Abstract
Effect of Mobile Application Usage on TIR and TAR within People with T1D and T2D: A Pilot Cohort Retrospective Study
Undermyfork Inc. Hermosa Beach, CA, USA
Understanding why the blood glucose levels are changing is often difficult for patients with diabetes. Combining postprandial continuous glucose monitoring (CGM) data with meal photos may assist in self-management and help reduce glycemic variability. We aimed to evaluate how a visualization of postprandial time in range (TIR) data via a mobile application impacts glycemic stability.
For this study, we chose the application users with type 1 (T1D) or type 2 diabetes (T2D) and a time in range (TIR) of <70%. The intervention was done via access to a photo-based food diary and CGM data history. We assessed the patients’ blood glucose parameters at the beginning of application usage and after 14 days. To exclude TIR improvement due to CGM use per se, we collected the preceding glucose data of each user in a control cohort.
Users (n=21) demonstrated significant improvements in TIR (11.0%±5.0, P=.001) and time above range (TAR, -12.0%±5.0, P=.001), with no significant reduction in time below range. A digital collection of meal photos connected to postprandial CGM data could help improve TIR and TAR.
A photo-based diabetes management application with visualized CGM postprandial data may improve postprandial TIR and TAR in individuals with T1D and T2D. Thus fluctuations in their blood glucose levels decrease. Further research is needed to evaluate TIR and TAR changes caused by the regular use of the digital food diary combined with CGM data.
Global Surveillance of Glucometer Interferences and Precision
Cavendish Laboratory University of Cambridge, Cambridg, United Kingdom
The reliability of glucometers is investigated pre-market, but post-market surveillance is not usually conducted. Earlier studies found that medical treatments can affect the measurement of some post-market glucometers in the United States. No study has investigated the issue at the global level yet. Here we designed a study to investigate the interference and precision of commonly available glucometers in 19 countries.
In each country, students and researchers will study three of the most common glucometers. Two plasma pools will be prepared to obtain normal and low blood sugar concentrations. Ascorbic acid (vitamin C), maltose, and acetaminophen as interferants will be tested per pool at the following concentrations: one sub-therapeutic, two therapeutics concentrations, and one at the over-dose range. Conducting the experiments will start in July 2022, and the majority of countries will finalize the experiments by October 2022.
A total of 48 professors from 25 global universities have been confirmed to enrol in experimenting. Experiment protocols have been optimized to meet the different conditions in different countries to match the different available resources in the 19 countries across 6 contents. We expect to see similar trends across the nearby regions, and larger variability across continents.
We have designed and organized the first global surveillance study to investigate post-market glucometer interference. The study aims to establish the variability in post-market glucometer reliability and global variability of the most commonly used ones to better guide commercial and clinical decisions to improve the well-being of diabetic patients.
Healthcare Professionals’ Beliefs, Experiences, Attitudes and Knowledge of Using Diabetes Mobile Applications in Supporting Diabetes Self-Management
Albaha University, Albaha, Saudi Arabia University of Birmingham, Birmingham, United Kingdom.
This study aimed to explore healthcare professionals’ beliefs, experience, attitudes and knowledge with regard to patients’ use of mobile applications in supporting diabetes self-management.
The TAM model was utilised to develop a web-based questionnaire. This questionnaire was administered online through different social media platforms and targeted Saudi Arabian healthcare and work directly with people with diabetes. Ethical approval was obtained and data collection started in February 2022.
Of the 89 survey participants, only 71 were eligible. Their various roles included physician (15%), pharmacist (22%), nurse (13%), dietician (25%) and diabetes educator (10%). Of those, only 30% had recommended the use of a mobile app to their patients. The reasons for the recommendation included that the apps are helpful for patients (46%), easier for counting carbohydrates and dose calculations (27%), cost-free (18%), and makes it possible to track patients’ health (20%). However, several reasons prevented healthcare professionals from recommending the apps, including that they have no effect (8%), they have no information about diabetes apps (25%), the effectiveness of the apps is not evidence-based (6%), they are busy and have no time to make recommendations (3%), and there are numerous apps and they do not know which one is suitable (18%).
A limited number of healthcare professionals recommend mobile applications for their patients. Further effort should be made to design an application that meets users’ expectations and fixes the issues that cause concern.
Patients’ Beliefs, Experiences, Attitudes and Knowledge of Using Diabetes Mobile Applications in Supporting Diabetes Self- Management
Albaha University, Saudi Arabia University of Birmingham, United Kingdom.
This study aimed to explore patients’ beliefs, experiences, attitudes and knowledge of using diabetes mobile applications in supporting diabetes self-management.
The AADE7 self-care, MARS and TAM models were utilised to develop a web-based questionnaire. This questionnaire was administered online through different social media platforms and targeted the Saudi Arabian population with diabetes. Ethical approval was obtained and the data collection started in February 2022.
Of the 277 survey participants, only 179 were eligible. The majority of participants (62%) installed at least one application to help self-manage their diabetes, while 38% did not install any applications. Only 11% installed these apps following professional health care advice, while most reported different advice sources. Of the total number of participants, 24.5% used mobile applications more than once a day. Mobile applications were used for longer than 12 months by 20% of participants. Most application users faced issues while using the applications, such as application crashes, inaccurate results and problems in units. Regardless of these issues, on a 5-point scale, the majority considered the mobile applications to be very useful (Mean 4.38 SD 0.84). There were many reasons behind not using applications among participants.
Most participants used the mobile applications and considered them to be very useful regardless of the issues that some faced. Further efforts should be made to design an application that meets users’ expectations. Awareness needs to be raised for people with diabetes with regard to applications.
LIFELEAF: A Novel Device and Platform for Non-Invasive Continuous Measurement of Blood Glucose
AMCR Institute Escondido, California, USA
Optimal diabetes management requires frequent glucose measurements, most often performed via capillary fingerstick blood glucose monitors (BGMs) or minimally invasive continuous glucose monitors (CGMs) sampling interstitial fluid. While more convenient, economical, and non-invasive methods of glucose monitoring are needed, many previous attempts at non-invasive assessments have not produced sufficiently reliable results. In this study, we demonstrate the feasibility of performing non-invasive glucose measurements within a clinically relevant dynamic range of 70-240 mg/dL.
The LIFELEAF® wristwatch developed by LifePlus, Inc. amplifies glucose absorption peaks in the near infra-red spectrum via novel algorithms applied to photoplethysmography (PPG) signal from an off-the-shelf sensor. Device calibration is performed before shipping against CGM reference prior to device reporting of predicted glucose readings via machine learning models. A total of 164 subjects (diabetic and non-diabetic) contributing 2,838 paired datapoints were assessed in the first protocol comparing the wristwatch to BGM measurements. The second protocol assessed 66 patients with Type 1 diabetes and Dexcom G6 CGM against the wristwatch, with 48,333 total paired datapoints collected over 20 days. All datasets were split with 70% of paired data used for training and the remaining 30% for prediction.
All wristwatch-reported predicted glucose values fell within zones A and B of the Clarke Error Grid when compared against BGM- or CGM-based references in the 70-240 mg/dL dynamic range, with most values falling within zone A. Mean Absolute Relative Difference (MARD) was 10.6% against CGM reference, and 13.3% against BGM reference.
This study demonstrates the feasibility of conducting meaningful non-invasive continuous glucose measurements via a wristwatch-based PPG sensor within a clinically relevant dynamic range, facilitated by novel algorithms and specialized deep learning.
The First Real World Experience with Bigfoot Unity: A Six-Month Retrospective Analysis
East Alabama Endocrinology Opelika, Alabama, USA
The Bigfoot Unity® Diabetes Management System is a recently FDA-cleared smart pen cap system incorporating CGM data (Abbott FreeStyle Libre 2), real-time alerts, and clinician-directed dose recommendations. The objective was to analyze real world, 6-month use data for those with insulin requiring diabetes using Bigfoot Unity.
A retrospective analysis was performed using deidentifed data and a prespecified analysis plan. A cohort (N=57 at 13 clinics) managing MDI with Bigfoot Unity for >180 days is reported. At least 50% of CGM data were required in the first 2 weeks and in the 6th month of system use. An HbA1c (A1C) prior to starting Bigfoot Unity was obtained from clinic medical records (N=50).
The mean age was 61.7 years and 84.2% had T2D. Most used CGM previously (75.9%) while nearly all (94.4 and 98.1%) were new to using diabetes apps and smart pens, respectively. Mean A1C prior to Bigfoot Unity use was 8.4±1.8%. Mean GMI in the 6th month of use was 7.3±0.8%. In a subset of those with baseline A1C >8% (N=29), the mean A1C was 9.4±1.7% and GMI after 6-months was 7.5 ± 0.8%. TIR was 67.0±20.2% and 64.2±20.5% with TBR of 1.9±3.1% and 1.5±2.4% at 2 weeks and 6 months use, respectively.
For people using MDI with suboptimal glycemic control, using the Bigfoot Unity System has the potential for rapid and durable glycemic improvement. Considering the cohort’s older age, results demonstrate close adherence to established glycemic targets including a relatively short time spent in the hypoglycemia range. Limitations include the retrospective analysis, relatively small sample size and lack of direct adverse event data.
A New Competency Model for Patients with Diabetes
Humaginarium LLC Oak Park, Illinois, United States
An evidence-based competency model for adult patients facing prediabetes, diabetes, complications of diabetes, and related metabolic disorders (e.g. cardiometabolic syndrome).
The ’job” of every patient facing diabetes is to avoid, prevent, and mitigate disease with a plan of self-care and timely, affordable access to patient-centered healthcare. Yet no competency model exists for doing this job. To create one, it was necessary to cull practical knowledge, skills, attitudes, and techniques that every adult should have from the competency models of diabetes educators and medical practitioners, and reframe them for autonomous patients and others at risk who don’t identify as patients.
The CHC Model for Metabolic Disorders is a comprehensive, evidence-based index of what every adult patient (and others at risk) should be able to do for themselves in order to perform their job successfully. “Every patient” includes all ages, genders, races, literacies, communities, and socioeconomic classes; it also means autonomously — as in not being told what to do or how to do it — because competence is a product of self-determination rather than habit or conditioning.
The draft Model reported in this session is a framework for empowering patients of all stripes to control and improve the determinants of their metabolic health. The Model references determinants of health as a dynamic, nonlinear system that is everywhere experienced yet poorly understood and remembered by ordinary people even after they participate in diabetes education. The Model was developed to support novel healthcare simulation for patients, in which heuristic experience makes the system seem more accessible and addressable, less scary and overwhelming, more open to control and improvement.
A Novel Infusion Failure Detection Paradigm Based on Fluid Pressure and Supervised Learning
Diatech Diabetes, Inc. Memphis, TN, USA
To develop a novel insulin infusion failure detection model for pump therapy using fluid pressure, exploiting supervised learning techniques for binary classification.
