2021 Diabetes Technology Meeting Abstracts

Alharthi	“Pre-Ramadan” Telemedicine: Effect on Fasting Experience & Glycemic Control during Ramadan in People with Type 1 Diabetes	A1
Arnold	Smart Insulin Pen Integrates New Sensors and Connects to AI for Insulin Dose Calculation	A2
Arrieta	Real-World Outcomes of the MiniMed™ 780G System When Used with Optimal Setting	A3
Arunachalam	Impact of Real-World MiniMed™ 670G System Use on Glycemic Outcomes in the United States	A4
Baumann	Regression Modelling between Statistical Rigor and Clinical Implementation - The Benefits of Open Communication among Disciplines	A5
Beck	Improvements in Hypoglycemia and HbA1c with RT-CGM Adoption: Results from the COACH Study	A6
Bosoni	Relationship between Time in Range, Glycemic Variability Metrics and Daily Scan Rate in Children with Type 1 Diabetes	A7
Brew-Sam	Diabetes Technology Experiences of Young People Living with Type 1 Diabetes and Their Parents – Analysis Guided by a Hybrid Theoretical Foundation	A8
Bunsick	Inexpensive Regular Human Insulin Used with CGM Provides Similar Outcomes to More Expensive Insulin Lispro	A9
Button	1,5-Anhydroglucitol Is an Independent Predictor of Mortality in Patients with COVID-19	A10
Champakanath	Long-Term (7-Year) Glycemic Outcomes with Continuous Glucose Monitoring (CGM) Initiation within First Year of Type 1 Diabetes (T1D) Diagnosis	A11
Chattaraj	Rate of Unexplained Hyperglycemia and Infusion Set Occlusion: Medtronic Extended Infusion Set (EIS) Pivotal Trial and 3-Day Infusion Set Study Comparison	A12
Chien, A.	Potential Cost Savings from a Reduction in Sensor-Detected Severe Hypoglycemia among Users of the InPen Smart Insulin Pen System	A13
Chien, D.	Diabetic Eye Disease and Virtual Assistant AI: Do They Help Our Patients?	A14
Cox	Virtual Care in the COVID-19 Era: Assessing Telemedicine Experiences among Diabetes Care and Education Specialists	A15
Dugas	Development of Self-Management Behavior Scores and Profiles with Digital Health Data	A16
Fogle	Current Technology Use in Pediatric Patients with Type 1 Diabetes: An Equity-Focused Analysis	A17
Gadiraju	I Am Diabetic: Can I Speak to Virtual Assistants in Different Languages?	A18
Garelli	Outpatient Full Closed-Loop Trial Using an Open-Source Remote Monitoring and Artificial Pancreas Platform	A19
Haldimann	Increased Time-in-Range after Meals When Using the SNAQ App	A20
Idi	Unsupervised Anomaly Detection Algorithms to Identify Compression Artifacts in Continuous Glucose Monitoring	A21
Ilkowitz	Identifying Patients at High Risk for Cystic Fibrosis Related Diabetes through the Use of the Freestyle LibrePro CGM	A22
Jimenez	How Telemedicine Assists Adults with Type 2 Diabetes Obtain Adequate Glucose Control	A23
Kaiserman	Higher Dose of Inhaled Technosphere Insulin Provides Significant Reduction in Post Prandial Glucose Excursions, without Hypoglycemia	A24
Kamecke	Clinical Evaluation of the Lumee™ Glucose System – a Fluorescence-Based Long-Term Continuous Glucose Sensor	A25
Kesserwan	Phenolic Preservatives, Present in Commercial Insulins, Trigger Inflammation at Infusion Sites via Mast Cell Activation	A26
Kinzel	The Influence of Health Insurance on the CGM System Choice of a Type 1 Diabetic	A27
Lalani	Evaluation of the Safety and Efficacy of the Lalani Insulin Infusion Protocol	A28
McHugh	Determining Losses in Jet-Injection Subcutaneous Insulin Delivery	A29
Mujahid	Generation of Realistic Virtual Diabetes Patients Using a Pix2Pix GAN: In Pursuit of an AI-Based Diabetes Simulator	A30
Narayan	Predicting Postprandial Glucose Excursions with Nutrient Content Using an Interpretable Random Forest Augmented by a Digital Twin ODE Model	A31
Newson	Preferences for Connected Insulin Pens: A Discrete Choice Experiment among Type 1 and Type 2 Diabetes Patients	A32
Nosrati	Development of the Microfluidic Bio-Artificial “mPancreas”: A Scalable and Modular Microfluidic-Based Biomimetically-Designed Bio-Artificial Pancreas	A33
Oehme	Method to Test and to Validate ML/AI SaMD	A34
Ormsbee	Can EGP Be Estimated from Heart Rate?	A35
Owen	Representing Type 1 Diabetes Device Data Using CDISC Standards	A36
Pinsker	Significant Reductions in Adverse Events and Hospitalizations with Control-IQ Technology in Adults with Type 1 Diabetes: Data from the CLIO Study	A37
Potts	Wearable Electrochemical Platform for Glucose Monitoring and Diabetes Management	A38
Redding	Transdermal Insulin Patch-No Needles	A39
Rivera	Si Cuentas, Come Lo Que Quieras. Carbcounting Quiz App	A40
Romero-Ugalde	Hyperoptimized Deep Learning Model for Glucose Prediction on DBLG1	A41
Roversi	In Silico Trial to Quantitatively Measure How Much Different Behavioural Factors Influence Time in Hypoglycemia in Type 1 Diabetes Management	A42
Sears	The Effects of One Drop Digital Program on Glucose Control in Employees with Type 1 and 2 Diabetes	A43
Sears	The Effects of One Drop Digital Program on Weight Reduction in Overweight and Obese Employees with Prediabetes	A44
Setford	Post-market Clinical Surveillance of OneTouch Ultra® Plus Blood Glucose Test Strips	A45
Shah	A Data-Driven Approach to Predict Carbohydrate Counting Errors in Diabetes Management	A46
Siviero	Importance of Accounting for Correlation among Noise Samples When Filtering CGM Data: A Simulation Study	A47
Stroh	Point-of-Care Insulin Monitoring Using Electrochemical Sensing	A48
Taylor	User Experiences with an Insulin Pen Platform	A49
Wartenberg	CxM - Novel Sensors	A50
Wu	Can I Talk to Amazon Alexa and Google about Diabetic Kidney Disease?	A51
Wuttke	Upcycling of Used CGM to a Cool Chain Monitor for Insulin	A52
Zimmermann	Effect of Personalized Feedback Using Motivational Interviewing Strategies on the Frequency of Self-Monitoring of Blood Glucose and Subsequent Glycemic Control in Adults with Type 2 Diabetes	A53

“Pre-Ramadan” Telemedicine: Effect on Fasting Experience & Glycemic Control during Ramadan in People with Type 1 Diabetes

Sahar Alharthi, MBBS; Areej Alrajeh, MBBS; Ebtihal Alyusuf, MBBS; Abdullah M. Alguwaihes, MD, MPH; Anwar Jammah, MBBS; Mohammed E. Al-Sofiani, MD, MSc

Division of Endocrinology, Diabetes and Metabolism, King Saud University, College of Medicine Riyadh, Saudi Arabia

Moh.alsofiani@gmail.com

Objective:

People with type 1 diabetes (T1D) are advised to have a “pre-Ramadan” clinic visit to receive the appropriate assessment and education needed to safely fast during the holy month of Ramadan. The COVID-19 lockdown has interrupted this standard of care, particularly in Muslim-majority countries where telemedicine is not well-established. Here, we examined the effect of virtual “pre-Ramadan” visits, as an alternative to the traditional (in-person) visit, on fasting experience and glycemic control during Ramadan in people with T1D.

Method:

151 individuals with T1D were categorized into 3 groups according to the type of “pre-Ramadan” visit that they had in 2020: virtual (n=50), in-person (n=56), and no visit (n=45). Number of days fast was broken due to diabetes-related issues and CGM metrics were retrospectively compared across the 3 groups.

Result:

Patients who had a virtual “pre-Ramadan” visit were more likely to be women (60%) and use CGM during Ramadan (61.7%) than those who had no visit (37.8% and 38.6%, respectively, both p<0.05). Similarly, patients who had an in-person “pre-Ramadan” visit were more likely to use CGM (70.9%) than those who had no visit. Attending a virtual “pre-Ramadan” visit was associated with the least number of days fast was broken compared to those who had no visit (p<0.01) or in-person visit (p=0.02). The virtual group had the highest time in range (TIR) during Ramadan (59%) compared to the no visit group (44%, p<0.01) and in-person group (47%, p<0.01). After adjusting for age, gender, pre-Ramadan A1c, and use of CGM, the odds ratios of fasting most days of Ramadan (i.e., breaking the fast :2 days) were the highest in the virtual group [OR (CI): 9.13 (1.43, 58.22)] followed by the in-person group [3.02 (0.54,16.68)] compared to the no visit group.

Conclusion:

A virtual “pre-Ramadan” visit is an effective alternative to in-person visits when managing people with T1D who plan to fast during Ramadan.

Smart Insulin Pen Integrates New Sensors and Connects to AI for Insulin Dose Calculation

Nico Arnold; Ziyi Zhong; Thomas Wuttke, MS

Technische Universität Dresden Dresden, Germany

nico.arnold@ibta.de

In insulin therapy, diabetics use insulin pens, injection devices similar to a syringe with replaceable insulin cartridges, robust and designed for repeated use.

