Abstract
Assessment of Infusion Set Survival of the Newly Developed Lantern Catheter in Type 1 Diabetes by Glucose Clamp Technique
Medical University of Graz - Division of Endocrinology and Diabetology Graz, Styria, Austria
Objective:
To evaluate the efficacy and safety of the new coated subcutaneous infusion set with Lantern Technology over 7 days of wear and to determine whether the novel cannula design facilitates consistent insulin flow over an extended wear time.
Method:
Sixteen type 1 diabetes patients (age 44.2±15.4 years, BMI 24.5±2.3 kg/m2, A1c 55±8 mmol/mol, diabetes duration 20±9 years) underwent a 7-day continuous subcutaneous insulin infusion (CSII) treatment combined with flash glucose monitoring (FGM). The patients participated in three 8-hour euglycemic clamp experiments (Study Days 1, 4, and 7) and spent the days between the experiments (Study Days 3, 4, 5, and 6) at home routinely managing diabetes with CSII and FGM.
Result:
The catheter survival rate was 100%. There was no incidence of severe hypoglycemia or ketoacidosis during the study period. The percentage of time within the target range (70-180mg/dL) was similar over 24-hours for the three clamp experiments (66.5± 18.7 vs.55.8±21.2 vs. 53.5±25.2%). Time to reach 50% of the maximum glucose infusion rate (GIR) was comparable between clamps (31.5±1.6 vs. 29.3±1.3 vs. 27.3±1.3min; P=0.51). The log-transformed AUCGIR did not significantly differ for the first 2-hours of the clamp (geo mean 343.7 [228.6-516.7]; 421.3 [271.4-654.1]; 350.6 [200.3-613.6]mg/kg; p=0.14). Time to GIRMAX was reduced with increasing wear time (median 137.50 [72.5-147.5] vs. 50.0 [40.0-80.0] vs. 45.0 [35.0-62.5]min; P<0.002). However, there was a reduction in AUCGIR over the 8-hour over time (geo mean 874.2 [620.0-1232.7] vs. 744.5 [451.4-1228.0] vs. 509.2 [257.2-1008.2]mg/kg; p<0.05).
Conclusion:
The coated infusion set with Lantern Technology could be safely used during extended wear. There was a shift of GIRMAX profile towards faster onset and reduced insulin action over time. These findings need to be confirmed in a larger scale trial under routine conditions.
Using Machine Learning Methods for Dynamic Forecasting and Control of Type 2 Diabetes Using Mobile-Based Health Lifestyle Data
UT Health San Antonio San Antonio, TX, USA
Backgroud:
Type 2 diabetes mellitus (T2DM) not only creates a huge public health burden but also contributes to a tremendous patient burden due to intensive and complex self-management regimens. Adherence to these complex regimens on a day-to-day basis is particularly challenging. We propose using machine learning methods to analyze streams of daily captured data from individual patients’ mobile health data in order to help patients and clinicians manage T2DM more efficiently. The objective of this study is to dynamically forecast glucose levels in patients with T2DM based on their mobile health lifestyle data on diet, activity, and weight control and to alarm unusual lifestyle regimens in order to achieve better glucose control.
Method:
We used data from 10 T2DM patients who were overweight or obese in a behavioral lifestyle intervention using mobile tools for monitoring diet, physical activity, weight, and blood glucose daily over 6 months. We developed a deep learning model based on dynamic convolutional neural networks to estimate and forecast the trajectory of glucose levels in individual patients. The proposed neural network utilized several layers of computational nodes to model how mobile health data (i.e., diet, calories consumed, fat, carbs, breakfast, lunch, dinner, snack, exercise, and weight) were progressing from one day to another from noisy data. Given past lifestyle regimens data, the deep learning network monitored the patient daily regiments (e.g., diet, activity, weight control) and alarmed any unusual trend or patterns that led to undesirable glucose levels.
Result:
Using the cross-validation method based on a dataset of 10 patients who have been monitored on a daily basis for over 6 months, the proposed deep learning network demonstrates >90% accuracy in predicting the glucose level (major increase, minor increase, no-change, decrease) one day ahead. It also shows >83% accuracy for two days in advance prediction and >72% accuracy for three days in advance prediction.
Conclusion:
Using machine learning methodologies to analyze mobile health data can be effective in developing more individualized lifestyle interventions and glucose control for T2DM management.
Predictive Dosing Support for Long Acting Insulin in Type 2 Diabetes
Novo Nordisk A/S Bagsværd, Copenhagen, Denmark
Objective:
More than 60% of type 2 diabetes (T2D) patients on insulin treatment in the United States do not reach recommended treatment targets. The many reasons for this deficiency include the complexity of insulin titration and a lack of confidence and understanding. Titration algorithms for home use have slow convergence especially for patients who need large doses. Our objective is to develop an automatic decision support algorithm that enables efficient and safe long-acting insulin initiation.
Method:
We used a physiological model of fasting glucose and long-acting insulin dynamics in T2D to develop a model predictive control (MPC)-based decision support system. The MPC controller optimized the insulin administered once daily based on predicted fasting glucose values with a sampling period of 15 minutes. We performed an in silico simulation to test the sensitivity of the control to variations in insulin sensitivity and glucose sensitivity of endogenous insulin production.
Result:
We observe that, compared with current titration algorithms, the MPC-based decision support provides faster convergence to the glycemic target, and adapts faster to changes in parameters. Furthermore, the fast sampling period of the prediction provides safety in the case where we administer insulin at the same time as the glucose measurement.
Conclusion:
We conclude that the MPC-based decision support is a promising method for long-acting insulin titration compared to current titration algorithms designed for home use.
POPS! one Blood Glucose Monitoring System (BGMS) tightly Integrated with a Mobile App and Novel Lancing System, is Accurate and Easy to Use
AMCR Institute Escondido, CA, USA
Objective:
This study evaluated the accuracy and usability of the POPS! BGMS in the hands of intended users. POPS! was designed to make blood glucose testing and diabetes management easier and more convenient. It replaces the traditional BGMS with a meter and self-contained test module that utilizes a novel fixed lancet mechanism. The meter pairs with the app via Bluetooth, allowing the meter to perform the test and send results directly to the app where they can be viewed, stored (in the meter and the cloud), and shared.
Method:
This single center study was designed to meet the FDA 2016 Guidance. The primary endpoint was system accuracy as compared to YSI (95% results within +/-15%, 99% results within +/-20%). The secondary endpoint was system ease of use based upon questionnaire. Testing was performed across multiple test strip lots and phone models/operating systems.
Result:
A total of 362 subjects were enrolled; of these 76 were naïve to blood glucose testing. All 362 subjects were able to successfully complete a test with POPS! using only labeling. System accuracy met the study endpoint; 100% results were within +/-15% and 100% results were within +/-20% of the comparator results across the entire claimed measuring range of the device. Additionally, 93.9% (340/362) were within +/-10%. Subject feedback on the system was very positive, with 96.4% responding that the system was easy to use and 100% responding that viewing the result on the app was easy.
Conclusion:
POPS! exceeds the FDA criteria for accuracy, regardless of subjects’ blood glucose testing experience. These results, paired with painless lancing and convenient system design, may result in more frequent testing. Future studies should investigate diabetes outcomes and long-term system use and compliance.
Safety and Efficacy of an Electronic Glycemic Management System among Labor and Delivery Patients at a not-for-profit Hospital in Hawaii
Kapi’olani Medical Center for Women and Children Honolulu, Hawai’i, USA
Objective:
To evaluate if an electronic glycemic management system (eGMS®) safely supports tight glycemic control with minimal episodes of hypoglycemia among antepartum patients receiving intravenous (IV) insulin therapy.
Method:
This retrospective quality improvement project examined the rates and percentages of hyperglycemia >180mg/dL, hypoglycemia <70mg/dL, hypoglycemia <54mg/dL and extreme hypoglycemia <40mg/dL for patients treated with IV eGMS® on a labor and delivery (L&D) unit at a 243 bed, not-for-profit hospital in Honolulu, Hawaii. Time to target data were extracted to determine how quickly patients were brought into their targeted blood glucose (BG) range among patients with a prescribed range of 80-120mg/dL. Four thousand, one hundred and seventy-four BG values were collected from 204 patients over a 31-month period, from November, 2015 through June, 2018.
Result:
There was a 7.19% rate of hyperglycemia >180mg/dL (n=300 BG values). There was a 2.35% rate of hypoglycemia <70mg/dL (n=98 BG values); 0.53% rate of hypoglycemia <54mg/dL (n=22 BG values); and a 0.14% rate of extreme hypoglycemia <40mg/dL (n=6 BG values), respectively. The median time for patients to achieve the target range of 80-120mg/dL from the onset of IV eGMS® insulin therapy was six hours.
Conclusion:
Non-invasive Raman-based Glucose Monitoring with weeks of Sustained Calibration
RSP Systems Odense, Denmark
Objective:
Stability of calibration is an important quality of non-invasive and continuous glucose monitoring devices. Drift and other variation due to the device, as well as variation due to measurement conditions, are typically compensated by calibration to reference measurements at appropriate time intervals. Herein, we assess the sustainability of calibration models of non-invasive Raman-based glucose monitors. The assessment is carried out by comparing the predictive performance of calibration models without and with weekly recalibration. The analysis period covers the range of 4 weeks to more than three months, depending on the subject under study.
Method:
Raman spectra were recorded from the thenar of the diabetic subjects using RSP Systems Working Model WM34 with accompanying glucose references obtained using Hemocue 201. Data from two protocols RSP-07/10, both outpatient studies, were pooled and the dataset comprised 7 diabetic subjects. Calibration models were based on 40 days of analysis, where each day contained four measurements with corresponding reference blood glucose values. The validation period was dependent on the subject under study, but it covered a minimum of 21 analysis days that were distributed over a minimum of 4 weeks. In the scenario with recalibration, two additional reference measurements were added to the model every seventh day. The calibration models were subject-specific and based on partial least squares regression.
Result:
The scenario without recalibration shows an average inter-subject unified performance (ISUP) of 2.63 and mean average relative difference (MARD) of 25%, whereas the inclusion of weekly recalibration results in an ISUP of 2.71 and MARD of 25%. The independent validation set comprises 517 paired points.
Conclusion:
Despite the relatively small dataset, the results indicate an insignificant effect of weekly recalibrations, hereby emphasizing that non-invasive Raman-based glucose monitors can sustain the calibration for weeks.
New Clamp-PID Algorithm Outperforms Biostator Algorithm in Automated Glucose Clamps
Profil Neuss, Germany
Objective:
Automated glucose clamps (e.g. using the Biostator or ClampArt) use an implemented algorithm to determine glucose infusion rates (GIR) every minute based on continuous blood glucose (BG) measurements. This allows bias- free pharmacodynamic assessments of anti-diabetic agents. However, the Biostator algorithm was not developed nor validated for rapid acting insulin analogs. We, therefore, investigated the clamp quality achieved with the Biostator algorithm and a new Clamp-PID algorithm.
Method:
A new Clamp-PID algorithm was developed by numerical simulations. The coefficients for the proportional, the integrative, and the derivative part of the control algorithm were optimized with respect to the fluctuations in BG using a simulated rapid acting insulin profile. The best set of coefficients was tested in-vitro with a 5L container filled with glucose solution. The insulin profile was created by an external water infusion with appropriate infusion rates into a container which let the glucose solution overflow and thereby reduced glucose concentration. The glucose clamp quality parameters precision (standard deviation of all BG values) and absolute control deviation (mean of absolute differences between target level and BG values) were compared between both algorithms.
Result:
The new clamp PID algorithm reduced glucose fluctuations by nearly 60% in the in-vitro experiments (precision 1.4±0.3% vs. 3.3%±0.1% with the Biostator algorithm). In addition, glucose values were kept closer to target with mean absolute control deviation of 0.8mg/dl±0.2mg/dL with the new PID-algorithm versus 2.3mg/dl±0.4mg/dL with the Biostator algorithm).
Conclusion:
In-vitro experimental data show that the new Clamp-PID algorithm substantially improved glucose clamp quality for both precision and control deviation. The new Clamp-PID algorithm will now be investigated in clinical studies.
Glucose Profile Indicator – a Simple Combined Measure of Mean Glucose, Glycemic Variability and Hypoglycemia for Better Evaluation of Glycemia
International Diabetes Center Minneapolis, MN, USA
Objective:
A1C does not capture all of the relevant aspects of a glucose profile unlike continuous glucose monitoring (CGM). Mean glucose, glycemic variability, and time-in-hypoglycemia are all important factors influencing the time spent in and out of target range and are not revealed by A1C. However, evaluating multiple metrics at once can be challenging.
The objective was to develop a simple combined measure, the Glucose Profile Indicator (GPI), as an indicator of a clinically favorable glucose profile: most time-in-range without hypoglycemia.
Method:
GPI is a weighted measure of mean glucose and glycemic variability (coefficient of variation [CV]), with penalization for hypoglycemia [(%) time spent <54 mg/dL], standardized to a glucose of 154 mg/dL and a CV of 36%. The GPI metric was applied to data from Cornerstone4Care powered by Glooko, a diabetes management application capturing self-reported, real-world, CGM data, and tested via Spearman’s correlations to identify patients with the most time-in-range and the least time-in-hypoglycemia. Moreover, GPI was tested to rank ambulatory glucose profiles (AGPs) – a key diabetes management tool – according to the most time-in-range without hypoglycemia, compared with the individual components or time-in-range alone. Ranking was validated by subjective expert opinion. All calculations were based on 2-week periods with >70% of possible CGM-readings.
Result:
A total of 137 patients with self-reported type 1 diabetes uploaded CGM profiles for 790 2-week periods. Mean glucose correlated (r) to time-in-range (r=−0.96) and CV correlated to time-in-hypoglycemia (r=0.67). GPI correlated to both (r=−0.59; 0.71, respectively) and identified patients with the most time-in-range and with least time-in-hypoglycemia. GPI ranked the AGPs better than mean glucose, CV, or time-in-range alone.
Conclusion:
GPI is a simple combined measure that can help patients and healthcare professionals evaluate glucose control.
Activity Monitoring in Latino Adults with Type 2 Diabetes
Sansum Diabetes Research Institute Santa Barbara, California, USA
Objective:
Latino families in the United States bear a disproportionate cardio-metabolic disease burden. The goal of the initial year of the 10-year Mil Familias study was to establish methods to measure the impact of biological and socio- cultural factors for Latino adults with type 2 diabetes (T2D). We present pilot data on daily physical activity.
Method:
Latino adults ≥18 years with an established diagnosis of T2D wore a Fitbit Charge 2TM (FB) on the non-dominant wrist and an ActiGraph wGT3X-BT (AG) on the dominant hip, 24 hours a day for 7 days a week, except when bathing or swimming, to monitor daily activity (estimated by steps/day algorithms). An activity self-evaluation was also completed.
Result:
Fifty-one participants (n=36 women) completed activity monitoring (mean ± SD – age 53.9±11.9 yrs., weight 75.9±17.5 kg., waist circumference 113.8±75.9 cm, BMI 31.1±6.5 kg/m2). Daily step counts between two monitors were correlated for Mon – Sun (r=0.7–0.8, P<0.001). Median steps/day by AG were significantly less than by FB (p≤0.02) [Median AG & FB steps/day– Mon 7,416 & 9,219; Tue 7,347 & 10,023; Wed 7,766 & 10,996; Thu 7,548 & 9,472; Fri 7,935 & 10,481; Sat 5,608 & 8,405; Sun 5,340 & 8,735]. Significantly more activity occurred on weekdays compared to weekends (p<0.01, both AG & FB). Most participants reported working 2-3 days/week and being very active on those days [7.4±1.7 on a 1 to 10 scale (not active to very, very active)].
Conclusion:
Daily physical activity can be assessed with consumer-grade portable monitors like FB. Given that participants either met or approached the FB recommended goal of 10,000 steps/day on weekdays, measuring activity among Latinos with T2D via FB and AG is both feasible and challenges assumptions about lack of activity as a major factor in their T2D.
Real Time Blood Glucose Levels Control for Patients on Basal Insulin Injection Therapy
Insuline medical Jerusalem, Israel,
Objective:
In patients with type 1 diabetes (T1D), closed-loop systems provide excellent overnight fasting blood glucose control by automatically increasing or decreasing pump basal infusion rates. In patients on multiple daily insulin injections, such control has not been possible due to the inability to alter the rate of absorption of long-acting insulin analogs following injection. In this study, we tested the hypothesis that cooling or warming the skin around the injection site, can be used to prevent hypoglycemic and hyperglycemic events.
Method:
Type I diabetic subjects were admitted and injected their usual Insulin Glargine dose. A treatment device, attached to the skin over the injection site, applied cooling or heating to decrease or increase the Glargine release rates following controller decisions based on CGM readings. Each subject completed two procedures: a test case with device activation and a control test without device activation. The whole procedure was conducted under fasting conditions to mimic night time. We compared the number of hypoglycemic and hyperglycemic events under test or control conditions.
Result:
Seventeen type I subjects were included in this feasibility study. Overall, 4 patients in the test arm experienced 5 hypoglycemic events compared to 10 patients who experienced 15 hypoglycemic events in the control arm (P=0.0078 ). No patient experienced hyperglycemic events.
Conclusion:
The results indicate that the use of the treatment device can reduce the risk of hypoglycemic events. A larger study in home settings is needed to further evaluate the potential of this system.
Prevention of Severe Hypoglycemia by Use of the Electroencephalography-based Alarm Device, hyposafeTM SubQ
UNEEG medical A/S Lynge, Denmark
Objective:
Hypoglycemia is associated with characteristic changes of the electrical activity of the brain with highly varying glucose thresholds for the onset of these changes. The objective of our current study is to test the subcutaneous electroencephalography (EEG) alarm device, hyposafeTM SubQ using the brain as a biosensor to prevent severe hypoglycemia.
Method:
Eight patients with type 1 diabetes and impaired hypoglycemia awareness were included in a 3-month pilot study. EEG was recorded continuously and analyzed in real-time using an automated EEG algorithm. Daily use of the device was recorded.
Result:
Seven patients completed the study with a total of 659 recording days. Median daily use was 22.3 hours (range: 19.1-23.0) for 97 days (range: 60-109). One patient withdrew due to discomfort. The hyposafeTM SubQ detected a total of 16 events of hypoglycemia-related EEG changes in 5 patients - all during daytime. The alarm was triggered at a median blood glucose level of 2.4 mmol/L (range: 1.9-3.0). In all cases, the patients were able to take preventive actions to avoid severe hypoglycemia. No false negatives were reported, as no patients reported a need for third- person assistance. The patients experienced a median false detection rate of 2.9/week (range: 0.3-8.9) during daytime and 0.2/week (range: 0.0-2.0) during nighttime.
Conclusion:
Seven patients used the hyposafeTM SubQ over a 3-month period with high compliance. Hypoglycemia- related EEG changes were detected in time for the patients to take preventive action and avoid severe hypoglycemia. No false negatives, and few false positives, were detected, especially at night.
Insulin Sensitivity, a Function of Blood Glucose Level: Proof of Concept from the SP7 Multi-Centric Clinical Study
University of Grenoble Alpes CEA Leti Grenoble, France
Objective:
Compensation ratio (CR), a proxy of insulin sensitivity, is the number of insulin units required to reduce glycemia 1g/L. It is used on a daily basis to compute the amount of insulin (compensation bolus) required to compensate for hyperglycemia. In practice, this patient-specific setting is a constant value. However, it is recognized that CR is affected by the physiological environment such as the glycemic level, but not measured. This abstract proposes to investigate, using data from the NCT02987556 clinical study, the correlation between insulin sensitivity and glycemia.
Method:
The dataset included several factors [i.e., blood glucose (mg/dL), insulin infusion (UI), and CHO absorption (g)] from 32 patients with type 1 diabetes (n=20 female, age: 48 +/-13 years old, weight: 69 +/- 11 kg and total daily dose: 34 +/- 10 UI) wearing a Diabeloop DBLG1 for 3 months. Based on carefully selected events, CR was estimated, using the relationship: CR=InsulinConsumed/ΔGlycemia [UI/(g/L)]. Then we established a model to predict the CR with the glycemia measured at the time of the compensation bolus injection.
Result:
CR calculations are based on 38+/-21 events per patients. Based on these data, we found a mean correlation (0.59+/-0.15) between CR and glycemia, and the slope of the linear regression across all subject is 0.98+/-0.47 [UI/(g/L)2]. Gathering events (n=1229) from all patients and normalizing by patient body weight (CRBW), we found with glycaemia, a correlation of 0.38 (P<0.001) and linear regression model: CRBW=0.012 (glycurrent[g/L]-1.5)+0.107.
Conclusion:
In this work, we have demonstrated the correlation between CR, which is a surrogate of insulin sensitivity, and glycemia. Thus, this relationship could be useful for diabetes management in particular for determining the dosage of the compensation bolus.
Retrospective Evaluation of Abbott Precision Xceed Pro Glucose Meter in a Hospital Environment
Baptist Health System San Antonio, Texas, USA
Objective:
Due to the accuracy guidelines published for hospitals for the use of glucose meters by the United States Food and Drug Administration (FDA) and the FDA public meeting held on March 30, 2018 calling for public comment on this subject, we wanted to determine the accuracy in a real world situation with a variety of patients for the Abbott Precision Xceed Pro® glucose meter. The objective was to compare clinical accuracy to the accuracy data submitted to the FDA. Most studies of this nature compare critical patients only. This study is based on samples from all hospital areas. Little has been published on this meter type. Most of the published data is on critical patients and on other meters.
Method:
The method was a retrospective examination of the Electronic Medical Record (EMR) for the month of April 2018. This study used paired samples from the glucose meter and the laboratory instrumentation from 0-2 minutes of collection time difference for all patient units (n=323 results).
Result:
There were insufficient data to compare the results of samples <75 mg/dl to the FDA public meeting criteria or to enter the data into the Parkes error grid (N=12). For samples with glucose > 75 mg/dL, the data showed that 80.2% (259/323) were within +/- 12% of the laboratory glucose value.
Conclusion:
Use of the Parkes Consensus Error Grid showed that the paired samples did not result in possible insulin dosing errors because 100% of the values were within zones A and B. In this study, the Abbott Precision Xceed Pro® has lower accuracy than data submitted to the FDA.
Driving Behavior Change in a Large Population of People with Diabetes using Stochastic and Machined-Learned Algorithms
Livongo Health Mountain View, CA, USA
Objective:
To improve diabetes self-management behavior in a large real-world population of people living with diabetes through personalized messaging.
Method:
The Livongo for Diabetes Program leverages novel technology to offer: (1) a cellular enabled blood glucose meter, (2) real-time personalized analytics, (3) unlimited glucose test strips, and (4) access to certified diabetes educator coaches. We conducted a long-running randomized control trial to quantify the performance of a personalized messaging and behavior change platform that can support multiple machine-learned and stochastic policies. Insights about the resulting behavior change were analyzed using standard null-hypothesis significance testing (NHST) and machine learning approaches including supervised learning methods and Bayesian structural time series. This work is enabled by the self-monitoring blood glucose (SMBG) data continuously collected remotely from >80,000 people with diabetes in the Livongo for Diabetes program. The connected Livongo glucose meter provides data and also provides the surface for the behavior change messages.
Result:
Initial testing of our personalized messaging platform consisted of stochastic message selection from a clinically defined deterministic set of messages. This early experiment resulted in a statistically significant increase in SMBG checking frequency. In the second phase of this platform, machine-learned policies are being included and insights about model performance will be used to make improvements to behavior change policies.
Conclusion:
Personalized messages delivered via connected glucose meter based on stochastic and machine-learned models can improve diabetes self-management behaviors including BG checking frequency.