Three independent preclinical animal studies were conducted over a period of 24 months. Six non-insulin dependent female swine were anesthetized and infused using a 50/50 mixture of contrast and saline. In total, 503 boluses across 142 sites were administered and viewed under fluoroscopy using a Siemens Artis Pheno. Fluid pressure and flow rate were collected for every infusion using in-line sensors spliced into the infusion set. Infusions were performed using a MiniMed Paradigm (n=5), Harvard Apparatus Pump 11 Elite (n=131) and Medtronic 770G (n=289). Post study, 425 boluses (20.7 uL ± 3.11 uL) were accepted and manually labeled as normal (n=299) or malfunction (n=126), using the available x-ray scans. All malfunctions were observed within 30 mins of insertion. Python scikit-learn library was used to train/test (70/30) Support Vector Machine (SVM), Random Forest (RF), and Logistics Regression (LR) models. Hyperparameter tuning was done using GridSearchCV function with 10-fold 3 repeats stratified k-fold cross validation, F2-measure was used for scoring. Adaptive synthetic sampling was performed during training on minority class.
From the fluid pressure data, 221 features were created; later reduced to 28 using recursive feature elimination. Overall, the best classification performance was obtained with SVM showing 96.9% accuracy on the test set, Matthew’s Correlation Coefficient of 0.927 and weighted F2-measure of 0.97. This was followed by LR (95.3%, 0.893, 0.952) and RF (93.7%, 0.857, 0.94).
This study showed that bolus infusion malfunctions can be detected with high accuracy using fluid pressure data only, without the need for elevated blood glucose levels.
Smart Insulin Therapy Initiation In Insulin-Naive Subjects With Type 2 Diabetes Using Machine Learning
Department of Information Engineering, University of Padova Padova, Padova, Italy
Insulin-naive subjects with type 2 diabetes (T2D) are usually initiated to basal insulin therapy, starting from a low initial insulin dose (IID) that is progressively adjusted to reach the optimal insulin dose (OID) and glucose target (OTG). This procedure is time consuming, especially for very insulin resistant subjects. To speed up the achievement of the OID, it would be useful to guess, before insulin titration, if subject-specific OID will be high or low to roughly estimate a smarter IID (e.g. IID=10U for low, 44U for high).
We used 300 in-silico insulin-naive subjects of the T2D Padova simulator, who underwent a 52-week basal insulin titration trial, and for which demographics, fasting and post-prandial glucose, insulin and C-peptide concentrations were available.
We first classified each subject’s OID as high or low. Then, we tested three bootstrapping LASSO logistic regression models to predict OID class using body weight, sex, age, fasting and postprandial glucose, insulin and C-peptide (Model 1); same covariates of Model1 apart from postprandial concentrations (Model 2); same covariates of Model 2 apart from hormones measurements (Model3). Models were trained and tested in 70%-30% ratio. Model performance was assessed based on the area under the Receiver Operating Characteristic curve (AUC-ROC).
Model2 well predicted OID, with performance comparable to Model1 (AUC-ROC=92.74% and 94.22%, respectively),despite Model1 would require postprandial measurements of glucose, insulin and C-peptide. Performance of Model3, which would have the advantage of not requiring any hormone measurements, is a bit worse but still acceptable (AUC-ROC=88.4%).
Depending on the availability of fasting and/or post-prandial plasma glucose and hormone concentrations, one can use one of the three models to guess the IID to rapidly reach the OGT.
Diabetes Phenotypes Quantified via a Physiofunctional Model Fitted to Raw CGM Time Series Data
Teladoc Health Mountain View, California, USA
The aim for this work is to establish a robust quantitative definition of diabetes phenotype—type 1 (T1D) vs. type 2 (T2D)—with the additional constraint that raw continuous glucose monitoring (CGM) data is the only input. Further, the phenotypic definition should incorporate the key physiological mechanism(s) underlying the observed glucose dynamics—in this case, glucagon and insulin.
A computational glucose model was developed to reproduce the glucose dynamics observed in CGM recordings (6-10 days long), which were obtained from a random sample of people with diabetes
KS tests revealed multiple statistically significant differences between the parameters for diabetes subpopulations (T1D vs. T2D). Of note were the parameters P and y—the mean-centered and gradient-centered glucose sensitivity, respectively. Both were significantly higher for the T2D subpopulation (p=0.006, p=0.002).
The statistically significant parameter differences reveal phenotypic differences in the physiological mechanisms underlying glucose dynamics in people with T1D and T2D. Moreover, this result illustrates an important use case for “physiofunctional” modeling techniques—namely that the functional importance of specific physiological mechanisms can be inferred by post-hoc analysis of the model parameters. Further, this work paves the way for future experiments to leverage physiofunctional models to produce counterfactual forecasts—e.g., for medication optimization/adherence algorithms.
Smartphones for Telehealth: Can We Document Diabetic Complications?
Case Western Reserve University Cleveland, Ohio, United States
Can smartphones document diabetic complications for telehealth? Issues of concern are diabetic eye disease, diabetic peripheral vascular disease, diabetic skin and feet problems.
Smartphone cameras: Apple (iPhone 11, 12, 13) and Samsung (S22 ultra, S10E) vs in-office Heidelberg Spectralis (Heidelberg, GE).
Parameters: 1) Resolution, 2) Near Focus, 3) Field of View.
External anatomy: eyes, skin (toes, feet, legs)
Heidelberg Spectralis: 1600x1200 pixels at 764 ppi (pixels per inch)
iPhone 11: 1792x828 pixels at 326 ppi
iPhone 12: 2532x1170 pixels at 460 ppi
iPhone 13 Pro Max: 2532x1170 pixels at 460 ppi
Samsung Galaxy S10E: 3040x1440 pixels at 550 ppi
Samsung S22 Ultra: 3088x1440 pixels at 500 ppi
Heidelberg: 15 cm
iPhone 11: 8.2 cm
iPhone 12: 7.2 cm
iPhone 13 Pro Max: 2 cm
Samsung Galaxy S10E: 4.2 cm
Samsung S22 Ultra: 4.2 cm
Heidelberg: 30°
iPhone 11: 54°
iPhone 12: 54°
iPhone 13 Pro Max: 54°
Samsung Galaxy S10E: 66.6°
Samsung S22 Ultra: 36°
Heidelberg’s in-office camera has the highest resolution (greatest ppi) and narrowest field-of-view. Samsung phones are next. However, iPhones have a macro lens that compensates for its lower pixel count.
Smartphone imaging provides excellent photos with great detail, for lesions such as iris lesions, iris bleeding, necrobiosis lipoidica, feet ulcers, etc. As MDs, we can continue to use smartphones for visual examination of the skin, feet, and eyes of our diabetic patients.
Comparison of POCT Devices to Glucose Isotope Dilution Gas Chromatography-Mass Spectrometry Reference Measurement Procedure
Centers for Disease Control and Prevention Atlanta, Georgia, USA
Point-of-care testing (POCT) can be a cost-effective way to screen and identify individuals potentially at risk for diabetes. In addition to POCT devices dedicated for measuring only glucose, other devices that can measure a range of clinical analytes, including glucose, are used in clinical and public health settings. There are concerns about the accuracy of POCT devices, especially at hypoglycemic concentrations. In this study, agreement was evaluated between the CDC glucose isotope dilution-gas chromatography-mass spectrometry reference measurement procedure (RMP), and 2 blood POCT devices that measure lipids along with glucose and are typically used in clinical and public health settings.
Glucose was measured in 20 single donor serum samples covering a glucose concentration range from 23.11 to 376.53mg/dL using the GC/MS RMP and 2 POCT devices suitable for serum testing. Samples were analyzed in 2 replicates over 2 independent runs by the RMP and analyzed in 3 replicates by POCT devices (POCT-A and -B). Accuracy of both POCT devices was compared to reference values obtained by the RMP.
For POCT-A, the mean bias across all samples was 9.45%, and the sample-specific imprecision ranged from 0%-3.03%. For POCT-B, the mean bias was -1.55%, and the sample-specific imprecision ranged from 0%-4.18%. The mean bias for samples <70mg/dL was 16.02% for POCT-A and 2.93% for POCT-B. POCT-A showed a higher positive bias in hypoglycemic samples, which may lead to misinterpreting results. POCT-B was not able to measure glucose in the acute hypoglycemic concentrations range.
Data suggest that the accuracy ofPOCT measurements in the hypoglycemic concentration range needs further improvements. The CDC GC/MS RMP is highly sensitive and specific and is suited to help improve POCT accuracy.
Insulin Quality Monitoring by Infrared-Spectroscopy and Reversed-Phase HPLC Separation Method
South-Westphalia University of Applied Sciences, Interdisciplinary Center for Life Sciences Iserlohn, Germany
Human insulins and their analogs can undergo different degradation processes on a molecular level, when exposed to stress conditions such as temperatures or shear strain, deviating from recommended manufacturer’s storage recommendations. These molecular processes lead to a conformational reorganization of the insulin molecules, followed by irreversible agglomeration and fibrillation, most likely accompanied by decrease in biological potency.
Infrared spectroscopy offers a reliable approach for quantitative and qualitative insulin monitoring without the application of HPLC reference methods with protein digestion, as requested from international pharmacopoeias. Total insulin quantification and determination of misfolded hormone fractions of aged, formulated insulin samples were carried out using infrared spectroscopy. Comparisons were done with reference measurements for original and aged insulins by reversed-phase HPLC with focus on formed aggregates.
Changes in the molecular structure of short-and long-acting insulins were observed after several weeks when stored at 37 °C. When stored at ambient temperatures or just above 0 °C, no significant changes have been found for formulated insulins, but for samples purified by ultrafiltration. An analysis of the insulins secondary structure reveals early molecular conformational changes as identified by IR-spectroscopy. Agglomerates were detected by using a novel HPLC protocol and comparative IR measurements confirming extensive misfolding.
IR-spectroscopy offer a fast and reliable method for quality assurance and quantification of commercial insulins and could replace the current pharmacopoeial methods. Information on insulin aggregates can be obtained by a special rapid separation method (HPLC).
Simulating the Impact of Diabetes Therapy Engagement on Outcomes
Nudge BG, Inc Thousand Oaks, CA, USA
Glucose management activities and behaviors play a very large role in glycemic outcomes for people with diabetes. The level of support for these behaviors varies widely across individuals and cohorts; many people are unable to fully engage in self-management activities. Recruitment bias in clinical trials often results in a more engaged cohort being studied, which means outcomes may not translate to a less-engaged cohort, and positive clinical trial outcomes may be a result of engagement rather than the effects of therapy being studied. Computer modeling and simulation may provide a way to better understand the relationship between engagement and outcomes across therapies.
A composite “engagement” measure was modeled based on empirical data distributions of fifteen key behaviors related to management of hypoglycemia, hyperglycemia, and meals (i.e., someone with low engagement may take more time to respond to a continuous glucose monitoring alarm). A cohort of 500 people with type 1 diabetes (T1D) was simulated for 100 days across four treatment modalities: Multiple Daily Injections (MDI), Automated Insulin Delivery (AID), long-acting insulin+inhaled insulin, and AID+inhaled insulin.
Glycemic and therapy burden outcomes varied by therapy and by level of engagement: MDI was most sensitive to engagement overall, AID was most sensitive to hypoglycemia, and AID+inhaled insulin was least sensitive to engagement overall, with very little change in glycemic outcomes from minimum to maximum engagement.
In-silico evaluation of four diabetes therapies suggest that behavior engagement is a strong determinant of glycemic and therapy burden outcomes. Of the therapies simulated, AID+inhaled insulin was found to yield the most consistent results across all levels of engagement indicating the lowest therapy burden to achieve glycemic outcomes.
