Harald Kirchsteiger

Performance evaluations of continuous glucose monitoring systems: precision absolute relative deviation is part of the assessment.

Background: 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. Materials and Methods: 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. Results: 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. Conclusions: 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.

Estimating Interval Process Models for Type 1 Diabetes for Robust Control Design

Abstract A linear transfer function model comprising 4 parameters is used to describe the post-prandial breakfast excursions of a group of 10 type 1 diabetes patients who are treated with multiple daily insulin injections. The model is able to simulate the glucose concentration in blood and uses the information of carbohydrate content of the breakfast and the administered insulin injections as inputs. Additionally, a measurement of the actual blood glucose concentration at the time when the breakfast occurs is required. No additional information, in particular the use of a continuous glucose monitoring system is necessary. Based on a 3 day observational period, parameter intervals are calculated such that the measured glucose responses are inside the bounds given by the output of the model. The model together with the parameter intervals can be used for robust control design.

Time Delay of CGM Sensors: Relevance, Causes, and Countermeasures

Background: Continuous glucose monitoring (CGM) is a powerful tool to support the optimization of glucose control of patients with diabetes. However, CGM systems measure glucose in interstitial fluid but not in blood. Rapid changes in one compartment are not accompanied by similar changes in the other, but follow with some delay. Such time delays hamper detection of, for example, hypoglycemic events. Our aim is to discuss the causes and extent of time delays and approaches to compensate for these. Methods: CGM data were obtained in a clinical study with 37 patients with a prototype glucose sensor. The study was divided into 5 phases over 2 years. In all, 8 patients participated in 2 phases separated by 8 months. A total number of 108 CGM data sets including raw signals were used for data analysis and were processed by statistical methods to obtain estimates of the time delay. Results: Overall mean (SD) time delay of the raw signals with respect to blood glucose was 9.5 (3.7) min, median was 9 min (interquartile range 4 min). Analysis of time delays observed in the same patients separated by 8 months suggests a patient dependent delay. No significant correlation was observed between delay and anamnestic or anthropometric data. The use of a prediction algorithm reduced the delay by 4 minutes on average. Conclusions: Prediction algorithms should be used to provide real-time CGM readings more consistent with simultaneous measurements by SMBG. Patient specificity may play an important role in improving prediction quality.

Cooperative adaptive cruise control applying stochastic linear model predictive control strategies

In this paper a cooperative adaptive cruise control approach using stochastic, linear model predictive control strategies is presented. The presented approach deals with an urban traffic environment where vehicle to vehicle and vehicle to infrastructure communication systems are available. The goal is the minimization of a vehicles fuel consumption in a vehicle-following scenario. This is achieved by minimizing a piecewise linear approximation of the vehicles fuel consumption map. By means of a conditional Gaussian model the probability distribution of the upcoming velocity of the preceding vehicle is estimated based on current measurements and upcoming traffic light signals. The predicted distribution function of the predecessors velocity is used in two ways for stochastic model predictive control. On the one hand, individual chance constraints are introduced and subsequently reformulated to obtain an equivalent deterministic model predictive control problem. On the other hand, samples are drawn from the prediction model and used for a randomized optimization approach. Finally, the two developed stochastic control strategies are evaluated and compared against a deterministic model predictive control approach by means of a virtual traffic simulation. The evaluation of the controllers show a significant reduction of the fuel consumption compared to the predecessor while increasing safety and driving comfort.

Direct continuous time system identification of MISO transfer function models applied to type 1 diabetes

This paper shows an application of continuous time system identification methods to Type 1 diabetes. First, a general MISO transfer function structure with individual nominator and denominator polynomials for each input is assumed and a parameter estimation procedure via an iterative prediction error method presented. Then, the proposed identification method is evaluated on a simple simulation example and finally applied on real-life data from Type 1 diabetes patients with the purpose of modeling blood glucose dynamics. To this aim, the method was extended to consider the time-varying nature of the system.