Diabetics make mistakes in insulin administration: no insulin administration, double insulin administration, wrong insulin taken, wrong insulin dose taken. This leads to hyperglycemia, hypoglycemia and even death. Monitoring of insulin administration in the insulin pen could detect these errors and avoid the consequences.

Students at the University of Dresden have developed a new type of intelligent insulin pen together with the startup diafyt MedTech. To meet the requirements for reliability and accuracy, the latest sensor technology and microcontrollers were used in collaboration with Texas Instruments. The insulin pen accesses artificial intelligence via a Bluetooth connection, which can calculate the insulin dose with the data obtained. With the addition of a CGM, the solution has similar features to an AID but at much lower cost.

The sensor technology and novel data processing can be demonstrated on a working prototype. A patent has been filed.

Today, insulin pens are rather restrained in design. In 3D printing, however, unusual individual designs are also possible. The students have developed a 3D printable insulin pen design that allows individual looks. High-tech and over-the-top designs can be demonstrated in one package.

Real-World Outcomes of the MiniMed™ 780G System When Used with Optimal Setting

Arcelia Arrieta, MS; Javier Castaneda, MS; Julien Da Silva, MS; John Shin, PhD; Ohad Cohen, MD

Medtronic Bakken Research Center Maastricht, The Netherlands

arcelia.arrieta@medtronic.com

Objective:

MiniMed™780G system pivotal trial demonstrated highest time spent in target glucose range (70-180 mg/dL, TIR) of 78.8%, with a 100mg/dL glucose target and active insulin time (AIT) setting of 2 hours (Carlson et al., Diabetes. 2020;69[Supplement 1]). Real-world data from 4,120 MiniMed™780G systemusers (overall cohort) showed TIR of 76.2% (Da Silva J, et al. Diabetes Technol Ther. 2021;23(S2). Oral 012 / #494). The glycemic control achieved by real-world individuals using the optimal settings was evaluated.

Method:

MiniMed™ 780G system data uploaded (27August2020 to 03March2021) voluntarily to CareLink™ personal software by individuals living in Belgium, Finland, Italy, the Netherlands, Qatar, South Africa, Sweden, Switzerland and the United Kingdom, and who provided their consent, were aggregated and retrospectively analyzed. The GMI, TIR, TBR<70 and TBR<54 was determined for those with ≥10 days of SG data post-AHCL initiation and who used the optimal target and AIT for more than 95% of the time.

Result:

Individuals (N=446) using optimal settings achieved a TIR of 80.3±7.7% and a GMI of 6.6±0.3% (overall cohort: 6.8%). The percentage of users achieving a TIR of >70% and GMI of <7.0% was 90.8% and 93.3%, respectively (overall cohort: 77.3% and 79.0%). Their TBR<70 and TBR<54 were 3.0±2.2% and 0.7±0.8%, respectively (overall cohort: 2.5% and 0.5%).

Conclusion:

Individuals using the MiniMed^TM 780G system with the optimal settings (100mg/dL target and 2-hours AIT), in real-world conditions, achieved better glycemic outcomes compared to the overall real-world cohort. This real-world analysis confirms findings from the pivotal trial and demonstrates opportunity for improving glycemia with the MiniMed™ 780G system.

Impact of Real-World MiniMed™ 670G System Use on Glycemic Outcomes in the United States

Siddharth Arunachalam, MSc; Kevin Velado, BS;Toni L. Cordero, PhD; Robert A. Vigersky, MD

Medtronic Northridge, California

siddharth.arunachalam@medtronic.com

Objective:

Retrospective analyses of real-world hybrid closed-loop system use on glycemic outcomes provide a pragmatic and important report on diabetes technology therapy performance and its impact on diabetes management. This study assessed real-world glycemic outcomes of individuals with type 1 diabetes (T1D) using the MiniMed™ 670G system in the United States.

Method:

A total of 123,555 individuals who reported type 1 diabetes (T1D) diagnosis uploaded MiniMed™ 670G system data between March 2017 to November 2020 to CareLink™ personal software. These data were analyzed to determine percentage of time spent between 70-180mg/dL (%TIR), at <70mg/dL (%TBR) and at >180mg/dL (%TAR) senor glucose (SG) range and the glucose management indicator (GMI). Similar metrics in users (N=52,941) with ≥10 days of SG data pre- and post-Auto Mode initiation were also analyzed and proportions of users meeting consensus-recommended glycemic targets were assessed. Statistical analyses were conducted with paired-sample t-tests.

Result:

The overall group who spent 87.9±38.9% of the time in Auto Mode had a GMI of 7.0±0.4%. The %TIR, %TBR and %TAR were 70.4±11.2%, 2.2±2.1% and 27.5±11.6%, respectively. Pre- and post-Auto Mode initiation values for GMI were 7.3±0.6% versus 7.1±0.4% (p<0.001); for %TIR, were 61.5±15.1% versus 68.1±11.9% (p<0.001), for %TBR, were 2.11±2.4% versus 2.07±2.3% (p=0.002); and for %TAR, were 36.3±15.7% versus 29.8±12.2% (p<0.001). The proportion of users (pre- versus post-Auto Mode initiation) with GMI <7.0% was 30.2% versus 41.7%; with %TIR >70% was 30.4% versus 47.2%; and with %TBR <4% was 85.7% versus 87.4%.

Conclusion:

Real-world MiniMed™ 670G Auto Mode use in the United States improves %TIR, %TBR and %TAR and allows more users to reach recommended glycemic control targets.

Regression Modelling between Statistical Rigor and Clinical Implementation - The Benefits of Open Communication among Disciplines

Petra M. Baumann, MA, MA; Daniel A Hochfellner, MD; Haris Ziko, BSc;Amra Simic, MA; Peter Beck, PhD; Julia K Mader, MD; Johannes Plank, MD, MBA

Medical University of Graz Graz, Styria, Austria

petra.baumann@medunigraz.at

Objective:

Regression analysis offers advantages over standard statistical tests commonly used in medical research (e.g., t-test, ANOVA). It can include multiple independent variables and is flexible regarding complex data structures and non-normal distributions. However, clinicians who are not well versed in regression techniques might be reluctant to accept results and rather go with (supposedly) simpler methods that may not best represent data. Clinicians and statisticians should therefore work together closely and acknowledge the needs for statistical rigor and clinical usefulness.

Method:

We analyzed glycemic control during fasting vs. regular nutrition in hospitalized patients. Data were hierarchically structured (repeated measures within two conditions) and number of measures differed by patient and condition. Outcomes (BG, insulin) were not normally distributed, and covariates pertained to both measurements as well as patients. This required advanced multilevel models whose results are not easy to grasp, thus impeding clinical interpretation.

Result:

Diligent, open-minded communication between statistician and clinicians led to a methodologically sound as well as clinically valuable approach: We used linear models (after extensive sensitivity analysis) and untransformed outcomes. The main effect (-0.5±0.2 mmol/L mean daily BG and -11.1 IU total dose for fasting days) was therefore in the unit of the outcome and easy to interpret. It was also adjusted for covariates which were interpretable as well. To further simplify presentation of results, we created a summary table that highlights results.

Conclusion:

Reconciling diverging objectives of statisticians and clinicians promotes scientific rigor as well as clinical implementation of results. Statisticians who are open to clinicians’ needs make better modelling choices and clinicians who are comfortable with complex models play an important role in promoting the most accurate statistical techniques.

Improvements in Hypoglycemia and HbA1c with RT-CGM Adoption: Results from the COACH Study

Stayce E. Beck, PhD, MPH; Christy Chao, PhD; Diana Le, MS; David Price, MD

Dexcom, Inc. San Diego, California

stayce.beck@dexcom.com

Objective:

COACH was a post-approval study (NCT03340831) that assessed the risk of debilitating hypoglycemia and changes in HbA1c among people with T1 and insulin-requiring T2 diabetes resulting from non-adjunctive use of real-time continuous glucose monitoring (RT-CGM).

Method:

Adult participants used SMBG for diabetes management decisions for 6 months and then transitioned to RT-CGM for 6 months while receiving usual care. Moderate hypoglycemic episodes were defined as those that required assistance of another person to resolve; severe hypoglycemic episodes were those resulting in seizures or loss of consciousness. All events were documented via monthly phone calls with study staff. Participants were stratified by type of diabetes (T1 or T2), insulin delivery method (pump or MDI), hypoglycemia (un)awareness, and baseline HbA1c level. Hypoglycemic event frequencies between the SMBG and RT-CGM phases and changes in HbA1c level between the beginning and end of the RT-CGM phase were compared.

Result:

In the per-protocol population (n=519), the number of hypoglycemic events decreased from 42 (in 29 participants) during the SMBG phase to 16 (in 12 participants) during the RT-CGM phase. Decreased event rates with RT-CGM use were found for participants with T1 or T2 diabetes, participants using pumps or MDI, and participants with impaired or intact hypoglycemia awareness. The hypoglycemia effect size was larger for nocturnal events, MDI users, participants with baseline HbA1c <7.0%, and participants with impaired hypoglycemia awareness. Participants with starting HbA1c >10.0% experienced mean HbA1c reductions of >0.9 percentage points.

Conclusion:

In insulin-requiring diabetes, nonadjunctive use of RT-CGM reduced the frequency of debilitating hypoglycemic episodes regardless of type of diabetes, insulin delivery method, and hypoglycemia awareness. Favorable reductions in HbA1c were most apparent for participants with the highest baseline HbA1c levels.