Use Of Ultra-Rapid Insulin Lispro with A Full Closed-Loop Insulin Delivery System: An In Silico Analysis
Center for Diabetes Technology, University of Virginia Charlottesville, VA, USA
Effective postprandial control is still one of the main challenges for patients with type 1 diabetes. Insulins with faster pharmacokinetic (PK) and pharmacodynamic (PD) profiles could represent a natural means to tackle this problem, but the high uncertainty associated with carbohydrate counting and prandial dosing time impose barriers to exploiting their glycemic benefits. The aim of this work is to investigate how glucose control performance improves when a full closed-loop (FCL) automated insulin delivery (AID) system is properly adjusted to command a faster insulin analog.
The subcutaneous insulin transport models for insulin lispro (baseline) and ultra-rapid insulin lispro (URLi) are updated in the UVA/Padova simulator using clamp data reported in the literature. The tuning of our FCL AID algorithm (RocketAP) is automatically adjusted according to the PKPD properties of each insulin analog to allow for more aggressive control actions when URLi is commanded. Simulations capturing real-life variability are performed for performance evaluation of RocketAP when switching from lispro to URLi with and without adjusting the control law.
In silico results show that simply switching from lispro to URLi gave a 3 percent boost in the percentage of time in the range 70-180 mg/dL (TIR). However, clinically significance was only achieved when RocketAP was properly adjusted for URLi use, obtaining a change in TIR higher than 5% (overall: 70.2% vs 76.8%; 6am-12am: 62.2% vs 70.2%) with only a slight increase in time below 70 mg/dL (overall: 1.3%vs.1.6%; 6am-12am: 1.3%vs.1.6%).
This in silico analysis suggests that properly adjusting AID algorithms, especially FCL designs, can help leverage the benefits in glucose control of faster insulin analogs.
Open-Access, Light-Based, Wearable Glucose Sensor - a Low-Cost Equitable APS
Centre for Bioengineering, University of Canterbury Christchurch, Canterbury, New Zealand
Current blood glucose (BG) measurement techniques are infrequent, invasive, and painful. Semi-invasive continuous glucose monitors (CGM) are expensive and no non-invasive methods are available. This study presents first clinical results for alow-cost(<NZ$250), non-invasive, light-basedglucose sensor.
Heathy adult (ethics approval: University of Canterbury HEC) subjects drank 330mL of Coca-Cola™ (17.5g glucose). Glucometer and 3 light-based measurements (neck, wrist, finger) were made every 10mins from t=0-60, 90, and 120mins. NICU infant (ethics approval: NZ HDEC South) subjects underwent light-based glucose measurements, using a device measuring glucose spectroscopy in pulsatile blood like an oximeter, at 3 sites (foot, wrist, chest) every time a standard clinical BG measurement was taken. This trial assesses performance in low BG range. Reference and light-based BG values are compared using a modified Clarke Error Grid (CEG). The hand-held sensor design allows full access to raw data and will be freely-available open access.
N=27 subjects (22 neonates; 5 adults) yielded 173 valid BG measurement pairs (1.9-7.9mmol/L). The CEGcontains 98% and 2% in zones A+B and C. MARD was 19%, within clinical accuracy requirements (20%). Bland Altman analysis demonstrated slight overestimation of neonate BG, and slight underestimation for adults using a single calibration for both cohorts.
Results show good performance for a first prototype non-invasive BG monitor at <10% CGM annual cost. Testing athyperglycaemiaremains to be demonstrated and obtaining a good pulse waveform for analysis remains challenging.
The Continuous Glucose Deviation Interval and Variability Analysis (CG-DIVA): An Illustrative Method for Continuous Glucose Monitoring System Accuracy Assessment
Institut fur Diabetes-Technologie, Forschungs- und Entwicklungsgesellschaft mbH an der Universitat Ulm, Ulm, Germany
Comparison of CGM system accuracy is complicated by the heterogeneous and often redundant presentation of CGM evaluation results. We therefore propose a new approach for the characterization of CGM accuracy in replacement of existing parameters: the Continuous Glucose Deviation Interval and Variability Analysis (CG-DIVA).
The CG-DIVA is based on requirements for “integrated” CGM systems released by the FDA. Example performance data from two approved CGM systems were used to validate the approach.
The CG-DIVA analyzes the deviations between CGM and comparator measurements and is composed of two components. One component provides various intervals for different glucose ranges in which the deviations of a CGM system are expected to lie. Using one of the examined CGM systems as example, the CG-DIVA gives the result that 85% of deviations are expected to lie between -13 mg/dL and +40 mg/dL for glucose levels <70 mg/dL. This provides information on system bias and precision as well as the incidence of large errors of particular clinical relevance. The second component of the CG-DIVA characterizes the variability between individual sensors which can, e.g. be useful for comparing manually and factory calibrated systems. The results of the CG-DIVA are presented in an intuitive and straightforward graphical visualization and can be used to highlight clinically relevant differences between CGM systems.
The CG-DIVA provides a comprehensive characterization of CGM accuracy by illuminating the performance in clinically relevant glucose ranges and the variability between sensors. It thus simplifies the discussion and comparison of CGM accuracy and could be used instead of the high number of conventional approaches. Furthermore, it might be a valuable tool in the standardization of CGM performance testing.
Long-Term Glycemic and Patient-Reported Outcomes in Older Adults with Type 1 Diabetes Using Control-IQ Technology
Tandem Diabetes Care, Inc. San Diego, CA, USA
Recent publications have outlined holistic improvements in older adults using closed-loop systems. However, long term outcome data using these systems is lacking, especially in the older population reporting diabetes related complications.
We evaluated diabetes complications, glycemic metrics, and patient reported outcomes (PROs) in older adults (age >65) with T1D using the t:slim X2 insulin pump with Control-IQ technology who had completed first 6 months of the Control-IQ observational (CLIO) study. Participants completed online surveys and uploaded their insulin pump data to the Tandem t:connect web application. Those with >75% overall continuous glucose monitoring use were included in the analysis.
Participants included 148 older adults. Those reporting <3 diabetes-related complications at baseline were analyzed as Group A (n=93, White=93.5%, men=60.2%, prior pump users=91.4%, median baseline HbA1c=7.0 (6.6-7.7)). Group B included individuals with >3 complications (n=55, White=85.5%, men=47.3%, prior pump users=89.1%, HbA1c=7.2 (6.7-7.9)). High blood pressure, high cholesterol, and vision related issues were the most frequently reported diabetes complications in both groups. Median sensor time in range (70-80 mg/dL) at 6 months was 77.4% (70.3-83.1) for Group A and 75.7% (67.282.0) for Group B. Glucose Management Indicator at 6 months showed improved glucose control from baseline (Group A=6.8 (6.7-7.2) vs. 7.0 (6.6-7.7), Group B=7.0 (6.7-7.2) vs. 7.2 (6.7-7.9)). PROs at 6 months demonstrated improvement in device related satisfaction (Group A=7.9 (1.7) vs. 8.5 (1.4), p=0.002; Group B=7.3 (2.0) vs. 8.9 (1.1), p<0.001) and reduction in the impact of diabetes on daily life (Group A=4.3 (1.7) vs. 3.1 (1.4), p<0.001; Group B=5.1 (SD=2.0) vs. 3.1 (1.4), p<0.001).
Findings demonstrate long-term acceptability and efficacy of Control-IQ technology in older adults reporting diabetes complications.
Smartphone Camera to Image the Retina for Primary Care MDs and Diabetologists
University of California, Davis San Jose, California, United States
Sensor Controlled Diabetic Foot Ulcer Therapy - Next Generation Therapy of the Most Devastating Complication of Diabetes
CID GmbH Cologne, NRW, Germany
We developed a complex, e-health integrating approach to optimize diabetic foot ulcer therapy to preserve mobility and accelerate ulcer closure. We monitored pressure, temperature, humidity, and steps; used smartwatch and web applications to alert patients and their therapists when pressure limits exceeded; and used customized, non-removable offloading soles made of felt or felt plus fiberglass or a total contact cast.
In a randomized, controlled, multicenter trial, patients with diabetic plantar pressure ulcers who used the entire system were compared to a control group without information provided to patients and staff. Patients were asked to walk as much as they usually did before pressure ulcer onset. If necessary, PAD had to have been previously corrected and the ulcer surgically debrided.
20 patients participated in the study (12 intervention, 8 control). 3 had to be excluded because they did not use the system within the first 2 weeks (2 intervention, 1 control). 15 of 80 intervals between visits showed worsening, significantly less in the intervention group (4/42 vs. 11/38, p = 0.026). Median time to ulcer closure was reduced (232 to 40.5 days, p=0.03). The individual average of steps per day (mean 2,146, SD 1,427) corresponded to independent living and normal activity for many patients. Observed optimizations were related to offloading devices at the time of initial application and later after dressing changes, and behavioral changes of patients responding to alarms.
Integration of optimized offloading devices, sensor control, and alerts for patients and staff may enable rapid closure of plantar foot ulcers without limiting the number of steps by optimizing medical staff performance and improving patient’s health-related behaviors.
InPen™ System Use and Glycemic Outcomes in Older Adults
Medtronic Diabetes Northridge, California, USA
The InPen™ smart insulin pen allows for the calculation of rapid-acting insulin meal and correction doses and the tracking of active insulin. This study compared use-behaviors and glycemic outcomes between older (ages 65+yrs) and younger (ages 18-64yrs) individuals with type 1 diabetes (T1D) or type 2 diabetes (T2D) using the InPen with continuous glucose monitoring (CGM).
InPen+CGM data were obtained within a 30- to 60-day period of InPen start, from 2,735 adults (18-64yrs=2,422, 65+yrs=313; T1D=2,320, T2D=415) between July 2020 and June 2021. Use-behaviors (bolus adherence, basal logging adherence, therapy mode, calculator usage frequency, and report generation) and glycemic metrics were assessed. Student’s t-test or Mann-Whitney U test was used, depending on Shapiro-Wilk test for normality, to compare means for continuous variables. Fisher’s Exact test and Chi-square test were used for categorical variables. Users with insufficient CGM data and inconsistent insulin therapy settings were excluded.
The older group, compared to the younger, showed more consistent daily use of InPen in both bolus dosing (83.2% vs. 80.1%) and manually logging basal doses (48.8% vs. 41.4%), as well as higher frequency of InPen app dose calculator use (45.2% vs. 40.0%) and report generation (35.8% vs. 22.3%). They also set non-carb-counting therapy modes (meal estimation and fixed) more frequently (29.4% vs. 11.8%). Their glycemic metrics (TIR; 60.7% vs. 56.8%, TBR; 1.50% vs. 2.03%, TAR; 37.8% vs. 41.2%, and CV; 31.0% vs. 32.9%) were also better. All differences were significant (p<0.05).
These real-world data demonstrate older adults’ use-behavior of novel smart insulin pen technology and their improved glycemic outcomes compared with those of younger adult users.
Free Diabetes Apps and Health Literacy for Diabetic Patients
Purdue University West Lafayette, Illinois, United States
Free diabetes health apps:do they bridge the divide of health literacy in our diabetic patients?
Using search term, “diabetes” in GooglePlayStore(Android/AD) and AppleAppStore(iOS) to find the best, free diabetes apps. Exclusion criteria: less than 100,000 downloads (DL) on Google and <1600 reviews (RE) on Apple (Apple does not provide download data).