Ready-To-Use Liquid Glucagon Rescue Pen – A Phase 3 Study of Plasma Glucose Recovery in Pediatric Patients with Type 1 Diabetes (T1D)
Department of Pediatrics, Endocrinology and Diabetes Division, Stanford University School of Medicine Stanford, CA, USA
Objective:
A novel ready-to-use, room-temperature-stable liquid glucagon rescue pen (GRP, Xeris Pharmaceuticals) auto- injector was evaluated for the treatment of insulin-induced hypoglycemia in pediatric subjects with age-specific doses.
Method:
Three cohorts were studied: children ages 2 to < 6 years and 6 to < 12 years received one dose of Xeris Rescue Pen 0.5mg and adolescents ages 12 to < 18 years received 0.5 and 1 mg doses of the Rescue Pen on two separate visits. Glucagon was administered when plasma glucose (PG) was lowered to < 80mg/dL.
Result:
A total of 31 children with T1D (ages 2 to < 18 years) were studied. All evaluable participants had a glucose elevation of ≥ 25 mg/dL from baseline. Across age groups there were no notable differences with regards to mean glucose AUC(0-90min), Cmax (199-208 mg/dL), and Tmax (66-82 min). Plasma glucagon AUC(0-90min), Cmax, and Tmax were similar across the age groups. Across all age groups the mean increase in glucose at 15 minutes was 23.3±20.6 mg/dL and at 20 minutes was 42.2±22.7 mg/dL. Mild and moderate nausea (43%) and vomiting (14%) were the most commonly reported adverse events. No serious adverse events occurred.
Conclusion:
Currently available glucagon kits for treatment of severe hypoglycemia require a complex multi-step reconstitution process, making them difficult to administer in an emergency. A ready-to-use, easy, two-step administration GRP using a pre-measured pediatric glucagon dose has been developed and tested. Results from this Phase 3 study demonstrate that age-appropriate doses of GRP were effective and safe in pediatric participants and support the use of this ready-to-use liquid glucagon formulation for the treatment of severe hypoglycemia.
Lag Time of a Sixth-Generation Continuous Glucose Monitoring System
Dexcom, Inc. San Diego, CA, USA
Objective:
Diffusion of glucose from the vascular to the interstitial compartment causes an unavoidable delay in detecting changes in blood glucose concentration with commercially-available continuous glucose monitoring (CGM) systems. We estimated the lag time of a sixth-generation CGM system (“G6”, Dexcom, Inc., San Diego, CA) which includes a novel sensor, applicator, and signal-processing algorithm.
Method:
Each of 62 participants had insulin-treated diabetes and wore one system. Lag time was estimated by comparing the system’s estimated glucose values (EGVs) to concurrent reference (YSI) venous glucose measurements. EGVs were reported every 5 minutes and interpolated to provide EGVs at 1-minute intervals. YSI values were obtained every 15±5 minutes. Each of the measured and interpolated EGVs was associated with a lag time between it and the most recent prior YSI value. The lag time with the lowest EGV-YSI absolute relative difference (ARD) was taken as the lag time for that sensor. Lag times for sensors inserted in the abdomen and upper buttocks regions were assessed separately. All analyses were performed using SAS® software, version 9.3 (SAS Institute, Inc., Cary, NC).
Result:
The median lag time for all 62 sensors was 4 minutes (IQR, 1-6 minutes). The mean (SD) time lag of the 62 individually analyzed sensors was 3.7 (3.1) minutes. Fourteen (23%) of the sensors had estimated lag times of <1 minute. The mean (SD) lag time for the 47 abdominally-inserted sensors was slightly lower than for the 15 sensors inserted in the upper buttocks region [3.4 (3.0) vs. 4.7 (3.3) minutes, respectively].
Conclusion:
Changes in blood glucose concentrations are rapidly reflected in EGVs provided by the G6 CGM system.
A New CGM-based Algorithm to Generate Preventive Hypotreatments: an in-silico Assessment
Department of Information Engineering, University of Padova Padova, Italy
Objective:
Mitigation of the risk of prolonged hypoglycemia in T1D management requires patients to consume small doses of fast-acting carbohydrates, the so-called hypotreatments (HTs), as soon as hypoglycemia is detected. The present work evaluates, in an ideal noise-free simulated environment, the margins of improvement in HT administration granted by the possibility of predicting the occurrence of hypoglycemia ahead of time using CGM sensors.
Method:
A simulation framework relying on a well-established mathematical model of T1D metabolism has been devised to generate hypoglycemic events with different severity classes in 100 virtual patients.
Result:
An algorithm to administer preventive HTs was developed by resorting to the “dynamic risk” non-linear function.
Conclusion:
An algorithm that combines current glycaemia with its rate-of-change provided by CGM can be adapted to distinguish the severity of the about-to-happen hypoglycemia.
Using Patient Activity Trends to Predict Impending Exercise and Sleep
Rensselaer Polytechnic Institute Troy, NY, USA
Objective:
Anticipating exercise allows for a reduction in insulin administration before exercise, mitigating the chance of hypoglycemia. Anticipating meals allows for more accurate and earlier meal detection and for user prompts or for fully closed-loop glucose control. This work explores the use of person-specific trends (e.g. high school schedules, biking to work, league sports, etc.) as revealed in activity logs to improve the prediction of future activity with an emphasis on meals and exercise
Method:
We synthetically generated a dataset for a high-school student over 12 weeks. The data included a school schedule with physical education every other day. Variability was included in wakeup time, sleep time, breakfast time, dinner time, existence of league sports games, weekend exercise, afternoon snack, and to a lesser extent: lunch. The prediction method used locally-weighted regression (LWR) against sleep/wake logs from the American Time Use Survey and user logs. LWR adapts by also regressing on the past activity logs. The algorithm heavily weighted distance for the current activity. We compared to an algorithm that predicts based on the prior probabilities of each activity with another that assumes that the current activity continues forever.
Result:
The prior algorithm produced overall/eating/exercise root mean square error (RMSE) of 0.73/1.2/1.2 across all prediction horizons, weeks, and activities. The trend predictions are most accurate for short prediction horizons with exercise and meal prediction RMSEs saturating after an hour at 1.4. For 1, 6, and 12 weeks, the LWR algorithm reduces overall prediction RMSE to 0.39/0.34/0.34, 30-minute meal prediction RMSEs to 1.2/0.99/1.0, and 2-hour exercise prediction RMSEs to 1.2/0.71/0.66.
Conclusion:
With 6 weeks of data, the method reduced overall/meal/exercise prediction error by 46/18/45%, respectively. Our current effort is applying this approach to free-living data.
CGM-based Improvement and Personalization of the Standard Formula for Insulin Meal-Bolus Calculation in Type 1 Diabetes
Department of Information Engineering, University of Padova Padova, Italy
Objective:
In type 1 diabetes (T1D) therapy, the insulin meal bolus (IMB) is normally computed using an empirical “standard formula” (SF) originally developed for blood glucose (BG) measured by finger prick devices. When CGM sensors measure BG, the SF can still be used, but a finer IMB calculation could be obtained by exploiting glucose dynamics information, with potential improvement of glycemic control. Our aim is to develop a machine learning methodology to adjust and individualize the computation of IMB by taking into account CGM-based information and easily accessible patient specific parameters.
Method:
The UVa/Padova T1D Simulator was used to generate data for100 virtual subjects in single-meal, noise-free, scenarios with different preprandial BG and glucose rate of change (ROC), opportunely divided into training and test sets. In the training set, we calculated ΔIMB as the optimal IMB minus IMB obtained with SF. Then, we fitted a linear regression model (LRM) to predict ΔIMB using 10 patient features (carbohydrate-to-insulin ratio, correction factor, insulin-on-board, carbohydrate-on-board, body weight, meal carbohydrate intake, target BG, basal insulin, preprandial BG and ROC), by minimizing the BG risk index (BGRI). On the test set, we compared performance of SF and LRM in terms of time in hypo/hyperglycemia and BGRI.
Result:
On average, comparing LRM vs. SF, the glycemic performance improved: time in hypoglycemia 5.22% vs. 12.57%; time in hyperglycemia 21.93% vs. 21.92%; BGRI 6.94 vs. 8.75.
Conclusion:
We developed a machine learning method to adapt to CGM and individualize the IMB computation. Compared to SF, results obtained in an ideal scenario show a consistent improvement, which encourages future development of the methodology including an assessment under more challenging simulations.
Evaluating the Impact of eGMS- Glucommander on Length of Stay, Hypoglycemia, and Glucose Control Used in a Regional Medical Center
Riverside Medical Center Kankakee, IL, USA
Background:
Healthcare organizations face numerous challenges when implementing glycemic management improvement initiatives. This study will examine differences in patient outcomes with a focus on COPD, CHF and DKA patient populations between patients whose insulin titrations are managed with or without the GlucommanderTM (GM) electronic glycemic management system (eGMS®).
Method:
The aim of this retrospective quality improvement study is to evaluate the clinical and financial outcomes of the eGMS® system compared to standard (paper) protocols (SP) in the Critical Care Units of a 335 bed Regional Medical Center. Data were collected and analyzed from November 1, 2016 through October 31, 2017 on patients requiring glucose management with IV or subcutaneous insulin.
Result:
Length of stay (LOS) index for GM was 1.12 vs 1.37 days for SP. The severe hypoglycemia point of care (POC) rate was 0.11% for GM vs 0.41% for SP and mild-moderate hypoglycemia POC rate was 2.34% for GM vs 3.86% for SP. Average BG for GM was 166.61 mg/dL and 187.33 mg/dL for SP with an ICU average final BG of 154.65 mg/dL for GM and 185.85 mg/dL for SP.
Conclusion:
GM provided 0.25 days lower LOS index compared to SP. GM patients experienced less severe and overall hypoglycemia and GM was more effective at reaching ADA targets for average and final BG compared to SP.
Advanced Glycemic Care Approach by Identifying Diurnal Glycemic Patterns and A1c Factorization Utilizing Flash Glucose Monitoring
Lina Diabetes Care Centre Mumbai, Maharashtra, India
Objective:
The Flash Glucose Monitoring (FGM) system provides a visual snapshot of glycemic patterns for up to 14 days without any SMBG calibration. We aimed to analyze the diurnal glycemic patterns across various categories of A1c.
Method:
We retrospectively analyzed Continuous Glucose Monitoring (CGM) reports of 181 type 2 diabetes patients conducted using FreeStyle Libre ProTM during January 2017- December 2017. The groups were categorized based on estimated A1c (eA1c), <7% (n= 76), 7-7.9% (n= 44), 8-10% (n= 45), >10% (n=16) and 2-hour factorized glycemic patterns were analyzed. Statistical analysis was conducted using Mann Whitney test, Kruskal-Wallis test and ANOVA.
Result:
The lowest blood glucose (BG) values were recorded during 4am-6 am, irrespective of the A1c category (mean 157 mg/dL, SD 66.57, SEM 33.28, min 98, max 251, 95% CI 51.08 to 262.9) followed by 6am-8am (mean 162.5 mg/dL, SD 67.32, SEM 33.66, min 101, max 257, 95% CI 55.38 to 269.6) and 2am-4 am (mean 197.8 mg/dL, SD 72.84, SEM 36.42, min 125, max 296, 95% CI 81.84 to 313.7). The mean BG was consistently > 200 mg/dL from 10am –12am. The highest BG for <7% and 7-7.9% was 147.3 and 203.6 mg/dL respectively, noted at post-breakfast 10am-12 pm, whereas for the uncontrolled group >10 %, the highest BG was at pre-dinner during 6pm-8pm (323.18 mg/dL)
Conclusion:
We postulate that efficacious therapeutic interventions should be targeted towards the management of hyperglycemia during the post-breakfast and pre-dinner period. There is high risk of low blood glucose levels in the later part of the nocturnal period. Our study results provide a precise novel perspective with FGM for advanced glycemic care which needs further validation through prospectively designed, larger, multi-centric population studies.
Activity Detection and Activity Level Categorization in Free-Living Subjects with Type 1 Diabetes
Harvard John A. Paulson School of Engineering and Applied Sciences, Harvard University Cambridge, MA, USA
Objective:
To develop a method to perform physical activity detection and physical activity level classification using unlabeled accelerometer data collected from patients with type 1 diabetes (T1D) in a free-living setting.
Method:
Twenty subjects with T1D (12F/8M; mean age 42.6 years) participated in a 5-week outpatient study performing their daily life activities while wearing an Empatica E4 watch on their non-dominant wrists, annotating activity times and corresponding intensity on a personal diary. The 3-axis-accelerometer data recorded by the watch at 32 Hz, bandpass filtered between 0.5 and 10 Hz to eliminate noise and unwanted hand motion, were used to calculate the average magnitude of the displacement vector in 3 dimensions for every second in time. Three activity regions and their relative thresholds corresponding to sedentary, mild and moderate activity were empirically determined by analyzing over 5,200 hours of the computed displacements. Our algorithm detected and classified activity levels at 5 minute intervals based on the mean displacements threshold crossing.
Result:
The proposed method for activity detection and classification was tested retrospectively on a 315 days-long blinded dataset from 10 subjects and its performances qualitatively assessed against patients’ annotations. Categorization between two classes (sedentary/non-sedentary) revealed 94% specificity; 74% sensitivity and 90% accuracy, while classification between three classes (sedentary/ mild/ moderate activity level) gave 84% specificity; 92% sensitivity and 89% accuracy.
Conclusion:
The study showed that our method can detect and categorize daily life activities with good accuracy from a wrist- worn accelerometer without user intervention in patients with T1D. Our proposed method can potentially be utilized in an automated insulin delivery system by adapting insulin dosage to the intensity and duration of the performed activity.
Effectiveness and Safety of a Novel Percutaneous Optical Fiber Continuous Glucose Sensor (FiberSense) in Clinic and Home Use in Patients with Diabetes
Phase 1 Clinical Trial Centre, Department of Medicine and Therapeutics, Prince of Wales Hospital, The Chinese University of Hong Kong Hong Kong, SAR
Objective:
To evaluate the effectiveness and safety of the FiberSense system, a percutaneous fiber-optic real-time continuous glucose sensor for home use in diabetes patients, when used for up to 29 days.
Method:
This was a single-centre clinical trial involving type 1 and type 2 diabetes patients on insulin. Each subject wore a FiberSense sensor either on the upper arm or abdomen for up to 29 days. Subjects attended four in-clinic measurement sessions on days 1, 7, 14 and 28 where FiberSense readings were compared against a standard laboratory method (YSI glucose) every 10 minutes during a glucose challenge. During home use, patients performed self-monitored capillary blood glucose at least four times a day. Subjects also wore a commercial comparator CGM (Dexcom G4) during one of the study weeks.
Result:
In an analysis of 15 subjects who completed up to 29 days of sensor wear (28.4 days on average; n=7 on abdomen and n=8 on upper arm), the pooled mean absolute relative difference (MARD) against YSI glucose was 15.0% (15.7%; 95%CI upper bound) across the full glycemic range (n=1,172) as compared to MARD 16.8% (18.6%, 95%CI upper bound) (n=336) with Dexcom CGM. Consensus Error Grid (CEG) analysis yielded 99.7% of paired FiberSense measurements in zones A plus B. Surveillance error grid showed 73.9% of values within the ‘no clinical risk’ zone, followed by 18.5%/7.2% of slight risk of hyper/hypoglycemia and only 0.5% for moderate risk of hyperglycemia. The system was well tolerated and there were no serious device-related adverse events or sensor site reactions.
Conclusion:
These early results provide encouraging evidence that FiberSense CGM is acceptable and comparable in clinical accuracy to existing commercial CGM over 29-days of home use.
A Phase 3 Comparison of a Ready-To-Use Liquid Glucagon Rescue Pen to Glucagon Emergency Kit for the Treatment of Severe Hypoglycemia
Diablo Clinical Research Walnut Creek, CA, USA
Objective:
To prevent medical complications of severe hypoglycemic emergencies, prompt and reliable rescue intervention is critically important. A ready-to-use stable liquid Glucagon Rescue Pen (GRP; Xeris Pharmaceuticals) auto-injector was evaluated for the rescue treatment of severe hypoglycemia.
Method:
A Phase 3 randomized, controlled, single-blind, crossover study was conducted in 81 adults with type 1 diabetes to compare 1 mg doses of the GRP versus the Glucagon Emergency Kit (GEK; Eli Lilly) for the treatment of insulin- induced severe hypoglycemia. Efficacy was evaluated as either a return of plasma glucose to >70 mg/dL or an increase in plasma glucose of ≥20 mg/dL from a baseline glucose <50 mg/dL within 30 min of dosing.
Result:
On an ITT basis, all subjects achieved successful plasma glucose recovery within both groups. From a decision to dose, the mean time for plasma glucose > 70 mg/dL after drug administration was 13.3
Conclusion:
A ready-to-use, easy two-step administration GRP achieved plasma glucose recovery reliably, was therapeutically equivalent to GEK, and had an incidence of nausea and vomiting comparable to GEK. These results demonstrate that GRP is an effective, safe, and well tolerated rescue treatment for severe hypoglycemia and is a viable alternative to GEK.
A Phase 3 Comparison of a Ready-To-Use Liquid Glucagon Rescue Pen to Glucagon Emergency Kit for the Symptomatic Relief of Severe Hypoglycemia
Diablo Clinical Research Walnut Creek, CA, USA
Objective:
A novel ready-to-use, stable liquid Glucagon Rescue Pen (GRP; Xeris Pharmaceuticals) auto-injector was evaluated for relief of symptoms during rescue treatment of severe hypoglycemia.
Method:
A Phase 3 randomized, controlled, single-blind, crossover clinical trial enrolled 81 adults with type 1 diabetes to compare subcutaneous 1 mg doses of the GRP versus Glucagon Emergency Kit (GEK; Eli Lilly) for the treatment of insulin-induced severe hypoglycemia in adults. Serial assessments of four autonomic, four neuroglycopenic, average total symptoms, and sensation of hypoglycemia were performed at each treatment visit.
Result:
The mean time to symptom relief from the time receiving glucagon treatment was comparable between the GRP and GEK for autonomic symptoms (9.9
Conclusion:
The prompt relief of neurologic symptoms is critical in the rescue of severe hypoglycemic emergencies. The GRP achieved both autonomic, neuroglycopenic, and average total symptom relief during induced severe hypoglycemia. GRP achieved successful plasma glucose recovery in a reliable manner, was safe and well tolerated, and had an incidence of nausea and vomiting comparable to GEK. These results demonstrate that ready-to-use GRP is a viable alternative to currently available GEK.
Minimum CGM Data Requirements to Assess Risk in Type 1 Diabetes
University of Virginia Charlottesville, Virginia, USA
Objective:
Low Blood Glucose Index (LBGI), High Blood Glucose Index (HBGI), Risk Index (RI), and Average Daily Risk Range (ADRR) are indices of glucose variability and glycemic risks, for individuals with type 1 diabetes (T1D), with unique capacity to predict significant clinical events (e.g. severe hypoglycemia). Though originally intended to be computed from sparse self-monitoring blood glucose (SMBG) values, these risk measures have been modified for continuous glucose monitoring (CGM), and this work aims to determine the minimum number of days of CGM data required to calculate LBGI, HBGI, RI, and ADRR.
Method:
LBGI, HBGI, Risk Index, and ADRR were calculated retrospectively using 4 weeks of data from 19 participants in clinical trials at the University of Virginia Medical Center. For each studied length of time (1 to 27 days), indices were computed over randomly selected days of CGM (up to 1,000 iterations) and compared to 28 day assessments. The risk scores from 4 week data were considered the saturation values. The median amount of data across all the subjects that fell within a 95% confidence interval of the mean at 28 days was compared for each amount of data. Ninety-percent of the estimates within the mean confidence interval was considered sufficient.
Result:
The threshold of 90% of the values within the confidence interval was reached at 19, 12, 19, and 18 days for LBGI, HBGI, RI, and ADRR calculations, respectively. Only 67% of estimates were deemed accurate enough with data from only one week.
Conclusion:
We conclude that 3 weeks of CGM data is sufficient to calculate a stable estimate of LBGI, HBGI, RI, and ADRR in subjects with T1D.
An Assessment of Usability and Drug Preparation Time for a Ready-To-Use Liquid Glucagon Rescue Pen
Xeris Pharmaceuticals Chicago, IL, USA
Objective:
Prompt and reliable intervention is critically important when responding to severe hypoglycemic emergencies. Time-to-treatment and usability of a novel, ready-to-use, liquid Glucagon Rescue Pen (GRP) was evaluated and compared to current emergency glucagon rescue kits (GEK) which requires a complex multi-step reconstitution and administration process.
Method:
A simulated-use human factors study was conducted with the GRP compared to GEK in 16 participants experienced with GEK, consisting of first responders and caregivers of diabetic patients. A validation study was conducted with 75 adults and adolescent caregivers both experienced and naïve to GEK. In addition, a Phase 3 randomized, controlled, single-blind, crossover clinical trial enrolled 81 adults with type 1 diabetes compared 1 mg doses of GRP versus GEK and a comparison of study drug preparation time was performed.
Result:
In the formative usability study, 14/16 (87.5%) of participants successfully administered a rescue injection using the GRP vs. 5/16 (31.3%) using the GEK (p<0.05). Mean total rescue time of was 47.9 seconds with the GRP vs 109.0 seconds with GEK (p<0.05). In the summative validation study, 74/75 (98.7%) of subjects successfully administered the GRP. Overall, there were no patterns of differences between user groups. In the Phase 3 study, drug preparation and administration time was significantly shorter (p<0.0001) for GRP (27.3
Conclusion:
These results demonstrate that the GRP and associated instructional materials can be correctly, safely, and effectively used by the intended user populations. There was a significant difference in drug preparation time in favor of the GRP. These results support the GRP as an alternative to GEK.
Increased Time in Range and Improved Glycemic Variability with Sotagliflozin in Combination with Insulin in Adults with Type 1 Diabetes: A Pooled Analysis of 24- Week Continuous Glucose Monitoring Data
Diabetes Center for Children and Adolescents, Auf Der Bult, Kinder-und Jugendkrankenhaus Hannover, Germany
Objective:
To evaluate effects of the dual SGLT1/SGLT2 inhibitor sotagliflozin in combination with insulin on glycemic metrics in adults with type 1 diabetes using masked continuous glucose monitoring (CGM).
Method:
Pooled data of 278 participants in a masked CGM substudy of the inTandem1 (NCT02384941) and inTandem2 (NCT02421510) trials were analyzed, comprising patients randomized to placebo (nr), sotagliflozin 200mg (nr), or sotagliflozin 400mg (nr). The primary outcome was the 24-week change of time in range (TIR) in target glucose levels (70–180mg/dL). Secondary outcomes were time below/above target range, area under the curve (AUC) outside target range, mean amplitude of glucose excursions (MAGE), and 2-hour postprandial glucose (PPG) after a standardized mixed meal.
Result:
Treatment with sotagliflozin 200mg or 400mg versus placebo significantly increased TIR at target (1.3±0.6hr/day, P=0.026 and +2.8±0.6hr/day, P<0.001, respectively), with no significant changes observed for hypoglycemia TIR (<70mg/dL) for sotagliflozin 200mg (−0.1±0.2hr/day, P=0.70) and 400mg (−0.02±0.2hr/day, P=0.93). TIR in hyperglycemia (>250mg/dL) was significantly reduced with sotagliflozin 200mg or 400mg versus placebo (−0.9±0.4hr/day, P=0.045 and −1.9±0.4hr/day, P<0.001, respectively). Significant reductions in 2-hr PPG were observed with both sotagliflozin 200mg (−34.6±13.2mg/dL, P=0.009) and 400mg (−49.7±13.1mg/dL, P<0.001) versus placebo. CGM AUC (>180mg/dL), standard deviation (SD), and MAGE (all P<0.05), but not CGM AUC (>250mg/dL) or mean daily glucose, showed significant improvements with sotagliflozin 200mg versus placebo. Treatment with sotagliflozin 400mg vs placebo resulted in significant improvements in CGM AUC (both >180mg/dL and >250mg/dL), mean daily glucose, SD, and MAGE (all P≤0.002).