Relationship between Time in Range, Glycemic Variability Metrics and Daily Scan Rate in Children with Type 1 Diabetes

Pietro Bosoni, MSc; Lucia Sacchi, PhD; Valeria Calcaterra, MD; Cristiana Larizza, PhD; Riccardo Bellazzi, PhD

Department of Electrical, Computer and Biomedical Engineering, University of Pavia Pavia, Italy

pietro.bosoni02@universitadipavia.it

Objective:

The adoption of intermittently scanned continuous glucose monitoring (isCGM) systems can improve glycemic control and quality of life in type 1 diabetes (T1D) individuals, but few studies have involved children and adolescents. The aim of this research was to explore the relationship between time in range (TIR: 70-180 mg/dL), glycemic variability (GV) metrics and daily scan rate in a pediatric population with T1D during real-life conditions.

Method:

Monitoring data were collected in a pilot study between January 2018 and February 2020 within the Advanced Intelligent Distant-Glucose Monitoring (AID-GM) project, which involved 27 T1Dpediatric patients under multiple daily injection insulin-therapy. All participants used the isCGM Abbott’s FreeStyle Libre system. 14-days monitoring windows were split into 10 equally sized rank ordered groups considering different ranking variables in separate analyses: glucose standard deviation (SD), continuous overall net glycemic action at 1-hour (CONGA_l), average daily risk range (ADRR), and daily scan rate. For each ranking, a Dunn’s test was performed to verify if TIR was significantly different among groups (p-value<0.05); relationships were investigated through Spearman’s correlation coefficient.

Result:

Each patient was monitored on average for eight months. GV metrics showed a negative relationship with TIR (SD: r=-0.98; CONGA1 and ADRR: r=-0.99; p-value<0.01), which was significantly higher in groups with the lowest variability. Instead, daily scan rate revealed a positive relationship with TIR (r=0.75; p-value<0.02), which was significantly higher in the highest scan rate group (30.02 scans/day) compared to the lowest group (2.88 scans/day).

Conclusion:

GV reduction and elevated scan rate through isCGM systems are associated with high time spent in the euglycemic range in children and adolescent with T1D.

Diabetes Technology Experiences of Young People Living with Type 1 Diabetes and Their Parents - Analysis Guided by a Hybrid Theoretical Foundation

Nicola Brew-Sam, PhD; Madhur Chhabra, BDS, MPH; Anne Parkinson, PhD, AFHEA; Adam Henschke, MA, PhD; Ellen Brown; Lachlan Pedley; Elizabeth Pedley, RM, RN; Kristal Hannon; Karen Brown, BA, RN; Kristine Wright, BSc, RN, CDE; Christine Phillips, MBBS, MA, MPH, FRACGP, MD; Antonio Tricoli, MSc, PhD; Christopher J. Nolan, MBBS, PhD, FRACP; Hanna Suominen, MSc, PhD, MEDL; Jane Desborough, RN, RM, MPH, PhD

Department of Health Services Research and Policy, Research School of Population Health, College of Health and Medicine, Australian National University Canberra, Australia

nicola.brew-sam@anu.edu.au

Objective:

A strategy to understand young people’s needs regarding technologies for Type 1 Diabetes (T1D) management is to examine their day-to-day technology experiences. This study aimed to describe young people and their caregivers’ experiences and preferences regarding insulin pumps, sensor technologies, and related communication technologies based on a hybrid theoretical foundation, as well as to describe derived ideal device characteristics.

Method:

Sixteen face-to-face interviews were conducted with young people with T1D and their parents about their diabetes technology use. A combination of data-driven thematic analysis in a first stage, and theory-driven analysis in a second stage was used to incorporate in-depth study analysis and existing literature. Relevant theories included technology adoption and value sensitive design (VSD) models. Based on this approach ideal device characteristics were summarized.

Results:

Initial interview themes included aspects of self-management, device use, device-related technological characteristics, and feelings associated with specific device types. The data delivered information congruent with most theoretical technology adoption and VSD factors. Participants described their expectations of devices and a variety of ideal device characteristics. However, in line with previous research the analysis indicated that reality deviated from expectations, with inaccuracy problems and technical failures reported for most devices.

Conclusion:

Technologies for diabetes self-management require continual advancement to meet the needs and expectations of young people with T1D and their caregivers. Understanding their experiences, including challenges with devices is essential to ensure that future development of technologies aligns with the needs and expectations of the young people who will use them. The use of theory as a foundation for structuring the evaluation of technologies used to manage T1D provides a valuable and evidence-based approach to research and development.

Inexpensive Regular Human Insulin Used with CGM Provides Similar Outcomes to More Expensive Insulin Lispro

Philip Bunsick, MBA

Managing Director, Gemini Medical Inc. Shrewsbury, Massachusetts

phil@geminimedicalinc.com

Objective:

To compare the effectiveness of older, inexpensive, regular human insulin (Novolin-R, Wal-Mart, $25) to the more expensive insulin lispro (Humalog) for Time-in-Range and Average Glucose using an MDI regimen combined with continuous glucose monitoring (CGM).

Method:

A 57-year-old male with T1D used his usual lispro insulin and captured time in range (TIR) and average glucose (AG) data over 8 weeks using a Freestyle Libre 2 CGM system with alarms, followed by 8 weeks on regular human insulin. Glargine was used as the basal insulin at a stable dose for both parts of the study and MDI using disposable insulin syringes was the insulin delivery method.

Result:

The TIR and AG were similar with the two insulins. TIR was 81% with regular insulin and 80% with lispro. AG was lower with regular insulin at 116mg/dL, compared to 120mg/dL with lispro. The lower average glucose was due, in part, to increased time below 70mg/dL with regular insulin at 12% vs. 10% with lispro, seen more during overnight hours. The average daily dose of regular insulin was higher than the dose of lispro (64 units/day vs. 51 units/day).

Conclusion:

An MDI regimen using regular insulin can provide comparable results to the more expensive Lispro insulin with regards to glycemia control, when combined with an advanced CGM system. There was a marginal increase in hypoglycemia with regular insulin and the dose required was higher. When considering the costs and benefits of new technologies and insulins, MDI regimens with older less expensive insulin can provide excellent glucose control when used with a CGM system.

1,5-Anhydroglucitol Is an Independent Predictor of Mortality in Patients with COVID-19

Eric Button, MS, MBA; Hirotaka Ishibashi, MA; Elizabeth Clark, MS; Stewart Holt, PhD; Kevin Bliden, MBA; Udaya Tantry, PhD; Jeffrey R. Dahlen, PhD; Paul A. Gurbel, MD

Precision Diabetes, Inc. Releigh, North Carolina

ebutton@precisiondiabetesinc.com

Objective:

To assess the relationship between serum 1,5-anhydroglucitol (1,5-AG), a marker of glycemic variability, and mortality in COVID-19 patients.

Method:

Data from 64 hospitalized COVID-19 patients were collected between June 2020-February 2021 at Sinai Hospital (Baltimore, MD), including 9 patients who died in the hospital. Medical history, demographic variables, and biochemical measurements were taken at time of admission. Baseline means for fasting blood glucose and 1,5-AG were 143.5 mg/dL (SD 68.9) and 14.7 ug/mL (SD 8.8), respectively. HbA1c available in 40 patients – mean value 6.9% (SD 2.3).

Result:

Multivariate logistic regression analysis showed that only 1,5-AG (n=64) was an independent predictor of mortality (AUC = 0.69, p value 0.017). Fasting glucose (n=64) and HbA1c (n=40) were not statsically significant with AUCs of 0.60 (p value 0.322) and 0.58 (p value 0.464), respectively. In an analysis of clinical variables, a combination of BMI and Age was predictive of mortality (AUC = 0.77, p value 0.004). Interestingly, when 1,5-AG was added to BMI and Age, the AUC increased to 0.94 (p value <0.0001). When fasting glucose was added to BMI and Age the AUC was 0.79 (p value 0.001). A cox regression analysis showed an OR (1,5-AG < 10 ug/mL) for mortality of 0.44 (95% CI 0.11, 1.85).

Conclusion:

1,5-AG was an independent predictor of COVID-19 mortality. Fasting glucose and HbA1c showed no statistical significance to outcomes. The HbA1c findings confirm the results from other COVID-19 studies, but the finding that 1,5-AG outperformed fasting glucose in predicting mortality is new. 1,5-AG may provide important and unique information in the COVID-19 clinical setting. An algorithm of BMI, Age, and 1,5-AG may also be clinically useful.

Long-Term (7-Year) Glycemic Outcomes with Continuous Glucose Monitoring (CGM) Initiation within First Year of Type 1 Diabetes (T1D) Diagnosis

Anagha Champakanath, MBBS; Janet Snell-Gergeon, PhD; Viral Shah, MD

Barbara Davis Center for Diabetes Aurora, Colorado

anagha.sc25@gmail.com

Objective:

This study was aimed to evaluate long-term glycemic control in patients with T1D who initiated CGM within first year of T1D diagnosis.

Method:

We collected HbA1c at each visit from electronic health records of patients with T1D (1-35 years of age) diagnosed between 1/2013 and 12/2015 and started on CGM within 1 year of T1D diagnosis (CGM group) or not initiated CGM within 1 year from T1D diagnosis (Non-CGM group) and followed at the Barbara Davis Center till 12/30/2020. Linear mixed model was used to examine A1c levels by time points and by CGM use adjusted for age, gender, and insulin delivery method (pump vs multiple daily injections).