The app introduction was “pasted” into readabilityformulas.com and webfX.com for analysis via Flesch-Kincaid Reading Ease (FRE), Flesch-Kincaid Grade Level Score (FGL), Simple Measure of Gobbledygook Index (SI), and the Dale-Chall Adjusted Grade Level (DCGL).
Push notification=19/20 apps(AD=9, iOS=10). Connect users to GoodRx=0/20 apps. Connect users to professional care team/diabetes education=8/20 apps(AD=3, iOS=5). Connect users to free/discounted insulin/health products=0/20 apps. Spanish=11/20, Chinese=5/20 apps (AD=3, iOS=8, some apps have both). Only 2/20 apps were at 7-8th grade and the rest were high school and beyond. Free test strips were offered only on 1/20 apps(Livongo).
Free diabetes apps are helpful but the majority are at high school level. Thus Diabetes Healthcare teams are still needed as a partner for ehealth for our diabetic patients.
Machine Learning-Based Models for Estimating CGM Accuracy to Venous Plasma Glucose from Accuracy to Capillary Glucose
Senseonics Inc., Germantown, MD, USA
The accuracy of a CGM system is typically compared to venous plasma glucose measurements from a Yellow Springs Instrument (YSI) glucose analyzer in clinical trials. However, development data may only include capillary fingerstick glucose (FSG) measurements as reference for trial design simplicity, especially during home use. In this work, we propose machine learning-based models for estimating CGM accuracy to YSI from accuracy to FSG.
The development dataset contained CGM glucose values from 183 sensors matched to YSI and FSG glucose measurements. The dataset was divided into 50/50 train/test “splits.” The mean accuracy difference to YSI and FSG was computed in non-overlapping glucose bins encompassing the full glycemic range (40-400 mg/dL). This process was repeated 100 times to get 100 splits. Confidence intervals (C.I.) around the predictions on the test splits were computed using the root-mean-squared-error (rmse) of the predicted accuracy to YSI. This methodology was validated on an additional 43 sensors to derive models for both MARD and 15/15% agreement metrics.
For an implanted CGM system with 5-min sampling rate and sensors lasting up to 180 days, accuracy to YSI was correctly predicted to be within a 95% C.I. around the prediction in six (6) glucose bins (40-60 mg/dL, 61-80, 81-180, 181-300, 301-350, 351-400) and overall (40-400 mg/dL). Higher prediction accuracy was observed in glucose bins with the most matched pairs.
We derived and validated predictive models which estimate the accuracy of a CGM system to YSI based on accuracy to FSG. Such models can be used to predict the accuracy of any experimental CGM system to venous plasma glucose using its accuracy to capillary glucose in smaller development studies.
Social Network Analysis of Diabetes and Glucose Monitoring Systems on Twitter Informed by ADCES7™ Framework
University of Missouri Columbia, MO, USA
Using social network analysis guided by ADCES7™ guidelines, this study investigates Twitter conversational trends regarding diabetes monitoring and glucose monitoring systems.
Tweets and social network diagrams were retrieved using the tool NodeXL. Twitter data was collected over a period of ten days in June 2022, resulting in a dataset of 7161 users and 11059 tweets. Data was extracted using thirteen keywords related to diabetes monitoring drawn from ADCES7 guidelines (e.g., glucose monitoring, HbA1C etc.) and using brand names of seven glucose monitoring systems (e.g., Dexcom, Onetouch Verio, etc.).
The term “diabetes management” occurred most frequently on Twitter (tweets=2354) and “HbA1C” was the most popular term as measured by number of users (users=1448) while “Diabetes sensors” being the least popular term (users=37). A comparison of volume of posts and number of users posting within the same time frame for seven top-ranked glucose monitoring systems by brand, revealed that “Dexcom” was the most common glucose monitor mentioned on Twitter (users=953, tweets=1490), followed by “OneTouch Verio” (users=231, tweets=269). Further, “Ascensia Contour” (users 31, tweets= 37), was the least common glucose monitor mentioned on Twitter followed by “Accu-check” (users 41, tweets= 52). The top hashtags for “diabetes management” were #nephpearls and #digitalhealth. The top hashtags for “diabetes monitoring” were “livingfullywithdiabetes” and “usmedsupply”.
Our preliminary analysis suggests that popularity trends for diabetes monitoring topics tend to match concepts emphasized by ADCES7™ framework. Twitter can be a useful platform to study trends for keywords and brand names related to diabetes and glucose monitoring. Additionally, it can provide insights into social media marketing for glucose monitoring companies and help devise better strategies to influence online diabetes communities.
Improving Uncontrolled Glycated Hemoglobin (A1C) in Rural Areas Using an Algorithm for Weekly Telemedicine Check-In
Purdue Global University West Lafayette, IN
The purpose was to show how creating an algorithm based on A1C level and scheduling weekly check-ins with participants using telemedicine to review diet, medication compliance, glucose monitoring and exercise could improve health outcomes and A1C’s. This project assisted with gaps due to the lack of a diabetic educator at the practice site. It allowed patient communication and education about diabetes and decreased stress on patients who had transportation issues and or financial concerns.
Participants were given a pre and post survey to validate knowledge and understanding. Weekly telephone check-ins were completed based on algorithm of frequency of calls. Logs were kept for glucose monitoring, diet, exercise, and medication compliance.
Follow up surveys showed improvement in medication regimen, understanding of diet and glucose monitoring. Compliance with weekly check-ins were 79%. Out of 17 participants, 13 had improvements in follow up A1C.
Patients across the Unites States are faced with increasing challenges each day. Having healthily lifestyles is an opportunity for many to work on and especially those living in lower income areas, such as rural southern low-income areas in states like South Carolina (SCDHEC, 2020). Many want to do the right thing, they unfortunately lack the knowledge and understanding of what they need to do. Many face transportation issues getting to appointments. However, most have a telephone and being able to utilize that for education and check-ins with healthcare specialist to improve health disparities such as diabetes. This can also improve chronic comorbidities such as hypertension, coronary disease, and obesity (SCTA, 2021). Telemedicine has proven to become a part of today’s modern world of healthcare and decrease stress (Hunt et al., 2018).
Inhaled Insulin + AID Safely Reduced Glucose in the 2 Hours Following a Meal in Proof-of-Concept Study Data Subset
MannKind Corporation Westlake Village, CA USA
Technosphere Insulin (TI) is an ultra-rapid-acting inhaled insulin. This proof-of-concept study subset compares efficacy and safety 2 hours after a standardized meal of a higher (~2x) TI dose while wearing an automated insulin-delivery (AID) system compared to a meal-bolus with AID alone.
Twenty-three participants with T1D on an AID system have been enrolled at two clinical sites so far. Each participant consumed a standardized nutritional shake. All subjects in the TI group (n=18) received a pre-prandial dose of TI calculated by doubling the injected mealtime insulin dose and rounding down to the nearest TI cartridge; the AID system did not cover the meal. The AID group (n=5) entered a meal-bolus to cover the meal. TI was dosed at the start of the meal and the AID alone group dosed 15 minutes before. Capillary glucose was measured by Ascensia Contour™ meters at timepoints 0, 15, 30, 45, 60, 90, and 120 minutes relative to the meal.
Glucose was reduced significantly at timepoints 60-120 minutes (p<0.05) and glucose excursion 90-120 minutes (p<0.05) in the TI group following a meal with no hypoglycemia events under 70-mg/dL. Mean peak glucose was reduced by 61.6-mg/dL in the TI group and the mean peak glucose excursion was reduced by 58.8-mg/dL. Peak glucose was reached 30 minutes earlier in the TI group.
The proof-of-concept study subset indicates that TI with a higher converted dose (~2x), in combination with an AID system, significantly reduces glucose in the 2-hour period after the meal compared to an AID system alone. The trend indicates that hyperglycemia is reduced within 2-hours following a meal with no new safety signals, including hypoglycemia.
Hypoglycemia Anxiety as an Influence Factor on CGM System Use in a Type 1 Diabetic Patient
University of Applied Sciences Neu-Ulm, Germany
With regard to continuous glucose monitoring (CGM) system use in Germany, the aim was to investigate whether an existing hypoglycemia anxiety of a type 1 diabetic can be regarded as a significant acceptance factor of a CGM system. Therefore, we investigated whether hypoglycemia anxiety has an influence on CGM system use and the choice of a particular system.
A quantitative survey was conducted in which a total of 2,102 patients were interviewed via an online questionnaire, of which 1,642 patients (type 1 diabetics) responded completely. Non-parametric tests were performed, specifically the Mann-Whitney U test and the Kruskal-Wallis test.
Hypoglycemia anxiety counts as an acceptance factor regarding the use of CGM systems. Other acceptance factors were investigated in our study.
The Use of a Combined Real-Time CGM and Digital Health Solution Lowered A1C in People with Type 2 Diabetes
Welldoc Columbia, Maryland, USA
AI-driven digital health coaching has the potential to deliver real-time actionable insights from large amounts of raw data and may help individuals with diabetes make better “in the moment” decisions regarding food, medications, and activity. Recent studies have demonstrated that non-intensive insulin users benefit from real-time CGM (rtCGM). We therefore hypothesized that combining rtCGM with digital health coaching may significantly improve A1C in non-intensive insulin users.
We reviewed real-world data from 119 participants who were enrolled in a program that provided both a rtCGM (Dexcom G6) as well as a digital health coaching platform (BlueStar by Welldoc). Enrollment targeted individuals who were not prescribed insulin. We defined two groups of participants: those who wore the rtCGM intermittently and those who wore it continuously. We collected the A1C data at baseline, 12 weeks, and 24 weeks.
The baseline A1C was 9.5%. In the intermittent use group, the A1C decreased by 1.3 and 1.7 points at 12 weeks and 24 weeks, respectively. In the continuous use group, the A1C decreased by 1.8 and 2.2 points. In those continuous use participants whose baseline A1C was over 9%, the A1C decreased by 3.1 points at 24 weeks. The changes in A1C from baseline for all groups was highly statistically significant.
The combination of rtCGM and a digital health solution significantly improved A1C after 12 and 24 weeks of use. The degree of A1C lowering was greater in the continuous use group and in users who had a high baseline A1C. These data highlight the important role that digital health tools may have for rtCGM users.
Free Diabetes Apps with ADCES7 Guidelines: Do They Help Our Patients?
Washington University in St. Louis St. Louis, Missouri, United States
ADCES7 Parameters:1) Goal-setting; 2) Calorie-counting; 3) Nutrition-education; 4) Exercise-reminder; 5) Medication-reminder; 6) Blood-glucose; 7) Weight; 8) Blood-pressure; 9) HbA1c; 10) Problem-solving; 11) Reducing-risk; 12) Insulin-information; 13) Insulin-pumps-information; 14) Non-English options.