Conclusion:
Sotagliflozin, combined with insulin, significantly increased time spent in target range (70–180mg/dL) and reduced PPG and glycemic variability. These outcomes suggest efficacy beyond A1C reduction, without increasing time spent in hypoglycemia.
Fast and Reliable Insulin Identification in Insulin Formulations and Ultrafiltrates by MALDI TOF Mass Spectrometry
South-Westphalia University of Applied Sciences, Interdisciplinary Centre for Life Sciences Iserlohn, Germany
Objective:
Concerning the composition and quality of different insulins (human and insulin analogs), precise analytical methods are essential for satisfying international quality regulations for commercial formulations available at pharmacies. We developed a multi-analytical procedure for fast and reliable insulin identification by Matrix Assisted Laser Desorption Ionization-time of flight (MALDI TOF) mass spectrometry, quantification by HPLC with UV detection, and monomer/dimer/hexamer characterization by gel electrophoresis.
Method:
A total of 13 insulins from different manufacturers and the reference United States Pharmacopoeia (USP) standard insulin were analyzed using MALDI TOF mass spectrometry. For quantification, a reversed phase C18 HPLC column, with isocratic and gradient methods, was used. All insulins were stored under recommended conditions at 4°C in a refrigerator. Ultrafiltrates were collected using centrifugal filters with 3 kDa molecular cut-off; subsequently, pure insulin solutions and original formulations were characterized applying a sinapinic acid / acetonitrile TFA matrix. Time-of-Flight mass spectra were recorded using an enhanced ultrafleXtreme mass spectrometer from Bruker.
Result:
All insulin specimens were identified by comparing their mass spectra with those of the corresponding insulin class. Differentiation between human insulin and different analogs (lispro, glulisine, aspart, detemir and glargine) was realized by a classification of the analogs due to their differences in the peptide chains. Quantitative results have been achieved using a HPLC system with UV detection at 214 nm wavelength and calibration.
Conclusion:
A first draft of a multi-analytical insulin quality assay has been developed by using different methods for the characterization of insulins. Validation of the discussed analytical schemes has been realized by using ultrafiltrates of the distributed insulin formulations. The suggested methods have the potential of developing an efficient and fast quality control for reliable insulin identification and quantification.
Quality Control and Determination of Molecular Stability for Different Insulins using FTIR-ATR Spectroscopy
South-Westphalia University of Applied Sciences, Interdisciplinary Centre for Life Sciences Iserlohn, Germany
Objective:
For the composition analysis of insulin formulations and monitoring of long term activity of human and insulin analogs, analytical methods are required to determine the molecular stability of formulated insulin specimens from pharmacies, especially after long-term storage under different environmental conditions. Infrared spectroscopy has been successfully utilized for secondary structure analysis in cases of protein misfolding and fibril forming.
Method:
Thirteen different insulins were randomly purchased including the United States Pharmacopoeia human insulin standard (USP) from Sigma-Aldrich. Selected insulin samples were stored in climatic exposure test cabinets for different storage times. Using Fourier transform infrared (FTIR) attenuated total reflectance (ATR) spectroscopy of dry-film preparations, a systematic analysis of insulin spectra was performed with samples from original formulations and solutions of insulins purified by ultrafiltration, stored at 0 °C, 20 °C and 37 °C, respectively. Weekly ATR-measurements allowed the monitoring of secondary structure changes in the proteins, supposedly correlated with the insulin activity.
Result:
ATR spectroscopy provides the identification of different subclasses of insulins, which have been carried out by means of a spectral database and similarity testing. Secondary structural changes have been detected from band shape changes and peak shifts of Amide I- and II-bands and standardized curve fitting. There are individual spectral characteristics influenced by the sequence of amino acid side chains of the different insulin analogs by storage temperature and after purification of the insulins by ultrafiltration compared to the original formulations.
Conclusion:
Changes in the secondary structure of the proteins lead to the assumption of an inactivation process of insulin activity influenced by temperature. For further investigations concerning the quantification of intact insulins, an internal standard method for dry film ATR-spectra and an enzymatic essay will be applied.
Inaccurate Performance of a Specific Blood Glucose Meter in Recent Randomized Clinical Trials
Pfützner Science & Health Institute, Mainz, and Technical University Bingen, Germany
Objective:
In recent randomized clinical trials, an unexpected high number of severe hypoglycemic events with clinically serious symptoms was reported. A potential reason was suboptimal performance of the blood glucose (BG) meters that systematically led to high BG values, resulting in inaccurate treatment decisions.
Method:
In all affected trials, the same glucose measurement system was employed. A total of 120 devices and 18 strip lots used by patients across the trials were randomly selected and subjected to a standard repeatability procedure to sample 10 devices and 6 strip lots to be used for laboratory tests (repeatability and linearity) and in a clinical study. In all experiments YSI Stat2300plus was used as the reference.
Result:
The repeatability test performed with all combinations of strip lots and devices met general regulatory acceptance criteria (CV range: 3.4-8.4%, mean CV: 5.7%). In the laboratory linearity experiment, there was a bias towards high BG values (MARD: 38.5%) and only 14% of the measurements <100 mg/dL were within ±15 mg/dL. In addition, a clinical system accuracy evaluation in accordance with ISO15197:2015 (with additional 20 patients in the hypoglycemic range) was performed. At the time of abstract preparation, 86 clinical datasets were analyzed. In all, 178/1032 readings (17.2%) were outside the ISO acceptance criteria and 291/1032 (28.2%) readings were outside the FDA acceptance criteria (<15%). For glucose values < 100 mg/dL, there was a clear bias towards elevated values (MARD: 16.5%, range: -25-+61%).
Conclusion:
The results show that this specific BG meter, although approved according to standard regulatory guidelines, did not meet the level of analytical accuracy required for clinical treatment decisions. In general, caution should be exercised before selection of BG meters for patients and in clinical trials.
Impact of Fat Content on Postprandial Glucose Excursions While in a Hybrid Closed-Loop System
Department of Pediatrics, Endocrinology and Diabetes Division, Stanford University School of Medicine Stanford, CA, USA
Objective:
Currently, with most hybrid closed-loop insulin delivery systems, the meal bolus is based solely on the carbohydrate content. We compared the glycemic response to meals with both low and high fat content for 8 hours following dinners with subjects using a hybrid closed-loop (HCL) system.
Method:
Forty-three subjects aged 6-49 years old (37% female) completed supervised outpatient clinical trials of a HCL system. The carbohydrate (CHO), protein and fat content of each meal was recorded by the medical staff, using food labels, and information from chain restaurants. We divided 36 dinners (31 participants) into two categories based on their fat intake. We considered a meal low fat (LF) if it had < 16 grams of fat, n=19 and high fat (HF) if it had >30 grams of fat, n=17. We calculated the area under the curve (AUC) above 140 mg/dl for 8 hours following the start of the meals following a standard premeal insulin bolus.
Result:
There was a significant difference in AUC above 140 mg/dl between high fat and low-fat meals [HF = 7,036 ± 17,064 mg/min/dL and LF = -4478 ± 13554 mg/min/dL (p=0.03)]. There was a significant increase in glucose between 120 and 240 minutes following a HF comparing to a LF dinner (p value< 0.0001).
Conclusion:
Meals with high fat content can significantly increase the CGM values 2-4 hours following meal onset and the effect of delayed gastric emptying and insulin resistance can persist for hours even with closed-loop control. Further studies to determine the optimal insulin delivery based on nutritional inputs may improve overall glycemic control.
Assessment of Subject-Specific Insulin Sensitivity (SI) Daily Pattern in Adolescents and Young Adults with Type 1 Diabetes during Open-loop Treatment
Yale University New Haven, CT, USA
Background:
The oral glucose minimal model-derived insulin sensitivity (SI) is an important measure of insulin action. A new method to reliably estimate SI based on continuous glucose monitor (CGM) and insulin pump (CSII) data, the SISP, has been validated in adults with type 1 diabetes (T1D) and increased SISP is associated with higher SI. We assessed daily variability of SISP in adolescents and young adults with T1D.
Method:
CGM and CSII data were collected over 6 weeks in 4 adolescent/young adult subjects with T1D under free-living conditions. SISP was calculated from meals with appropriate pre-meal insulin bolus, without a correction insulin dose, and meals were separated by a minimum of 3-hour fasting periods. To examine the effect of time of day on SISP, each 24 hour period was divided into three separate intervals: morning (6a-11a), reflecting breakfast; mid-day (11a-6p), reflecting lunch; and night (6p-6a), reflecting dinner and late night snack; and analyzed using mixed effects regression model estimates.
Result:
Among four subjects (2F, age: 20.4±3.0y, A1c: 7.6±0.9%, BMI:22.2±5.2kg/m2, mean total daily insulin dose: 0.68±0.27U/kg/d) with 151 qualified meals (41:70:40 for morning, mid-day and night-time, respectively), SISP was 14.6±6.4 higher during mid-day than morning (p=0.024), while there was no significant difference between morning and night (5.4±6.6, p=0.412).
Conclusion:
Our preliminary findings suggest that SISP-determined insulin sensitivity is lower during the morning as compared to midday in young adults, suggesting a higher insulin requirement at breakfast as compared to lunch. Integration of SISP into individual insulin dose adjustment decision making may help to optimize glycemic control during CGM-augmented CSII treatment.
Exploitation of Physical Activity Data to Better Predict Glucose in Type 1 Diabetes
Department of Information Engineering, University of Padova Padova, Italy
Objective:
To investigate the benefit of adding physical activity data, collected by an off-the-shelf wearable device, to other physiologic signals [i.e. injected insulin, carbohydrate intake, continuous glucose monitoring (CGM) records] in an algorithm for prediction of future values of blood glucose (BG) for type 1 diabetes (T1D).
Method:
Experimental data were collected by six T1D subjects for five days during a clinical trial, where they wore a CGM sensor, an insulin pump and an activity tracker. We derived individualized linear predictors based on black-box models, using the prediction error method (PEM), the state-of-the-art identification technique, and the most general model parametrization (Box-Jenkins). First, we identified models using only meal and insulin information, and then using also the exercise information. The prediction accuracy of the two models, with or without physical activity, was compared with coefficient of determination (COD) and root mean squared error (RMSE).
Result:
The two models have similar performance for a short prediction horizon (PH) (up to 45 min): nevertheless, the inclusion of physical activity affects and improves the accuracy of longer PH. Indeed, the presence of physical activity improves the 3-hr prediction COD by mean ± standard deviation of 18.5% ± 30.1 % and the 3-hr prediction RMSE by mean ± standard deviation of 99.1 mg/dL ± 119.8 mg/dL.
Conclusion:
This preliminary investigation shows that the consideration of additional physiological information, like the physical exercise, improves the performance of models, leading to better predictions for longer PH.
Predicting Blood Glucose in T1D Patients Using Recursive Neural Networks
Canyon Crest Academy San Diego, CA, USA
Objective:
This work studies the application of recurrent-neural-network (RNN) based methods to predict blood glucose (BG) levels from continuous glucose monitoring (CGM) for type-1 diabetes (T1D) patients. The proposed solution applies a 30-minute sliding window consisting of 5-minute samples with an RNN based on long-short-term memory (LSTM) to predict the future reading over 5 to 30 minutes. The study also evaluates transferability of the model across patients and its effectiveness compared to conventional approaches.
Method:
The RNN network is based on LSTM designed for time-series prediction. Both training and testing data were from DirecNet, a public dataset with CGM data recorded every 5 minutes over a period up to 48 hours for each of the 110 T1D patients between the ages of 3 and 17 years. The hyperparameters and window size at the network input was determined empirically by considering statistics of patient data including autocorrelation. The input data were normalized and pre-processed properly a-priori. Traditional baselines including logistic regression and Autoregressive Integrated Moving Average (ARIMA) were evaluated for comparison. We also evaluated the performance of the RNN model over a portion of data set with rapid variation in BG values, where classical methods have limitations. Transferability of the model across patients was investigated.
Result:
With the network being trained over 100 epochs per patient and run over 10 trials, the resulting mean absolute relative difference (MARD) score for LSTM is 10.86, while the MARD scores for linear regression and ARIMA are 9.66 and 35.14, respectively.
Conclusion:
The proposed model and associated evaluation demonstrate early potential of RNN model using a large public patient data set with significant data variation. More work is on the horizon including further optimization and application to additional data sets.
Artificial Pancreas
OpenAPS not for profit corporation Pleasanton, CA, USA
Objective:
The objective is the development of a closed loop system with cloud data sharing utilizing continuous glucose monitoring (CGM) technology
Method:
OpenAps has created a closed loop system that currently works with all types of CGM available in the marketplace and has successfully tested the algorithm with simulated data.
Result:
Our data show that the method and algorithm is working like a charm with simulated data and is closed loop.
Conclusion:
The OpenAPS closed loop system is performing excellently in a clinical trial with simulated results. The project is an open source model in order for the larger community to test and survey the device. We are currently pairing our technology with other open source APPs and enabling the reading of the data from a MongoDB database.
Evaluation of NYU’s Pilot Course, Introduction to Biomedical Entrepreneurship
Tufts University School of Medicine Boston, MA, USA
Objective:
The main goal of the Biomedical Entrepreneurship Program is to support and educate early stage scientists (at postdoctoral, graduate, and junior faculty levels) in biomedical venture creation, specifically the commercialization of novel discoveries and inventions. The Introduction to Biomedical Entrepreneurship (IBE) course was designed to address the translational science training gap and promote the development of feasible and innovative healthcare solutions for both preventative health and disease management including diabetes.
Method:
IBE was piloted in 2017. Enrollment was open to all NYU postgraduate and graduate students. Twelve lectures were given by experts in pertinent, specialized fields. Students completed pre- and post-course surveys (knowledge self- assessment) as well as lecture ratings. Students who completed the course were also surveyed 1-year post completion.
Result:
Thirty-three students initially enrolled in the IBE; 19 (58%) completed the post course survey and 14 (42%) completed both the pre- and post-course surveys. Lectures were highly rated (3.7/4 ±0.2). Students demonstrated improved knowledge in the following areas: understanding business concepts and startup terminology (p=0.006), identifying business opportunities in the life sciences (p=0.003), evaluating business opportunities (p=0.003), understanding business models (p=0.003), understanding protection and licensing of university developed intellectual property (p=0.004), understanding the process of bringing a biomedical product to market (p=0.005), understanding legal ramifications of a new venture (p=0.003), and familiarity with career opportunities in “The Business of Science” (p=0.004). At time of writing, 14 students (54%) completed the one-year follow-up survey. Eighty-five percent of respondents reported that IBE affected the way they approached biomedical research.
Conclusion:
IBE enables researchers to better understand the business side of science, thus helping to bridge the translational science gap in diabetes and other diseases.
Sensing Solutions on the Nanoscale to Allow for Improved Functionality of Medical Devices and Diagnostic Procedures
CantiMed UG, Darmstadt, and Pfützner Science & Health Mainz, Germany
Background:
Modern medical devices can benefit from addition of sensing functions for achieving improved functionality or enhanced safety. However, existing sensors do not scale down easily, prohibiting further increase in their sensitivity and detection speed without changing device form factors.
Method:
We developed a nanomechanical sensor readout based on electron co-tunneling through a nanogranular metal. The sensors can be deposited with lateral dimensions down to tens of nm, allowing, for example, the readout of nanoscale cantilevers without constraints on their size, geometry, or material. By modifying the inter-granular tunnel-coupling strength, the sensors’ conductivity can be tuned by up to four orders of magnitude, to optimize their performance. The core sensors are 3D-printed using a modified electron-grid microscope and their sensitivity is suited even for demanding applications such as atomic force microscopy (AFM). To test the functionality of the nanosensors in a sensing application, we have equipped a small, highspeed silicon nitride (SiN) AFM cantilever with a full Wheatstone sensor bridge, a design which eliminates all potential temperature effects.
Result:
The small active sensor lengths allow for the use of a small bridge voltages of 0.1–0.5V, while still giving sufficient readout signal. In this experimental setting, a sensor movement results in an easy way to measure changes in electric resistance that is related to the magnitude of sensor movement in a linear way. Current research and development projects include development of implantable sensors for measurement of arterial and ventricular blood pressure, implantable sensors for interstitial glucose assessment and measurement of a variety of other metabolic parameters, as stand-alone devices, or as additional applications to upgrade existing and approved devices. Further applications of interest are point-of-care test platforms for protein and gene determination, for example, from capillary blood samples.
Conclusion:
The new nanosensor technology may be helpful to lift existing devices and sensor systems to a next level of sensitivity and functionality.
Non-invasive Control of Blood Glucose In Vivo Using a Photo-Activated Insulin Depot
Division of Pharmaceutical Sciences, School of Pharmacy, University of Missouri-Kansas City Kansas City, MO, USA
Objective:
Our overall objective is to allow for continuously variable insulin release, like that provided by an insulin pump, but without the physical connection of the outside and inside of the patient that is required by the pump. By doing this, we hope to avoid the myriad of problems associated with this physical connection: biofouling, crimping, and snagging. Our approach depends on using a small LED light source to trigger and meter insulin release from an injected photoactivated depot (PAD) of insulin.
Method:
We accomplished the desired light control through multiple different mechanisms. First generation materials linked insulin to an injectable but insoluble polymer that retained the PAD material at the site of injection. Second generation materials linked small, property-modulating moieties to insulin, to allow toggling of insulin solubility with light. New materials were assessed in-vitro, and, ultimately, in-vivo using a diabetic rat model.
Result:
We demonstrate that insulin release can be triggered both in-vitro and in-vivo from the injected insulin PAD using a small LED light source. We show that in the absence of light, no insulin is released. Furthermore, we demonstrate that we can control the amount of insulin released by varying the amount of light. We confirm that the released insulin is bioactive and effectively reduces blood glucose to normal levels using second generation PAD materials.
Conclusion:
We have shown that we can synthesize and characterize photoactivated depots of insulin, and that these have the properties required of non-invasive controlled insulin release. With the combined performance standards met, we have established a viable new approach for the variable controlled delivery of insulin that does not rely on a physical connection between the outside and inside of the patient.
Use of a Novel Mobile Platform, POPS! one, Improves Diabetes Management in Adolescent T1D Patients over a 6-month Trial
Children’s Minnesota Research Institute and McNeely Pediatric Diabetes Center, Children’s Minnesota St. Paul, MN, USA
Objective:
POPS! one is a mobile platform designed to simplify diabetes self-management. The platform integrates a self- contained glucose meter and test modules with an interactive mobile application. Preliminary testing validated the accuracy of blood glucose measurements and less painful lancet experience compared to traditional devices. In this study, we examine the impact of sustained use of POPS! on A1c, test frequency, and blood glucose.
Method:
Patients from an upper Midwest children’s hospital were recruited into a prospective, single-arm, 50 subject clinical trial. Enrolled subjects, all injection users, used POPS! for 6 months with routine quarterly follow-up visits. Differences between A1c, blood glucose measures, and test frequency at baseline and 6 months were compared using paired t-test and generalized mixed regression modeling procedures. Data presented here reflect the first 5 patients to complete the trial.
Result:
Of completed patients, 80% are male and 60% White/Caucasian. The mean age is 15.6±3.7 years, with a mean duration of T1D diagnosis of 6.6±2.9 years. Over 6 months, patients’ A1c improved significantly from 9.40±0.70% at baseline to 7.90±0.54% at 6 months (p-value<0.0001). In addition, blood glucose measures improved significantly from 248.2±54.9 mg/dL at baseline to 176.8±33.8 mg/dL (p-value=0.0298). Frequency of testing increased from 3.15±0.76 times/day to 3.80±1.30 times/ day, however, this change was not statistically significant.
Conclusion:
Our findings suggest that use of POPS! enhances diabetes management with observed improvements in blood glucose and A1c. Analyses of all enrolled patients are warranted before drawing definitive conclusions; final study results are expected in late 2018. Future research of the platform should include examining the impact on the percentage of blood glucose data within, below, and above the target range and also quality of life metrics.
Hormones and their Impact on Changes in Insulin Action in Relation to Menstrual Cycle for Females with Type 1 Diabetes
Yale School of Medicine New Haven, CT, USA
Objective:
We previously illustrated blunting of rapid acting insulin (RAI) peak action and reduction of overall RAI glucodynamic action during luteal phase (LP) of menstrual cycle (MC) compared to follicular phase (FP). Our current objective was to ascertain the potential effect of physiologic changes in hormone levels during different phases of MC on RAI action.
Method:
Ten non-obese female subjects with type 1 diabetes (T1D) (mean age 21.8±6.1y, A1c7.3±0.9%, T1D duration13.1±7.4yrs,) underwent two insulin action studies after a subcutaneously administered, standard 0.2u/kg/dose RAI during FP and LP of their MC in random order. Plasma estradiol, progesterone, Dehydroepiandrosterone sulphate, sex hormone binding globulin, testosterone, and androstenedione levels were measured at the beginning of each insulin action study and analyzed by linear mixed-effects model with comparisons made for pharmacodynamic (PD) and pharmacokinetic (PK) measures between FP and LP. The primary PD and PK metrics were: area under curve (AUC) for glucose infusion rate (GIR0-300min) and plasma insulin concentration (Cins) during a five-hour insulin action study.
Result: The
AUC GIR0-300min was 39% lower (p=0.001) during LP compared to FP despite similar AUC Cins 0-300min (p=0.07). The plasma progesterone level was higher during FP compared to LP (1.2ng/mL vs. 0.76 ng/mL; p=0.01). No significant differences were detected between other hormone levels during FP and LP. The Spearman correlation coefficient values of hormone levels to AUC GIR0-300min during FP and LP ranged between -0.22 to 0.6 across a range of hormones without reaching statistical significance (p=0.1-0.9).
Conclusion:
Our preliminary data analysis has not shown any significant associations between hormone levels of interest and the RAI PK/PD. Further investigation is required to determine factors mediating the impact of MC phase on RAI PK/PD in females with T1D.
Automatic Insulin Dose Monitoring: an Essential Technology for Optimizing Time in Range (TiR) for Multiple Daily Injection (MDI) Patients
Common Sensing Inc. Cambridge, MA, USA
Objective:
The A1c value is a coarse, retrospective measure for the assessment of glycemic control. For day-to-day management purposes, Time in Range (TiR) is more informative. We present a method for understanding the interaction between dosing behavior and TiR. This method allows clinicians to rationally modify treatment regimens to optimize TiR.
Method:
We used an electronic insulin dose capture device (Gocap) to record basal and bolus insulin injection data from two groups of multiple daily injection (MDI) patients wearing a continuous glucose monitor (CGM) (younger group: 18-35 years, n=32; older group: age≥65 years, n=24).
Result:
We document actual adherence rates by comparing the prescribed regimen versus the actual doses administered, as measured by Gocap.
Conclusion:
We report on prescription deviations and their impact on TiR.
Does Continuous Glucose Monitoring(CGM) Influence Illness Perception in Patients with Type 2 Diabetes Mellitus? A Pilot Study
Rush University Medical Center, Chicago, IL, USA
Objective:
Adherence to diabetes self-care recommendations is strongly influenced by patients’ illness perceptions. Although studies assessing the role of CGM in management of patients with type 2 diabetes (T2DM) have been performed, very few have addressed its influence on illness perception. This study aims to assess the influence of CGM on illness perception in patients diagnosed with T2DM, especially in non-insulin treated patients.
Method:
In this cross-sectional, pilot study, 13 patients with T2DM completed the revised Illness Perception Questionnaire (IPQ-R). CGM was used in all insulin-treated patients (n=4) and 44% (4/9) of non-insulin treated patients. Domain scores were compared between a) insulin- and non-insulin treated patients and b) patients on, and not on, CGM.