Result:

Eighty T1D patients (mean age at diagnosis 10.4 years, 60% female, 81% Whites, 96% private insurance) were included in the CGM group while 341 T1D patients (mean age 10.2 years, 52% female, 60% White, 56% private insurance) were included in the Non-CGM group. A1c at diagnosis was similar between two groups (11.4 ± 2.3 vs 11.6 ± 2.3, P=0.52). There was significant improvement in A1c in the CGM group at 6 months and maintained throughout 7 years of follow-up period [least square mean A1cs: 6 months 7.1% vs 8.0%, 1-year 7.3% vs 8.5%, 2-year 7.5% vs 9.0%, 3-year 7.4% vs 9.3%, 4-year 7.4% vs 9.6%, 5-year 7.6% vs 9.6%, 6-year 7.3% vs 10.0% and 7-year 7.5% vs 9.8%, all p<0.001) after adjusting for age at diagnosis, sex, and insulin delivery method.

Conclusion:

This is the longest CGM follow-up study to our knowledge demonstrating that early initiation of CGM from T1D diagnosis results in sustained improvement in A1c over 7-years.

Rate of Unexplained Hyperglycemia and Infusion Set Occlusion: Medtronic Extended Infusion Set (EIS) Pivotal Trial and 3-Day Infusion Set Study Comparison

Sarnath Chattaraj, PhD; John Shin, PhD; Gina Zhang, PhD; Jenny Fusselman, MS; Vivian Chen, MS; Evan Anselmo, BS; Cheryl Chambers, BS; Scott W. Lee, MD; Robert A. Vigersky, MD

Medtronic Diabetes Northridge, California

sarnath.chattaraj@medtronic.com

Objective:

A common problem with continuous subcutaneous insulin infusion therapy is development of unexplained hyperglycemia (UH) and infusion set (IS) occlusion. In a previous multicenter 13-week study in 256 participants wearing 3-day ISs for a mean of 71hrs, UH and/or IS occlusion occurred in >60% (van Bon et al.,Diabetes Technol Ther. 2011;13:607-614). Preclinical study demonstrated insulin preservative loss associated with significant increase in insulin aggregate formation and IS occlusion (Anselmo et al., J. Diabetes Sci. Technol. 2020;14:2[A3]). An EIS was developed to reduce preservative loss and improve insulin formulation stability for up to 7 days. The present study examined UH and IS occlusion rates during the EIS pivotal trial and those published in van Bon et al.

Method:

Participants (N=259, 18-80yrs) in the EIS pivotal trial had T1D, used the MiniMed™ 670G system with Humalog™ or Novolog™ insulin, and wore the EIS 12 consecutive times with each wear lasting for ≥174hrs or until device failure. The EIS pivotal UH (>250mg/dL glucose unamenable to correction) and IS occlusion data were analyzed and compared with UH (>300mg/dL glucose unamenable to correction) and IS occlusion data from van Bon et al.

Result:

The UH and IS occlusion rates, on day 7, in the EIS trial were 2.2% and 5.0% of participants, respectively, which were lower than the ~56% and ~27%, respectively, reported for the 3-day IS study.

Conclusion:

The EIS pivotal trial rates of UH and IS occlusion at 7 days were very low compared to those published for 3-day infusion sets, and with a stricter UH definition. These findings indicate EIS reduces burden and improves safety for those managing their diabetes with an insulin pump.

Potential Cost Savings from a Reduction in Sensor-Detected Severe Hypoglycemia among Users of the InPen Smart Insulin Pen System

Albert Chien, MA, MPH; Sneha Thanasekaran, MS; Angela Gaetano, MS; Glen Im, MS; Kael Wherry, PhD; Janice MacLeod, MA, RD, CDCES, FADCES

Medtronic Northridge, California

albert.a.chien@medtronic.com

Objective:

To compare the frequency of sensor-detected severe hypoglycemic events among a population of continuous glucose monitoring (CGM) users on insulin therapy after initiation of the InPen™ smart insulin pen system and to estimate the potential hypoglycemia-related medical cost savings across a population of InPen users.

Method:

For study inclusion, InPen users were required to have at least 90 days of active InPen use with a connected CGM device. The last 14 days of complete sensor glucose (SG) data within the 30-day period prior to the start of InPen use (“pre-InPen”) and the last 14 days of complete SG data along with requirement of at least one bolus entry per day within the 61 to 90-day period after InPen start (“post-InPen”) were analyzed. Sensor-detected severe hypoglycemic events (defined as ≥10 minutes of consecutive SG readings less than 54 mg/dL) were determined. Once factored, the expected medical intervention rates and associated costs for severe hypoglycemia-related hospitalization, emergency room visits and ambulance transportation were calculated. Intervention rates and costs were obtained from the literature.

Result:

There were 1,681 InPen + CGM users from March 1, 2018 to April 30, 2021. The mean number of sensor-detected severe hypoglycemic events per week declined from 0.67 in the pre-InPen period to 0.58 in the post-InPen period (p=0.008), which represented a 13% reduction (31% reduction for age group ≥65 years, N=166). The estimated cost-reduction associated with reduced severe hypoglycemic events was $237 and $2,176 per InPen user per month and per year, respectively.

Conclusion:

Use of the InPen smart insulin pen system with a connected CGM is associated with reduced sensor-detected severe hypoglycemia, which can result in significant cost savings.

Diabetic Eye Disease and Virtual Assistant AI: Do They Help Our Patients?

Daniel Chien, BS; Gloria Wu, MD; Vincent Siu; Srija Gadiraju; Joselyn Alanzalon, BS; Sahej Sidhu; Ting-I Sung, BS; Justin Huynh, BS

University of California, Los Angeles Los Angeles, California

gwu2550@gmail.com

Objective:

Evaluate Virtual Assistant AI devices for diabetic patients with eye disease.

Methods:

1) Questions: “I have Type 2 Diabetes - what do I do?”, “I can’t see and I have diabetes. What do I do?”, “I have diabetes - what do I do?”, “Does diabetes cause me to have problems seeing?”, and “My child has Type 1 Diabetes. What should I do?”

2) Virtual Assistant Devices AI was used via the Alexa app on Google Play Store and Google Home Mini. Responses were recorded from the Alexa app and directly transcribed into text format. Google Home AI required a laptop with Google Document “speech to text” tool.

3) The texts were pasted into WebFx.com and analyzed for readability (RD) via the Flesch Kincaid scale (FK) and SMOG Index.

Results:

Reading Ease Scale (FK): Alexa (54.0 15.0), Google Home (65.1 12.1) (higher = easier reading)

Grade Level (FK): Alexa (8.72.4), Google Home (6.81.9)

Grade Level (SMOG): Alexa (9.12.0), Google Home (7.61.6)

Average Word Count / Response: Alexa (71.624.0), Google Home (51.67.2)

Average Response Time (sec): Alexa (22.79.9), Google Home (17.62.1)

Words ≥ 3 syllables (eg. diabetic retinopathy) lower the readability score. Alexa was unable to answer “Does diabetes cause me to have problems seeing?” In comparison, Google Home’s answer included “diabetic retinopathy”, “diabetic macular edema”, “cataracts”, and “glaucoma”.

Conclusion:

The voice assistant devices’ use of polysyllabic words in their AI responses may hamper understanding of diabetes and diabetic eye disease to the lay public.

Virtual Care in the COVID-19 Era: Assessing Telemedicine Experiences among Diabetes Care and Education Specialists

Evelyn Cox, BA; Jacqueline Tait, BA; Emily Ye, BA; Julia Stevenson, BA; Rebecca Gowen, BA; Emily Xu, BA; Edem Asamoa, BA; Erik Monroy-Spangenberg, BA; Richard Wood, BSc, MBA

dQ&A Diabetes Research San Francisco, California

evelyn.cox@d-qa.com

Objective:

The rapid rise of telemedicine, necessitated by the COVID-19 pandemic, has changed how care is administered and created additional burdens for healthcare professionals. This study investigated how Diabetes Care and Education Specialists (DCES) have implemented telemedicine in their practices to identify possible areas for improvement.

Method:

350 DCES from an opted-in US research panel were surveyed. Respondents were asked about their expected and present use of telemedicine, satisfaction with telemedicine versus in-person appointments, and perceived changes in the frequency of treatment decisions made over telemedicine compared to in-person.

Result:

On average, respondents reported that 57% of appointments were conducted using telemedicine this year and estimated a decrease in telemedicine use during 2021 (41%).

While 67% of respondents were satisfied with in-person visits (selecting a 9 or 10 on a 10-point scale), just 25% were satisfied with telemedicine visits. Analysis of verbatim comments revealed dissatisfaction with telemedicine stems from difficulty with teaching patients virtually (20%), technology use (18%), establishing personal connections with patients (16%), and obtaining patient device data (11%).

DCES indicated that larger treatment decisions like starting new therapies/devices occur less often over telemedicine whereas adjusting patient’s settings occurs more often. For example, relative to in-person visits, 64% report starting patients on new insulin pumps less often while 62% report adjusting pump settings just as often.

Conclusion:

Telemedicine has become a regular part of DCES’s practices despite notable gaps in satisfaction and ability to make certain treatment decisions when compared with in-person visits. Although telemedicine may be a useful tool for adjusting doses and device settings, these findings emphasize the need for improvements surrounding virtual diabetes care to alleviate the challenges experienced by providers.

Development of Self-Management Behavior Scores and Profiles with Digital Health Data

Michelle Dugas, PhD; Di Hu, MSIS; Abhimanyu Kumbara, MS, MBA; Shiping Liu, MS; Kenyon Crowley, MSIS, MBA; Anand K. Iyer, PhD, MBA; Malinda M. Peeples, RN, MS, CDE; Mansur Shomali, MD, CM; Guodong (Gordon) Gao, PhD

University of Maryland College Park, Maryland

mdugas@umd.edu

Objective:

This research aims to leverage patient-generated health data to develop measures that summarize a person’s self-management behavior in a standardized, easily understood metric. Furthermore, this research aims to identify segments of patients with distinct patterns of self-management behavior that could then be used for targeted outreach and support.