Features:
Goal-setting=9/20 apps (AD=4, iOS=5);
Calorie-counting=10/20 apps (AD=5, iOS=5);
Nutrition-education=9/20 apps (AD=3, iOS=6);
Exercise-reminder=7/20 apps (AD=2, iOS=5);
Medication-reminder=11/20 apps (AD=5, iOS=6);
Blood-glucose=18/20 apps (AD=10, iOS=8);
Weight=9/20 apps (AD=6, iOS=3);
Blood-pressure=10/20 apps (AD=6, iOS=4);
HbA1c=11/20 apps (AD=6, iOS=5);
Problem solving=12/20 apps (AD=8, iOS=4);
Reducing-risk=7/20 apps (AD=5, iOS=2);
Insulin-info=7/20 apps (AD=3, iOS=4);
Insulin-pumps info=1/20 apps (AD=0, iOS=1);
Non-English options=12/20 apps (AD=3, iOS=9)
Behavior Engagement and Activation Model (BEAM) for Digital Healthcare Solutions in Adults with Type 2 Diabetes: A Cross-Sectional Survey
Sanofi UK Reading, United Kingdom
Many adults living with T2D using Digital Healthcare Solutions (DHCS), like mobile apps, do not use them long enough to improve their condition. To characterize heterogeneity among these individuals, we aimed to identify underlying subgroups of adults who do and do not use DHCS for T2D management.
Individuals were recruited between December 2021 and March 2022 and were eligible if they self-reported a T2D diagnosis, were >18 years of age, lived in the US, read and understood English, and had internet access. Two cohorts were constructed: 1) Participants: individuals who have used DHCS for T2D management and 2) Non-participants: individuals who have never used DHCS. Respondents completed a survey on demographics, psychosocial and clinical characteristics, and barriers to using DHCS. We used latent class analysis and descriptive statistics to identify and characterize subgroups within each cohort.
A total of N=637 respondents completed the survey. The Participant cohort was 48±11 (mean±sd) years of age, 34% female, and 62% White. Three subgroups were identified based on psychosocial and clinical characteristics — 21% (81/380) “Motivated Self-Managers,” 42% (159/380) “Room for Improvement,” 37% (140/380) “Needs Support to Continue”. The Non-Participant cohort was 56±12 (mean±sd) years of age, 56% female, and 69% White. Two subgroups were identified based on psychosocial and clinical characteristics — 31% (79/257) “Convincible Non-Users,” and 69% (178/257) “Needs Support to Initiate”.
Self-efficacy for managing diabetes, health activation, perceived barriers, and HbA1c emerged as characteristics that may help identify different segments of Participants and Non-Participants. Future research should include objective measures of clinical characteristics and adherence to specific types of DHCS to better understand adoption and engagement across subgroups.
Significant Reduction in Hyperglycemia and High Satisfaction with Use of Control-IQ Technology in Prior MDI and Basal Only Insulin Users with Type 2 Diabetes (T2D)
Division of Endocrinology, Icahn School of Medicine at Mount Sinai New York City, NY USA
To assess safety and glycemic outcomes with use of the t:slim X2 insulin pump with Control-IQ advanced hybrid closed loop system (CIQ) in adults with T2D.
Thirty adults with T2D (mean age 54±12 years, mean HbA1c 8.6±1.2, median BMI 31) using either MDI (N=15), pump (N=2) or basal without bolus insulin (N=13) collected unblinded CGM data (baseline) followed by an open-loop period prior to initiating use of CIQ for 6 weeks. Primary outcomes were superiority for CGM time >180 mg/dL and non-inferiority for time <54 mg/dL.
Mean time >180 mg/dL decreased from 44% at baseline to 29% with CIQ, a reduction of 3.6 hours per day (P=0.004), time 70-180 mg/dL increased from 56% to 71% (P= 0.003), and mean glucose decreased from 184 to 162 mg/dL (P=0.007). Time >250 mg/dL decreased by almost one hour per day. Mean time <54 mg/dL was 0.12% at baseline and 0.04% during CIQ use (mean change= -0.09%, upper end of onesided 97.5% CI =+0.01%). Median time in closed loop was 96% (interquartile range 91% to 97%). There were no severe hypoglycemia or DKA events. On the Diabetes Impact and Device Satisfaction Scale, participants indicated a high level of satisfaction with CIQ (mean item score 8.8±1.9 on scale of 1-10). Both prior MDI or pump and prior basal insulin only participants showed similar levels of improvement in all outcomes.
In this prospective single-arm trial of adults with T2D, the use of CIQ was demonstrated to be safe during 6 weeks of use, with no increase in CGM-measured hypoglycemia and a substantial improvement in time in range and mean glucose related to a reduction in hyperglycemia.
Delivering Insulin by Glucose-Responsive Polymer-Based Microneedle Patch
Advanced Pharmaceutics and Drug Delivery Laboratory, University of Toronto Toronto, Canada
All people with type 1 diabetes (T1D) and advanced type 2 diabetes require insulin therapy to control their blood glucose levels (BGLs). Improved glycemic control currently relies on hybrid closed-loop insulin delivery systems that combine a continuous glucose monitoring device connected to an insulin pump delivering insulin at hyperglycemia. However, the cost of the device and accessories and needle insertion-associated side effects limit the wide application of such devices. Although glucose-responsive microneedle (MN) patches have been developed to deliver insulin, harsh conditions used for fabrication could adversely affect the bioactivity and native structure of encapsulated insulin. Hence, we designed a novel self-crosslinkable polymer-based microneedle patch (mMN) to load highly concentrated insulin under mild fabrication condition for delivering insulin at hyperglycemia.
Glucose-responsive polymers were synthesized by conjugating functional groups that could form cross-linkages under physiological pH and respond to changes in glucose concentration. Insulin-containing mMN was made from these polymers using a mild self-crosslinking method. Mechanical strength of the mMN was measured with a tensile tester. Glucose-responsive insulin release was determined in vitro. Bioactivity of released insulin and biocompatibility of the mMN patch were evaluated in STZ-induced T1D rats.
The mMN provided adequate mechanical strength for skin penetration. In vitro insulin release showed glucose-responsiveness. The mMN maintained BGLs in T1D rats within a tight glycemic range over 7 hours. The mMN treated rats tolerated high doses of glucose injection. Histological analysis of rat skin treated with mMN indicated undetectable local inflammation and tissue damage.
The new insulin-mMN patch demonstrated high efficacy against hyperglycemia in a T1D rat model. The simple fabrication process of the mMN patch offers great potential for clinical translation.
Real-World Digital Health Data Demonstrate the Utility of the Glycemia Risk Index (GRI) as a Composite CGM Metric
Carey Business School of Johns Hopkins University Baltimore, MD, USA
The GRI has been proposed1 as a single-number metric to simplify the analysis of complex CGM data, but there are few examples of its use in real-world populations. Digital health tools can integrate with CGM devices and may record thousands of data points each week. We were interested in evaluating the utility of GRI in users of a digital health tool2.
A real-world data set of 499 CGM users was created. The data were de-identified according to standard procedures. The GRI was calculated for the first 3 days and last 3 days of a 14-day observation period in those with at least 80% sensor wear time (n=381). People with diabetes can be mapped to GRI zones A (best) to E (worst).
55% of individuals had type 1 diabetes. 51% were female. Individuals whose baseline GRI started in zones A and B ended in Zones A and B. Individuals whose baseline GRI started in higher zones were the ones whose GRI improved. The mean GRI of older individuals (>=65 years) stayed in zone B, and the mean GRI of younger individuals (18-39 years) stayed in zone C. Using the F-test, GRI was significantly lower in older individuals and in females, but GRI did not differ significantly by type of diabetes.
These data demonstrate the utility of GRI in uncovering both individual and population level insights for people with diabetes using a digital health tool. GRI may be used as an outcome variable in clinical studies. In addition, GRI’s simplicity may be useful in the clinical management of patients and populations of patients by health care professionals.
1 Klonoff DC, Wang J, Rodbard D et al. A glycemia risk index (GRI) of hypoglycemia and hyperglycemia for continuous glucose monitoring validated by clinician ratings. J Diabetes Sci Technol. 2022 Mar 29.
2Quinn CC, Shardell MD, Terrin ML, et al. Cluster-randomized trial of a mobile phone personalized behavioral intervention for blood glucose control. Diabetes Care. 2011 Sep;34(9): 1934-42.
Next Generation Glucose Reference Platform - Data Supporting the Use of the Radiometer ABL90 Analyzer
ProSciento Chula Vista, CA 91911 USA
The aim of this study is to evaluate the Radiometer ABL90 analyzer as to the analytical equivalence to the YSI 2300 and YSI 2500 instruments, the current standard in the industry.
A dataset of 2113 blood glucose (BG) samples collected across 30 T1DM and T2DM subjects during a device study conducted at ProSciento’s Clinical Research Unit (CRU) was used for this analysis. Each sample was measured once in three different Radiometer ABL90, three different YSI 2300, and three different YSI 2500 devices.
Mean BG values from the triplicate ABL90 measurements were compared to the mean BG from the YSI 2300 (reference method) and YSI 2500 for correlation, accuracy, and clinical performance using the Clark error grid.
Coefficient of Variation (CV) was calculated for all 3 instruments independently.
BG values ranged from 48.6 mg/dl to 342.6 mg/dl for the ABL90, 48.1 mg/dl to 350.8 mg/dl for YSI 2300, and from 46.6 mg/dl to 371.2 mg/dl for YSI 2300.
ABL90 vs YSI 2300
Paired results from the ABL90 and YSI 2300 (n=2017) showed a high correlation coefficient (R>0.99) and a Mean Absolute Error of 2.4%. Clark error gird analysis shows 100% of results in Zone A (figure 1).
ABL90 vs YSI 2500
Paired results from the ABL90 and YSI 2500 (n=2018) showed a high correlation coefficient (R>0.99) and a Mean Absolute Error of 2.6%. Clark error gird analysis shows 100% of results in Zone A (figure 2).
CV across the three devices was 1.67% for ABL90, 1.74% for YSI 2300, and 1.97% for YSI 2500.
Data presented here support the analytical equivalency of the ABL90 instrument to the YSI 2300 and YSI 2500.
Patient Perspectives on the Ambulatory Glucose Profile (AGP) Report for Type 1 Diabetes (T1D) Management: A National Online Survey
McMaster University Hamilton, Ontario, Canada
The Ambulatory Glucose Profile (AGP) is a standardized report produced by continuous/flash glucose monitoring devices. The utility of the AGP report is emerging in the literature, however the current Diabetes Canada Clinical Practice Guidelines do not provide recommendations about its use. Making recommendations about the AGP would involve consideration of the perspectives of people living with type 1 diabetes (T1D).
We conducted a self-administered online survey for people living with T1D to understand factors related to the use of this digital intervention. The survey was pre- and pilot-tested, with expert opinion incorporated. The survey was advertised through online public and patient communities. Statistical analyses were completed using SPSS.
A total of 402 consented to participate, with 291 responses included. Most respondents were female, <40 years of age, and have lived with T1D >15 years. Respondents with greater understanding of the AGP were more motivated towards diabetes management (aOR:2.61, 95%CI:1.45-4.71). More than 75% review their AGP and consider it important, yet only 51% (primarily younger with less education) often discuss their AGP with health-care providers (HCPs). Consistent discussion and support from HCPs increased the likelihood of independent review of the AGP (aOR:7.23, 95%CI:3.22-16.24; aOR:2.17, 95%CI:1.18-4.01, respectively). Only 48% are satisfied with the AGP’s clarity and 40% with the layout, however >90% agree the AGP is useful and accurate. Additionally, there is low satisfaction with insurance coverage for this technology (27%).