Result:
Mean age of our cohort was 55.7 ±14.2y with 69% (n=9) being females. We found that insulin-treated patients had higher mean identity (4.75 vs 2.22) and emotional representation (17.75 vs 16.56) scores while non-insulin treated patients had higher scores on personal and treatment control. Similarly, patients on CGM had higher mean identity (3.75 vs 1.8), coherence (20.9 vs 19.8), and emotional representation scores (17.9 vs 15.4) compared to those not on CGM. Similar trends were observed for non-insulin treated patients with higher identity (2.8 vs 1.8), coherence (21.5 vs 19.8), and personal control (26.3 vs 24.4) scores on CGM.
Conclusion:
Insulin administration and CGM in patients with T2DM might be associated with better illness perception. CGM might increase illness perception in non-insulin treated T2DM patients; indicating a potential role for its early initiation in the course of the disease.
MiniMed 670G: Results from a User Experience Survey
Department of Biomedical Informatics, Arizona State University Scottsdale, AZ, USA
Objective:
Recent studies regarding the MiniMed 670G hybrid closed-loop system (670G) focus on blood glucose (BG) control. There is a lack of research from the perspective of 670G users for this device. The goal of this study was to administer a user experience survey to patients with type 1 diabetes (T1D) using the 670G.
Method:
We designed a survey that contained three structured and six unstructured questions. Adult patients with T1D using the 670G were recruited from an outpatient endocrinology clinic. Demographic and diabetes treatment information was obtained from the medical record of each participant.
Result:
Fifteen participants completed the survey. Mean age was 45.2 years, mean A1c was 6.8%%, and mean duration of experience using the 670G was four months. Survey responses revealed that most (10/15) respondents expressed a high level of satisfaction with the 670 G technology. The most liked features were improved BG control (n=4) and the automode function (n=4); and the least liked features were perceived need for increased user interaction (n=4) and alarm frequency (n=3). There was a high frequency of manual mode usage, with n=9 participants using that function either often or always, n=3 rarely, and n=3 never using manual mode. In addition, there was mixed satisfaction with the auto-mode feature, with n=7 being satisfied or very satisfied, n=3 neither satisfied or dissatisfied, but n=5 reporting being either dissatisfied or very dissatisfied. Some (n=6) perceived there was better/more consistent BG control.
Conclusion:
Personal experiences with the 670G varied. Several patients were not taking advantage of the auto mode function. Further data is required on patient’s perception of this new technology and how use and confidence changes over time.
Self-Reported Behaviors Among Insulin Pump Users
Department of Biomedical Informatics, Arizona State University Scottsdale, AZ, USA
Objective:
Optimizing glycemic control in patients with type 1 diabetes mellitus (T1D) will require characterizing individual behaviors in order to devise personalized interventions.
Method:
We conducted a 4-week prospective study among patients with T1D using insulin pumps and continuous glucose monitors. A survey was completed at the beginning of the study. Subjects were asked to use an app to track techniques used for compensating for meals, alcohol intake, and exercise. Quantitative and qualitative analyses were performed on the survey data. Programs were used to compute behaviors from device data.
Result:
Twenty-two participants completed the study. Mean age was 49.3 years, A1c 7.0%, and pump duration 16.4 years. Survey responses revealed that when compensating for alcohol intake, 13 participants reported eating a snack, 10 bolused using their own insulin estimation, while 8 adjusted the basal rate. To accommodate exercise, 24 participants reported eating a snack, 19 adjusted the pump’s basal rate, and 10 removed the pump. To compensate for a meal, 26 bolused using the pumps bolus wizard, 19 bolused based on their own estimations, 12 adjusted the basal rate, and 9 bolused following advice from a different source, such as a software app
Conclusion:
There is intra- and inter-patient variability in how patients compensate for critical behaviors such as alcohol, exercise, and meals. We plan to use this information to develop personalized, data-driven behavioral-change educational interventions to target patient-specific disease adherence barriers to improve glycemic control.
Developing a Technology Enabled Model to Support Medication Taking Behaviors
Mytonomy, Inc Bethesda, Maryland, USA
Objective:
Medication taking is a critical behavior for successful diabetes clinical outcomes, however, about 25-40% of patients do not fill their primary insulin prescription. Health care providers report that less than 30% of their patients are successfully using insulin therapy. Self-reported use of insulin at 6 months is roughly 45% with 38% of patients self- reporting either missed, mistimed, or reduced doses. Digital solutions are one opportunity to improve medication taking behaviors.
Method:
A review of the literature was conducted to understand how technology solutions can support medication taking behaviors. Critical challenges identified were: patient-provider communication, knowledge deficit, health beliefs, patient/provider clinical inertia, complex medication schedules, polypharmacy, social determinants, and financial costs.
Result:
Two existing models in the literature support the development of a framework to improve medication taking behaviors. The e-Health Enhanced Chronic Care Model (CCM) incorporates informatics principles of collecting data that turns into information, knowledge, and ultimately provides the opportunity for developing wisdom. The Technology Enabled Self-Management (TES) Feedback Loop identifies four key features of a digital health solution to improve A1C in people with diabetes: 1) two-way communication, 2) use of and analysis of patient generated health data (PGHD), 3) customized education, and 4) tailored feedback. Toolkits, designed to support healthcare provider’s actions and patient behavior changes, during three phases of medication taking behaviors (initiation, implementation, and persistence) are driven by the TES engine. The proposed outcomes are increased engagement, improved experience, and empowerment.
Conclusion:
The Technology Enabled Patient Action Model for Medication Taking is a framework that can support the development of a digital health strategy to improve medication taking behaviors and clinical outcomes and it needs to be tested in clinical trials.
Patient Engagement with Digital Therapeutic Leads to Reduction of A1C and Costs in T2DM patients; Cost Savings are Correlated to both A1C Drops as well as Patients’ Starting A1C Levels
Truven Health Analytics, an IBM company Bethesda, Maryland, USA
Objective:
To estimate economic savings associated with A1C reduction in type 2 diabetes mellitus (T2DM) patients achieved by engaging with FDA-approved T2DM digital therapeutic BlueStar®.
Method:
Adults aged 40 years and older with ≥1 medical claim for T2DM between 1/1/2014-12/31/2014 were identified from the Truven Health MarketScan Commercial and Medicare Supplemental Databases. Patients were continuously enrolled with medical and pharmacy benefits from 1/1/2014-12/31/2015 and had ≥1 A1C test with results during this period. A1C test results were separated by value ranges (‘in control’ (6-6.99), ‘elevated’ (7-7.99), ‘high’ (8-8.99), ‘not controlled’ (≥9)); patients with four results in the same range and no results in another range were included in our analysis. Total healthcare costs were assessed for each patient for 12 months following the first A1C test and reassessed with a 1-point A1C reduction.
Result:
A total of 2,168 T2DM patients had 4 A1C tests within 365 days of the first test, where results fell into the same A1C range. The distribution of patient A1C ranges differed between insurance types; 70% of patients with commercial insurance were above target levels as compared to 59% of Medicare patients. Annual patient costs were lowest for ‘elevated’ patients and progressively increased across ranges. Starting A1C level was found to be equally important when accurately calculating an achievable patient cost-savings. Annual cost-savings were found to be $1,218, $738, and $4,944 for commercially-insured elevated, high level, and uncontrolled patients, respectively. For the Medicare segment, annual cost-savings were found to be $2,134, $2,940, and $232, respectively.
Conclusion:
Economic savings can result from A1C reduction; assisted by integrating an FDA-cleared digital therapeutic BlueStar® and are statistically different depending on the starting A1C range of the patient and insurance coverage type.
Technological Interventions in Patients with Type 2 Diabetes by Primary Care Physicians and Endocrinologists in the United States
Grunberger Diabetes Institute Bloomfield Hills, MI, USA
Objective:
Assess current practice trends of diabetes technology (DT) for the management of type 2 diabetes (T2D) among primary care physicians (PCPs) and endocrinologists (ENDOs) in the United States.
Method:
An online survey to assess the perspectives and use of DT (e.g., insulin delivery devices [IDD]) was conducted in a sample of PCPs and ENDOs nationwide. Results were analyzed by descriptive statistics.
Result:
Participants included 102 PCPs and 100 ENDOs. PCPs treated ≥20 T2D patients/month, ≥25% of them receiving insulin; ENDOs treated ≥80 T2D patients/month, ≥50% receiving insulin. Fifty-percent of ENDOs and 48% of PCPs prescribed a traditional or wearable tube-free patch insulin pump when patients failed to reach A1c targets with basal therapy plus ≥3 prandial injections of insulin/day. With ranks score (RS) ranging 4.04–48.48, the highest ranked deterrents for IDDs by PCPs were: device complexity (16.12), extra office time requirement (14.27), patient training and monitoring difficulty (14.24), and ranked deterrents for IDDs by ENDOs were: cost/insurance coverage (RS=20.06), patient acceptance (16.60), and device complexity (14.38) . All participants ranked requiring fewer injections (20.42), glucose data vs insulin dosing graphical representation (19.48), and objectively capturing insulin dosing data (18.16) as the most useful technologies adjunct with IDDs. Most important perceived features of an IDD were ease of use (18.71), flexible dosing (13.79), and large insulin reservoirs (12.38). Patients’ blood glucose control motivation (60%), health literacy and/or cognitive ability (49%), and previous good adherence (48%) were considered the strongest predictors for success with DT.
Conclusion:
Despite differences in perspectives and usage of DT among PCPs and ENDOs, both specialties highly utilized these interventions in patients with T2D whose glucose control has not been optimized on basal-bolus insulin therapy.
Multivariable Adaptive Artificial Pancreas Using Personalized Plasma Insulin Estimates
Chemical and Biological Engineering Department, Illinois Institute of Technology Chicago, IL, USA
Objective:
A multivariable adaptive artificial pancreas (MAAP) using personalized plasma insulin estimates is proposed to efficiently accommodate the major disturbances to the blood glucose concentration, such as meals, exercise, and stress.
Method:
An accurate adaptive glycemic model was developed by recursive subspace identification using wearable physiological measurements and estimates of unannounced meal effect and plasma insulin concentration (PIC) along with measured glucose signal to characterize the glucose concentration dynamics under various conditions such as meals, physical activity and stress. A model predictive control (MPC) system was designed using these adaptive models to effectively compute the optimal exogenous insulin infusion without any need to enter manually inputs for meal and exercise announcements. The MPC was equipped with a safety constraint derived from a personalized PIC estimator to prevent insulin stacking and a meal detection technique to detect rapid deviations from the reference trajectory of glucose concentration caused by disturbances like meals to give the required amount of insulin infusion in a correct time.
Result:
The MAAP efficacy is illustrated using a multivariable simulator developed by our research group at Illinois Institute of Technology that can take into account the effects of physical activities and meals. Twenty virtual subjects were simulated comprehensively with different meals and physical activities. No hypoglycemia was observed and there was a significant high percentage of time spent in the target range especially during exercise times (~ 71%).
Conclusion:
An MAAP cognizant of personalized plasma insulin estimates could be a reliable step towards improved glycemic control by individualizing the insulin computations and reducing the risk of post-exercise hypoglycemia in the next generation of artificial pancreas algorithms.
“Dude, where’s my insulin?” – Modelling Insulin Adsorption in Infusion Sets
Department of Mechanical Engineering, Centre for Bio-Engineering, University of Canterbury, Christchurch, Canterbury, New Zealand
Objective:
Material adsorption of insulin has been previously studied but is not widely accounted for in glycemic control in the ICU. Clinical studies of adsorption by infusion sets have varied in methodology, results, and comprehensiveness, but indicate that over the first couple of hours as much as 80% of insulin can be lost to the infusion sets. This has significant implications for the quality of glycemic control and the use of model-based protocols.
Method:
A compartment model describing free insulin and insulin bound to infusion set material surfaces was developed to conserve mass. Two parameters described insulin adsorption (K1) to a surface and its release back into free flow (K2). Laplace analysis provided time domain solutions as two decaying exponentials. The model was fit to literature data.
Result:
The model fit the experimental data for polyethylene (PE) and polyvinyl chloride (PVC) tubes well, with a max percentage difference of 8.54% between the fitted and experimental data. The model was more inaccurate for PVC tubes than PE tubes. K1 values identified were similar magnitude for both materials, K2 was a similar magnitude for PVC, but not PE. Limited data from other studies meant they could not be used to test the model more completely.
Conclusion:
Initial validation of a compartment model with literature data suggests promise for modeling insulin adsorption in infusion sets. Different materials have different model parameters, as would be expected. Future work will look to collect a comprehensive set of experimental data to further validate and refine the model and its usefulness clinically. If successful, such a model could be used to significantly improve glycemic control in the first few hours of care.
Importance of Foreign Body Reaction in Intraperitoneal Insulin Catheter Obstruction: Development and Evaluation in an In Vivo Animal Model
Department of Pharmaceutical Sciences, University of Connecticut, Storrs, CT, USA
Objective:
Continuous intraperitoneal insulin infusion (CIPII) shows many clinical advantages for type 1 diabetic patients. However, the ongoing incidence of catheter obstructions remains a barrier to patient acceptance. One cause of catheter obstruction may be the foreign body reaction (FBR). Developing an in vivo animal model to reproduce catheter obstruction will be helpful in understanding the root cause of catheter obstruction.
Method:
Catheters (PhysioLogic Devices, Inc. Intraperitoneal Catheter) were inserted into the peritoneal cavity of the rats and tunneled subcutaneously to the dorsal scapular region. The catheter diameter was the same as marketed human-use intraperitoneal catheters. The catheters were explanted after different time-points and examined histologically by hematoxylin and eosin (H&E) and immunofluorescence staining.
Result:
Catheter obstructions were observed in all rats. One of the rats had only tip blockage, while the others showed both catheter tip blockage and encapsulation at the tip end. Encapsulations were observed in the area of suturing bands. H&E staining showed that the components of the tip blockage vary for different durations of implantation. The major components were neutrophils 3 months post-implantation, while macrophages and fibroblasts became dominant 4 months post-implantation. Moreover, for the case of 4 months implantation, fibroblasts were located at the edge of blocked samples and a broad distribution of macrophages was observed.
Conclusion:
In vivo catheter obstruction models have been successfully developed using a rat model. Different phases of the FBR observed indicate the importance of neutrophils, macrophages, and fibroblasts in catheter tip obstruction. Animal studies will be continued to understand the correlation between the incidence of catheter blockage and time.
Acknowledgement:
Support was provided by a grant from the Juvenile Diabetes Research Foundation (JDRF) (2-SRA-2017-292-S-B).
Low-Cost Sensing and Connectivity for Injection Devices
PA Consulting Cambridge, UK
Objective:
Creating smart, connected, insulin injectors which measure when the patient uses the device, and the dose delivered, are an attractive proposition. Minimal change to a device that people are familiar with will maintain user compliance whilst delivering enhanced device capability such as data sharing for artificial pancreas applications. However, re- designing existing devices to include this function, and the resulting additional cost, are significant barriers. Emerging technologies in disposable, low-cost flexible electronics will allow adaptation into the existing user experience without the need to retrain users and increase the complexity of operation. We identify approaches and possible roadmaps that will enable this function to be added to existing devices at low cost.
Method:
We reviewed available and emerging wireless communications and electronics assembly technologies.
Result:
We identify novel methods by which electronic function may be integrated into pen injectors with minimum impact upon the device design, and with a roadmap to low cost. This integration included new technologies.
Conclusion:
Connected, smart devices are feasible at low cost. New electronics technologies are available to do this but development of them is required.
Strong Customer Satisfaction among Users of Mobile Diabetes Management
mySugr Inc. Encinitas, CA, USA
Objective:
The mySugr App is the most widely used mobile solution in the field of digital diabetes care reaching 1.4M patients in over 61 countries. The mySugr Bundle service introduces convenient, unlimited test strip delivery and Certified Diabetes Educator-led coaching. In this study we reviewed customer satisfaction amongst a United States-based population with type 2 diabetes (T2D).
Method:
T2D mySugr Bundle users who received at least one shipment of test strips were sent a satisfaction survey via email. One of the best established tools for exploring customer satisfaction is the Net Promoters Score (NPS), a tool well deployed in populations above 100 people, which also defined the number of participants aimed for in this study.
Result:
The survey was sent out in May 2018 and resulted in 105 replies. Of the participants, 54.29% lived with diabetes for more than 5 years and 45.71% measured their blood glucose (BG) 3-5 times a day. Of those reaching out to the CDE via smartphone, 86% were satisfied with the coaching service. Overall, the results showed 9 detractors, 80 promoters, and 13 passives resulting in a NPS score of 69.61.
Conclusion:
Customer satisfaction is often not observed or reported on in the field of digital diabetes care, making it difficult to estimate or benchmark the quality of services. The mySugr Bundle was primarily designed with people with type 1 diabetes in mind, but has shown growth and potential in T2D as well. A NPS-score of 69.61 places the service at the level of Netflix, which comes in at 68. We encourage a neutral institute to perform a study on customer satisfaction in the near future in order to eliminate self-reporting bias.
Clinically Relevant Improvement in Quality of Blood Glucose Control in Well-Controlled Users of mySugr's Mobile Diabetes Management Tool
ProSciento Inc. Chula Vista, CA, USA
Objective:
The mySugr App is the most widespread mobile health applications in the diabetes industry reaching 1.4M patients in 61 countries. The mySugr Bundle introduces unlimited test strip delivery and Certified Diabetes Educator-led coaching. In this retrospective study, we explored real world changes in blood glucose (BG) in a United States population of mySugr Bundle users.
Method:
We analyzed changes in BG control [mean ± SD, tests in range (TIR), estimated A1c (eA1c)] and frequency of testing in a random sample of users. Participants monitored BG ≥3 times/day during the observation period. Data from the first two weeks of use (t0), two months before (t1), and two months after (t2) initiation of Bundle usage were aggregated and statistically compared using two-sided t-tests.
Result:
Study participants were 52 users; 55.8% with type 1 diabetes, 36.5% with type 2 diabetes, 5.8% with LADA, and 3 with unreported diabetes type. Of these, 77.1% used insulin, 19.4% used insulin pumps, and 22.9% used non-insulin therapies. Significant (p<0.05) improvements were observed in mean BG (-16 mg/dl), TIR (+8.5%), readings above target (-8.85%) and eA1c (-0.43%) between t0 and t2. Significant improvement was also observed in monitoring frequency (+17.75%) at t2. An indicated clinically relevant change in eA1c (defined as ≥0.3% according to EMA guidelines) was achieved by 30.77% of the population.
Conclusion:
Significant changes in the frequency of BG testing and quality of glucose control in mySugr Bundle users indicate the potential benefits of mobile interventions that combine complementary treatment strategies. Given that improvements were observed in well-controlled patients (indicated by t0 eA1c), findings support further prospective studies in this field.
The ABCs Outcomes after Using a Smart Phone-Based Lifestyle Application in Subjects with Type 2 Diabetes
EDC McDonough, GA, USA
Background:
We developed a smart phone application encouraging people with type 2 diabetes to walk, measure daily steps, calculate their daily caloric intake, and measure their A1C, blood pressure and total cholesterol.
Method:
A randomized clinical trial was conducted that involved 186 type 2 diabetic patients whose glycemic status was not under good control (A1C more than 7%). The intervention enhanced group used the smart phone application for 6 months while the control group received regular care for the same time period. The intervention was 30 minutes of walking each day with an average of 4000 steps and a caloric intake average of 25 calorie/kg with a maximum of 1,800 calories per day. In the intervention group, assessments (i.e., A1C, blood pressure and total cholesterol ) were measured before the start, at three months, and at six months. The ABC results were reviewed by a diabetes educator.
Result:
The study was fully completed by 162 subjects (n=92 in the intervention group and n=70 in the control group). On follow-up examination 3 months later, A1C level was significantly decreased in the intervention group (from 8.1% to 7.3%) compared to the control group (from 8.0% to 7.8%), Blood pressure level was significantly decreased in the intervention group (from 148/88 to 140/84) compared to the control group (from 150/88 to 148/88). Also after 3 months, serum cholesterol level was significantly decreased in the intervention group (from 196 mg/dL to 164 mg/dL) compared to the control group (from 202mg/dL to 194mg/dL). At 6 months, A1C level was also significantly decreased in the intervention group (from 8.1% to 7.1%) compared to the control group (from 8.0% to 7.7%) and blood pressure level was also significantly decreased in the intervention group (from 148/88 to 134/82) compared to the control group (from 150/88 to 144/86). Also after 6 months, serum cholesterol level was significantly decreased in the intervention group (from 196 mg/dL to 144 mg/dL) compared to the control group (from 202mg/dL to 186mg/dL)
Conclusion :
Smart phone lifestyle applications could help in optimizing the ABCs outcomes in people with type 2 diabetes
Development of an Advanced Continuous Subcutaneous Insulin Infusion (CSII) Catheter with Extended Duration of Use and More Reliable Insulin Delivery
Jefferson Artificial Pancreas Center, Department of Anesthesiology, Sidney Kimmel Medical College of Thomas Jefferson University Philadelphia, PA, USA
Background:
Insertion of a Continuous Subcutaneous Insulin Infusion (CSII) cannula through the skin into the subcutaneous tissue damages cells, connective tissue, and the extracellular matrix. A layer of thrombus and acute inflammatory tissue develops around the cannula due to insertion trauma, motion induced trauma, and the pro-inflammatory effects of the foreign body and insulin excipients.
Method:
We compared the performance of CapBio investigational CSII catheters (with a soft/flexible polymer wire- reinforced kink-proof cannula with multiple orifices) to commercial CSII catheters (with a Teflon cannula with one distal orifice) in large swine. An investigational and commercial CSII were inserted into the subcutaneous tissue of the abdomen every other day for 14 days. Insulin lispro (U-5) was infused through the CSII catheters using the same basal/bolus pattern. On day 14, a 70 ul bolus of insulin/x-ray contrast agent was infused through each CSII catheter. The tissue surrounding each CSII was excised 5 minutes later, frozen, and imaged using a high-resolution micro-CT scanner. The tissue was then processed and stained to produce high-resolution histology images.
Result:
Histology images revealed a statistically significantly thinner and smaller area of inflammatory tissue around the CapBio CSII cannulas compared with the commercial CSII Teflon cannulas. Micro-CT images revealed a statistically significantly larger volume and surface area of insulin/x-ray contrast agent delivered into the subcutaneous tissue when infused through the CapBio CSII compared with the commercial CSII catheters. Statistical analysis is currently being completed.
Conclusion:
There is great clinical need for an insulin infusion set that functions reliably for more than 2-3 days with improved consistency of insulin absorption pharmacokinetics (PK).
Physical Chemical Considerations Associated with the Development of a Wearable, Disposable Closed Loop AP System
EOFlow Co LTD Seongnam-Si, Gyeonggi-Do, Korea
Background:
The development of a wearable, fully integrated, disposable closed-loop Artificial Pancreas (AP) system entails certain challenges associated with adequate compensation for the accuracy and reliability of blood glucose control due to the physical chemical challenges at work in the near vicinity of the continuous glucose monitor (CGM) sensor and insulin infusion cannula.
Method:
Using a modified dielectric body phantom model, the authors developed a set of compensation algorithms designed to work with available AP algorithms such as TypeZero, which adjust accuracy for a sensor in very close proximity (within 1 cm) of an insulin infusion cannula.
Result:
The algorithm utilizes entropic considerations by applying the dielectric Kluitenberg's nonequilibrium thermodynamic theory to varying tissue locations with differing local properties including subcutaneous fat content to modify the generic model of subcutaneous insulin absorption within the AIDA v4 diabetes simulator.
