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

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

Using Iterative Learning for Insulin Dosage Optimization in Multiple-Daily-Injections Therapy for People With Type 1 Diabetes.

OBJECTIVE: In this work, we design iterative algorithms for the delivery of long-acting (basal) and rapid-acting (bolus) insulin, respectively, for people with type 1 diabetes (T1D) on multiple-daily-injections (MDIs) therapy using feedback from self-monitoring of blood glucose (SMBG) measurements. METHODS: Iterative learning control (ILC) updates basal therapy consisting of one long-acting insulin injection per day, while run-to-run (R2R) adapts meal bolus therapy via the update of the mealtime-specific insulin-to-carbohydrate ratio (CR). Updates are due weekly and are based upon sparse SMBG measurements. RESULTS: Upon termination of the 20 weeks long in-silico trial, in a scenario characterized by meal carbohydrate (CHO) normally distributed with mean µ = [50, 75, 75] grams and standard deviation σ = [5, 7, 7] grams, our strategy produced statistically significant improvements in time in range (70--180) [mg/dl], from 66.9(33.1) % to 93.6(6.7) %, p = 0.02. CONCLUSIONS: Iterative learning shows potential to improve glycemic regulation over time by driving blood glucose closer to the recommended glycemic targets. SIGNIFICANCE: Decision support systems (DSSs) and automated therapy advisors such as the one proposed here are expected to improve glycemic outcomes reducing the burden on patients on MDI therapy.

Activity detection and classification from wristband accelerometer data collected on people with type 1 diabetes in free-living conditions.

This paper introduces methods to estimate aspects of physical activity and sedentary behavior from three-axis accelerometer data collected with a wrist-worn device at a sampling rate of 32 [Hz] on adults with type 1 diabetes (T1D) in free-living conditions. In particular, we present two methods able to detect and grade activity based on its intensity and individual fitness as sedentary, mild, moderate or vigorous, and a method that performs activity classification in a supervised learning framework to predict specific user behaviors. Population results for activity level grading show multi-class average accuracy of 99.99%, precision of 98.0 ± 2.2%, recall of 97.9 ± 3.5% and F1 score of 0.9 ± 0.0. As for the specific behavior prediction, our best performing classifier, gave population multi-class average accuracy of 92.43 ± 10.32%, precision of 92.94 ± 9.80%, recall of 92.20 ± 10.16% and F1 score of 92.56 ± 9.94%. Our investigation showed that physical activity and sedentary behavior can be detected, graded and classified with good accuracy and precision from three-axial accelerometer data collected in free-living conditions on people with T1D. This is particularly significant in the context of automated glucose control systems for diabetes, in that the methods we propose have the potential to inform changes in treatment parameters in response to the intensity of physical activity, allowing patients to meet their glycemic targets.

Early Detection of Infusion Set Failure During Insulin Pump Therapy in Type 1 Diabetes.

BACKGROUND: Insulin infusion set failure resulting in prolonged hyperglycemia or diabetic ketoacidosis can occur with pump therapy in type 1 diabetes. Set failures are frequently characterized by variable and unpredictable patterns of increasing glucose values despite increased insulin infusion. Early detection may minimize the risk of prolonged hyperglycemia, an important consideration for automated insulin delivery and closed-loop applications. METHODS: A novel algorithm designed to alert the patient to the onset of infusion set failure was developed based upon continuous glucose sensor values and insulin delivered from an insulin pump. The method was calibrated on 12 weeks of infusion set wear without failures recorded by 4 patients in ambulatory conditions and prospectively validated on 18 weeks of infusion set wear with and without failures belonging to 9 other subjects in ambulatory conditions. RESULTS: The algorithm, evaluated retrospectively, identified a failure 2.52 ± 1.91 days ahead of the actual event as recorded by the clinical team, corresponding to 50% sensitivity, 66% specificity and 55% accuracy. If set failure alarms had been activated in real time, the average time >180 mg/dl would be reduced from 82.7 ± 40.9 hours/week/subject (without alarm) to 58.8 ± 31.1 hours/week/subject (with alarm), corresponding to a potential 29% reduction in time spent >180mg/dl. CONCLUSION: The proposed method for early detection of infusion set failure based on glucose sensor and insulin data demonstrated favorable results on retrospective data and may be implemented as an additional safeguard in a future fully automated closed-loop system.