Insurance coverage was a barrier to respondents, while facilitators were motivation and support, especially from HCPs. Given their perceptions, efforts to increase discussion between HCPs and patients may be helpful to improve the use and potential benefits of the AGP.
Comparing Different Individualized Black-Box Models for Insulin Pump Faults Detection on Artificial Pancreas Data
Department of Information Engineering, University of Padova Padova, Italy
Prompt and accurate detection of insulin pump faults could be key in preventing sustained hyperglycemia and possibly ketoacidosis. Model-based fault detection techniques relay on patient models to warn the patient of a possible malfunctioning. Here, we want to compare four individualized models to understand if there is a preferable choice in terms of fault detection ability.
Individualized, linear, black-box parametric models (ARX, ARMAX, ARIMAX, BJ) are identified with BIC-based optimal order on 7 days of fault-free closed-loop data for 100 virtual subjects, using one of the most recent versions of the UVA-Padova T1 Diabetic Patient Simulator, that accounts for intra/inter-patient variability. An online prediction of up to PH=3h of glucose concentration, along with its confidence interval, is calculatedthrough a Kalman filter based on the individualized model, running on data of past infused insulin (possibly affected by an unknown 6h insulin suppression), ingested meals and CGM values for 10 days. The real time fault detection algorithm raises an alarm if the predicted residual portion stays out its confidence interval for more than 15min.
The performance of the detection method is evaluated in terms of false positives per day and recall (FP/days, RE%), also taking into consideration the detection time. The performance is (0.12, 70%), (0.13, 85%), (0.17, 80%), (0.16, 78%), while the detection time is 236min, 242min, 237min, 238min for ARX, ARMAX, ARIMAX and BJ, respectively. If we compute the Euclidian distance from the optimal point (0, 100%), this metric will suggest ARMAX as the best model.
Although ARMAX appears to be the best choice, the use of the other models only slightly impacts the fault detection performance and detection time.
Insulin Pump Disruption: Cause and Effect
Washington University School of Medicine Division of Endocrinology, Metabolism & Lipid Research Saint Louis, MO, USA
Insulin pumps have led to improved glycemic control and quality of life for individuals with T1D and T2D. Pump removal is common in the setting of medical care. Furthermore, pumps or infusion sets may malfunction or become dislodged, leading to reductions in wear time. We investigated the prevalence of insulin pump disruption and the adverse consequences of reduced wear time. Additionally, we examined the out-of-pocket costs incurred and patient satisfaction with the pump replacement process.
A web-based survey of patients with T1D or T2D followed in the Washington University Diabetes Center was conducted.
Of 375 surveys sent, 56 were completed. Participants had a mean age 52, T1D 84%, female 71%, White 98%, non-Hispanic 93%, and mean duration of DM 29 years. Insulin pump breakdown was 19% Medtronic (630 8.6%, 670G 6.9%, 770G 2.3%), 19% Omnipod, 60% Tandem, 2% other. Pump disruptions occurred most frequently in the setting of insertion problems (76.5%)followed by device displacement (48.9%). Medical care related disruption occurred most frequently in the setting of imaging (57.4%),followed by surgery (17%), and hospitalization (6.4%). Off-pump days due to pump disruption occurred in 40.6% of respondents (1-3d 25.5%, 4-6d 8.5%, 7d+ 6.4%). Adverse events attributed to insulin pump disruption such as hyperglycemia and hypoglycemia occurred four or more times per year in 14.9% and 6.4% of the cohort respectively. All insulin pump replacement requests were fulfilled with 75% of respondents finding manufacturers to be very or fairly responsive to their needs. Out-of-pocket costs were incurred by 15.3% of those surveyed.
Disruption in insulin pump therapy is common. Prompt insulin pump replacement by manufacturers minimizes off-pump time and mitigates adverse events.
Correlation of Continuous Glucose Monitoring with Hemoglobin A1c in Pediatric Patients with CFRD
University of Alabama at Birmingham Heersink School of Medicine Birmingham, Alabama, United States
Cystic fibrosis-related diabetes (CFRD) is associated with increased mortality in patients with cystic fibrosis (CF). Annual CFRD screening with a 2-hour oral glucose tolerance test (OGTT) is recommended in those with CF age 10+, however this is burdensome for patients. Although highly sensitive for other forms of diabetes, hemoglobin A1c (A1c) isn’t recommended for CFRD screening because of its low sensitivity in detecting CFRD compared to OGTT. The low sensitivity of A1c in diagnosing CFRD is thought to be partly due to increased red blood cell (RBC) turnover. Use of continuous glucose monitoring (CGM) in the diagnosis and management of CFRD is an area of interest, but ideal CGM metrics in CFRD haven’t been established. We aimed to evaluate the correlation between A1c and CGM measures of glycemic variability in patients with CFRD.
This retrospective review examined A1c measurements and 14-, 30-, and 90-day DexCom G6 CGM metrics, average glucose (AG), glucose management index, standard deviation, coefficient of variation, time in range (TIR), time above 180mg/dL, time above 250mg/dL, time <70mg/dL, and time <54mg/dL in pediatric patients with CFRD.
Twenty A1c measurements and their respective CGM measures in 4 patients with CFRD were evaluated. On correlation analysis, the strongest correlation was seen in A1c and 30-day AG and 30-day TIR (r=0.973, P<0.001 and r=-0.916, P<0.001, respectively).
CGM-derived measurements of glycemic variability, particularly 30-day AG and TIR, are highly correlated with A1c in pediatric patients with CFRD. The stronger correlation with 30-day vs 90-day metrics is potentially related to increased RBC turnover in CF patients. Prospective studies are needed to determine how CGM may best be used to diagnose and manage CFRD.
Once-Weekly Insulin Icodec vs Once-Daily Insulin Glargine U100 in Combination with Bolus Insulin in Individuals with Type 2 Diabetes on Basal-Bolus Regimens (ONWARDS 4)
Clinical and Experimental Endocrinology, University of Leuven Leuven, Belgium
ONWARDS 4 was a 26-week, randomised, treat-to-target, phase 3 trial, designed to assess efficacy and safety of once-weekly insulin icodec (icodec) vs once-daily insulin glargine U100 (glargine) in a basal-bolus insulin regimen in type 2 diabetes (T2D).
Adults (n=582) with T2D (HbAic 7.0-10.0%) on a basal-bolus regimen were randomised 1:1 to icodec or glargine, both in combination with multiple daily injections of insulin aspart . Primary endpoint was change in HbAic from baseline to week 26 (non-inferiority margin 0.3%).
At week 26, from a mean baseline of 8.3%, estimated HbAic reductions were -1.16% (icodec) and -1.18% (glargine), confirming non-inferiority for icodec vs glargine (estimated treatment difference [ETD]: 0.02% [95% CI, -0.11 to 0.15]; p<0.0001). Overall rates of combined level 2 or level 3 hypoglycaemia were comparable between treatments (5.64 [icodec] and 5.62 [glargine] per patient year of exposure; estimated rate ratio: 0.99 [95% CI, 0.73 to 1.33]; p=0.929). Mean total weekly basal component of the total insulin dose from week 24-26 was significantly higher in the icodec arm vs the glargine arm (305 U/week vs 279 U/week; p=0.0286), while the mean total weekly bolus insulin component was significantly lower (197 U/week vs 255 U/week; p<0.0001). Estimated mean body weight change was 2.73 kg (icodec) and 2.16 kg (glargine) (ETD: 0.57 kg [95% CI, -0.39 to 1.54]; p=0.2444). No new safety concerns were identified for icodec.
Once-weekly icodec demonstrated comparable improvements in glycaemic control with fewer basal insulin injections, with a comparable hypoglycaemia profile vs once-daily glargine in individuals with T2D on a basal-bolus regimen.
Clinical Trial Registration Number: NCT04880850
Supported by Novo Nordisk A/S.
Molecular Simulation of Functionalized Hyaluronic Acid on Protecting Insulin in a Transdermal Microneedle Patch
Advanced Pharmaceutics and Drug Delivery Laboratory, Leslie L. Dan Faculty of Pharmacy, University of Toronto Toronto, Canada
Recently, our lab developed a modified hyaluronic acid (HA)-based transdermal microneedle patch to deliver insulin at hyperglycemia. To identify the possible stabilization effects of the hydrogel matrix-based microneedle patch (mMN) on the incorporated insulin molecules, the interaction mode between insulin and the functionalized polymers was investigated by all-atom molecular dynamics (MD) simulations.
For simulating the interaction between the polymers and insulin, a hormone molecule with the PDB ID of 4INS was downloaded from the protein data bank. Two independent systems, i.e., insulin with or without the polymers, were introduced to the 500 ns MD studies using the Desmond package (Schrodinger Inc.). For simulating a 200-disaccharides polymer, 10 chains of HA, each containing 20 disaccharide units were constructed. According to the ratio between polymer and insulin used for making the MN patch experimentally, the disaccharide units of HA, HA-derivative 1, and HA-derivative 2 were randomly integrated into each chain. Simulation results were analyzed by python scripts.
The flexibility and movement of polymer-bound insulin molecules were decreased as a result of polymer interaction with the protein. The radius of gyration values suggested that the polymers had no remarkable effects on the folding of insulin. The average number of intra-H-bonds in both free and bound insulin was nearly the same, whereas a significant amount of H-bonds formed between insulin and the polymers as found in the simulation. This hydrogen bonding pattern may explain the experimental findings that the insulin incorporated in the HA-based polymers was more stable than its free form.
Our result suggests that the HA-functionalized polymers can prevent insulin denaturation through a shielding effect and are a good candidate for making the insulin-loaded MN patch.
Correlation of Hemoglobin A1c and Continuous Glucose Monitor Metrics in Diverse Populations of Youth with Diabetes
UAB Heersink School of Medicine Birmingham, AL, USA
Impact of Implementing Continuous Glucose Monitoring Upon Hospital Discharge: A National Survey
Abbott Diabetes Care, Inc. Alameda, CA, USA
Hospital discharge is a crucial yet challenging transition of care for patients with diabetes. Implementing continuous glucose monitoring (CGM) may facilitate the discharge process. US hospitals have provided FreeStyle Libre™ system CGMs to patients with diabetes upon discharge and we sought to evaluate the effects of this approach on discharge processes.
We assessed the impact of providing these CGMs with a 3 category (positive, negative, undecided) 9-item questionnaire survey focusing on key elements of the discharge processsuch as: developing criteria to identify patients to be discharged with a CGM, updating protocol for discharge of patients with diabetes, changing the approach for education for diabetes education, improving communication within the hospital diabetes care team, improving communication without outpatient care teams, tracking of new outcomes in diabetes, and initiatingquality improvement projects in diabetes care. The fractions and corresponding confidence intervals were calculated and compared.
A total of 133 (23%) US hospital institutions responded via email. Using CGM in hospital discharge is associated with development of criteria for initiating CGM, changes to diabetes education, and improved communication within the hospital diabetes care team. Responses on changes in diabetes discharge protocol and communication between hospital team and outpatient providers were mixed. Most hospitals have not yet tracked diabetes outcomes or initiated quality improvement projects.
Overall, responses of hospitals using CGM samples upon discharge were positive for actions that may lead toimproving the diabetes discharge processes.The study is limited by the sampling size and bias, as it is within the US only and in the early stages of program development.
