T1D patients from twice-daily glargine or detemir to degludec.In this prospective observational study, T1D patients on twice-daily glargine or detemir were enrolled. At baseline and 12 weeks after switching to degludec, we recorded HbA1c, insulin dose, 30-day blood glucose self monitoring (SMBG) or 14-day continuous glucose monitoring (CGM), treatment satisfaction (DTSQ), fear of hypoglycemia (FHS). We included 29 patients (mean age 34 ± 11 years; diabetes duration 18 ± 10 years). After switching to degludec, HbA1c decreased from 7.9 ± 0.6% (63 ± 6 mmol/mol) to 7.7 ± 0.6% (61 ± 6 mmol/mol; p = 0.028). SMBG showed significant reductions in the percent and number of blood glucose values70 mg/dl and in the low blood glucose index (LBGI) during nighttime. CGM showed a significant reduction of time spent in hypoglycemia, an increase in daytime spent in target 70-180 mg/dl, and a reduction in glucose variability. Total insulin dose declined by 17% (p 0.001), with 24% reduction in basal and 10% reduction in prandial insulin. DTSQ and FHS significantly improved.Switching from twice-daily glargine or detemir to once daily degludec improved HbA1c, glucose profile, hypoglycemia risk and treatment satisfaction, while insulin doses decreased. ClinicalTrials.govNCT02360254.

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.

The quantitative analysis of glucose time-series can greatly help the management of diabetes. In particular, a static nonlinear transformation, which symmetrizes the distribution of glucose levels by bringing them in the so-called risk space, was proposed previously for both self-monitoring blood glucose and continuous glucose monitoring (CGM) and extensively used in the literature. The continuous nature of CGM data allows us to further refine the risk space concept in order to account for glucose dynamics.A new dynamic risk (DR) is proposed to explicitly consider the rate of change of glucose as a threat factor for the patient (e.g., risk levels in hypoglycemia and hyperglycemia are amplified in the presence of a decreasing and increasing glucose trend, respectively). The practical calculation of DR is made possible by the use of a regularized deconvolution algorithm that is able to deal with noise in CGM data and with the ill-conditioning of the time-derivative calculation, even in online applications.Results on simulated and real data show that DR can be effectively computed and fruitfully used in real time (e.g., to generate early warnings of hypo-/hyperglycemic threshold crossings). Further applications of DR in the quantification of the efficiency of glucose control are also suggested.Exploiting the information on glucose trends empowers the strength of risk measures in interpreting CGM time-series.

Knowing the statistical properties of continuous glucose monitoring (CGM) sensor errors can be important in several practical applications, e.g., in both open- and closed-loop control algorithms. Unfortunately, modeling the accuracy of CGM sensors is very difficult for both experimental and methodological reasons. It has been suggested that the time series of CGM sensor errors can be described as realization of the output of an autoregressive (AR) model of first order driven by a white noise process. The AR model was identified exploiting several reference blood glucose (BG) samples (collected frequently in parallel to the CGM signal), a procedure to recalibrate CGM data, and a linear time-invariant model of blood-to-interstitium glucose (BG-to-IG) kinetics. By resorting to simulation, this work shows that some assumptions made in the Breton and Kovatchev modeling approach may significantly affect the estimated sensor error and its statistical properties.Three simulation studies were performed. The first simulation was devoted to assessing the influence of CGM data recalibration, whereas the second and third simulations examined the role of the BG-to-IG kinetic model. Analysis was performed by comparing the "original" (synthetically generated) time series of sensor errors vs its "reconstructed" version in both time and frequency domains.Even small errors either in CGM data recalibration or in the description of BG-to-IG dynamics can severely affect the possibility of correctly reconstructing the statistical properties of sensor error. In particular, even if CGM sensor error is a white noise process, a spurious correlation among its samples originates from suboptimal recalibration or from imperfect knowledge of the BG-to-IG kinetics.Modeling the statistical properties of CGM sensor errors from data collected in vivo is difficult because it requires perfect calibration and perfect knowledge of BG-to-IG dynamics. Results suggest that correct characterization of CGM sensor error is still an open issue and requires further development upon the pioneering contribution of Breton and Kovatchev.