Conclusion:
These analyses have found that insulin formulation-specific adjustments are required to be incorporated into the control software to insure the target mean absolute relative difference (MARD) with the device
Continuous Subcutaneous Insulin Infusion (CSII) in Vivo: CSII Biocompatibility and Effective Insulin Delivery
Wayne State University Detroit, MI, USA, 48202
Objective:
CSII provides better blood glucose regulation (BGR) when compared to insulin injections. Unfortunately, unexpected/unexplained failures occur frequently with CSII that limit even the 3-day CSII lifespan. We hypothesize that the limited lifespan seen in CSII is the result of poor biocompatibility of CSII, i.e. CSII induced infusion site inflammation limits both immediate and long-term BGR. We believe that insulin sub-components, particular diluents (phenol and/or m-creosol), as well as insulin by-products (fibrils), trigger both acute and chronic inflammation resulting in fibrosis that leads to both immediate and long-term infusion site impairment and loss of insulin mediated BGR.
Method:
Tissue reactions to insulin, diluents, and fibrils were assessed with murine and swine models. Using a modified murine air pouch model (APM), tissue toxicity and inflammation was determined using APM lavage and fluorescence-activated cell sorting (FACS) analysis of cell numbers and cell subpopulations including histopathology evaluations at various time intervals. The impact of pre-induced inflammation in APM on insulin regulation of blood glucose was also assessed using a standard irritant (thioglycolate).
Result:
Our data demonstrate that commercial insulin, including diluent and insulin derived fibrils, trigger inflammation by day 3 of CSII. This inflammation is characterized but an influx of neutrophils and macrophages. Additional studies demonstrated that induction of inflammation at the insulin infusion site, prior to initiation of insulin infusion, prevented insulin from regulating BG in diabetic animals. Finally, we also found that insulin was rapidly taken up by neutrophils and macrophages, but not lymphocytes, and degraded into non-functional peptides.
Conclusion:
These studies demonstrate that insulin and related products are cell and tissue toxic resulting in tissue inflammation and reduced effectiveness of insulin-mediated BG regulation during CSII.
Relationship between Baseline BMI and Clinical Outcomes with V-Go® Wearable Insulin Delivery Device in Patients with Type 2 Diabetes Previously Prescribed Basal-Bolus Regimens
Endocrinology Specialists Greensburg, PA, USA
Objective:
According to the CDC, 87.5% of adult patients with diabetes are overweight or obese. Obesity is associated with higher insulin requirements, therefore, evaluating the impact of body mass index (BMI) on insulin treatment is important. V-Go a wearable insulin delivery device has demonstrated improved glycemic control with less insulin, however, use across different BMI strata and in particular obese patients has not been clarified. This evaluation aimed to investigate the relationship between baseline BMI, change in A1C, and insulin total daily dose (TDD) when switched from basal-bolus regimens to V-Go.
Method:
Electronic medical records across 9 diabetes specialized centers were evaluated. Prior to switching to V-Go, inclusion required a baseline A1C ≥ 7% and a regimen of ≥ 1 basal and ≥ 2 prandial insulin injections/day. Patients were stratified according to established BMI strata (normal-weight, overweight, obesity class I, II or III). An ANCOVA model that included BMI strata as the factor and baseline measurement as the covariate was performed to test for between group differences.
Result:
Of the 186 patients evaluated, 75% of patients were classified as obese (I, II, or III) and 22% as overweight. A higher baseline BMI was associated with both a higher baseline weight and insulin TDD. After 3 months of V-Go use, all strata benefitted from reductions in A1C and TDD. Comparing between strata (normal-weight, overweight, obesity class I, II and III), mean A1C changes adjusted for mean baseline were -0.8, -1.0, -1.1, -1.0, -0.7%, respectively (p=0.648) and mean TDD changes were -36, -30, -26, -27, -22 units/day, respectively (p=0.149).
Conclusion:
V-Go improved A1C and reduced insulin dose similarly across all BMI strata with significant reductions compared to baseline in overweight and obese strata.
GIR-based categorization of glucose clamps
Profil Neuss, Germany
Objective:
The glucose clamp technique is the gold standard to describe pharmacokinetic and pharmacodynamic parameters of blood glucose lowering agents. However, established glucose clamp quality parameters characterize the clamp quality solely based on blood glucose concentration (BG) measurements. We focused on the quality of glucose clamp outcome parameters other than BG measurements when investigating the impact of downtimes in automated glucose clamps.
Method:
Numerical simulations were performed to calculate the glucose infusion rates (GIR) needed to keep BG levels close to a pre-defined target considering the BG lowering effects of different insulins. To induce standardized downtimes minute-to-minute, GIR adaptation was stopped at different times for variable durations. By analyzing potential critical downtimes leading to deviations between BG and pre-defined targets, and therefore, to statistical changed clamp outcome parameters, the clamp-downtime-rating-curve (CDR-curve) was determined. The CDR-curve defines tolerable downtimes during automated glucose clamps.
Result:
The CDR-curve was applied to the technical downtimes that occurred in an automated glucose clamp study with 26 subjects in a cross-over design (37,492 minutes of automated clamp data). Three-hundred sixty-seven technical downtimes were found with an overall downtime-duration of 1,965 minutes and a duration of downtimes between 1 and 32 minutes. Six of the 367 downtimes had durations of 7 minutes up to 20 minutes (clamp-utilities from 76.2% to 93.7%) and were classified as being critical by means of the CDR-curve.
Conclusion:
The CDR-curve applied to technical downtimes during automated glucose clamps determines whether or not glucose clamp outcome parameters can be influenced due to a deficit in automation. As CDR-curve is based on GIR-values it is a better predictor of clamp quality than clamp utility.
HAART-induced Hyperglycemic Hyperosmolar Syndrome (A Rare manifestation of HAART therapy)
Saint Joseph University Hospital Paterson, NJ, USA
Objective:
The objective is to increase the awareness of hyperglycemic complications secondary to HAART therapy.
Method:
A Case Study is presented involving a 59-year-old Hispanic male, with a past medical history of pre-diabetes and recently diagnosed with HIV, who came to Saint Joseph emergency department with the complaint of fatigue for 3 days associated with some nausea over the same time. He denied having any other complaints. Patients’ blood sugar was 1900 mg/dL, with serum bicarbonate level: 22, blood pH: 7.38, serum acetone negative, urine ketones negative, serum osmolarity was 340 mmol/kg, and the A1c was 19%.
Result:
Based on the above laboratory values, a diagnosis of hyperglycemic hyperosmolar syndrome (HHS) was made. Patient was admitted to the Medical ICU and was started with intravenous (IV) fluids and IV insulin infusion. After normalization of blood sugar, the patient was transferred to the medical floor, was provided with diabetic education, and was also started with subcutaneous insulin injections. On review of past history, patient had been diagnosed with HIV 6 weeks ago and was started with highly active antiretroviral therapy (HAART) therapy containing a protease inhibitor (ritonavir) as one of the HIV drugs in combination. Blood work performed 8 weeks ago showed an A1C of 6.2 % and a serum blood sugar of 115 mg/dL. With a temporal relation to the of start of HAART therapy together with the sudden increase in blood sugar and increase A1c, the hospital’s infectious disease team was consulted and the patient’s HIV regimen was changed and the protease inhibitor ( Ritonavir) was stopped. The patient was discharged with new HAART therapy and with also subcutaneous insulin. We followed the patient closely in the outpatient department over the course of 3 months. The A1C and blood sugar came back to baseline, his insulin was stopped, and the patient is being managed without any diabetic medications.
Conclusion.
First, although protease inhibitors are commonly associated with the metabolic syndrome, this is the first case to the best of our knowledge in which protease inhibitors have been linked to a pure case of hyperglycemic hyperosmolar syndrome. Second, patients started with HAART therapy should be further investigated for possible hyperglycemia secondary to protease inhibitors.
Direct Electron Transfer Type Open Circuit Potential-Based Continuous Glucose Monitoring System
Joint Department of Biomedical Engineering, University of North Carolina at Chapel Hill and North Carolina State University Chapel Hill, N, USA
Objective:
The authors’ group has been engaged in the development of an electrochemical glucose sensing system based on the principle, a direct electron transfer (DET) principle, employing an innovative engineered glucose dehydrogenase enzyme. The principle is also referred to as the 3rd Generation and is the ideal principle for a continuous glucose monitoring (CGM) system. We previously reported on an amperometric measurement method using the 3rd generation principle. Here, we report on a new electrochemical method using the 3rd generation principle, the open circuit potential based sensing system, for use as a long term and stable CGM system.
Method:
The engineered glucose dehydrogenase was immobilized on gold electrodes using a self-assembled monolayer (SAM) and its open circuit potential (OCP) change was monitored in samples containing various concentrations of glucose against a silver/silver chloride (AG/AgCl) reference electrode.
Result:
The sensor showed high reproducibility of measurement after the sensor preparation was achieved. The sensor signal was not affected by the presence of either 380 µM ascorbic acid or 2.6 mM of acetaminophen in the sample solution - which are 2 times higher than the FDA-recommended concentrations to be investigated for their impact on glucose measurements. The third-generation OCP-based glucose sensor could be continuously operated for more than 9 days without significant change in the signal, sensitivity, or dynamic range.
Conclusion:
The combination of 3rd generation principle and OCP measurement produced highly reproducible, interference-free, and stable CGM. This third-generation OCP-based glucose sensor will be the innovative platform technology for future enzyme-based long term CGM system.
Non-Invasive Glucose Measurement in Human Skin Using Machine Learning and Mid-IR Spectroscopy
Princeton University Princeton, New Jersey, USA
Objective:
We have developed a mobile glucose sensor using mid-infrared spectroscopy to detect and accurately predict blood sugar concentrations within individuals using a quantum cascade laser. Glucose has strong absorption features in the mid-infrared range which differentiate it from other saccharides. This work has significant potential to provide a non-invasive, in vivo, glucose sensor that would benefit millions by providing a painless alternative to current invasive measurements.
Method:
A quantum cascade laser (QCL) emitted mid-infrared light onto a 2x2mm section of skin and the light interacted with the glucose present in the interstitial fluid within the dermis. Partially backscattered light was collected by an integrating sphere and measured by a mercury cadmium telluride (MCT) detector. This dataset was combined with readings from temperature and pressure sensors and given to a trained machine learning algorithm that produced a glucose measurement.
Result:
Prior testing indicates that our system predicts glucose concentrations with an accuracy of 83% on a human subject using principal component analysis. Since then, we have developed a smaller system with a higher power laser and a larger area detector. Our new data processing methods also incorporate multivariable regression models and neural networks. Lastly, we have added pressure and temperature sensors in addition to a finger stabilization clamp. These new additions provide a more comprehensive data set, noise reduction, and improved accuracy.
Conclusion:
We are currently conducting a usability study to gather more data to train the machine learning algorithm and improve the system’s accuracy. We believe the more complex data processing method will produce greater accuracy and consistency. We are further miniaturizing the system to allow for continuous glucose sensing in order to allow diabetics to better understand how external factors impact their glucose levels.
Vascular Endothelial Growth Factor (VEGF) Sensor as a Versatile Tool for Early Diagnosis of Diabetic Kidney Disease
Joint Department of Biomedical Engineering, University of North Carolina at Chapel Hill and North Carolina State University, Chapel Hill, NC27599, USA
Objective:
Diabetes is the leading cause of chronic kidney disease. To prevent diabetic kidney disease (DKD), the development of early diagnosis principles and devices for DKD is necessary. Recently, several reports have suggested that the circulating levels of vascular endothelial growth factor (VEGF) could be utilized for a novel biomarker for DKD. The authors’ group has been engaging in the development of novel biomolecular recognition elements (MREs) for VEGF sensing, especially the development of aptamers that specifically recognize VEGF. In this paper, we introduce our recent progress in the development of highly sensitive and selective aptamers to VEGF and their potential application for VEGF sensing.
Method:
The VEGF aptamers, those single stranded oligonucleotides that specifically recognize and bind to VEGF, were first identified and screened using an in silico maturation system. After this selection procedure, the engineered aptamers were designed based on structural modeling and also by further use of a multimerization strategy.
Result:
The results of this in silico maturation screening process resulted in VEGF aptamers with the dissociation constant (Kd) of about several hundred molar (M) units. Further improvement of VEGF aptamer by multimerization resulted in the creation of aptamers with more than 10 times lower Kd value. Using this developed VEGF aptamer, electrochemical and optical VEGF sensing systems were developed.
Conclusion:
The VEGF aptamers were developed by the combination of an in silco maturation method and multimerization strategy, which can be used for the electrochemical and optical VERGF sensing systems. This system can be used for a future early diagnostic for DKD.
Long Term In Vitro CGM characterization with controlled glycemic profile
Biorasis, Inc. Mansfield, CT, USA
Objective:
The objective of this work was to develop an extended bench-testing unit to interrogate the Biorasis’ miniaturized continuous glucose monitoring (CGM) system (GlucowizzardTM) and assess its long-term reliability. The system was designed to vary the concentration of glucose in an up and down fashion, from a hypoglycemic range (54 mg/dL) to a hyperglycemic range (180 mg/dL) and repeat the cycle every 90 minutes over the duration of multiple hours and eventually days and months.
Method:
Two peristaltic pumps with variable flow rates were driven by stepper motors controlled using an Arduino board. Pump #1 controlled the flow of a low glycemic content solution (54 mg/dl) and pump #2 controlled the flow of a high glycemic content solution (180 mg/dl). By varying the ratio between the two pumps we could maintain a constant flow rate while at the same time altering the concentration of glucose anywhere between the two concentrations. The output of the two pumps were feed into a mixing chamber and Biorasis’ CGM was placed downstream. The GlucowizzardTM output frequency (which is directly proportional to the glucose concentration) was measured by an Arduino board and values were exported to a Matlab program in order to save and analyze the data.
Result:
The sensor was operated based on the following pump routine: 1. sensor equilibrates at low glucose concentration (10 min.); 2. glucose concentration is ramped up from low to high (45 min.) by varying pump speed while the overall flow rate remains the same; 3. glucose concentration equilibrates at the high value (10 min.); 4. pump speed ramps down in a similar fashion; and 5. repeat steps 1-4 again. The GlucowizzardTM sensor response closely followed the glycemic up and down routine of the pump system. Data were collected with a 4-minute interrogation interval and showed good reproducibility over extended periods of time. Delay in sensor response can be readily extracted from the aforementioned routine.
Conclusion:
An extended bench-testing unit was developed for continuous interrogation of various CGM platforms. This system can assess long-term sensor reliability as well as extract a number of key sensor characteristics (i.e. sensor delay, sensitivity, and linearity drifts, etc.) over an extended period of time. The system can readily change glucose concentration profiles and the rate in which glucose values change to closely mimic typical in vivo glycemic excursions.
Diabetes Care and Management Using Electronic Health Records: A Systematic Review
University of Massachusetts, Boston Boston, MA, USA
Objective:
Diabetes treatment and management provides a unique opportunity for examination of the effectiveness of electronic health records (EHRs) on patient health outcomes, continuity of care, and areas for further development. This systematic literature review was designed to identify the strengths and limitations in EHR and opportunities for improvement proposed in original research and recent rigorous systematic reviews.
Method:
This review utilized methodology adapted from Preferred Reporting Items for Systematic Reviews and Meta- Analyses (PRISMA). Inclusion criteria for original research were: published between March 2003 and November 2017; included randomized controlled trial design with participants >/= 18 years of age with diabetes diagnosis >/= 1 year; and measured outcomes included A1c, blood pressure, and LDL-cholesterol levels. Criteria for systematic reviews included research focused on EHR outcomes, improvement of care for diabetic patients, prevention of adverse outcomes, web-based communication, and limitations of EHR regarding chronic disease management.
Result:
Thirteen articles qualified for inclusion. Meta-synthesis of articles indicates that the operationalization of EHRs in individual clinical practices demonstrate considerable variability. Factors influencing uptake of EHRs include office culture, attitudes toward adoption, and costs of hardware and implementation. Results suggest that chronic disease patients benefit most by decision support tools that alert physicians of drug interactions, communication tools that keep them informed and engaged in their treatment regimens, and detailed reporting and tracking designed to inform progress.
Conclusion:
Collective results suggest that EHR technology is advancing rapidly; however, patient outcomes documented via EHR systems remain largely unknown. A fertile area for inquiry designed to enhance patient outcomes in diabetes and chronic disease management is determining how EHR systems can be utilized for new drug and treatment options in addition to enhancing the quality, cost-effectiveness, and continuity of care.
The Development of Engineered Direct Electron Transfer Type Glucose Dehydrogenase with Miniaturized Electron Transfer Subunit
Joint Department of Biomedical Engineering, The University of North Carolina at Chapel Hill and North Carolina State University Chapel Hill, NC, USA
Objective:
The flavin adenine dinucleotide dependent glucose dehydrogenase (FADGDH) complex derived from bacteria consists of three subunits: 1) a catalytic subunit; 2) a small subunit; and 3) a three-heme c electron transfer subunit. The presence of the electron transfer subunit allows direct electron transfer (DET) to electrodes which is exploited in third-generation-type glucose sensing systems. In this paper we report on the engineering of the electron transfer subunit by creating a miniaturized subunit and its application of low potential operating continuous glucose monitoring (CGM) sensing system.
Method:
The electron transfer subunit was engineered by truncation of two of its three heme binding domains to construct a “miniaturized electron transfer subunit”. The designed and constructed protein was recombinantly produced to form an engineered DET-type FADGDH complex. The enzyme electrode was constructed using the engineered FADGDH complex by immobilization via a self-assembled monolayer on gold electrodes.
Result:
The engineered FADGDH complex had a lower relevant redox potential and was capable of DET. The constructed glucose sensor was successfully operated at a voltage potential as low as 0 mV versus a silver/silver chloride (Ag/AgCl) reference. At this extremely low operating potential, no bias due to ascorbic acid or acetaminophen was observed on the response current.
Conclusion:
The engineered DET-type FADGDH complex was constructed by truncating two heme c binding regions to construct a miniaturized electron transfer subunit. Consequently, a glucose sensor was developed with an extremely low operating potential which did not show interferences in the presence of ascorbic acid or acetaminophen.
Closed-Loop Composite Microneedle Patch for the Prevention of Hypoglycemia
Leslie L. Dan Faculty of Pharmacy, University of Toronto Toronto, Ontario, Canada
Objective:
Insulin-dependent diabetic patients require multiple daily injections of exogenous insulin to combat hyperglycemia and maintain glycemic targets. However, hypoglycemia, a life-threatening condition, often occurs due to inaccurate estimations of insulin dose, missed meal intake, and lack of glucose counter-regulation. This work is thus aimed to design a closed-loop composite microneedle (MN) patch to deliver glucagon automatically at low glucose levels for the prevention of hypoglycemia.
Method:
The composite system was composed of photo-crosslinked hyaluronic acid (mHA)-based MNs embedded with glucose-responsive microgels. The microgels were synthesized via dispersion polymerization of three functional comonomers for glucagon stabilization, glucose-sensing, and for enhanced glucagon loading. Glucose-induced microgel shrinking was measured by a particle sizer. Composite MN arrays were characterized using fluorescence microscopy and compression testing. Stability of released glucagon was examined by HPLC and far-UV circular dichroism (CD) spectropolarimetry, and its bioactivity was evaluated using streptozotocin (STZ)-induced type 1 diabetic rats. In vivo glucose-responsiveness and hypoglycemia prevention studies were carried out in the same type 1 diabetic rat model.
Result:
Glucose-responsive microgels were successfully prepared with a relatively uniform size. The microgels changed swelling degree in response to glucose levels at physiological pH and released loaded glucagon at hypoglycemia. The glucagon released from the microgels and MN patch was unchanged and bioactive. The MN patch provided adequate amounts of glucagon in the type 1 diabetic rats which successfully prevented hypoglycemia after an insulin overdose injection.
Conclusion:
A prototype of composite MN patch embedded with novel glucagon-loaded glucose-responsive microgels has been successfully developed for the prevention of hypoglycemia in a type 1 diabetic rat model. This system holds great potential for improving the quality of life of diabetic patients under insulin therapy.
High Usability and Satisfaction Reported by Diverse Patients with Diabetes Engaged in a Microlearning Video-based Cloud Platform to Increase Access to Insulin Education
Medstar Diabetes Institute Washington, DC, USA
Objective:
Type 2 diabetes (T2D) requires ongoing self-management. Barriers to in-person education prevent patients from learning self–care. Improved 24/7 access to education when patients are ready to learn is needed. This study assessed the usability of and satisfaction with Mytonomy’s cloud-based, microlearning video self-management education platform in a diverse T2D population newly prescribed insulin.
Method:
A playlist of 56 tailored microlearning videos (30-180 seconds each) addressed educational needs of individuals new to insulin, stigma of diabetes, and peer-to-peer support. Participants watched 4 self-selected videos, then answered 5 questions to assess usability and satisfaction with the program. Open access was then offered. One-week later participants answered 4 additional questions. Frequencies (percentages) were calculated for survey responses 5 (agree) or 6 (strongly agree) on a 6-point Likert scale. A non-parametric test (Wilcoxon rank-sum) compared responses of study completers to non-completers.
Result:
Participants (n=180) were 39% African American, 58% male, mean (SD) age 58 (13) yrs. After watching 4 or more videos, 93% answered agree/strongly agree for the 3 usability questions and 92% answered agree/strongly agree for the 2 satisfaction questions. Completers scored significantly higher than non-completers for all 5 interim usability and satisfaction questions (p = 0.002, p = 0.004, p = 0.002, p = 0.026, p = 0.002). Participants watched on average 24 videos. At study completion, 90% and 86 % answered agree/strongly agree for the final usability and satisfaction questions, respectively.
Conclusion:
Mytonomy’s cloud-based microlearning education platform was successfully used by diverse adults with T2D newly initiated on insulin. Completers scored significantly higher on usability and satisfaction with the technology. An innovative microlearning self-management education strategy shows promise for promoting access to learning.
Detecting Insulin Pump Failures without the Need of a Patient Model
University of Padova, Department of Information Engineering Padova, Italy
Objective:
The efficacy of closed-loop (CL) systems depends on the reliability of insulin pumps but they are subject to failures that can be critical for patient safety. Most literature methods for automated failure detection require the identification of a model of the patient’s physiology, which has proven to be a complicated procedure. Supervised machine learning algorithms are an alternative but they require labeled data that is often unavailable in practice. Here we present the in-silico validation of a new, unsupervised, model-free approach for the automated detection of insulin pump failures relying on modern data-driven techniques for anomaly detection (AD).
Method:
Using n=100 virtual subjects of the Padova/UVA type 1 diabetes (T1D) simulator, we tested the algorithms using a 30-day dataset. In each experiment, we simulated one insulin pump failure per patient. From the data, we extracted a feature set designed to account for the dynamics in T1D physiology and to highlight the anomalous behaviors associated with pump failures. Using this feature set, we applied a traditional control-chart (CC) method and two advanced anomaly detection methods, i.e., Local Outlier Factor (LOF) and Connectivity-Based Outlier Factor (COF).
Result:
We observed the best performance with LOF showing 64% sensitivity and only 8 false positives among all the simulated subjects, which amounts to a precision of 89%. Our method outperformed the traditional CC approach, which showed similar precision (91%), but considerably lower sensitivity (43%). COF exhibited inferior performance compared to LOF (53% sensitivity and 90% precision) although still outperforming CC.
Conclusion:
We propose an innovative approach for improving the reliability of CL systems that does not require the complex sub-task of identifying a model of the patient’s physiology.
Cellular-Enabled, Near, Real-time Blood Glucose Monitoring Supports Virtual Telemedicine Clinic in Delivery of Successful Care Management for Adults with Uncontrolled Type 2 Diabetes
Objective
Diabetes care management (DCM) interventions, including those that are technology-enabled, usually have a modest impact on glycemic control. We have assessed the impact of a DCM “Boot Camp” where concurrent use of telemedicine and a novel cellular enabled glucometer were central features. We report on findings relative to the cellular-enabled glucometer.