Imputation Model for Glucose Values Out of Measuring Range for Continuous Glucose Monitoring
Department of Endocrinology and Nephrology, Copenhagen University Hospital - North Zealand Hilleroed, North Zealand, Denmark
All continuous glucose monitors (CGMs) have an upper detection limit, typically of 22.2 mmol/L (400 mg/dL). This might bias CGM metrics. This study aimed to develop and validate a statistical model for imputing censored CGM values above the upper detection limit.
We analysed 358 days of in-hospital CGM data (Dexcom G6 and iPro 2) from 105 patients with diabetes. A Bayesian non-parametric latent Gaussian process regression model was applied to the CGM data which were intentionally censored at the 70% and 80% percentile (corresponding to 30% and 20% censoring during a day) and compared to the uncensored CGM data. Three different covariance functions for the latent Gaussian process were examined. The performanceof the imputation model wasassessed by the bias and the mean squared error (MSE) of standard CGM metrics, i.e., mean glucose level, standard deviation (SD) and coefficient of variation (CV).
For 20% censoring, the performance of the imputation model was very high across the three covariance models with MSE (range) 0.0008 to 0.2 and bias (range) -0.02 to 0.07 of standard CGM metrics. For 30% censoring, the performance of the imputation model was lower with MSE (range) 0.002 to 1.7 and bias (range) -0.03 to 0.3 of standard CGM metrics. Generally, one covariance model was superior with MSE< 0.09 and <0.2 and absolute values of bias <0.03 and <0.03 of all CGM metrics for 20% and 30% censoring, respectively.
An imputation model for glucose values above the upper detection limit of CGMs was succesfully developed and validated. This enables a more unbiased quantification of CGM metrics for patients with moderate to severe hyperglycemia.
SimDex Enables First-in-Human Trials of a Novel Automated Insulin Delivery System at Home
Dexcom, Inc. Charlottesville, Virginia, USA
Preclinical in silico trials have long been accepted as a substitute for animal studies in the run-up to first-in-human trials of automated insulin delivery (AID) systems. Generally, these first-in-human trials have taken place in either highly supervised clinical settings or semi-supervised “hotel” studies that approximate free-living conditions at home. The objective of this work was to demonstrate the feasibility of using a proprietary free-living simulator, SimDex, to support regulatory approval for unsupervised first-in-human trials of a novel AID system at home, bypassing supervised studies.
SimDex was developed by members of this research team to predict outcomes of novel diabetes treatments. It features the ability to emulate inter- and intra-subject physiological and behavioral variability, along with detailed characteristics of CGM devices and insulin pumps.
In preparation for a 40-person feasibility trial outside of the US, SimDex was configured to run the source code of a novel hands-free AID system. In silico tests exposed the system to the full spectrum of real-life behavioral interactions (meals in rapid succession, risk-compensated eating, and worst-case system ON/OFF behaviors). The results predicted low exposure to out-of-range blood glucose and were included in the submission to the regulatory body.
Approval for the feasibility study was granted in January 2022, six months after the SimDex trials were run. Based on preliminary analysis of the home trial data, experimental percentage time below 70 mg/dl agreed with the predicted value to within .2% on average. Experimental percentage time above 250 mg/dl agreed with the predicted value to within 6.2%.
SimDex enabled first-in-human studies of a novel AID system at home and provided an accurate assessment of the risk of out-of-range excursions.
In-Vitro Validation of an Open Source, Ultra-Low-Cost (<US$100) Insulin-Pump
University of Canterbury Christchurch, Canterbury, New Zealand
Improving the Quality of Glucose Concentration Measurements by Recalibration
Institut fur Diabetes-Technologie, Forschungs- und Entwicklungsgesellschaft mbH an der Universitat Ulm, Ulm, Germany
In analytical performance studies, the choice of comparator method plays an important role as studies have shown that there exist relevant systematic differences between laboratory analyzers. The feasibility of retrospective recalibration of measurement results through comparison with methods or materials of higher metrological order was therefore assessed.
Existing data were retrospectively analyzed to assess feasibility. Comparison with a higher-order method was performed based on mass spectrometry for two laboratory analyzers (dataset 1) and on a laboratory analyzer for two blood glucose monitoring systems (BGMS) (dataset 2). Part of the patients samples that were measured with both laboratory analyzers were also measured with mass spectrometry (dataset 1). The same approach was used for the patient samples in dataset 2 (BGMS vs laboratory analyzer). Linear regression equations were calculated according to Passing and Bablok for each analyzer/BGMS separately. These equations were then applied to the respective complete dataset of patient samples. Bias was assessed across all patient samples between analyzers/BGMS before and after recalibration.
In dataset 1, the bias of glucose concentrations in the same patient samples measured with the two laboratory analyzers was -3.6% before recalibration based on mass spectrometry results. Regression equations were yi=0.95*x+2.15 mg/dl for one laboratory analyzer and y2=1.03*x-1.24 mg/dl for the other. After recalibration, bias was +0.6%. For the two BGMS in dataset 2, bias in simultaneously measured patient samples was +10.4% before recalibration (regression equations were y3=1.02*x+1.59 mg/dl and y4=1.00*x-9.53 mg/dl), and +0.6% after recalibration based on laboratory analyzer results.
Recalibration was feasible when based on values from a method of higher metrological order. The procedure should be verified in a prospectively designed setting.
User Performance Evaluation and System Accuracy Assessment of Four Blood Glucose Monitoring Systems with Color Coding of Results
Institut fur Diabetes-Technologie, Forschungs- und Entwicklungsgesellschaft mbH an der Universitat Ulm, Ulm, Germany
Blood glucose monitoring systems (BGMSs) are a cornerstone in diabetes management. They have to provide sufficiently accurate results in the hands of lay-users, particularly in insulin-treated patients. Color coding can help users identify a suitable course of action more quickly, and it enables healthcare professionals an easier assessment of glycemic control. The aim of this study was user performance evaluation and system accuracy assessment of 4 BGMSs with color coding of results.
Study procedures were based on ISO 15197:2013. User performance evaluation was based on data from 100 participants, each of whom used every BGMS with one reagent lot per BGMS. Study personnel observed user techniques. For system accuracy, 100 capillary samples were obtained for measurement in duplicate with each of three reagent lots per BGMS, resulting in 600 results per BGMS. Comparison measurements were performed using a hexokinase-based method.
All assessed BGMSs exhibited a sufficient level of accuracy, with small differences between them. In the user performance evaluation, study personnel observed the smallest number of user errors with Contour® Next (Ascensia), followed by Accu-Chek® Instant (Roche), Medisafe Fit Smile™ (Terumo) and OneTouch Ultra Plus Reflect™ (Lifescan). In 11 attempts, users were unable to generate a measurement result with OneTouch Ultra Plus Reflect. Approximately 90% of participants stated that the same color scheme, e.g., for low values, should be used across all BGMSs. There was no clear preference among the four tested BGMSs regarding the specific way of displaying color coding.
The four BGMSs assessed in this study showed only small differences in an overall sufficient level of accuracy. User handling errors, as observed by study personnel, differed between the systems.
Ultra-Low-Cost, Ultra-Low-Power, Open-Source Clockwork Insulin Pump
University of Canterbury Christchurch, Canterbury, New Zealand
Insulin pumps enable better control, but are highly inaccessible due to cost (US$5000 per pump), creating significant inequity, particularly for indigenous peoples who often have higher risk and worse outcomes. This study presents in-vitro validation test results for an open-source, ultra-low-cost, ultra-low-power (<US$90 hardware) insulin-pump design.
A spring-drive clockwork insulin pump prototype with an escapement mechanism provides low-power dosage control to 0.05U resolution, and 1-2 years estimated battery life. Bolus accuracy was tested to the IEC 60601-2-24 standard for boluses of 0.05U, 0.2U, and 1U using the standard-specified approach, measuring mass displaced by the pump with an analytical balance (grams to 5d.p). N=25 measurements for each bolus volume were weighed and presented as mean volume (percent deviation)[max-negative-percent-deviation, max-positive-percent-deviation].Results are compared to a Medtronic 640G sourced in New Zealand.
The clockwork pump delivered mean volume (percent deviation) [max-negative-percent-deviation, max-positive-percent-deviation] 0.0503U(+0.5)[-36.0, +28.0]% for 0.050U boluses, and 0.1962U(-1.9)[-22.0, +14.5]%, and 0.9690U(-3.1)[-12.0, + 6.0]% for 0.2U and 1.0U, respectively. Results are comparable to the Medtronic 640G which yields 0.0450U(-10.1)[-38.0, +44.0]%, 0.1963U(-1.9)[-7.0, +13.0]%, and 0.9818U(-1.8)[-4.3, + 11.9]% for 0.05U, 0.2U, and 1.0U respectively.
A mechanically driven insulin pump controlled through a clockwork escapement mechanism provides comparable bolus delivery accuracy to a commercial motor-driven device, and offers significant cost reduction (<US$90 hardware), along with (estimated) 13X greater battery life. Assuming manufacture and regulatory costs of 3-5X hardware, a <US$300-500 pump would enable ~10X more pumps to be publicly funded in New Zealand and greatly increase equity of access. A patch-pump design would further reduce patient costs. Full IEC standard testing, clinical trials, CE mark, and phone control software with Bluetooth are next steps.
Yoga Practice Improves Cerebral Oxygenation and Mental Workload in Type 2 Diabetes Mellitus Patients: A Randomized Controlled Trial
Department of Yoga and Life Sciences, Swami Vivekananda Yoga Anusandhana Samsthana (S-VYASA, deemed-to-be University) Bangalore, Karnataka, India
Background: Type 2 diabetes mellitus (T2DM) is associated with cognitive decline and lifestyle behaviors such as physical activity and mind-body practices play an important role in improving cognitive function and preventing cognitive decline.
Purpose: The goal of this study was to see how a 12-week yoga intervention affected working memory and associated prefrontal cortex (PFC) activation, heart rate variability (HRV) and rumination in T2DM patients.
Methods: Fifty participants were randomized into yoga practice (n=25) and waitlist control group (n=25). The n-back task was administered to evaluate working memory before, after 6 weeks and 12 weeks of intervention. While performing the working memory task, the prefrontal cortex oxygenation was monitored using functional near-infrared spectroscopy (fNIRS). Rumination was assessed using the Rumination Response Scale (RRS).
Results: The yoga group showed a significant improvement in accuracy and reaction time in the 2-back (p < 0.05) task condition, while the control group showed no improvement in any task conditions. In the yoga group, oxyhemoglobin was significantly increased during the 2-back task, post-yoga intervention in the PFC region (p < 0.05). However, no significant changes were noticed in the control group. Significant improvement in HRV and RRS was observed in the yoga group.
Conclusions: These results suggest that yoga practice may improve working memory performance associated with efficient prefrontal oxygenation in patients with T2DM. Improved heart rate variability and reduced rumination may be the influencing factors for improving working memory performance. Further studies with a larger sample and robust design are needed to strengthen the findings.
A Patient Healthcare Literacy Model on Continuous Glucose Monitors and Insulin Pumps
University of Missouri Kansas City School of Medicine Kansas City, Missouri, United States of America
A key tool that patients often utilize to develop their understanding of diabetic technology is internet-based searches for educational materials. Moreover, the current RankBrain machine learning algorithm employed by Google Internet searches comprises a large part of this data traffic. However, there remains a need to explore the behavior of this algorithm when employed by patients. The aim of this study is to determine the readability of these materials initially produced by the algorithm to develop a healthcare literacy model.