Method:

With data from 270 users of a digital health solution, we generated scoring systems to assess the quality of users’ diabetes self-management based on three core behavioral dimensions: physical activity, carbohydrate intake, and medication use. Scores capture both self-monitoring and the extent to which behavior meets personal guidelines, and range from 0 (no information/behavior) to 100 (meets or exceeds guidelines). We then perform k-means clustering to identify distinct subgroups of user profiles based on their baseline blood glucose (BG) readings (first four weeks from activation) and eight weeks of behavioral scores.

Result:

Cluster analyses identified four distinct profiles of self-management. Cluster 1 (n = 102) consisted of low engagers with low scores across all behaviors and weeks. Cluster 2 (n = 62) exhibited low scores for physical activity but moderately high scores for carb intake and medication use. Cluster 3 (n = 73) showed high average scores on all three behaviors throughout the 8 weeks. Finally, Cluster 4 (n = 3) exhibited low scores for carb intake and medication use but very high scores on physical activity. Clusters 3 and 4 tended to have lower baseline BG than Clusters 2 and 1.

Conclusion:

This research highlights the promise of developing summary scores of diabetes self-management behavior by leveraging patient-generated digital health data, and potential benefit from differentiated messaging and support for their self-management.

Current Technology Use in Pediatric Patients with Type 1 Diabetes: An Equity-Focused Analysis

Katie Fogle, BS; Jessica Schmitt, MD

University of Alabama at Birmingham School of Medicine Birmingham, Alabama

kfogle@uab.edu

Objective:

To evaluate current use of continuous subcutaneous insulin infusion pumps in pediatric patients with type 1 diabetes (T1D) treated at an academic center in the southeast.

Methods:

Data was collected from an IRB-approved Diabetes Registry. Variables included demographics, hemoglobin A1c, and technology use. Patients with T1D for more than 12 months seen in-person or by telehealth from August 2020 through May 2021 were included. Publicly insured patients were defined as those insured through Medicaid, Medicare, or the Children’s Health Insurance Program. Analysis was done in GraphPad Prism 9.1.0.

Results:

In all, 602 of 1670 unique patients (36%) used a pump. Rates were lower in publicly relative to privately insured patients (25% vs 46%, p < 0.0001). Rates were lower in non-white patients (20.6% vs 42.5%, p < 0.0001). Median A1c for pump users was 8.2% compared to 9.1% for nonusers (p < 0.0001). In total, 65% of patients using a pump had an A1c <9% while only 47.8% of nonusers met that target (p < 0.0001).

Conclusion:

In our cohort, pump use was lower in minority and publicly insured patients. Barriers to pump use may include insurance coverage as well as the need to attend to several training sessions. Need to attend several visits may disproportionally affect those with fewer resources. With significant differences in A1c, we must ask: in real-world settings, does A1c improve due to pump usage, or is an A1c > 9% an effect of socioeconomic factors that impact glycemic control? Is disproportionate pump use due to biased prescribing and/or obstacles to pump coverage that disproportionately affect publicly insured and minority patients? Longer prospective cohort analysis may help answer this further.

I Am Diabetic: Can I Speak to Virtual Assistants in Different Languages?

Srija Gadiraju; Gloria Wu, MD; Vincent Siu; Daniel Chien, BS; Sahej Sidhu; Justin Huynh, BS; Joselyn Alanzalon, BS

California Northstate University San Jose, California

srijasgadiraju@gmail.com

Background:

7.7 M Hisp and 2 M Asians have Diabetes in the US. Many don’t speak English as their native language.

Objective:

Evaluation of Alexa and Google Home as a patient resource for non-English speakers.

Methods:

Virtual Assistant Device (VAD) Alexa was used to answer questions in English(E), Spanish(S), and Hindi(H). Google Home was used for Chinese(C). Responses were recorded from the device using a computer.

8 questions:

1) I have diabetes - what do I do?

2) I have Type 2 Diabetes - what do I do?

3) I can’t see and I have diabetes. What do I do?

4) I can’t see. Is this because of diabetes?

5) Does diabetes cause me to have problems seeing?

6) Does diabetes cause kidney problems?

7) Does diabetes cause problems with my feet?

8) My child has Type 1 Diabetes. What should I do?

We counted the number of words to evaluate multilingual replies.

Results:

(Word Count) = (WC), NR = no reply

Q1: E(88), S(102), C(NR), H(41)

Q2: E(91), S(59), C(NR), H(17)

Q3: E(88), S(102), C(87), H(NR)

Q4: E(88), S(NR), C(134) H(19)

Q5: E(42), S(46), C(114), H(NR)

Q6: E(24), S(36), C(103), H(27)

Q7: E(37), S(149), C(128), H(20)

Q8: E(49), S(33), C(80), H(49)

Avg WC: E(63.38), S(65.88), C(107.7), H(21.63). “No replies” occurred only in Spanish, Chinese and Hindi. Hindi had the most non-Medical responses.

Conclusion:

Virtual Assistants are useful for diabetics but could improve their non-English interface.

Outpatient Full Closed-Loop Trial Using an Open-Source Remote Monitoring and Artificial Pancreas Platform

Fabricio Garelli, PhD; Emilia Fushimi, ENG; Nicolás Rosales, PhD; Delfina Arambarri, ENG; Cecilia Serafini, ENG; Hernán De Battista, PhD; Ricardo Sánchez Peña, PhD; Julia García-Arabehety, MD; Javier Giunta, MD; Luis Grosembacher, MD

GCA, LEICI (UNLP-CONICET), Facultad de Ingeniería, UNLP La Plata, Buenos Aires, Argentina

fabricio@ing.unlp.edu.ar

Objective:

The first artificial pancreas (AP) ambulatory clinical trial in Argentina conducted in March 2021 in the COVID-19 pandemic context is reported. The main objective of this trial was to evaluate the feasibility of running advanced full closed-loop (FCL) algorithms for glycemic control in an own and free platform developed by the team at UNLP from open-source resources (www.insumate.com.ar).

Method:

The ARG algorithm, a FCL algorithm previously evaluated in-vivo [1], was implemented in InsuMate-AP. This platform was then connected to Dexcom G6 CGMs and Accu-Chek Spirit Combo insulin pumps. After configuring conventional open-loop (OL) treatment in the pumps, five adults with Type 1 Diabetes Mellitus completed one week of an ambulatory clinical trial, consisting in 3 days of OL treatment followed by 3 days of FCL glycemic control (i.e., without delivering meal priming insulin boluses). The participants performed regular activities in a hotel, including networking, eating without carbohydrates counting and extensive walking (3-5 km) around the neighborhood.

Result:

The last 24 hours of each phase were compared, finding a 25,3% increase in time in range (TIR, 70 mg/dl<G<180mg/dl), a 24,5% reduction of the time below range (TBR, G<70 mg/dl) and a 21,1% decrease in time above range (TAR, G>180 mg/dl) during FCL vs OL treatment. The InsuMate system performed reliably and showed good connectivity performance with both the multiple remote monitoring interface and the peripheral devices, achieving adequate FCL operation for 95.4% of time.

Conclusion:

The InsuMate system and the ARG algorithm were evaluated under challenging and patient-relaxed conditions showing safe and effective behaviour.

[1] Sánchez-Peña et al., “Artificial pancreas: Clinical study in Latin America without premeal insulin boluses,” JDST, vol. 12, no. 5, pp. 914–925, 2018.

Increased Time-in-Range after Meals When Using the SNAQ App

David Haldimann, MSc; Aurelian Briner, MA; Nico Previtali, MSc; Alexander Pfyffer, MA; Stefan Ebener, HF; Maurice Ducret, MAS

SNAQ AG Zurich, Switzerland

aurelian@snaq.io

Objective:

Learning the impact of meals on postprandial time in range (TIR) is an important element in the management of diabetes. The objective of this study was to compare the TIR of patients when they were using the SNAQ app against the TIR when they did not use the app. By visualizing glucose data with meals and nutritional information, the app facilitates users to gain insights and detect patterns to improve TIR.

Method:

During a collaborative study with Ypsomed, 67 people with T1D from Germany, the Netherlands, the UK, and Ireland using a “mylife YpsoPump” received the app to log their daily meals for 2 months. For those users with a Dexcom CGM, continuous glucose data was imported into the app to visualise glucose curves alongside the meals. The effect on postprandial TIR was analysed based on the 3 hour postprandial glucose data when using the app and standardized meal time slots when not (8am, 12pm and 6pm). Values between 4 and 10 mmol/l were considered as in range. A subset of 23 users with at least 1,000 glucose data points imported and 5 or more meals recorded were considered for this analysis.

Result:

On average the postprandial TIR was observed to be 9.5% higher after meals that were recorded using the app (n=1,603) versus standardized time slots when no meal was recorded in the app (n=2,181).

Conclusion:

The real-world findings suggest a benefit in postprandial TIR for patients due to the insights and information generated by the app when meals are logged. These results indicate the necessity for a controlled clinical study to further analyse the benefits of the app in managing and improving postprandial TIR.