Method
A 3-month intervention was offered to adults with uncontrolled type 2 diabetes (A1C
Result
A total of 732 patients were included (n=366 participants and n=366 propensity matched controls). Mean age was 56.1; 63% female; 79% Black; and 42% had Medicaid. On average, 1.36 BGs were uploaded daily. Providers accessed the Telcare dashboard an average of 3.9 times/patient/ month. A1C improved from 11.2% to 8.1% for cases and 11.3 to 9.9% for controls (mean difference of 1.8; p<0.001). Cases experienced a 77% drop in risk for all- cause hospitalizations (p<0.001). There was no association between total number of BG tests and A1C change. The rate of BG<40mg/dL was 0.21% for all BG values.
Conclusion
Integration of concurrent use of telemedicine and a real-time BG monitoring system in a DCM intervention was associated with significant improvement in glycemic control and decreased utilization. These results were obtained using on average just 1.36 BGs daily (obtained at different times of day) and weekly telehealth visits.
The degree “Master” was assumed to be equivalent to ‘MS’ Evaluation of High Accurate BGMS based on an Innovative Optical Transmission Absorbance System
Terumo Corporation Nakakoma-gun, Yamanashi, Japan
Objective:
Because the blood glucose monitoring system (BGMS) is essential for diabetes care and management, reasonable prices for test strips and further improvement of the accuracy of BGMS are mandatory to achieve better glycemic control of diabetes. In last year’s Diabetes Technology Meeting, we proposed the principle, and the evaluation results, of a BGMS based on an innovative optical transmission absorbance system that combined a new enzyme (GDH(FAD)) and an original absorption dye that is capable of accurate hematocrit measurements using multiple wavelengths. In this paper, we report the evaluation results for accuracy, and the hematocrit effect, of the newly developed compact BGMS.
Method:
To realize the highly accurate and convenient BGMS, we developed a highly sensitive reagent, with a brand new enzyme and an original absorption dye for electrode-less low-cost test strip, and accurate hematocrit detection technology, with multi wavelengths detection by small LED module for the compact meter. The accuracy and hematocrit effect of the new BGMS were evaluated according to the guideline of ISO 15197:2013.
Result:
Our new BGMS achieved accuracy within ±5% in addition to a hematocrit effect that met the ISO 15197:2013 requirement.
Conclusion:
We developed a novel BGMS with a highly sensitive reagent and an innovative optical transmission absorbance system using a small LED module. Our new BGMS achieved accuracy within ± 5% and a hematocrit effect meet the ISO 15197:2013.
An Investigation into the Causes of Insulin Aggregation
Arizona State University Tempe, Arizona, USA
Objective:
Recent research has indicated that prescription insulin is not at its labeled concentration when it reaches the patient. This decrease in concentration is also present in research grade supplies and could be due to insulin denaturation and aggregation which causes the insulin to lose its activity. As insulin molecules aggregate in solution, their size will increase which is detectable by dynamic light scattering. By studying the effects that concentration, pH, temperature, and time have on insulin aggregation, ideal storage conditions can be identified.
Method:
Lyophilized insulin was first reconstituted at the desired concentration and then titrated to the desired pH or treated with the desired temperature or storage duration prior to analysis. The hydrodynamic radius of the sample was then measured using dynamic light scattering.
Result:
Temperature and time were found to have the largest effects on the aggregation of insulin. Room temperature and 4 C was found to have minimal aggregation whereas -20 C caused aggregation to occur. As time progressed, the insulin became more aggregated with the radius increasing by an average of approximately 100 nm after one hour. The pH had a moderate effect on aggregation with the optimal pH being between 2.3 and 2.7 for long term storage. There was no relationship found between insulin concentration and the amount of aggregation.
Conclusion:
The optimal storage conditions for insulin were identified by measuring the hydrodynamic radius of a sample. To minimize aggregation, insulin should be reconstituted at 4 C and at a pH of approximately 2.5. Additionally, the sample should be used promptly to further minimize aggregation. Future work includes concurrent testing with an enzyme-linked immunosorbent assay (ELISA) assay to correlate aggregation with a decrease in activity.
Towards Automated Logging using Smart Devices for Meal and Activity Trend Detection
Rensselaer Polytechnic Institute Troy, NY, USA
Objective:
Since 65% of adolescents in a 2004 study by Burdick missed 1 or more meal boluses a week, automating activity detection would help patients by providing meal prompts or by helping to predict future exercise. This work automates collecting patient logs using smart device-based activity detection. We envision daily patient review to remove false positive detections.
Method:
Sleep, eating, and exercise were logged by a single person while collecting data on a smartphone and smartwatch. Recurrent neural networks (RNNs) were trained on watch accelerometer and gyroscope data to classify exercise, eating, sleep, and other activities. Location data from the phone’s GPS were used to generate a lookup table that returns activity probabilities for a given location. A decision tree was implemented that incorporates the hour of day and day of week to detect activity based on time. The outputs from these individual algorithms were combined with Bayes rule and merged with activity transition probabilities over time (trend-correction). The activity detection algorithms assumed an equal number of each activity.
Result:
The accelerometer / gyroscope / location lookup/ time algorithms achieve overall cross-validated accuracies of 76%/70%/56%/57%. Eating detection accuracies of 85%/77%/17%/50% were obtained, as were sleep detection accuracies of 76%/57%/97%/76%, and exercise detection accuracies of 70%/40%/85%/70%. The final trend- corrected cross-validated accuracy is 61%.
Conclusion:
Progress has been made to automate patient activity announcements. An overall activity detection accuracy of 61% is achieved. We seek to improve the accuracy of the detection algorithms by altering neural net architectures and training on larger datasets.
Postprandial Hypoglycemia Prediction in Patients with Type 1 Diabetes using Insulin Pump Therapy
Institut d'Informàtica i Aplicacions - Universitat de Girona, Girona, Girona, Spain
Objective:
Despite the sophisticated devices available for continuous glucose monitoring (CGM), the majority of patients with type 1 diabetes (T1D) treated with an insulin pump do not have access to this technology yet. This work proposes a method addressing the prediction of postprandial hypoglycemia which can be easily used in daily life by patients who cannot afford the use of CGM. We aimed to develop a personalized predictor sensitive enough to raise a reliable event prediction allowing for the correction of the insulin bolus before the meal.
Method:
Several machine learning-based classifiers were developed for a 240 min postprandial prediction window using insulin infusion and meal informative features from insulin pump data. We developed a custom risk-based labeling scheme for training the models and a custom assessing scheme using 70 mg/dL and 54 mg/dL as thresholds, according to the consensus for Level 1 and Level 2 hypoglycemia. Individual models were produced for 10 real patients in normal living conditions.
Result:
The best models in the test phase show a median specificity and sensitivity of 58% and 69%, respectively, for postprandial hypoglycemia <70 mg/dL. In the case of < 54 mg/dL, the best models show a median specificity and sensitivity of 65% and 71%, respectively.
Conclusion:
Our model for the prediction of hypoglycemia shows a performance (sensitivity and specificity) which satisfactorily allows the identification or rejection of forthcoming hypoglycemia in the postprandial period. The goodness of the results and the trade-off between the performances of the metrics could be used in decision support systems for patients using insulin pump therapy.
Glucose Sensor and Electronics Integrated on a Single Flexible Substrate
Zimmer & Peacock, Ltd Royston, Hertfordshire, UK
Objective:
We present a continuous glucose monitoring/SMBG system where the glucose sensor and electronics are constructed on the same substrate.
Method:
We developed a continuous glucose monitoring/SMBG system that eliminated the typical need for a connector between the sensors and electronics. This design reduced both the overall size of the device and when worn reduced one of the weak points of the system’s robustness.
Result:
The overall device is fabricated using scalable techniques from the electronics and SMBG industries and so can be manufactured in volume.
Conclusion:
We present the new device and glucose data gathered using the device.
PromarkerD: A Novel Test for Predicting Rapid Decline in Renal Function in Type 2 Diabetes
Proteomics International & University of Western Australia Perth, Western Australia, Australia
Objective:
Chronic kidney disease (CKD) occurs in one in three people with diabetes and, in the United States, accounts for 40,000 deaths and $100 billion of healthcare spending annually. Current usual-care tests urinary albumin:creatinine ratio (ACR) and estimated glomerular filtration rate (eGFR) and, while simple, are limited diagnostically. PromarkerD, a novel predictive test for the onset of diabetic kidney disease (DKD), was evaluated.
Method:
PromarkerD is a blood test that measures three protein biomarkers (CD5L, ApoA4, IBP3) and a combination of clinical factors (age, HDL-cholesterol, eGFR) to predict the risk of renal decline. In a longitudinal analysis of 750 community-based patients with type 2 diabetes, the PromarkerD score was assessed for individuals transitioning from no renal disease to those with renal disease (incident CKD: eGFR≥60 to eGFR<60mL/min/1.73m2) over a four year period. The test has been previously validated using targeted mass spectrometry and was further compared with an immunoassay platform.
Result:
The incidence of new CKD within the cohort was 11.2% (N=84/750). For predicting incident CKD, ROC curve analysis of PromarkerD gave an AUC of 0.90 with 84.5% sensitivity and 80.8% specificity. The PromarkerD biomarkers outperform ACR and eGFR for predicting future CKD with an odds ratio (OR) of OR=1.98 (95%CI 1.31-3.01) versus OR=1.16 (0.94-1.42) and OR=0.87 (0.85-0.90), respectively. Comparison of the different test formats by Bland-Altman analysis shows acceptable agreement between mass spectrometry and immunoassay.
Conclusion:
PromarkerD can be used for risk stratification in clinical trials and offers the potential for a new personalized approach to managing DKD. PromarkerD allows for early detection, intervention, monitoring and management of at-risk individuals via tighter monitoring and control of blood glucose and insulin level, alongside routine PromarkerD follow-up tests, or targeted treatments and medications to manage disease progression.
Autofluorescence Analysis for Non-Invasive Detection of Cellular Tissue Oxygen Supply and Consumption
Pfützner Science & Health Institute, Mainz; MfD Diagnostics Wendelsheim & Luckenwalde, Germany
Objective:
The development of microcirculation disorders in patient with diabetes mellitus is associated with changes in tissue oxygen supply and oxygen consumption. The redox-equilibrium between NAD+ and NADH indicates the level of oxygen supply on a cellular level. An increase in NADH indicates the switch from aerobic to anaerobic energy production. Chronic exposure of the tissue to low oxygen supply results in an accumulation of cellular and interstitial lactate, which is further enhanced during physical exercise and muscle work.
Method:
The non-invasive skin analyzer NADHJA Lab 355 measures NADH autofluorescence in the stratum spinosum of the epidermis after excitement by means of laser pulses at a wavelength of 355 nm. A probe was placed on the skin and allowed for non-invasive, continuous, and simple assessment of the NADH signal. In a study of patients with type 2 diabetes, NADH autofluorescence was measured in addition to plasma lactate concentrations and other diabetes parameters.
Result:
First results indicate impaired cellular oxygen supply in the skin in diabetes patients even in early stages as compared to a non-diabetic control group.
Conclusion:
Current devices measure oxygen supply and effects (tissue oxygen pressure, blood pressure, etc.) and the metabolic lactate load. The new non-invasive assessment of cellular oxygen turnover may provide a more comprehensive understanding of cellular impairment and metabolic burden in the course of diabetes development. It may also help to detect further pathways, which are important for development of microcirculatory secondary complications at later clinical disease stages. The new device allows for reliable non-invasive assessment of cellular oxygen supply and cellular vitality in the skin and the underlying tissue.
Results with the Non-Invasive Tissue TensorTip Combo Glucometer for Glucose Prediction in Patients with Type 1 Diabetes
Pfützner Science & Health Institute Mainz, Germany
Objective:
Access to pain-free and unlimited glucose information required for treatment monitoring is one of the most desired unmet medical needs of insulin treated patients with type 1 diabetes. TensorTip CoG, is a non-invasive tissue glucose monitoring device (NI-CoG), that operates based on optical methods in the near-infrared and visible wavelength range and employs an additional built-in invasive glucose meter (Inv-CoG). We performed two clinical studies where we enrolled patients with type 1 and type 2 diabetes and have performed a special post-hoc analysis with the type 1 patient population.
Method:
Six patients with type 1 diabetes (3 male/3 female, age: 43±16 years, A1c: 8.0±0.5%) participated in a standard meal study experiment and 29 type 1 patients participated in a system accuracy evaluation in accordance with ISO15197:2015 (13 male/16 female, age: 42±14 years, A1c: 7.5±0.5 %). The results were combined for this analysis.
Result:
The meal study patients (n=66 data points) and the ISO study patients (n=29) showed good agreement of the NI-CoG predicted tissue glucose with the capillary reference method (YSI Stat2300 plus). Mean absolute relative difference (values >100 mg/dL) was calculated to be 13.3% (mean absolute difference for values<100 mg/dL: 16 mg/dL). The consensus error grid (version for type 1 diabetes) revealed 78% of the data points to be in zone A (B: 18 %, C: 4 %). Zone C values were only observed in patients without proper device calibration over the entire measurement range of the device.
Conclusion:
The results of this new non-invasive glucose prediction device are encouraging for potential routine use in patients with type 1 diabetes. In our trials, this non-invasive glucose prediction technology has reached an accuracy level that is comparable to results previously published for needle sensors, for flash glucose measurement, or continuous glucose monitoring.
VIVI Cap Protects Insulin in Disposable and Reusable Pen Devices from Degradation During Exposure to Extreme Temperature Conditions
Pfützner Science & Health Institute, Mainz, Germany
Background:
Exposure to extreme cold or heat leads to insulin degradation in a very short time with loss of glucose-lowering efficacy. Therefore, insulin vials and cartridges have to be kept at 4-8° C (~ 39°F to 47°F) until use. Once opened, the insulin is supposed to be stable for 28 to 30 days at room temperature. VIVI Cap (TempraMed Inc.) is a reusable storage device that is available for the majority of existing disposable (D-Cap) and reusable (R-Cap) pen devices in the United States and Europe. VIVI Cap works without an external power supply and the technology is based on an ultra-high vacuum insulation and a material that uses heat consumption and a calorimetry-based phase-change.
Method:
Five disposable insulin pens (FlexPen, Insulin aspart) were kept under extreme temperature conditions for 5 days (each day: 8 h at 50 °C and 16 h at 22 °C) either without protection, in a Frio cooling case device (freshly prepared each day) and in the VIVI Cap. Samples were taken every day. Insulin degradation was determined in accordance with US pharmacopeia by appropriate HPLC methods for insulin aspart and high and low molecular weight degradation molecules. Each experiment was performed in triplicate.
Result:
Insulin aspart without protection was shown to have more than 2% impurities already after one day (Frio: 2 days, VIVI Cap: >5 days). High molecular weight products appeared in measurable quantities after two days without protection (Frio: 3 days, VIVI Cap > 5 days, p<0.01 against the other 2 storage conditions). In a second experiment, unprotected insulin aspart pens were exposed to -20°C and reached the critical temperature of 2°C after 37 min. When protected by the VIVI Cap device, this temperature was reached after 150 min.
Conclusion:
VIVI Cap was superior to the other tested conditions with respect to stabilizing insulin aspart during storage under extreme temperature conditions. The device provides an easy to use solution for maintaining insulin efficacy under daily life conditions.
Personalizing a Mathematical Model of T2DM via an Evolutionary Algorithm
Universidad Autónoma de Nuevo León San Nicolás de los Garza, Nuevo León, México.
Objective:
The aim is to develop a methodology to personalize a mathematical model of type 2 diabetes mellitus (T2DM). From a previous reported model validated using nominal parameters, the objective is to resolve a parametric fitting problem; such that the model output tracks blood glucose data measured with a continuous glucose monitoring system.
Method:
A set of T2DM patients belonging to a glycemic control program at the Hospital of the University of Nuevo León in México was asked to provide their metabolic data. Blood glucose concentrations were measured by an iPro™2 MCG Professional, a continuous glucose monitoring system by Medtronic. The nominal mathematical model was a physiological one presented by Vahidi et al. in 2016. This model is a set of ordinary differential equations divided into three subsystems: glucose, insulin, and glucagon dynamics. To personalize the nominal model, a parametric adjustment scheme, based in an evolutionary algorithm, was presented. This algorithm, presented by Luis Torres in 2006, originally took a randomly created population (parameters to be tested) to compute the model (evaluation), and the result of this calculation was compared to the reference data (blood glucose data) so that the error could be calculated. After that, the best set of parameters was chosen and a new population group was created to start a new iteration. Finally, the statistics of the search space of parameters for each solution was analyzed.
Result:
A set of eighteen parameters corresponding to metabolic sources and sinks for glucose, insulin, and glucagon compartments form the set of parameters to be adjusted by the evolutionary algorithm. As an example of the performance of the methodology, a set of six days of monitoring of a single patient was considered in the personalization of the model. The mean absolute error (~20 mg/dL) shows that the scheme can resolve the problem to adjust the response of a physiological model to a set of data. Furthermore, the boxplots of the normalized values of the computed solutions show they remain in a bounded subset of the search space.
Conclusion:
The glycemic dynamic of a specific T2DM patient can be estimated by a mathematical model and an adjustment scheme based on an evolutionary algorithm. The interest of using a physiological model as the structural basis of this personalized scheme is to have an interpretation of metabolic changes according to the glycemic variation measured by the continuous glucose monitoring system. Unlike black-box models that do not provide physiological information and only input/output prediction, the proposed scheme could provide useful information on how metabolic processes degrade as diabetes progresses. Thus, this scheme could be used as an analytical tool in prevention programs of long-term complications of T2DM.
Efficacy of Advanced Carbohydrate Counting and Automated Insulin Bolus Calculators in Type 2 Diabetes: The BolusCal2 Study, an Open-Label, Randomized Controlled Trial
Steno Diabetes Center Copenhagen Gentofte, The Capital Region, Denmark
Objective:
Carbohydrate counting and the use of an automated bolus calculator can reduce A1C in type 1 diabetes but this assertion has never been tested in type 2 diabetes. We evaluated the efficacy of advanced carbohydrate counting and the use of an automated bolus calculator (ABC) compared with mental insulin bolus calculation (MC) in persons with type 2 diabetes (PWT2D).
Method:
The study was a 24-week, open-label, randomized controlled study in PWT2D treated with basal-bolus insulin. Seventy-nine participants were randomized into two groups. The ABC group received training in carbohydrate counting and the use of an automated bolus calculator. The MC group received training in carbohydrate counting and the mental calculation of insulin bolus. Blinded continuous glucose monitors (CGM) were worn for 6 days at baseline and study end.
Result:
Baseline A1C was similar in both groups. At 24 weeks, A1C decreased by 8.1 mmol/mol in the ABC group and 8.0 mmol/mol in the MC group. There was no difference in the change in A1C between groups at study end. CGM data showed time in glycemic range increased significantly in the ABC group but there was no significant difference in the increase between the two groups. Time in hyperglycemia decreased in the ABC group but was unchanged in the MC group. Time in hypoglycemia decreased insignificantly in both groups with no difference between groups. Coefficient of variance decreased in both groups without any difference in decrease between groups.
Conclusion:
Advanced carbohydrate counting and insulin bolus calculation is an efficient, low-cost, tool to reduce A1C in PWT2D treated with basal-bolus insulin. Blinded CGM revealed decreased glycemic variability with both options and a reduction in time in hypoglycemia, but only the group using automated bolus calculation increased their time in glycemic range.
The Use of Mobile Applications for Diabetes Self-Management Education: Does Gender Matter?
Steno Diabetes Center Copenhagen (SDCC) Gentofte, Denmark
Objective:
The use of mobile applications for Diabetes Self-Management Education (DSME) significantly lowers A1c levels, reduces weight, has positive effects on health behavior, and should be part of the diabetes treatment plan. However, with more than 1500 applications for DSME, it can be difficult for healthcare professionals (HCP) to know which ones to recommend and who uses them. This pilot study investigates which types of applications people with diabetes use continuously for different aspects of DSME and if gender influences the choices.
Method:
An online survey questionnaire included 25 closed ended questions addressing the choice of application for DSME, duration of use, and demographic data. After each question, a comment and application names could be added. A link to the survey was posted on Social Media in diabetes online communities. The questionnaire was open for 3 months. The survey was filtered to avoid multiple people answering from the same device. For the analysis, univariate and multiple linear regression was used.
Result:
Sixty-five people answered the questionnaire, mean age 39±4.2 years, male 51%, type 1 diabetes 74%, mean diabetes duration 10 ±2.7 years. Eight different types of applications were chosen for DSME. Eighty percent of subjects used them longer than 3 months. Differences in use between men and women were identified. Men used applications for carbohydrate counting, interoperability, networking/online, community support, logbooks, calculation of insulin dosage, and information about diabetes. Women used applications for meal planning, weight loss, and exercise more than men.
Conclusion:
The results indicate that specific topics are requested in mobile applications used continuously for at least three months for DSME and that the choice of application differs between men and women. Future studies will investigate these findings further.
Insulin-Only Blood Glucose Control during Unannounced Aerobic Exercise
Universitat de Girona Girona, Girona, Spain
Objective:
The objective of this study is to use an exercise detection algorithm that does not rely on external physiological signals in order to employ strategies to reduce exercise-induced hypoglycemia and improve outcomes caused by unannounced aerobic exercise in an insulin-only closed-loop controller.
Method:
Exercise-induced hypoglycemia minimization strategies were employed in an insulin-only proportional-derivative (PD) controller with sliding mode reference conditioning. The performance of the controller was compared to the same closed-loop setting without exercise announcement. The exercise detection algorithm collects glucose and insulin infusion rate values and computes ‘D(t)’ using an unscented Kalman filter (UKF). A patient-specific exercise detection threshold was found as the minimum value of ‘D’ during a 15-day scenario without exercise. After exercise detection, the algorithm suggests the amount of carbohydrates to be ingested, reduces the maximum allowed insulin-on-board, reduces the basal insulin to zero, and reduces the next insulin bolus by 30%. This methodology was tested in silico using the UVA/Padova simulator in a 15-day scenario with 10 adult patients, 8 exercise sessions at 60% VO2max on alternate days, and additional variability in insulin sensitivity, meal estimation and insulin pharmacokinetics.
Result:
The results, with and without exercise announcement, were: median CGM: 121 (117-123) mg/dL vs. 133 (127-136) mg/dL (p=0.002), time in range (70-180mg/dL): 91 (90-93)% vs. 94 (91-94)% (p=0.006), time above 180 mg/dL: 4 (2-5)% vs. 6 (4-9)% (p=0.002), time below 70 mg/dL: 4 (3-7)% vs. 0 (0-0)% (p =0.002), time below 54 mg/dL: 2 (1-2)% vs. 0 (0-0)% (p =0.002), and the total number of hypoglycemic events: 131 vs. 6 (p = 0.002).
Conclusion:
A significant improvement in glucose control has been observed when incorporating an exercise detection algorithm coupled with exercise-induced hypoglycemia mitigation measures into an insulin-only PD control strategy.
Nurse-delivered digital-health interventions: A systematic review and meta-analysis
Sanofi, US, Innovative Solutions Bridgewater, NJ, USA
Objective:
With 425 million adults struggling with diabetes, technology-guided educational programs and data monitoring delivered by nurses could help improve clinical and behavioral outcomes.
Method:
A systematic literature review and meta-analysis were conducted to determine whether nurse-delivered digital- health interventions improve clinical and patient outcomes. We searched MEDLINE, Embase, and CENTRAL for comparative studies where nurses served as primary intervention providers. Meta-analysis of randomized controlled trials (RCTs) compared the impact of nurse interventions to standard of care (SoC). Patient-reported outcomes (PROs) were reviewed qualitatively. Analyses were performed on the DOC Data v2.0 advanced web-based platform.