This study explored the Google RankBrain machine-learning algorithm to extract internet-search analytics to determine the most popular internet search queries regarding continuous glucose monitors (CGMs) and insulin pumps (IPs). A Flesch Kincaid Grade Level (FKG) and Flesch Reading Ease (FRE) were calculated for all extracted educational materials.
Among the first 20 queried items for CGMs and IPs, the average FKG was higher for IPs (11.28) compared to CGMs (10.95). The average FRE for CGMs (53.12) was higher than IPs (48.85). 10% (n= 2) of the queried CGMs materials were classified as institutional (non-commercial developer) whereas 30% (n= 6) of queried IPs were institutional. All sources for CGM and all except for 1 for IPs were written in the United States.
Educational materials regarding CGMs were more difficult to read by nearly a grade level than compared to IPs. Additionally, IPs materials were more often written by institutional bodies (i.e., universities, and non-profit organizations) than CGMs. CGMs require more attention in simplify their readability comprehension compared to IPs. Greater sample size data and investigation is needed in future studies to further support these findings.
Translating Event History to Expected Glucose Distribution
Department of Chemical and Biological Engineering, Rensselaer Polytechnic Institute Troy, NY, USA
Inertial measurement units from wearable devices can provide information about a user’s activity state, enabling automatic prompting and event-logging. However, one cannot wholly depend on these sensors to identify actions affecting a patient’s blood glucose levels. This work is a step towards extending an existing event detection algorithm to include continuous glucose monitor (CGM) measurements, aiming to improve the sensitivity of auto-logging.
A particle filter algorithm that connects CGM data to a patient’s event history is proposed. The filter employs gold standard data from triple-tracer studies to evaluate posited mealtimes probabilistically. Collected studies were classified into fast, medium, and slow absorption meals. Classified curves were then normalized, scaled according to a population-level meal size distribution provided by National Health and Nutrition Examination Survey (NHANES), and combined to get an effective probability distribution for meal response. Similar data collection was performed for insulin response curves. Noise associated with the CGM sensor was modeled as a Gaussian distribution. Convolution was employed to get a resultant probability distribution for change in CGM.
The method was checked and verified on test case scenarios using combinations of triple-tracers. Real patient data is also analyzed using the Tidepool dataset for meal detection and CGM prediction using a particle filter. The advantage of this method is that it is data-driven and can handle multi-modal distributions for a non-linear process of the human body.
A method is developed to convert a patient’s event history to an expected change in CGM values. This method can be further extended to include other disturbances to blood glucose. Efforts are being made to combine the detections from wrist motion (accelerometer) and the CGM method presented here.
Development of Derivatization-Free ID-LC/MS/MS Method for Simultaneous Measurement of Human Serum Monosaccharides
Centers for Disease Control and Prevention Atlanta, Georgia, USA
Glucose is the major biomarker for diabetes diagnosis and treatment. Consumption of fructose and other monosaccharides are common in diabetic patients to aid glycemic control. However, high fructose consumption has been associated with increased de novo lipogenesis in the liver leading to non-alcoholic fatty liver disease (NAFLD). We aim to develop a high-throughput analytical method for measuring serum monosaccharides that is helpful for understanding their associations with various metabolic disorders and other diseases.
NIST SRM 917c was used for assay calibration for glucose, with 13C labeled internal standards spiked in all serum samples and monosaccharides calibrators. Monosaccharides were isolated from serum matrix by acetonitrile protein precipitation and filtrated through 96-well protein removal plate on automated liquid handler system. Monosaccharide filtrate was directly analyzed using selective reaction monitoring on a Triple Quad LC/MS/MS system after eluted from Unison UK Amino column under acetonitrile and water gradient.
The analytical measurement range covers 10 - 379 mg/dL for glucose and linear between measurement range of 0.3 - 12.2 mg/dL (all at R2 = 0.999) for fructose and mannose. The method demonstrated average accuracy and imprecision of 97% and <3% for serum glucose, respectively, using 4 levels of NIST SRM 965b. Spiked recovery of other monosaccharides was above 95% with overall inter-day imprecision less than 9% across multiple QC levels. Glucose, fructose and mannose were detected in all 83 human serum samples. All monosaccharides were fully resolved to demonstrate high level of specificity.
The current analytical method can be used to simultaneously measure serum levels of monosaccharides with appropriate accuracy and precision, using simple, derivatization-free sample procedure. The method is applicable for large biomonitoring studies.
Prevention of Rebound Hypoglycemia in an Advanced Automated Insulin Delivery System
Center for Diabetes Technology, University of Virginia Charlottesville, VA, United States of America
Automated insulin delivery (AID) systems, especially full closed-loop (FCL) designs, can increase the risk for rebound hypoglycemia (RH) after reacting to rescue carbohydrates. The aim of this work is to present and evaluate a controller-agnostic, cross-module RH prevention layer (RHPL) that can be easily integrated into an AID system.
The proposed RHPL applies an exponential decay function to constrain the maximum allowable insulin delivery dose based on the minimum glucose measurement in the last hour (Gmin) and the current glucose concentration (Gc). The lower Gmin and Gc the stricter the constraint. This approach is integrated into our FCL AID system, RocketAP, to constrain its Model Predictive Control (MPC) system that regulates the basal insulin infusion and its Bolus Priming System (BPS) that triggers priming doses when a rapid increase in glucose is detected. Performance is assessed on the 100-adult cohort of the UVA/Padova simulator, considering one-day simulations that include three unannounced meals with 70, 50, and 80 grams of carbohydrates each.
Comparing RocketAP with RocketAP+RHPL, RH events were eliminated (from 39 to 0) by effectively attenuating the commanded MPC and BPS doses within the 30-minute interval after treating a hypoglycemia event (MPC: 0.27U vs 0.04U; BPS: 1.02U vs 0.05U). Time below 70 mg/dL was slightly reduced (2.0% vs 1.7%) with no clinically significant changes in time in the range 70-180 mg/dL (TIR: 80.4% vs 79.8%) and a slight increase in time above 180 mg/dL (17.6% vs 18.5%).
The proposed RHPL was shown to be effective in eliminating RH events without affecting achieved TIR when combined with an advanced FCL system.
Smartphone Built-in Features: Do They Help Our Diabetic Patients?
University of California, San Francisco San Jose, California, United States
To evaluate the pre-set features of smartphones and the inclusion of ADCES7 guidelines for Diabetes self-care during the COVID-19 era.
The pre-loaded applications of the iPhone 13 and Samsung S22 Ultra were tested with the ADCES7 parameters: 1) Healthy eating, 2) Being Active, 3) Monitoring, 4) Taking Medication, 5) Problem Solving, 6 )Healthy coping, 7) Reducing Risks.
We evaluated the smartphones for the following subcategorties of the ADCES7 parameters: 1a) Nutrition, 1b) Calorie-counting; 2a) Step-counting, 2b) Calorie-burning, 3a): Blood pressure, 3b) Blood glucose, 3c) HbA1c, 3d) Weight, 3e) Push notification, 4a) Med Log, 5a) Contact healthcare team, 5b) Contact family; 6a) Mental health outreach, 7a) Goal-Setting, 7b) Graph analysis.
We added additional categories 8) Insulin pumps and 9) Language.
1. Nutrition facts= Apple(Yes), Samsung(No);
2. Calorie counting= Apple(Yes), Samsung(Yes);
3. Step counting= Apple(Yes), Samsung(Yes);
4. Calorie Burning= Apple
5. Monitoring BP= Apple
6. Monitoring Blood Glucose= Apple
7. Monitoring HbA1c= Apple
8. Weight Monitoring= Apple(Yes), Samsung(Yes);
9. Push Notifications and Med Reminders= Apple(Yes), Samsung(Y es);
10. Goal setting= Apple
11. Med Log= Apple
12. Facetime, Samsung Video Call= Apple(Yes), Samsung(Yes);
13. Healthy Coping= Apple
14. Goal Setting for Steps= Apple
15. Graph Analysis= Apple(Yes), Samsung(Yes);
16. Insulin pumps= Apple
17. Foreign languages= Apple
Samsung provides more ADCES7 parameters and more language options than Apple.
The smartphones have internal resources for patients to self-manage their Diabetes Mellitus. As MDs, we can continue to partner with our patients for diabetes management during COVID-19 and beyond.
Indications for Inpatient Glucose Telemetry
Rush University Medical Center Chicago, IL 60612, USA
The primary objective was to explore indications for inpatient glucose telemetry.
The inpatient glucose telemetry (IGT) has been instituted at the peak of COVID cases at an urban academic medical center. Besides remote glucose monitoring due to infection isolation, feasibility, reliability and indications for IGT were investigated in hospitalized patients.
IGT was used in n=75 patients in critical and non-critical care inpatient settings. In addition to remote glucose monitoring due to infection isolation, feasibility, reliability and indications for IGT were investigated in cases requiring hypoglycemia prevention, multimorbidity, fingertip bruising, transplant, cancer, intensive insulin management, brain/psychiatric disease/injury, and inpatient rehabilitation.IGT was used in patients hospitalized with personal home continuous glucose monitoring system and patients with recurrent diabetes hospitalizations.
Inpatient glucose telemetry indications must be expanded beyond hypoglycemia prevention.
Feasibility of a Prototype Dual Function Glucose Sensing and Insulin Delivering Cannula
Diabetes Technology Research Group, The University of Melbourne Fitzroy, Victoria, Australia
To assess the feasibility of a prototype glucose-sensing and insulin delivering cannula; a dual-function/ single-insertion device in type 1 diabetes adults.
Following a 48-hour run-in using a commercial insulin infusion set and study pump (Medtronic 780G), the study cannula was inserted subcutaneously in the subject’s abdomen and connected to the pump. A standardized meal was then eaten, and an insulin bolus administered as per the pump settings through the study cannula. Venous glucose was measured every 10 minutes for 60 minutes pre-meal and 15-minutely post-meal by YSI and SMBG (Roche Accu Chek Guide). Following two days of free living where subjects performed at least 10 SMBG readings each day, a second meal test was performed on Day 4. Throughout the study a Dexcom G6 sensor was worn, and insulin dosing was manually determined. Sensor data from the study canula was processed using a preliminary algorithm to calculate glucose with an initial one-point calibration at post-processing.
Eight subjects (median [IQR] age 42.5 [35, 50.5] years, mean±SD HbA1c 7.2±0.4%) have been studied to date. Comparing run-in vs the study cannula TDD was 59 [35.9, 74.7] units vs 66.3 [36.7, 73.6] units (p=0.401); mean TIR 57.3±20.7% vs 46.4±25.1% (p=0.44); and mean sensor glucose 167.4±30.6mg/dL vs 185.4±32.4mg/dL (p=0.476) respectively.
A total of 718 paired sensor data points with SMBG as comparator have been analyzed. For the four day duration, individual device MARD ranged between 13.3% to 19.7%. No artifactual calculated sensor glucose spikes were observed.
This ongoing study provides initial feasibility data supporting a glucose-sensing and insulin delivering cannula. Algorithm development is underway aimed at improving sensor accuracy.