Unsupervised Anomaly Detection Algorithms to Identify Compression Artifacts in Continuous Glucose Monitoring

Elena Idi, MS; Andrea Facchinetti, PhD; Giovanni Sparacino, PhD; Simone Del Favero, PhD

Department of Information Engineering, University of Padova Padova, Italy

idielena@dei.unipd.it

Background:

Continuous glucose monitors (CGMs) provide real-time blood glucose concentrations that are essential for the automated treatment of type 1 diabetes (T1D). Compression artifacts, known as Pressure Induced Sensor Attenuations (PISAs), are transient failures of the sensors due to the application of a pressure to the site: the resulting underestimation of the blood glucose concentrations may lead to inappropriate insulin dosing decisions, which can be prevented with the automated detection of the faults.

Methods:

We generate 100 CGM traces using the UVA/Padova type 1 diabetes simulator, and we corrupt them with compression artifacts. From the data, we extract suitable features to make more noticeable the malfunctions and we perform the feature selection to determine the optimal feature set. We employ unsupervised anomaly detection algorithms: in this way, we remove the need of labeled data for training, that are hard to obtain in practice, and we avoid the sub-task of identifying a physiological model. The performance of different methods is analyzed and compared.

Results:

The best performance is achieved with the Histogram-based Outlier Score (HBOS) and the Isolation Forest (IF), which detect respectively the 64% and 60% of the failures (with precision equal to 68% and 67%). These methods outperform the rule-based approach of the state-of-art (Baysal et al., JDSD 2014).

Conclusions:

The unsupervised anomaly detection algorithms can be used effectively to identify pressure-induced malfunctionings. Future studies will test the proposed detection strategy on real-data.

Identifying Patients at High Risk for Cystic Fibrosis Related Diabetes through the Use of the Freestyle LibrePro CGM

Jeniece Ilkowitz, RN, MA, CDE; Chris Mavaro, NP; Annie Katzenberg, RN; Chris Lally, RN, CDE; Robert Giusti, MD; Mary Pat Gallagher, MD; Emily Breidbart, MD

NYU Langone, NYU Pediatric Diabetes Center New York, New York

Jeniece.ilkowitz@nyulangone.org

Objectives:

No literature investigates use of the Freestyle LibrePro Continuous Glucose Monitoring (LibrePro CGM) System in Cystic Fibrosis (CF), although studies report CGM reliability in evaluation for glucose excursions predicting CF-related diabetes (CFRD). In an effort to decrease the burden of yearly Oral Glucose Tolerance Testing (OGTT), we proposed a pilot study using LibrePro in a CF clinic.

Method:

Patients six and older, without history of hyperglycemia or recent steroids, were recruited. Subjects consented and LibrePro was placed. Fructosamine, OGTT, HbA1c and CGM data were collected and reviewed using descriptive analysis.

Results:

Ten subjects (3 male) enrolled [average age = 16 years (range 10-21), HbA1c = 5.5% (4.7-6), fructosamine = 239.6g/dl (220-264), OGTT results: fasting = 88.6 mg/dl (77-103), 60min =167.7 mg/dl (86-281), 120min = 109mg/dl (72-147)]. Average sensor wear was 12.9 days (8-15), time-in-range: 93.8%, time over 180mg/dl: 2%, and time less than 70mg/dl: 4.2%. In subjects (n=6) with CGM glucose excursions over 180mg/dl, four had abnormal 60 minute OGTT results.

Conclusion:

The LibrePro provides an opportunity for glucose assessment in CF. The ease of this one-time use device, which requires no calibrations and is easily placed with a one button insertion, may help identify patients at risk for CFRD. Our findings indicate the LibrePro identified all subjects with an abnormal OGTT glucose level plus two subjects without OGTT abnormalities. The 60 minute OGTT value was most sensitive in identifying dysglycemia. By introducing this CGM use as part of CFRD screening, it may increase comfort of use when remote monitoring is beneficial (e.g. acute illness, need for telehealth visits). Future studies must recruit larger cohorts and assess remote options and user preferences.

How Telemedicine Assists Adults with Type 2 Diabetes Obtain Adequate Glucose Control

Maribel Jimenez, DNP, FNP-BC, AMB-BC, RN

New York Presbyterian Hospital New York, New York

Maj9036@nyp.org

Objective:

Diabetes is one of the leading causes globally of morbidity and mortality. Its growing incidence is alarming and the impact that its comorbidities are inflicting upon patients of all ages is detrimental. The cost of healthcare is outrageously increasing while access to care seems to be deteriorating.

Providing patients with additional support in between office visits promises earlier adoption of better self-care behaviors, identifying knowledge gaps, and meeting patient’s needs in real-time. Patients will have the opportunity to interact with a healthcare professional who will guide them in taking the necessary steps to achieve optimum control of their disease.

Method:

An integrative review and synthesis of 15 articles were done to further investigate how the use of technology was compared and contrasted and to confirm the impact its integration has on patients diagnosed with type 2 diabetes. This review allowed investigation of the specifics of its design and purpose as well as the opportunity to identify any gaps interfering with the desired patient outcomes.

Results:

The final results of this integrative review were mixed. All of the studies presented valuable information to consider and utilize in the further development of telemedicine.

Conclusion:

Incorporating telemedicine into the patient’s usual care is a promising venue to help patients obtain adequate control of their diabetes. This integrative review allowed confirmation for the need for reassessments of the program designs and identified gaps addressed.

Higher Dose of Inhaled Technosphere Insulin Provides Significant Reduction in Post Prandial Glucose Excursions, without Hypoglycemia

Kevin Kaiserman, MD; Mark Christiansen, MD; Sunil Bhavsar, MS; Johanna Ulloa, BSE; Brandi Santogatta, BA; Timothy S. Bailey, MD, FACE, CPI

MannKind Corporation Westlake Village, California

kkaiserman@mannkindcorp.com

Objective:

Technosphere Insulin (TI), is an ultra-rapid-acting inhaled insulin. This proof of concept study compared the efficacy of the TI dose per current US labeling (USPI) to a higher (~2x) dose as measured by mean change in postprandial glucose excursion (PPGE) from baseline over 2 hours for an identical standardized meal. Secondary objectives were to evaluate hypoglycemia, change in FEV_l, and monitor for any other adverse events.

Methods:

Twenty patients with T1D or T2D on basal-bolus insulin therapy were enrolled. Each subject received an initial pre-prandial dose of TI based on the instructions for switching from injected mealtime insulin in the current USPI. At a subsequent visit, all subjects received a pre-prandial dose of TI calculated by doubling the injected mealtime insulin dose and rounding down to the nearest TI cartridge. The protocol specified administering TI doses immediately prior to consuming an identical standardized meal. Capillary glucose was measured by Ascensia Contour™ meters.

Results:

19 of the 20 enrolled subjects completed the study per protocol. The higher (~2x) conversion dose provided significant reductions in PPGE from 45 minutes post-meal (50.5 vs 29.0 mg/dL @ 45 min, P=0.031; 94.1 vs 41.5 mg/dL @ 120 min, P=0.022). Peak glucose was reduced from 234 to 186 mg/dL (p=0.034). There were no hypoglycemic events observed during the 2-hour post-meal period. No significant changes in FEV_l measured from before to after either dose of TI.

Conclusions:

A higher (~2x) pre-prandial dose of TI significantly reduced post-prandial glucose excursion vs. the current USPI recommended dose without any new safety concerns. When confirmed, this new TI dosing recommendation may help patients and clinicians to minimize postprandial hyperglycemia without fear of increased hypoglycemia.

Clinical Evaluation of the Lumee™ Glucose System - a Fluorescence-Based Long-Term Continuous Glucose Sensor

Ulrike Kamecke, Dipl-Ing(FH); Sayanti Banerjee, PhD; Kerstin Rebrin, MD, PhD; Nina Jendrike, MD; Manuela Link, Dipl-Ing(FH), MD; Eva Zschornack, MD; Collin Krauß, Dipl-troph; Guido Freckmann, MD

Ulrike Kamecke, Profusa, Inc. Emeryville, California

ulrike.kamecke@profusa.com

Objective:

The Lumee™ Glucose System is designed to continuously measure glucose concentrations in subcutaneous tissue. A small hydrogel sensor (2.5x0.4x0.3mm, dehydrated) is injected remaining permanently. A wireless optical reader is placed onto the skin for monitoring. The reader interrogates the sensor with red/near infrared LED light and collects fluorescent signals proportionally to glucose. Besides establishing safety and functionality, this feasibility study is intended to inform algorithm development for calibration eliminating potential disturbances such as motion and temperature.

Method:

Two hydrogels have been injected into the upper arm of subjects with insulin dependent diabetes. Monitoring sessions from 10 subjects during a 3 month period are reported. Functionality is determined by the ability of sensors to track meal-induced glucose changes with capillary BG references during multiple 8-hour in-clinic visits allowing to mimic daily activities. A data quality filter has been applied with consideration of proper reader location and attachment. A three-point retrospective calibration employing an automated machine-learning (ML) approach has been introduced.

Result:

Lumee™ glucose values have been estimated based on a preliminary calibration algorithm trained by machine learning without separate temperature and movement corrections. Analysis of Lumee™/reference glucose data pairs reveal a MARD below 20% with lower and stable numbers past the two-week time point. Values presented in the Parkes Error Grid result in 97.1% within zones A and B (63.3% and 33.8% respectively) and 2.9% in zone C.

Conclusion:

The Lumee™ Glucose System does track glucose excursions for a minimum of 3 months. Applying a preliminary ML calibration algorithm generates glucose values with reasonable accuracy. Additional efforts are ongoing to improve reliability of measurement results by implementing specific corrections for temperature changes and movement disturbances.