Continuous-time interval model identification of blood glucose dynamics for type 1 diabetes

While good physiological models of the glucose metabolism in type 1 diabetic patients are well known, their parameterisation is difficult. The high intra-patient variability observed is a further major obstacle. This holds for data-based models too, so that no good patient-specific models are available. Against this background, this paper proposes the use of interval models to cover the different metabolic conditions. The control-oriented models contain a carbohydrate and insulin sensitivity factor to be used for insulin bolus calculators directly. Available clinical measurements were sampled on an irregular schedule which prompts the use of continuous-time identification, also for the direct estimation of the clinically interpretable factors mentioned above. An identification method is derived and applied to real data from 28 diabetic patients. Model estimation was done on a clinical data-set, whereas validation results shown were done on an out-of-clinic, everyday life data-set. The results show that the interval model approach allows a much more regular estimation of the parameters and avoids physiologically incompatible parameter estimates.

Innovative approach for online prediction of blood glucose profile in type 1 diabetes patients

Recursive identification techniques are used to estimate predictions for the human glucose-insulin subsystem. By replacing a constant gain with a physiologically inspired adaptation rule and adding as additional inputs the two variables ingested meal and administered insulin-which have the highest impact on the glucose concentration-the overall performance of a 45 min glucose prediction could be increased compared to standard identification and prediction methods. The results were analyzed from a system theoretical, and also from a clinical point of view using the CG-EGA.

Robustness properties of optimal insulin bolus administrations for Type 1 diabetes

Type 1 diabetic patients compensate the lack of endogenous insulin by basal delivery and bolus injections at meal-times. Exact dosage of the bolus amount is critical to keep the blood glucose both below the maximum limits and above the hypoglycaemia critical values. Determination of the optimal dosage would require information which in general is not available to the patient, who uses empirical rules of thumb to choose the dosage. Although closed loop control obtained by linking insulin delivery from insulin pumps and continuous glucose monitoring systems may be considered as the ultimate solution, multiple daily insulin injections and finger stick glucose measurements remain the current mode of therapy. This paper is concerned with this conventional insulin treatment and is based on the use of model predictive techniques extended to approximate continuous control output signal by single control moves in time. The paper shows that substituting continuous measurement and insulin delivery with discrete values leads to a suboptimal control performance, but that this residual defect is not essential if compared with estimation errors of model parameters, patient inputs and/or measurements. Furthermore, the approach proposed shows in simulation sufficient robustness margins. Computations are done with an extended Bergman model tuned on available data of Type 1 diabetic patients.

Performance Comparison of CGM Systems MARD Values Are Not Always a Reliable Indicator of CGM System Accuracy

Background: The ongoing progress of continuous glucose monitoring (CGM) systems results in an increasing interest in comparing their performance, in particular in terms of accuracy, that is, matching CGM readings with reference values measured at the same time. Most often accuracy is evaluated by the mean absolute relative difference (MARD). It is frequently overseen that MARD does not only reflect accuracy, but also the study protocol and evaluation procedure, making a cross-study comparison problematic. Methods: We evaluate the effect of several factors on the MARD statistical properties: number of paired reference and CGM values, distribution of the paired values, accuracy of the reference measurement device itself and the time delay between data pairs. All analysis is done using clinical data from 12 patients wearing 6 sensors each. Results: We have found that a few paired points can have a potentially high impact on MARD. Leaving out those points for evaluation thus reduces the MARD. Similarly, accuracy of the reference measurements greatly affects the MARD as numerical and graphical data show. Results also show that a log-normal distribution of the paired references provides a significantly different MARD than, for example, a uniform distribution. Conclusions: MARD is a reasonable parameter to characterize the performance of CGM systems when keeping its limitations in mind. To support clinicians and patients in selecting which CGM system to use in a clinical setting, care should be taken to make MARD more comparable by employing a standardized evaluation procedure.

LMI-Based Approaches for the Calibration of Continuous Glucose Measurement Sensors

The problem of online calibration and recalibration of continuous glucose monitoring (CGM) devices is considered. Two different parametric relations between interstitial and blood glucose are investigated and constructive algorithms to adaptively estimate the parameters within those relations are proposed. One characteristic is the explicit consideration of measurement uncertainty of the device used to collect the calibration measurements. Another feature is the automatic detection of fingerstick measurements that are not suitable to be used for calibration. Since the methods rely on the solution of linear matrix inequalities resulting in convex optimization problems, the algorithms are of moderate computational complexity and could be implemented on a CGM device. The methods were assessed on clinical data from 17 diabetic patients and the improvements with respect to the current state of the art is shown.