Result:
The search returned 3,649 studies of which 38 addressed nurse-delivered digital-health interventions. Interventions included mobile or web-based technology guiding educational programs. The Meta-analysis revealed digital-health interventions delivered by nurses improved glycemic control and patient outcomes compared to SoC. Individualized education with nurse engagement proved most effective for lowering A1c [Mean Difference: -0.29%, 95% CI (-0.48, -0.09)] over interventions without active nurse engagement [-0.17% (-0.32, -0.02)]. Incorporating glycemic monitoring with active nurse engagement further reduced A1c [-0.44% (-0.48, -0.17)]. Follow up at both 6 and 12 months showed positive improvements, however this lessened over time possibly indicating reduction in long-term adherence. Patients were also more likely to achieve A1c target levels when receiving nurse-delivered digital-health interventions compared to SoC (p=0.023). Seventy-seven percent of PRO studies reported favorable outcomes for nurse interventions.
Conclusion:
Digital health technology programs led by nurses demonstrated statistically significant overall improvement and obtainment of target A1C over SoC. Integrating digital self-monitoring devices with nursing interventions should be encouraged but further research is needed.
Low-cost Built-in Self-Calibration Method for Blood Glucose Test Strips
PA Consulting Cambridge, UK
Objective:
Electrochemical blood glucose meter systems rely on factory batch calibration to provide glucose values from a current signal produced by the test strip. Improvements on this have included sophisticated signal processing and electrochemical techniques to give a more accurate result. However, these methods often require the use of expensive materials and electronics such as gold with impedance-based sensors. We report on a novel, low cost, on strip calibration technology that can be applied to current test strip architectures based upon a controlled deposit of glucose within the test strip. Measurement of the signal from this additional glucose allows a correction for confounding factors that might affect the signal from the user’s blood sample: such as: hematocrit, temperature, and manufacturing variation.
Method:
Commercial blood glucose meter (BGM) test strips were modified by replacing the capillary lid with a lid coated with a fast dissolving formulation including a known amount of glucose. These test strips were characterized for different error conditions (i.e., hematocrit and temperature) using a bespoke glucose meter to collect chronoamperometric current signals and from this a calibration algorithm was developed. The modified test strips were then used in a clinical evaluation involving direct fingerstick blood samples from 169 subjects
Result:
We present results of a clinical evaluation of our system where we demonstrated an accuracy improvement as measured with mean absolute relative difference (MARD) from 11.1% to 5.9%.
Conclusion:
A simple, low-cost modification to traditional BGM test strips allows a significant improvement in accuracy by directly correcting for effects on the glucose signal measured.
Unannounced Meal Control through Sliding Mode Techniques in an Artificial Pancreas
Universitat Politècnica de València València, Spain
Objective:
Carbohydrate counting in hybrid artificial pancreas systems remains a burden for patients with type 1 diabetes and fully automatic systems without meal announcement are desirable. Besides that, missed prandial boluses could degrade the performance of hybrid systems. Here, a meal detector module and a bolusing algorithm are presented for unannounced meal control.
Method:
The meal detection module employed a super-twisting-based residual generator to detect the meal depending on a certain threshold criteria. After meal detection, the bolusing algorithm delivered a series of insulin boluses following an exponential decay in magnitude. Bolus timing relied on the area-under-the-curve of the rate of glucose appearance, which was estimated via a first-order sliding-mode observer. Finally, a comparative analysis of the proposed fully closed-loop system, its hybrid counterpart, and a missed bolus scenario was performed with UVa- Padova simulator under a 42-meal scenario with variability in insulin sensitivity and in-meal absorption dynamics.
Result:
The proposed fully closed-loop system achieves a reduced time in hyperglycemia when it is compared with the missed bolus case (16.73 (9.86)% vs. 32.26 (19.56)%, mean (SD), p=2.11·10-8), but it is slightly higher with respect to the hybrid case (16.73 (9.86)% vs. 12.347 (11.744)%, p=1.36·10-6). Although the number of false positives in the meal detector is low (1.50 (1.08)), its occurrence led to an increased time in hypoglycemia compared to the hybrid case (0.048 (0.103)% vs. 0.007 (0.023)%, p=0.042), whereas there are no significant differences regarding the missed bolus case: (0.048 (0.103)% vs. 0.007 (0.033)%, p=0.052).
Conclusion:
The suggested meal detection module and bolusing algorithm manage to reduce the time in hyperglycemia when meals are not announced and it had a low occurrence of false positives.
Multivariable Simulation Platform for Type 1 Diabetes Mellitus
Illinois Institute of Technology Chicago, IL, USA
Objective:
Multivariable artificial pancreas (mAP) systems are designed to improve glycemic control by utilizing physiological signals from wearable devices in addition to continuous glucose monitoring (CGM) measurements. The in-silico evaluation of mAP systems is inhibited by the available metabolic simulators that compute glucose and insulin concentrations in response to meals and a limited subset of exercises. A new physiologic and metabolic simulation platform involving 20 virtual subjects with type 1 diabetes mellitus (T1DM) is proposed. In addition to the CGM values, it generates physiological variable signals reported by noninvasive wearable devices such as heart rate, energy expenditure, skin conductance and temperature, and accelerometer readings.
Method:
Hovorka’s glucose-insulin dynamic model was extended to explicitly consider the effects of exercise in glycemic dynamics. The proposed glycemic model was integrated with exercise models to derive metabolic variations in response to physical activity. Additional models are included to compute physiological signals affected by physical activity. Experimental data from 17 subjects with T1DM in open-loop clinical studies were used to characterize the interpatient variability and determine realistic virtual subject parameters. Users of the multivariable simulator provided the meal, exercise scenarios, and insulin infusions. Model equations were solved using this information to yield the output variables, including physiological biosignals, blood glucose and plasma insulin concentrations, and CGM measurements.
Result:
Across 17 subjects with over 440-hour simulations, the proposed model significantly reduces the average root-mean- square error between the actual and predicted CGM from 21.4±18.0 mg/dL(for the reference model without explicit consideration of exercise) to 11.4±6.6 mg/dL (p-value<0.05 [8.34*10-5]).
Conclusion:
The multivariable simulator for virtual subjects with T1DM will enable the in-silico evaluation and accelerate the development of mAP systems.
Accuracy and Precision of the Afinion HbA1c Dx Test System Compared to Three Laboratory Methods
Abbott Laboratories, Rapid Diagnostics Division San Diego, CA, USA
Objective:
This study compared the accuracy and precision of the point-of-care Afinion HbA1c Dx test to laboratory A1c methods routinely used for the diagnosis of diabetes.
Method:
To evaluate accuracy, remnant whole blood samples (n=618) were tested in singleton with the Afinion HbA1c Dx and one of three laboratory methods (Roche Tina-quant HbA1c Gen. 3, BIO-RAD Variant II Turbo, Siemens Dimension Vista) at each of five clinical sites. These samples were also tested in duplicate on the Tosoh G8 at a National Glycohemoglobin Standardization Program (NGSP) Secondary Reference Laboratory (SRL). The correlation, bias, and number of results within ±6% compared to the NGSP method were calculated. To evaluate precision, at each site three samples (low, medium, and high %A1c) were tested on both the Afinion and the laboratory method (four replicates, twice per day, for 8-10 days). The between-day, between-run, within-run and total coefficient of variation (CV) for each method were calculated.
Result:
Correlation (r) was 0.994 for the Afinion HbA1c Dx and ranged from 0.984-0.994 for the laboratory methods. Across the assay range, the mean relative bias of the Afinion was -1.38% to -0.04% (absolute bias: -0.09 to 0.00 %A1c) and for the aggregate of the laboratory methods -1.23% to 0.01% (absolute bias: -0.11 to 0.00 %A1c). Total imprecision for the Afinion was 0.85%-1.46% CV and for the laboratory methods it was 0.83%-3.23% CV. It was found that 97.1% of Afinion results were within ±6% of the NGSP method, compared to 94.5% of laboratory results.
Conclusion:
The accuracy and precision of the Afinion HbA1c Dx test are comparable to laboratory methods used for the diagnosis of diabetes.
Accuracy, Precision, and Ease-of-Use of the Point-of-Care Afinion HbA1c Dx Test* when used by Self-Trained Operators: Results from a Pilot Study
Abbott Laboratories, Rapid Diagnostics Division San Diego, CA USA
Objective:
This pilot study evaluated the accuracy, precision, and ease-of-use of the Afinion HbA1c Dx test* in the hands of self-trained operators who are representative of those in a CLIA-waived setting.
Method:
Self-trained operators of the Afinion HbA1c Dx test collected and tested fingerstick whole blood samples from 131 subjects at a single study site. For comparison, trained operators performed testing on the same subjects in parallel. Matching venous whole blood samples were also collected and tested on both the Afinion and on the Tosoh G8 at a National Glycohemoglobin Standardization Program (NGSP) Secondary Reference Laboratory (SRL). The accuracy of the Afinion results was assessed via correlation and bias with respect to the NGSP SRL method. The Afinion precision was estimated by calculating the coefficient of variation (CV) for duplicate measurements and ease-of-use was assessed by surveying the self-trained operators.
Result:
Afinion results from self-trained operators were highly correlated with the NGSP SRL method (r ≥ 0.994). The mean absolute bias was -0.016% A1c (-0.2% relative bias) using fingerstick samples and -0.015% A1c (-0.2% relative bias) using venous samples. Self-trained operators produced precise results using both fingerstick and venous blood, with CVs ranging from 1.08%-1.18%. Overall the results from self-trained operators were similar to those obtained by trained operators. All self-trained operators “agreed/strongly agreed” that the Afinion was easy to use in 6/7 usability parameters.
Conclusion:
The results of this study suggest that self-trained, non-laboratory professional users of the Afinion HbA1c Dx test can obtain equally accurate and precise results as trained laboratory professionals and that they also find the test easy to use.
*The Afinion HbA1c Dx test is currently moderate complexity and has not been granted a CLIA waived categorization
Rescue Carbohydrates Suggestion Algorithm During Exercise for People with Type 1 Diabetes
Illinois of Technology Chicago, Illinois, USA
Objective:
The objective of this study is to develop a new algorithm for artificial pancreas systems that suggests the appropriate amount of carbohydrates (CHO) to people with type 1 diabetes (T1D) during exercise with different speeds and intensities to reduce the risk of hypoglycemia and avoid inadvertent hyperglycemia while minimizing the total CHO consumed.
Method:
An algorithm based on real-time variables, including energy expenditure (EE), continuous glucose monitoring (CGM) measurements, slope of CGM, plasma insulin concentration estimates, and patient demographic information was used to compute the optimal CHO suggestion during exercise. A fuzzy logic estimator determined the rescue CHO amount. A multivariable simulator involving in silico subjects with T1D was used to test the approach. Several indices were employed to evaluate the performance of the proposed algorithm in averting physical activity-induced hypoglycemia without adverse consequences. Simulation case studies with 20 virtual subjects during various unannounced physical activity protocols involving treadmill and bike exercises were used for evaluation.
Result:
Compared to the previous rules-based algorithm with average CGM values of 229±19 mg/dL, the new proposed approach reduced the hyperglycemia risk and maintained the CGM values closer to the safe target range with mean CGM of 159±10 mg/dL. Total consumed rescue CHO for each subject was reduced by 48.4% without increasing hypoglycemia risk. The CGM measurements remained in the safe target range (90-180 mg/dL) for 4% longer time duration during exercise with the new approach compared to the previous method.
Conclusion:
The proposed algorithm for rescue CHO suggestion will improve the benefits of exercise in people with T1D by reducing the amount of rescue CHO required for maintaining euglycemia without increasing the risk of hypo- and hyperglycemia.
Bedtime Scanning is Associated with Improved Nocturnal Glycemic Control: A Real-World Observational Analysis of Flash Glucose Monitoring
Barbara Davis Center for Diabetes Aurora, CO, USA.
Objective:
Use of the flash glucose monitoring (FreeStyle LibreTM system) has been shown to reduce hyperglycemia and hypoglycemia. This study was aimed to evaluate the effect of bedtime scanning on nocturnal glycemic control in real-life settings.
Method:
De-identified glucose data from 6,802 users, involving 818,608 nights over six months, were studied. The population was divided into nights where scanning was performed (n=636,555) and nights where scanning was not performed (n=182,053), during the hours of 9pm to 11pm. Furthermore, the scanning group was divided into nights where the glucose was low at time of scanning (n=80,832) and nights where it was not (n=667,286). Time below 70mg/dL and 54mg/dL and time above 180mg/dL and 250mg/dL were then compared between the groups during the hours of 11pm to 8am using an independent student t-test at a significance level of 0.05.
Result:
Comparing the scanning and non-scanning groups, scanners had significantly less time spent above 180mg/dL (2.55h/night vs 3.09h/night; p<0.001) and time spent above 250mg/dL (0.73h/night vs 1.09h/night; p<0.001). Minor differences were observed in the time spent below 70 mg/dl (45min/night vs 44min/night; p = 0.03) and time spent below 54 mg/dl (19min/night vs 21min/night; p < 0.001). However, on nights where the glucose was low at the time of scanning, scanners had significantly less time below 70mg/dL (95min/night vs 136min/night; p<0.001) and time below 54mg/dL (48min/night vs 76min/night; p<0.001).
Conclusion:
In real-life settings, bedtime scanning with flash glucose monitoring is associated with improved hyperglycemic outcomes at night. There is an association with improved hypoglycemic outcomes in those with low glucose at time of scanning.
Disclosure Statement:
All work was funded by Abbott Diabetes Care.
Minimally Invasive Continuous Glucose Monitoring Sensors based on Microneedle Arrays
Swansea University, College of Engineering, Bay Campus Swansea, Glamorgan, UK
Objective:
The objective of this study is the development of a minimally invasive microneedle array device for continuous glucose monitoring in type 1 diabetes (T1D).
Method:
To address the problems associated with the commercially available CGM sensors (such as biofouling and high costs of manufacturing), we developed polycarbonate based, minimally invasive, microneedle arrays. These polycarbonate microneedle arrays were fabricated using scalable injection molding. The microneedle devices incorporate glucose oxidase entrapped in electropolymerized polyphenol films and were tested in healthy volunteers over 6 hours (Phase 1), over 24 hours (Phase 2), and with participants with T1D (Phase 3, over 24 hours).
Result:
We will report on the performance of the microneedle array based continuous glucose monitoring sensors in healthy volunteers and in participants with T1D. A summary of the data points obtained from the study plotted using the Clark Error Grid indicates that the microneedle CGM performance is clinically acceptable. The analysis showed that 96.4% points fall within clinically acceptable zones A and B and 3.6% fall in zone C. The mean absolute relative difference (MARD) values calculated for the phase 3 studies in subjects with T1D was 9%.
Conclusion:
This study presents a strategy that helps in developing a minimally invasive continuous glucose monitoring device.
Validation of a novel algorithm for interpreting glycemic control from CGM data
Glooko, Inc. Mountain View, California, USA
Objective:
Although there are clear benefits for the use of continuous glucose monitoring (CGM) in diabetes, the interpretation of dense, high-dimensional, glucose data can be subjective and burdensome. This is an area where digital health can potentially help decision support. Here we compare the performance of a novel algorithm that integrates multiple glycemic data features with the interpretations of expert clinicians.
Method:
We applied the Glooko algorithm to a dataset consisting of CGM data from 10 people with diabetes (PWDs) with each PWD dataset spanning 7 days, and validated the outputs with the assessment of 57 clinicians familiar with CGM interpretation (29 endocrinologists, 28 diabetes educators). Algorithm performance was assessed as the identification of the correct “best day” of glycemic control and three similarity metrics: Cohen’s Kappa (weighted), R2, and mean absolute error (MAE).
Result:
The algorithm correctly identified the “best day” for 9 of the 10 PWDs. For the PWD on which the algorithm did not agree with clinicians on the “best day”, the algorithm and clinicians agreed on the top two ranked days. Compared to clinicians’ assessments, the algorithm had mean R2 = 0.77 (P = 0.011; IQR: [0.66, 0.97]), mean weighted Kappa = 0.86 (IQR: [0.79, 0.97]), and mean MAE = 0.64 (P = 0.0063; IQR: [0.21, 0.96]).
Conclusion:
The algorithm performed well in inferring glycemic control from CGM data across multiple metrics and its rank decisions were very similar to those of expert clinicians. In a healthcare setting, the algorithm has the potential to support clinical decisions by reducing subjectivity and the time spent evaluating complex CGM data.
Exploratory Analysis of Insulin Sensitivity Adaptation with the MiniMed™ 670G system in Children, Adolescents, and Adults
Medtronic Diabetes Northridge, California, USA
Objective:
The MiniMed™ 670G system establishes new algorithm parameters, based on historical data, daily. In-home use of this therapy for 3 months by children, adolescents, and adults with type 1 diabetes (T1D) improved percentage of time in target range (TIR) and A1c. In this exploratory analysis, changes in the device-calculated insulin sensitivity factor (ISF) and glycemic control were assessed.
Method:
Participants aged 7-13 years (n=105) enrolled at 9 centers (8 US,1 Israel) and participants aged 14-21 years (n=30) and 22-75 years (n=94) enrolled at 10 centers (9 US,1 Israel) used the MiniMed™ 670G system in Manual Mode during a baseline 2-week run-in phase followed by a 3-month study phase in which Auto Mode was enabled. Changes in glycemic metrics and weekly-averaged ISF were calculated during baseline run-in and throughout the study period and were evaluated by age group.
Result:
From run-in to end of study, mean ISF decreased from 84.4±44.5 to 56.9±22.9 (P<0.001) and 43.2±13.6 to 34.2±11.2 (P<0.001) in children and adolescents, respectively; and decreased slightly in adults (50.2±19.4 to 47.7±21.4 [P=0.110]). Basal insulin delivery increased from 15.6±6.9 to 17.0±7.5 (P<0.001) and 26.2±9.2 to 27.0±8.3 (P=0.381) for children and adolescents, respectively; and decreased in adults (23.4±12.3 to 22.7±16.0 [P=0.005]). The percentage of TIR (24 hours) increased from 56.2±11.4% to 65.0±7.7% (P<0.001), 60.4±10.9% to 67.2±8.2% (P<0.001), and from 68.8±11.9% to 73.8±8.4% (P<0.001), respectively. The percentage of TIR (early morning, 3am-6am) increased from 60.1±19.3% to 80.2±10.2% (P<0.001), 65.9±17.5% to 79.3±11.7% (P<0.001), and from 69.5±15.6% to 83.0±9.7% (P<0.001), respectively.
Conclusion:
These findings show that necessary adjustment of MiniMed™ 670G basal insulin delivery with insulin sensitivity adaptation resulted in improved glycemic outcomes of children, adolescents, and adults with T1D.
Determinants for Digital Health Engagement
MedStar Union Memorial Hospital Baltimore, MD, USA
Objective:
Not all digital interventions are appropriate for all individuals with a particular disease state. User engagement with digital health technology is critical for successful clinical outcomes. We sought to understand the characteristics of users of the diabetes digital therapeutic BlueStar that would predict successful engagement. We hypothesized that users could be categorized as possessing low or high diabetes self-management knowledge/skills and low or high technology capabilities.
Method:
A performance improvement project for people with type 2 diabetes was launched in a primary care practice in early 2018. The key element of the project was the implementation of the digital therapeutic BlueStar. Users were recruited by the practice based on the FDA-approved intended use of the device. The downloading, activation, and usage of the BlueStar app on the users’ phones were monitored by the study team. The providers caring for the patient users recorded their clinical characteristics.
Result:
A qualitive review of the first 24 users revealed the following characteristics that were associated with engagement:
1. General health motivation
2. Salience of diabetes specifically to the user
3. New diagnosis (last 12 months)
4. Technology savviness
5. Relational connectedness with the provider
6. Richness of communication during the onboarding process
7. Degree of follow up prodding
Conclusion:
The determinants listed above are a first effort to identify characteristics of users who would be most successful with a digital health tool. These data can help guide providers to optimally implement digital health technologies into clinical practice.
Tablet-based Hospital Education for Diabetes Care – A Panacea or Work as Imagined?
MedStar Health Quality & Safety
Objective:
Diabetes self-management education and support (DSMES) is an essential part of the care continuum for people with type 2 diabetes (T2DM) but optimizing delivery and uptake of DSMES is challenging for hospitals. Emerging technology has the potential to address this challenge. Diabetes to Go Inpatient (D2Go-IN) is a multi-modal survival skills DSMES program that leverages technology to deliver video-based education on a touchscreen tablet. We describe our experience with an adaptive pragmatic trial to examine the feasibility of delivering the D2Go-IN program to hospitalized T2DM patients in a large US urban hospital.
Method:
Focus groups and observations by human factors scientists informed the process design for tablet delivery within nursing unit workflow. D2Go-IN was implemented November-2017 through May-2018 on three nursing units in one hospital. In-services oriented unit staff to the web-based platform and delivery process. Implementation challenges were assessed every 4-6 weeks and rapid-cycle adaptations to the platform and to the process were made to further support adoption.
Result:
A total of 538 T2DM patients were identified with 330 of those being tablet-eligible. Sixty-two patients (19%) accessed D2Go-IN and, of those, 39 (63%) completed the program. Barriers to implementation were revealed at multiple levels: staff (receptivity, time, production pressures, culture, context); process (patient identification, tracking, tablet delivery); technology (timing out, touch sensitivity, discomfort with technology), and patient (receptivity, illness acuity, availability, accessibility). Most D2Go-IN completers required significant assistance. Successful tablet delivery required highly engaged staff to make multiple attempts per patient.
Conclusion:
Uptake and adoption of the technology in this acute care setting was limited. Despite staff interest, time constraints and high patient barriers limited the ability to deliver DSMES using tablets at the bedside.
Prevention of Inpatient Hypoglycemia Using CGM devices: The Glucose Telemetry System
Division of Endocrinology, Baltimore Veterans Administration Medical Center Baltimore, MD, USA, 21201
Objective:
We recently reported that glucose values from continuous glucose monitoring (CGM) devices can be successfully transmitted to a monitoring system in the nursing station (Glucose Telemetry System-GTS).
Method:
To determine if the GTS can reduce inpatient hypoglycemia, we randomized ten insulin treated patients with type 2 diabetes mellitus (T2D) to GTS (intervention group) or standard of care (control group). Participants in the GTS were monitored by real-time CGM and those in standard group used a “blinded’ CGM. The lower glucose alarm was set at <85mg/dL and, when this alarm occurred, nurses confirmed low blood glucose by capillary point of care testing and provided hypoglycemia treatment. Study outcomes included the difference in hypoglycemia (≤70 mg/dL) between groups.
Result:
Nine patients, five monitored by GTS and four by standard of care, completed the study. There was a non-statistical difference between the number of hypoglycemic episodes (2 vs. 5, p=0.44) and the hypoglycemia event rate (0.1 vs 0.26, p=0.33) between groups. Two subjects in GTS group had three interventions for impending hypoglycemia. GTS average glucose was 167.5±19.6 mg/dL, vs. 160.1±22.6 mg/dL (p=0.66) with standard of care. Overall percent time spent in hypoglycemia was 0.30±0.39% vs. 1.59±1.59% (p=0.26), and clinically significant hypoglycemia (<54mg/dL) was 0% vs. 0.27±0.36% (p=0.28) in the intervention and control groups. Percent time within normoglycemia (70-179 mg/dL) was 64.68±15.39% vs. 70.11±19.23% (p=0.70) and hyperglycemia (≥180 mg/dL) was 35.02±15.5% vs. 28.30±19.55% (p=0.64), respectively. Sample size calculations demonstrate that 244 patients (122 subjects in each group) need to be recruited to identify a statistically significant difference between the two groups.
Conclusion:
In this pilot study, subjects monitored by GTS had a non-statistically significant difference in inpatient hypoglycemia compared to standard care.