Self-monitoring of blood glucose (SMBG) devices are portable systems that allow measuring glucose concentration in a small drop of blood obtained via finger-prick. SMBG measurements are key in type 1 diabetes (T1D) management, e.g. for tuning insulin dosing. A reliable model of SMBG accuracy would be important in several applications, e.g. in in silico design and optimization of insulin therapy. In the literature, the most used model to describe SMBG error is the Gaussian distribution, which however is simplistic to properly account for the observed variability. Here, a methodology to derive a stochastic model of SMBG accuracy is presented. The method consists in dividing the glucose range into zones in which absolute/relative error presents constant standard deviation (SD) and, then, fitting by maximum-likelihood a skew-normal distribution model to absolute/relative error distribution in each zone. The method was tested on a database of SMBG measurements collected by the One Touch Ultra 2 (Lifescan Inc., Milpitas, CA). In particular, two zones were identified: zone 1 (BG≤75 mg/dl) with constant-SD absolute error and zone 2 (BG>75mg/dl) with constant-SD relative error. Mean and SD of the identified skew-normal distributions are, respectively, 2.03 and 6.51 in zone 1, 4.78% and 10.09% in zone 2. Visual predictive check validation showed that the derived two-zone model accurately reproduces SMBG measurement error distribution, performing significantly better than the single-zone Gaussian model used previously in the literature. This stochastic model allows a more realistic SMBG scenario for in silico design and optimization of T1D insulin therapy.

Goal: Minimally invasive continuous glucose monitoring (CGM) sensors measure in the subcutis a current signal, which is converted into interstitial glucose (IG) concentration by a calibration process periodically updated using fingerstick blood glucose (BG) references. Though important in diabetes management, CGM sensors still suffer from accuracy problems. Here, we propose a new online calibration method improving accuracy of CGM glucose profiles with respect to manufacturer calibration. Method: The proposed method fits CGM current signal against the BG references collected twice a day for calibration purposes, by a time-varying calibration function whose parameters are identified in a Bayesian framework using a priori second-order statistical knowledge. The distortion introduced by BG-to-IG kinetics is compensated before parameter identification via nonparametric deconvolution. Results: The method was tested on a database where 108 CGM signals were collected for 7 days by the Dexcom G4 Platinum sensor. Results show the new method drives to a statistically significant accuracy improvement as measured by three commonly used metrics: mean absolute relative difference reduced from 12.73% to 11.47%; percentage of accurate glucose estimates increased from 82.00% to 89.19%; and percentage of values falling in the “A” zone of the Clark error grid increased from 82.22% to 88.86%. Conclusion: The new calibration method significantly improves CGM glucose profiles accuracy with respect to manufacturer calibration. Significance: The proposed algorithm provides a real-time improvement of CGM accuracy, which can be crucial in several CGM-based applications, including the artificial pancreas, thus providing a potential great impact in the diabetes technology research community.

In type 1 diabetes (T1D) therapy, continuous glucose monitoring (CGM) sensors, which provide glucose concentration in the subcutis every 1-5 min for 7 consecutive days, should allow in principle a more efficient insulin dosing than that based on the conventional 3-4 self-monitoring of blood glucose (SMBG) measurements per day. However, CGM, at variance with SMBG, is still not approved for insulin dosing in T1D management because regulatory agencies, e.g. FDA, are looking for more factual evidence on its safety. An in silico assessment of SMBG- vs CGM-driven insulin therapy can be a first step. Here we present a simulation model of T1D patient decision-making obtained by interconnecting models of glucose-insulin dynamics, SMBG and CGM measurement errors, carbohydrates-counting errors, insulin boluses time variability and forgetfulness, and subcutaneous insulin pump delivery. Inter- and intra- patient variability of model parameters are considered. The T1D patient decision-making model allows to run realistic multi-day simulations scenarios in a population of virtual subjects. We present the first results of simulations run in 20 virtual subjects over a 7-day period, which demonstrates that additional information brought by CGM (trend and hypo/hyperglycemic warnings) with respect to SMBG produces a statistically significant increment (about of 9%) of time spent by the patient in the euglycemic range (70-180 mg/dl).