Phenolic Preservatives, Present in Commercial Insulins, Trigger Inflammation at Infusion Sites via Mast Cell Activation

Shereen Kesserwan, MS; Li Mao, BS; Don Kreutzer, PhD; Ulrike Klueh, PhD

Wayne State University, Biomedical Engineering Detroit, Michigan

el6832@wayne.edu

Objective:

A major obstacle to prolonging the useful life of insulin infusion sets is the phenolic preservative (IPP), that are in commercial insulins (CI), which induce skin inflammation. However, the mediators and mechanisms underlying these tissue reactions remain to be elucidated. Skin mast cells (MC) and their products play a major role as they are the “first responders” in skin breach/injury and are key effector cells in skin inflammation. We hypothesized MC trigger IPP induced skin inflammation. Our goal is to determine the contribution of MC to IPP induced inflammation seen during insulin infusion therapy.

Methods:

Mouse and human MC cell lines evaluated time and concentration-dependent cytotoxicity, as well as MC degranulation, following in vitro treatment with IPP. A modified mouse air pouch model was used to quantitatively assessed IPP induced inflammation (i.e., cell influx) in MC deficient and innovative transgenic (Cre/loxP) mouse models. MC activation was also assessed in the physiologically relevant porcine model.

Result:

These studies demonstrated that CI containing IPP or IPP alone, at concentrations >1% of CI formulations, induced significant MC degranulation in vitro. Subsequent MC deficient and transgenic mouse studies demonstrated a significant reduction in tissue inflammation (i.e., decreased cell influx) when compared to normal (MC+) mice. Swine histopathology analysis demonstrated intense inflammation at IPP infusion sites over a 7-day period including increased MC presence at these tissue sites.

Conclusion:

These studies provide insights into the key role(s) of MC and MC products in IPP induced inflammation during insulin infusion. Furthermore, these studies indicate the therapeutic potential of targeting MC, to attenuate insulin infusion induced inflammation, to enhance and extend effective glycemic control for individuals with diabetes.

The Influence of Health Insurance on the CGM System Choice of a Type 1 Diabetic

Carolin Kinzel, MSc

Neu-Ulm Bavaria, Germany

carolin.kinzel@hnu.de

Objective:

With regard to continuous glucose monitoring (CGM) systems use by type 1 diabetics in Germany, the aim of the study was to investigate whether health insurance has an influence on getting the CGM system which the patient wants to have or which the physician has recommended. So beside the major stakeholders physicians and patients, also the health insurances might control the CGM system choice.

Therefor the study analyzed if there is a connection between different types of health insurances (“Bei welcher Krankenkasse sind Sie versichert?”) and the use of CGM systems (“Nutzen Sie aktuell ein CGM-System (Continuous Glucose Monitoring System)?”).

Method:

A quantitative survey was conducted with an online questionnaire. A total of 2,102 patients were interviewed, of which 1,642 patients (type 1 diabetics) completed the questionnaire. Statistical analyzes were performed by using the statistics software SPSS. A cross table was first generated to represent the usage figures (relative frequencies–proportions) between the two items under study. A chi-square test was used to prove statistical significance.

Result:

The cross table showed an unequal proportion ratio between the two items under investigation. The chi-square test showed a significant result with a two-sided p-value of 0.013 at a significance level of 5 percent (p ≤ 0.05).

Conclusion:

The type of health insurance has an influence on the choice of CGM system and thus the economic consideration must be included in the approval process.

Evaluation of the Safety and Efficacy of the Lalani Insulin Infusion Protocol

Benjamin Lalani; Kevin P. Gosselin, PhD; Sunil Lalani, BE; Horacio Rilo, MD; Atul Lalani, MD

Johns Hopkins University Baltimore, Maryland

blalani2@jhu.edu

Objective:

Although it is imperative to control blood glucose in hospitalized patients, maintaining tight glucose control poses the risk of hypoglycemia. This study’s objective is to evaluate the safety and efficacy of the novel Lalani Insulin Infusion Protocol (LIIP), a computerized insulin infusion protocol that uses a unique algorithm with dynamic targets based on glycemic status in critically ill patients requiring IV insulin.

Method:

We conducted a retrospective analysis of 1172 patients in whom LIIP was used (8/18/2020 to 6/25/2021) at six HonorHealth Hospitals (Arizona). This sample was compared to patients who used the previous computerized protocol, Glucommander™ (1/1/2018 to 8/17/2020, n=4700). Primary endpoints of the study include time to euglycemia, % of time in target range (70-180 mg/dl), % of time in hyperglycemia, and time in hypoglycemia.

Result:

We found that the LIIP protocol had a faster time to euglycemia (180 vs. 223 minutes), similar percentage of time in euglycemia (88.12% vs. 88.34%), similar percentage time in hyperglycemia (11.70% vs. 11.51%), and similar amount of time in hypoglycemia after euglycemia (3 vs. 3 minutes) compared to Glucommander™. The magnitude of hyperglycemic and hypoglycemic events was lower in LIIP. LIIP was used in 148 patients with COVID-19 treated with high-dose steroids.

Conclusion:

For HonorHealth Hospitals, LIIP was a safe and effective method for titrating insulin infusion and resulted in a quicker time to euglycemia compared to the previous method. Contributing factors to the success of LIIP could be the ease of use and learnability (as stated by the hospital nursing staff), ability to be incorporated in the EHR and dynamic and adaptive algorithm.

Determining Losses in Jet-Injection Subcutaneous Insulin Delivery

Alexander D. McHugh, BE(Hons); J. Geoffrey Chase, PhD; Jennifer L. Knopp, PhD; Lui Holder-Pearson, BE(Hons); Jennifer J. Ormsbee, BS; Diana G. Kulawiec, BS

Centre for Bioengineering, Department of Mechanical Engineering, University of Canterbury Christchurch, New Zealand

alex.mchugh@pg.canterbury.ac.nz

Objective:

Accurate and safe glycaemic management requires reliable delivery of insulin doses. Insulin can be delivered subcutaneously for action over a longer period of time. Needle-free jet injectors provide subcutaneous (SC) delivery without requiring needle use, but the volume of insulin absorbed varies due to losses associated with the delivery method. This study employs model-based methods to determine the expected proportion of exogenous insulin delivered from a needle-free subcutaneous dose.

Method:

Insulin, C-peptide, and glucose assay data from an oral glucose tolerance test (OGTT) trial with a 2 U SC insulin delivery, paired with a well-validated metabolic model, are used to predict metabolic outcomes for N=8 healthy adults. Subject-specific non-linear hepatic clearance profiles are modelled over time using second-order b-splines with knots at assay times. Hepatic clearance profiles are constrained both within physiological values and relative to glucose profiles. Values of insulin loss proportion and hepatic clearance are determined based on optimal insulin assay fit.

Result:

Parameter identification shows 20% to 30% of a nominal subcutaneous dose is lost and does not appear as plasma insulin. Hepatic clearance was on average 0.15 per minute over all patients, well within physiological expectations, and varied inversely to plasma glucose as also expected physiologically.

Conclusion:

Model-based analysis shows needle-free subcutaneous injection of a nominal dose of insulin does not provide metabolic action equivalent to the total dose, indicating losses up to 30%. By quantifying and accounting for this dose variability, better glycaemic management outcomes using needle-free SC injection may be achieved.

Generation of Realistic Virtual Diabetes Patients Using a Pix2Pix GAN: In Pursuit of an AI-Based Diabetes Simulator

Omer Mujahid, MSc; Ivan Contreras, PhD; Josep Vehi, PhD

Model Identification and Control Laboratory, Institut d’Informatica i Applicacions, Universitat de Girona Girona, Spain

omer.mujahid@udg.edu

Objective:

Simulators play a pivotal role in the in silico clinical trials for diabetes healthcare. Almost all of the simulators till now have been designed using physiological models of the human body. It is; however, impossible to incorporate the entire physiology of the human body in a single simulator. This calls for an artificial intelligence (AI) based diabetes simulator that could learn from data the entire physiology of the human body as well as hidden behaviors or medication.

Method:

Blood glucose (BG) profile and estimated plasma insulin are used as output and input respectively by a Pix2Pix generative adversarial network (GAN). 1119 days of data from 29 patients was used for training and validation. Six hours of input and output samples are first converted into images using the HSV color map to obtain a dataset of 4478 input-output image pairs. Plasma Insulin images from the validation set were used to obtain distinct glucose outputs.

Result:

Primary and secondary outcomes indicating the BG profiles of diabetic patients were compared against clinical data. The mean BG value of 168.69 mg/dL against 162.8 mg/dL, percentage time in range (70 – 80 mg/dL) of 62.03 % against 63.26 % and the coefficient of variation of 32.6 against 37.2 were obtained.

Conclusion:

The results showed that the unique quality of generative networks to learn from data and generate outputs that follow the original distributions but are completely novel may prove to be a good tool for realistic diabetes data generation. The model’s outputs depicted that the generated BG values were conditioned on input insulin values and followed patterns that were similar to the actual patients’ BG profiles.

Predicting Postprandial Glucose Excursions with Nutrient Content Using an Interpretable Random Forest Augmented by a Digital Twin ODE Model

Rahul Narayan, MS; Leah Wilson, MD; Jessica R. Castle, MD; Mark Clements, MD PhD; Francis J. Doyle, III, PhD; Melanie Gillingham, PhD; Michael R. Rickels, MD; Michael Riddell, PhD; Corby K. Martin, PhD; and Peter G. Jacobs, PhD;

Oregon Health & Science University Portland, Oregon

narayanr@ohsu.edu

Objective:

We explored how macronutrient intake (carbohydrates, fats, proteins) as well as fiber, water, and caffeine may improve the prediction accuracy of the postprandial glucose response in people with type 1 diabetes (T1D) using a random forest augmented by an ordinary differential equation (ODE) physical model of metabolism (digital twin).

