Probabilistic Model of Transition between Categories of Glucose Profiles in Patients with Type 1 Diabetes Using a Compositional Data Analysis Approach.

The time spent in glucose ranges is a common metric in type 1 diabetes (T1D). As the time in one day is finite and limited, Compositional Data (CoDa) analysis is appropriate to deal with times spent in different glucose ranges in one day. This work proposes a CoDa approach applied to glucose profiles obtained from six T1D patients using continuous glucose monitor (CGM). Glucose profiles of 24-h and 6-h duration were categorized according to the relative interpretation of time spent in different glucose ranges, with the objective of presenting a probabilistic model of prediction of category of the next 6-h period based on the category of the previous 24-h period. A discriminant model for determining the category of the 24-h periods was obtained, achieving an average above 94% of correct classification. A probabilistic model of transition between the category of the past 24-h of glucose to the category of the future 6-h period was obtained. Results show that the approach based on CoDa is suitable for the categorization of glucose profiles giving rise to a new analysis tool. This tool could be very helpful for patients, to anticipate the occurrence of potential adverse events or undesirable variability and for physicians to assess patients' outcomes and then tailor their therapies.

Dynamic Rule-Based Algorithm to Tune Insulin-on-Board Constraints for a Hybrid Artificial Pancreas System.

The artificial pancreas (AP) is a system intended to control blood glucose levels through automated insulin infusion, reducing the burden of subjects with type 1 diabetes to manage their condition. To increase patients' safety, some systems limit the allowed amount of insulin active in the body, known as insulin-on-board (IOB). The safety auxiliary feedback element (SAFE) layer has been designed previously to avoid overreaction of the controller and thus avoiding hypoglycemia. In this work, a new method, so-called "dynamic rule-based algorithm," is presented in order to adjust the limits of IOB in real time. The algorithm is an extension of a previously designed method which aimed to adjust the limits of IOB for a meal with 60 grams of carbohydrates (CHO). The proposed method is intended to be applied on hybrid AP systems during 24 h operation. It has been designed by combining two different strategies to set IOB limits for different situations: (1) fasting periods and (2) postprandial periods, regardless of the size of the meal. The UVa/Padova simulator is considered to assess the performance of the method, considering challenging scenarios. In silico results showed that the method is able to reduce the time spent in hypoglycemic range, improving patients' safety, which reveals the feasibility of the approach to be included in different control algorithms.

Prediction of Nocturnal Hypoglycemia in Adults with Type 1 Diabetes under Multiple Daily Injections Using Continuous Glucose Monitoring and Physical Activity Monitor.

(1) Background: nocturnal hypoglycemia (NH) is one of the most challenging side effects of multiple doses of insulin (MDI) therapy in type 1 diabetes (T1D). This work aimed to investigate the feasibility of a machine-learning-based prediction model to anticipate NH in T1D patients on MDI. (2) Methods: ten T1D adults were studied during 12 weeks. Information regarding T1D management, continuous glucose monitoring (CGM), and from a physical activity tracker were obtained under free-living conditions at home. Supervised machine-learning algorithms were applied to the data, and prediction models were created to forecast the occurrence of NH. Individualized prediction models were generated using multilayer perceptron (MLP) and a support vector machine (SVM). (3) Results: population outcomes indicated that more than 70% of the NH may be avoided with the proposed methodology. The predictions performed by the SVM achieved the best population outcomes, with a sensitivity and specificity of 78.75% and 82.15%, respectively. (4) Conclusions: our study supports the feasibility of using ML techniques to address the prediction of nocturnal hypoglycemia in the daily life of patients with T1D on MDI, using CGM and a physical activity tracker.

Prediction and prevention of hypoglycaemic events in type-1 diabetic patients using machine learning.

Tight blood glucose control reduces the risk of microvascular and macrovascular complications in patients with type 1 diabetes. However, this is very difficult due to the large intra-individual variability and other factors that affect glycaemic control. The main limiting factor to achieve strict control of glucose levels in patients on intensive insulin therapy is the risk of severe hypoglycaemia. Therefore, hypoglycaemia is the main safety problem in the treatment of type 1 diabetes, negatively affecting the quality of life of patients suffering from this disease. Decision support tools based on machine learning methods have become a viable way to enhance patient safety by anticipating adverse glycaemic events. This study proposes the application of four machine learning algorithms to tackle the problem of safety in diabetes management: (1) grammatical evolution for the mid-term continuous prediction of blood glucose levels, (2) support vector machines to predict hypoglycaemic events during postprandial periods, (3) artificial neural networks to predict hypoglycaemic episodes overnight, and (4) data mining to profile diabetes management scenarios. The proposal consists of the combination of prediction and classification capabilities of the implemented approaches. The resulting system significantly reduces the number of episodes of hypoglycaemia, improving safety and providing patients with greater confidence in decision-making.

Individual categorisation of glucose profiles using compositional data analysis.