Impact of Real-World Use of the CONTOUR®DIABETES App on Glycemic Control and Testing Frequency
Ascensia Diabetes Care Parsippany, NJ, USA
Objective:
The CONTOUR®NEXT ONE smart meter and app system includes a wireless-enabled blood glucose (BG) meter that links to the CONTOUR®DIABETES app installed on a mobile device. The app reports patterns of BG readings and provides guidance for diabetes self-management based on the Information-Motivation-Behavioral Skills (IMB) model, which is a well-researched and well-validated approach to enable health behavior change.
Method:
Anonymized data from 5,870 people who used the app for >180 days in North America were analyzed. Two user subsets were identified with BG measurements both within the first 30 days and after 180 days of app use: A) ≥1 BG ≤70 mg/dL and >180 mg/dL (n=1,253); B) ≥1 BG ≤50 mg/dL and >250 mg/dL (n=654). Odds ratios (OR) were compared for users with hypoglycemia (BG <50 or <70 mg/dL) and hyperglycemia (BG >180 or >250 mg/dL) within the first 30 days versus after 180 days of app use. Testing frequency was also evaluated.
Result:
App use decreased the likelihood of ≥1 BG event ≤50 mg/dL (OR:2.47; 95% CI:2.02-3.07) and >250 mg/dL (OR:1.56; 95% CI:1.32-1.91) after 180 days of use. Similar results were obtained for hypo/hyperglycemic events defined as ≤70 mg/dL and >180 mg/dL, respectively, and occurrence of ≥3 and ≥5 events per user. BG testing frequency significantly increased from 2 times daily within the first 30 days to 4.5 times daily after 180 days of app use (P<0.0001).
Conclusion:
App use over 180 days significantly reduced the chances of hypo/hyperglycemic events. Users were more engaged in diabetes management as demonstrated by more frequent BG testing. These results suggest that app use supports active involvement in diabetes self-management, which may lead to improved glycemic outcomes.
Microdialysis Testing for Multi-analyte Monitoring
University of Connecticut, Department of Pharmaceutical Storrs, CT, USA
Objective:
This research focuses on expanding beyond glucose to examine other analytes to facilitate accurate insulin dosing. The objective is to develop a robust in vitro and in vivo microdialysis method to detect multiple analytes at standard interstitial concentrations.
Method:
A CMA20 microdialysis probe with a 20kDa molecular weight cutoff was used. In vitro microdialysis was performed at five different concentrations for the test analytes: glucose; glycerol; and lactate. The dialysate samples were collected every 10 minutes for 2h. Glucose, lactate and glycerol dialysate were detected using enzymatic assays. Standard test analytes were measured to calculate % recovery. For the in vivo studies, the microdialysis probe was inserted subcutaneously into Sprague Dawley rats. Blood was collected from the lateral saphenous vein (fed state) prior to administering Ringer’s solution through the microdialysis probe. The recovery (%) of glucose in the microdialysis samples was calculated using the formula (Cout/Cin)*100, where Cin is the glucose concentration in the fed state and Cout is the glucose concentration of individual samples collected.
Result:
In vitro, recovery for all concentrations of glucose, lactate and glycerol were observed to be between 80-100%, 40-75% and 30-50%, respectively. In the mixture containing glucose, lactate and glycerol (5.5 mmol/L of each analyte), the recovery (%) was 100-110%, 40-60% and 30-60%, respectively in 2h.The recovery (%) for both glucose and lactate were observed to be directly proportional to their in vivo concentrations. The microdialysis data indicated a decrease in concentration of both analytes between the fed and fasted state.
Conclusion:
Microdialysis was successfully used to analyze glucose, lactate and glycerol from a mixture in vitro and to track glucose and lactate concentrations in vitro and in vivo.
Sterilization of Implantable Glucose Biosensors
Department of Pharmaceutical Sciences, University of Connecticut Storrs, CT, USA
Objective:
The main objective of this research was to investigate the effect of two terminal sterilization techniques: gamma and ethylene oxide sterilization on the stability of the various coatings of miniaturized, implantable glucose sensors. In addition, the effect of sterilization to the overall in vitro performance of the glucose-sensing element was investigated post sterilization.
Method:
Sterilization of individual composite coatings, the various components of the coatings, and the fully-constructed glucose-sensing element of the implantable sensor was investigated using both sterilization techniques. The sterilized samples were compared with non-sterilized controls. The coatings containing dexamethasone-loaded PLGA (poly lactic-co-glycolic acid) microspheres in whole as well as the individual components of the coatings were characterized for drug loading, morphology, glass transition temperature, and in vitro drug release. The glucose-sensing element was characterized in vitro for its sensor linearity before and after sterilization.
Result:
Dexamethasone crystals showed significant peak shifts in X-ray diffraction patterns post gamma sterilization. PLGA microspheres underwent significant plasticization as a result of the exposure to ethylene oxide. The dexamethasone drug loading of the PLGA microspheres remained unchanged post sterilization using both techniques. The glucose-sensing element exhibited a small loss of its linearity post gamma sterilization whereas the linearity was maintained post EO sterilization.
Conclusion:
In order to facilitate advancement of fully-implantable glucose biosensors to the clinic, the stability of the various materials and coatings along with the glucose-sensing linearity post sterilization was successfully evaluated for the first time.
How Much Do We Gain from Greater Personalization?
University of Canterbury Christchurch, Canterbury, NZ
Objective:
STAR (Stochastic TARgeted) is risk-based glycemic control (GC) using prediction of future insulin sensitivity (SI) variability to safely dose insulin and nutrition, where SI variability is the key driver in GC difficulty and hypoglycemia. Currently, STAR uses a 2D stochastic model where current identified patient-specific SI is used to predict future SI variability in a cohort-specific sense. This study assesses the impact on GC performance of a new, more patient-specific 3D stochastic model, using previous and current SI values to predict metabolic variability.
Method:
Bi-variate and tri-variate Gaussian kernel density methods were used to estimate conditional probability estimation of future SI knowing current SI (2D model) and also previous SI (3D model). Models were built randomly using 411 (70%) of retrospective GC episodes. They were tested using clinically validated virtual trials on the 176 (30%) remaining patients, repeating 3 times (n=528 episodes). Safety, performance, and workload were compared.
Result:
Out of the total 528 simulated episodes, workload was similar (11.6 measures/day). Performance was similar (90% in 80-145mg/dL band), but tighter for the 3D model (78% vs 74% in 80-125mg/dL band). Median blood glucose (BG) level was lower for the 3D model (108 [99, 120] vs. 113 [103, 124]mg/dL), with higher insulin (3.0 [1.5, 5.0] vs. 2.5 [1.5, 4.0] U/h) and nutrition (99 [66, 100] vs. 92 [70, 100] % goal feed). Safety was very slightly better for the 2D model (2% vs. 3% BG<72mg/dL; 1% vs. 1.4% BG<40mg/dL).
Conclusion:
The new, more personalized 3D stochastic model provides moderately improved performance and similar safety and workload. Overall, the results suggest greater personalization in predicting variability can improve STAR GC performance and justify implementation to see if it improves outcomes.
Insulin-Only STAR: Liège Clinical Trial Interim Results on Safety and Efficacy
University of Canterbury Christchurch, Canterbury, NZ
Objective:
Stress-induced hyperglycemia and insulin resistance are common in critically ill patients and are associated with an increased risk of adverse outcomes. STAR (Stochastic TARgeted) glycemic control (GC) has proven effective over different units and clinical practices. However, this risk-based dosing approach uses both insulin and nutrition to control glycemia, whereas virtually all other approaches use insulin only. This study uses STAR with insulin only (i.e., with nutrition set clinically) in the University Hospital of Liège, Belgium to assess the safety and efficacy of this technique.
Method:
STAR-Liège is an insulin-only version of STAR targeting 80-145mg/dL. Patients were included if two successive blood glucose (BG) measurements were >145mg/dL. Insulin was administered continuously through an IV catheter and nutrition was determined clinically. GC was stopped after 72h or if BG was stable at an insulin rate ≤2U/h. Safety was assessed by the %BG in severe (<40mg/dL) hypoglycemia, mild (<72mg/dL) hypoglycemia, and >180mg/dL hyperglycemia. Performance was evaluated by the %BG within the target band and the median BG. Clinical data from the first 11 patients was analyzed totaling 645 hours of control. Ethics approval was granted by the University Hospital of Liège Ethics Committee.
Result:
The insulin-only STAR-Liège protocol showed high performance, with a median [IQR] BG of 122 [106, 147] mg/dL and 78% BG in target band. Mild hypoglycemia occurred 1.6% of time, but there was no incidence of severe hypoglycemia. Additionally, only 9.8% of BG were >180mg/dL and administered insulin and nutrition was 4.0 [1.8, 4.6] U/h and 8.1 [4.9, 9.2] g/h.
Conclusion:
Insulin-only GC with the STAR-Liège protocol provided equally high control of safety and quality for all patients. These results are encouraging, comparable to previous studies, and supportive of the STAR risk-based dosing approach as a robust solution across different ICU setting, usages, and support continuation of the clinical trial.
Comparing the Glycemia Kinetics of Children and Adults to Adapt Predictive Low-Glucose Suspend
CEA LETI, MINATEC Campus Grenoble, France.
Objective:
Glycemia regulation models are generally developed for adult patients. In particular, predictive low-glucose suspend systems are tailored for adults’ glycemia kinetics. If the triggering parameters of these systems, such as the glycemia and/or time window are incorrect, then basal suspension may occur too late to prevent hypoglycemia. To verify if existing smart insulin pumps are suitable for children with type 1 diabetes (T1DM), this study compares the glycemia kinetics of adults and children patient groups.
Method:
A first dataset, provided by the Catholic University of Leuven (KUL), contained continuous glucose measurements and mealtimes from five children, aged 2-15 years, with T1DM under sensor augmented pump therapy. A second dataset, from the Diabeloop WP6.2 study, contained the same measurements from 35 adults with T1DM, during four weeks with open-loop regulation at home, and three days with closed-loop regulation in a hospital setting. A third data, from the Diabeloop WP7 study, contained the same measurements from 29 adults with T1DM, during 7 to 84 days with closed-loop regulation, in free-living conditions. Several metrics were computed for each dataset, to evaluate the patients’ glycemic balance and variability.
Result:
While adults and children have similar glycemic balance metrics, statistical tests show that their glycemic dynamics are significantly different. In particular, the children’s average negative rate-of-change (anGRC) is faster than the adults’ one: the median anGRC is -0.97 mg/dL/min for children and -0.76 for adults from the WP6.2 study in open loop. Such differences are even more evident after meals.
Conclusion:
Based on glycemic monitoring of 5 children with T1DM, the average negative rate-of-change is significantly faster for children than for adults.
Performance Comparison of the Assure® Platinum and Gluco NaviiTM Blood Glucose Monitoring Systems for Use in Professional Healthcare Settings against the ISO 15197:2015 Accuracy Criteria
ARKRAY USA, Minneapolis, MN, USA
Objective:
Blood Glucose Monitoring Systems (BGMS) are important tools utilized in professional healthcare settings for the management of diabetes mellitus. BGMS readings must be accurate to prevent potential microvascular and macrovascular complications due to uncontrolled blood glucose levels. ISO 15197:2015 is an accepted standard for measuring the accuracy of BGMS. It requires that 95% of results are within ±15mg/dL of the reference analyzer at glucose concentrations <100mg/dL and within ±15% of the reference analyzer at glucose concentrations ≥100mg/dL. Furthermore, 99% of results need to be within the A and B zones of the Consensus Error Grid. This study compared the performance of the Assure® Platinum and Gluco NaviiTM BGMS.
Method:
Two lots of test strips for each BGMS were evaluated side-by-side at ARKRAY in Minneapolis, MN with the same study participants. Blood samples were drawn from the fingertip of participants with confirmed diabetes (n=120) by laboratory professionals. Reference values were obtained using the YSI Model 2300 Analyzer. Data were evaluated against the accuracy boundaries of the ISO 15197:2015 standard and Consensus Error Grid.
Result:
Assure® Platinum demonstrated 100.0% of <100mg/dL samples (14/14) were within ±15mg/dL of reference and 96.2% of ≥100mg/dL samples (102/106) fell within ±15% of reference. Overall bias was -3.6% and correlation coefficient (r) = 0.98. For Gluco NaviiTM, 100.0% of <100mg/dL samples (14/14) provided values ±15mg/dL of reference and 96.2% of ≥100mg/dL samples (102/106) fell within ±15% of reference. Overall bias was -1.8% and correlation coefficient (r) = 0.98. All data for the Assure® Platinum and Gluco NaviiTM fell within the A and B zones of the Consensus Error Grid.
Conclusion:
Assure® Platinum and the Gluco NaviiTM had equivalent performance when assessed against the ISO 15197:2015 accuracy boundaries.
Performance of the Assure® Prism Multi Blood Glucose Monitoring System throughout Shelf-Life
ARKRAY USA Minneapolis, MN, USA
Objective:
Blood Glucose Monitoring Systems (BGMS) need to provide accurate results throughout their product life-cycle. ARKRAY employs a rigorous quality testing program that evaluates lots over the entire product life-cycle (24 months) and post-expiration. This study evaluated the performance of the Assure® Prism Multi BGMS throughout Shelf-Life against the accuracy boundaries of ISO 15197:2015 (the global standard for accuracy).
Method:
Three lots of test strips were evaluated throughout the product shelf-life and post-expiration at the ARKRAY factory. Blood samples were drawn directly onto test strips from the fingertip of individuals with diabetes (n = a minimum of 30/time points equaling a total of 241 samples) by laboratory professionals. Reference values were obtained using the YSI Model 2300 Analyzer. Data were analyzed against the accuracy boundaries of the ISO 15197:2015 Standard and the Consensus Error Grid. In addition, average bias with 95% confidence intervals (CI) throughout shelf-life were calculated.
Result:
When all data were combined, (all three lots at all time points, including those 5 months beyond expiration), 100% (19/19) of results <100 mg/dL fell within ±15 mg/dL of reference and 95.9% (213/222) of results ≥100 mg/dL fell within ±15% of reference. 100% of the values were within the A and B zones of the Consensus Error Grid. Average bias throughout shelf-life and post-expiration for all lots combined was 2.1% (95% CI of 1.3% to 3.0%) with the correlation coefficient (r) = 0.99. The results of each lot individually were also within the ISO requirements.
Conclusion:
Data collected on the Assure® Prism Multi BGMS performed within the accuracy boundaries of the Consensus Error Grid and the of the ISO 15197:2015 Standard. The lots demonstrated consistent performance throughout their product life-cycle and post-expiration.
Usability Evaluation of a Connected System to Link Patient Reported Lifestyle Data into Electronic Health Records for Diabetes Self- Management Education and Support: A 2- Week Pilot Study
UT Health San Antonio San Antonio, TX, USA
Objective:
We developed an interface within Chronicle Diabetes, a national electronic health record system for diabetes education to link patient self-monitoring data from patients’ smartphones and fitness trackers to their diabetes educators to facilitate diabetes self-management education and support. The goal of this study is to evaluate the usability of the system in real world practice settings before deployment in a multisite clinical trial.
Method:
A 2-week real-world usability evaluation study was conducted with four patients with type 2 diabetes recruited from a diabetes education program. Each patient was given a Jawbone UP24 wristband and a smartphone application for recording physical activity and dietary behaviors. After 2 weeks, their diabetes educator reviewed the information in Chronicle Diabetes and gave feedback on their progress. Participants also completed a survey that was adapted from the System Usability Scale, a 10-point Likert scale about the usability of the UP24 wristband and app, as well as their thoughts about sharing data with their diabetes educator.
Result:
After 2 weeks, patients agreed that they would like to use the Jawbone UP24 wristband to help them with diabetes (mean = 8.50 ± 1.0). They strongly agreed that the interface would help them to better communicate with their diabetes educator (mean = 9.00 ± 0.82) and that the interface helped them to set realistic behavioral goals (mean = 9.50 ± 0.58). They also strongly agreed that the interface helped them to receive more personalized care from their educator (mean = 9.25 ± 0.50) and that they felt comfortable knowing their diabetes educator could check on their progress using the interface (mean = 9.50 ± 0.58). When asked whether they felt the system was invasive or intrusive, they strongly disagreed (mean = 1.50 ± 0.58). Some usability issues were discovered regarding the mixing of patient accounts.
Conclusion:
The survey findings supported the high usability of the connected system as perceived by patients. Conducting real- world usability evaluations can identify issues not encountered in usability testing performed only in a lab setting.
Use of Connected Information Technology in A Community Clinical Collaborative Care Model for Uninsured Mexican American Patients with Uncontrolled Diabetes: Lessons Learned
UT Health San Antonio San Antonio, TX, USA
Objective:
Although Information technology (IT) growth presents opportunities to enable community clinical collaboratives beyond electronic health record, it is particularly challenging for low-income Mexican Americans with uncontrolled diabetes and no health insurance. The objective of this study is to present lessons learned in deploying connected information technology systems through health information exchange (HIE) to support connections between a dozen community partners and clinics in a collaborative care model involving social workers and community health workers (CHWs).
Method:
Individuals with A1c ≥ 8% were eligible to be enrolled in the program and received Diabetes Self-Management Education (DSME) provided by CHWs. CHWs were integrated into the care continuum, along with the mental health authority, community-based service organizations, clinics and hospitals, and have expanded service delivery to homes- and community-based settings. A connected system was adapted from a national diabetes education documentation system during the beginning of the study to facilitate the documentation of DSME along with other surveys and metrics. The system was later changed to a population health management tool supported by HIE to better facilitate connections with clinics. Spanish versions of both tools were developed to accommodate population needs.
Result:
The project enrolled 5,579 patients over 5 years. Lessons learned from the implementation of the connected system in addressing the project needs to connect community and clinical partners can be categorized into its benefits and lessons. Lessons included: 1. No existing IT tool, 2. Cost can be prohibitive, 3. Had to create workarounds for data cleaning and data management. 4. Time consuming for system set up, programming, waiting to access data, and incomplete control of the data that were entered into the system, 5. Technical issues can be disturbing, 6. CHWs may still use paper entry and enter data after visits. Benefits: 1. Forced communication among partners, 2. Flags and prompts in clinic electronic medical record systems to initiate and re-engage patient participation. 3. Provided tools for CHW for follow-up every 3 months and facilitated the efficiency of CHW with large caseloads. 4. Facilitates the implementation of a complex intervention that allow individualized care based on individual needs. 5. Allows for the use of validated metrics into delivery of standard clinical care, 6. Care coordination across different clinical and community sectors build upon each other to facilitate standard data sharing. 7. Created interdependence among the different stakeholders based on the value-based care model, 8. Facilitates communication, interdependence, and shared responsibilities for independent financial model. 9. Facilitates goal setting for DSME, 10. Allows individualization, notes, motivational interviewing, longitudinal follow-up to allow staff turnover and case workload shift in an easier way.
Conclusion:
Despite the technical and language barriers encountered, a connected information technology system based on health information exchange can function as a population health management tool to link community and clinical partners in providing collaborative care for underserved diabetes patients in the Mexican American community.
Evidence Based Continuous Probability Estimation for Hypoglycemic Excursion from Sparse Blood Glucose Data
mySugr GmbH Vienna, Austria
Objective:
The high data density inherent to Contiguous Glucose Measurements (CGM) allows the utilization of the Ambulatory Glucose Profile to depict the risk of glycemic events throughout the course of a day. With traditional Blood Glucose Measurements (BGM), the density of data is far lower and thus precludes the proper application of such methods. In this study we explored a new method that enables a continuous probability estimation of glycemic control even in cases of sparse data.
Method:
The proposed model used a Bayesian inspired periodic kernel density estimation method to model glycemic event probabilities throughout the day. Parameters were chosen based on a-priori knowledge of glycemic excursion duration. Distributions were estimated for 70 randomly selected users of the mySugr App based on 14 days of coherent CGM data traces. Model performance and data constraints were then assessed using artificially subsampled datasets.
Result:
Examination of the posterior probability distribution error revealed a reduction in root sum square (RSS) of approximately 75% by increasing the measurement frequency from 1 to 4. However, further increasing the frequency to 10 gave a comparably small effect (RSS=4.4%). We suggest a minimum of 14 days of data in order to achieve a reasonable tradeoff between estimation accuracy and data constraints.
Conclusion:
The developed model can be used to obtain continuous probabilities of glycemic excursions throughout the course of a day based on sparse glucose data. In contrast to frequency-based approaches, normalization by evidence reduces the bias introduced by varying measurement frequencies between days. The method further indicates areas with insufficient data to make a reliable glucose prediction. A minimum of 4 measurements per day over 14 days seems a reasonable choice between performance and data constraints.
Performance of a Glucose Sensor with an Alternative Sterilization Method
Medtronic Diabetes Northridge, California, USA
Objective:
Enzyme-based glucose sensors may be sensitive to ethylene oxide (EtO) gas exposure and other processing conditions associated with a typical EtO sterilization cycle. Thus, the effect of EtO sterilization on Medtronic continuous glucose monitoring (CGM) sensors was evaluated to assess its impact on sensor performance while maintaining sensor sterility.
Method:
A preliminary clinical feasibility study involving patients with type 1 and type 2 diabetes evaluated the performance of Medtronic sensors sterilized with unoptimized EtO sterilization parameters against a control group of CGM sensors undergoing a standard radiation-sterilization method. Performance was assessed by the mean absolute relative difference (MARD) between sensor glucose (SG) and self-monitoring of blood glucose (SMBG) values.
Result:
The overall MARD for the sensors that underwent EtO sterilization with unoptimized parameters was 10.80% compared to that of 10.07% for the control group. As next steps, the EtO sterilization process is being further characterized using the design of experiments (DOE) approach to identify optimum sterilization settings.
Conclusion:
Data from the preliminary feasibility study with unoptimized EtO parameters suggest that Medtronic CGM sensors retain acceptable performance after EtO sterilization. Upon completion of the DOE method to identify optimum sterilization settings, a follow-up feasibility study will be conducted to further evaluate sensor performance with EtO sterilization.
Using CGM Measurements for Glycemic State Analysis in Infants
Department of Mechanical Engineering, Centre for Bio-Engineering, University of Canterbury, Christchurch, Canterbury, NZ
Objective:
Continuous glucose monitoring (CGM) devices offer a way to continuously monitor patient conditions including helping to detect neonatal hypoglycemia in at-risk infants. Minimizing neonatal hypoglycemia is an important goal because its occurrence may result in adverse neurologic outcomes in later life. However, uncertainty remains as to whether hypoglycemia is a universal or relative state in a patient. CGM also provides insight into the underlying glycemic and metabolic state. Characterizing glycemic states can be easily done visually, but no simple, explicable, and clinically-relevant algorithm exists. This study presents a method to algorithmically characterize glycemic states and detect state changes.
Method:
The algorithm was developed using a cohort of 366 infants monitored up to 48 hours and comprising 12,356 hours. State changes were identified from the intersection between the average of the entire CGM trace and a 6 hour rolling average of the CGM trace. Clinically relevant defined conditions for a given state included a 5-hour minimum per glycemic state and a 5.4 mg/dL minimum difference between adjacent states. These two conditions can be changed to meet clinical conditions.
Result:
The state change algorithm that was developed generally matched expectations from visual analysis. The majority of infants had less than 2 state changes in the first 48 hours (279/366; 76%). The median number of state changes per day was 0.68 [IQR: 0.60, 1.14] while median absolute change in interstitial glucose per state change was 10.8 [IQR: 7.2, 16.2] mg/dL.
Conclusion:
Glycemic states and state changes were algorithmically determined in CGM data using a simple, explicable, and clinically-defined algorithm. Future use includes associating glucose states and state changes with clinical outcomes, based on the clinical inputs, for potential new diagnostics.