The aim of this study was to apply a methodology based on compositional data analysis (CoDA) to categorise glucose profiles obtained from continuous glucose monitoring systems. The methodology proposed considers complete daily glucose profiles obtained from six patients with type 1 diabetes (T1D) who had their glucose monitored for eight weeks. The glucose profiles were distributed into the time spent in six different ranges. The time in one day is finite and limited to 24 h, and the times spent in each of these different ranges are co-dependent and carry only relative information; therefore, CoDA is applied to these profiles. A K-means algorithm was applied to the coordinates obtained from the CoDA to obtain different patterns of days for each patient. Groups of days with relatively high time in the hypo and/or hyperglycaemic ranges and with different glucose variability were observed. Using CoDA of time in different ranges, individual glucose profiles were categorised into groups of days, which can be used by physicians to detect the different conditions of patients and personalise patient's insulin therapy according to each group. This approach can be useful to assist physicians and patients in managing the day-to-day variability that hinders glycaemic control.

Accuracy of Continuous Glucose Monitoring before, during, and after Aerobic and Anaerobic Exercise in Patients with Type 1 Diabetes Mellitus.

Continuous glucose monitoring (CGM) plays an important role in treatment decisions for patients with type 1 diabetes under conventional or closed-loop therapy. Physical activity represents a great challenge for diabetes management as well as for CGM systems. In this work, the accuracy of CGM in the context of exercise is addressed. Six adults performed aerobic and anaerobic exercise sessions and used two Medtronic Paradigm Enlite-2 sensors under closed-loop therapy. CGM readings were compared with plasma glucose during different periods: one hour before exercise, during exercise, and four hours after the end of exercise. In aerobic sessions, the median absolute relative difference (MARD) increased from 9.5% before the beginning of exercise to 16.5% during exercise (p < 0.001), and then decreased to 9.3% in the first hour after the end of exercise (p < 0.001). For the anaerobic sessions, the MARD before exercise was 15.5% and increased without statistical significance to 16.8% during exercise realisation (p = 0.993), and then decreased to 12.7% in the first hour after the cessation of anaerobic activities (p = 0.095). Results indicate that CGM might present lower accuracy during aerobic exercise, but return to regular operation a few hours after exercise cessation. No significant impact for anaerobic exercise was found.

Blood glucose monitoring during aerobic and anaerobic physical exercise using a new artificial pancreas system. / Control de la glucemia durante el ejercicio físico aeróbico y anaeróbico mediante un nuevo sistema de páncreas artificial.

AIM: To assess an artificial pancreas system during aerobic (AeE) and anaerobic exercise (AnE). METHODS: A pilot clinical trial on five subjects with type 1 diabetes (4 males) aged 37±10.9 years, diabetes diagnosed 21.2±12.2 years before, insulin pump users, and with a mean HbA1c level of 7.8±0.5%. Every subject did three AeE and three AnE sessions. Blood glucose levels were monitored by the artificial pancreas system during exercise and up to four hours later. Before the start of exercise, 23g of carbohydrates were administered orally. RESULTS: The mean glucose level was 124.0±25.1mg/dL in the AeE studies and 152.1±34.1mg/dL in the AnE studies. Percent times in the different glucose ranges of 70-180, >180 and <70mg/dL were 89.8±18.6% and 75.9±27.6%; 7.7±18.4% and 23.2±28.0%; and 2.5±6.3% and 1.0±3.6% during the AeE and AnE sessions, respectively. Only six rescues with carbohydrates (15g) were required during the studies (4 in AeE and 2 in AnE). Total insulin dose during the five hours of the study was 3.1±1.0IU in the AeE studies and 3.5±1.3IU in the AnE studies. CONCLUSIONS: Blood glucose response to AeE and AnE exercise is different. The evaluated artificial pancreas system appeared to achieve effective and safe blood glucose control during exercise and up to four hours later. However, new control strategies that minimize patient intervention should be designed.

Modeling the Error of the Medtronic Paradigm Veo Enlite Glucose Sensor.

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.

Extensive Assessment of Blood Glucose Monitoring During Postprandial Period and Its Impact on Closed-Loop Performance.

BACKGROUND: Closed-loop (CL) systems aims to outperform usual treatments in blood glucose control and continuous glucose monitors (CGM) are a key component in such systems. Meals represents one of the main disturbances in blood glucose control, and postprandial period (PP) is a challenging situation for both CL system and CGM accuracy. METHODS: We performed an extensive analysis of sensor's performance by numerical accuracy and precision during PP, as well as its influence in blood glucose control under CL therapy. RESULTS: During PP the mean absolute relative difference (MARD) for both sensors presented lower accuracy in the hypoglycemic range (19.4 ± 12.8%) than in other ranges (12.2 ± 8.6% in euglycemic range and 9.3 ± 9.3% in hyperglycemic range). The overall MARD was 12.1 ± 8.2%. We have also observed lower MARD for rates of change between 0 and 2 mg/dl. In CL therapy, the 10 trials with the best sensor spent less time in hypoglycemia (PG < 70 mg/dl) than the 10 trials with the worst sensors (2 ± 7 minutes vs 32 ± 38 minutes, respectively). CONCLUSIONS: In terms of accuracy, our results resemble to previously reported. Furthermore, our results showed that sensors with the lowest MARD spent less time in hypoglycemic range, indicating that the performance of CL algorithm to control PP was related to sensor accuracy.