Method:

Data used for model development was from 25 adults with T1D monitored for 28 days using continuous glucose monitors, insulin pumps/pens, exercise fitness trackers (Garmin), and food intake collected with a custom smartphone app. Nutrient content was estimated by nutritionists from photos of food consumed in four days/week (977 meals). A random forest model augmented by an ODE digital twin was trained to predict incremental area under the glucose curve (iAUC) 3-hours following meals; 89% of meals were used for training and 11% were used for validation. Effect of macronutrient content and other predictor variables were assessed using Shapley coefficients.

Result:

Shapley analysis revealed that carbohydrates, caffeine, and alcohol all tended to increase post-prandial iAUC, whereas larger amounts of fats, proteins, and fiber decreased iAUC; the impact was dependent on meal size. Rootmean-squared-error decreased by 1.4% when including protein and fat as predictors. The ODE model alone explained 6% of iAUC variance while random forest alone explained 25% of the variance and the combined ODE and random forest explained 30% of the variance.

Conclusion:

Including macronutrient inputs to postprandial glucose prediction algorithms improves prediction accuracy. Combining data-driven machine learning and ODE-based physical models is promising for improving prediction accuracy.

Preferences for Connected Insulin Pens: A Discrete Choice Experiment among Type 1 and Type 2 Diabetes Patients

Rachel S. Newson, PhD; Sebastian Heidenreich, PhD; Jaein Seo, PharmD; Esraa Aldalooj, PhD; Jiat-Ling Poon, PhD; Erik Spaepen, MSc; Elizabeth L. Eby, MPH

Eli Lily and Company Sydney, Australia

Newson_rachel@lilly.com

Objective:

To elicit preferences of people with diabetes (PwD) for connected insulin pens and associated mobile applications that automatically record and integrate insulin dosing information with relevant data from other devices (e.g. glucose).

Methods:

A literature review and interviews informed a survey in the US and UK, which included background questions, a video-introduction, and a discrete choice experiment (DCE). In the DCE, PwD repeatedly chose between a standard pen and two connected pens characterized by five attributes: 1) device type (i.e. attached vs integrated Bluetooth^® transmitter); 2) dosing support; 3) glucose monitoring; 4) additional features (i.e. physical activity and dietary trackers); and 5) data sharing functions. A logit model estimated relative attribute importance (RAI). Predicted choice probabilities (PCPs) were obtained for 15 specifications of insulin pens and corresponding mobile applications. This is an interim analysis.

Results:

143 T1DM and 361 T2DM PwD completed the survey. A connected pen was selected in 78.8% of DCE choices (T1DM: 72.9%; T2DM: 81.0%). Preferences were most affected by glucose monitoring (total: RAI = 41.4%; T1DM: RAI = 42.1%; T2DM: RAI = 40.9%), followed by additional features (total: RAI = 21.1%; T1DM: RAI = 23.3%; T2DM: RAI = 19.9%), data sharing (total: RAI = 19.6%; T1DM: RAI = 15.1%; T2DM: RAI = 22.4%), dosing support (total: RAI = 13.5%; T1DM: RAI = 13.6%; T2DM: RAI = 13.2%), and device type (total: RAI = 4.3%; T1DM: RAI = 5.8%; T2DM: RAI = 3.6%). A connected pen was preferred over a standard pen in 4812 out of 6048 scenarios (T1DM: 1442/1716; T2DM: 3370/4332).

Conclusion:

Our findings suggest that, PwD expect connected insulin pens to be valuable for supporting their insulin therapy.

Development of the Microfluidic Bio-Artificial “mPancreas”: A Scalable and Modular Microfluidic-Based Biomimetically-Designed Bio-Artificial Pancreas

Mordechai S. Nosrati, MD

Micromedics Inc. Tarzana, California

micromedicsinc26@gmail.com

Objective: Development of the wearable bio-artificial “mPancreas” utilizing the proprietary Modular “Microfluidic Capillaries And Lymphatics” (MCAL) Biochipset Technology for ex vivouse.

Method: We employed the MCAL Biochipset’s design and microfabrication to engineer multilayer-microfluidic-chipsets by sandwiching/sealing two semipermeable-membranes in between three layers of microfluidic-channels forming special conduits that mimic the capillaries required for tissue engineering. The Islet-Chipsets provide 3-D architectural support and reservoirs to hold the islets and allow use of the 20-30% discarded Islets called Mantled Islets and other components of the pancreas with regenerative potentials in the middle-layer while providing biomimetic blood supply and nutrients to the islets via the outer two layers.

For the initial phase of the proof-of-concept studies, prototypes of the biochipset were fabricated using a PES membrane with 20-kDa MWCO value. The chipsets underwent simple plasma oxygen surface modification to improve hemocompatibility. Blood was used to conduct the studies under the most stringent experimental condition to test diffusion of molecules hemocompatibility, clotting, and hemolysis.

Result: The two-hour studies flowing blood through the top and bottom layers of the chipsets, while collecting the dialysate fluid in the middle layer where islet cells will be placed to be bathed with the dialysate fluid that contains oxygen the islet cells that are located in this layer. This fluid was tested for different MW compounds including, BUN, Creatinine, Phosphorus and Vitamin-B12 showed 50% transference while preventing cellular movement into the middle layer. Excellent diffusion was achieved for important molecules lower than 20KDa while preventing cellular components to reach the middle layer without hemolysis.

Conclusion: These proof-of-concept studies demonstrate the feasibility of using MCAL Chipset design to develop Islet-Biochipsets using different membranes and compile 7-10 of these optimized Islet-Biochipsets connected in parallel-fashion to produce the Modular and Scalable mPancreas and connected to the patient’s vasculature.

Method to Test and to Validate ML/AI SaMD

Markus Oehme, PhD, Thomas Wuttke, Dipl-Ing

diafyt MedTech Leipzig, Germany

markus.oehme@pg40.com

Objective:

Determining the clinical effectiveness of self-learning systems using AI/ML software for insulin dosage calculation as an example. The results will be used for review and clinical evaluation for certification (FDA) of SaMD.

Method:

Insulin doses from preclinical data are calculated, compared and evaluated by experts and the device algorithm. The “knowing observer”, the expert, can see the future outcome (future data) with the assumption that he can determine the correct insulin dose with this knowledge. The algorithm, on the other hand, will calculate the insulin dose using only information at that time and beforehand as it would in reality.

n=1000

first expert: k1=n, practitioner

second and third expert: random sample of 30 control values will meet a significance level of 5%, k2,3=30, physician

Result:

The data show a strong correlation between machine-calculated insulin requirements and expert insulin requirements.

Conclusion:

The method is suitable ML/AI SaMD to verify its function and demonstrate its clinical relevance. The method can be used for certification (FDA).

Can EGP Be Estimated from Heart Rate?

Jennifer J. Ormsbee, MSc; Jennifer L. Knopp, PhD; Geoffrey Chase, PhD

University of Canterbury, Centre for Bioengineering Christchurch, Canterbury, New Zealand

jennifer.ormsbee@pg.canterbury.ac.nz

Objective:

Glycemic control is a balance between direct measurements and estimates allowing model simplicity and robustness. Endogenous glucose production (EGP) is a fundamental parameter in the glycemic balance. Although a population constant can be used, EGP can vary significantly during exercise. However, EGP is difficult to measure directly, requiring isotope tracers, and cannot be measured non-invasively. This study develops a model to estimate EGP from heart rate (HR) using literature data.

Method:

A literature search was conducted in PubMed using search terms including “EGP”, “glycogenolysis”, “gluconeogenesis”, “glucose production”, “exercise”, “liver”, and “hepatic”. Abstracts were manually reviewed and studies using splanchnic arteriovenous difference or isotope tracers to measure hepatic/endogenous glucose production were included. Studies must also have provided HR data to be included. All studies used cycling in their exercise protocol, and, with the exception of one study, used male subjects. These data were used to determine the relationship between HR and EGP.

Result:

EGP rises linearly at low to mid-level HR, rising exponentially from 160 bpm. The linear range can be described bythe equation EGP =0.040*HR (R² = 0.82).

Conclusion:

Exponentially increasing EGP at high HR may be due to physiological changes associated with high intensity exercise and different energy metabolism, such as an increase in lactate. EGP may be estimated with non-invasive devices such as HR monitors, which has implications for diabetes management and glycemic control during exercise.

Representing Type 1 Diabetes Device Data Using CDISC Standards

John Owen, BSC; Rebecca Baker MS, MHA, BSN, RN

CDISC Austin, Texas

jowen.external@cdisc.org

Objective:

Clinical research data is often “silo-ed”, reducing the extent the research community can leverage aggregated information from individuals and studies globally. This may limit possible discoveries within a studied disease and prevent the discovery of new scientific links between disease areas. Describing data consistently using data standards can bring clarity to data, amplifying its value, and bringing transparency, traceability, and reproducibility to the data.

• Method:

 With support from The Leona M. and Harry B. Helmsley Charitable Trust, the T1D Data Standards Project builds on existing CDISC diabetes standards with a team consisting of CDISC standards experts, T1D experts from CDISC member organizations and T1D subject matter experts from academia and industry, following a consensus-based, clinical data standards process, consisting of:

• Scoping – Identification of development topics.

• Concept Modeling – Deep dive understanding of topics.

• Standards Development – Development of standards content.

• Internal Review – Targeted review.

• Public Review – User community review.

• Publication – Freely available on the CDISC website.

Result:

To date, standards describing concepts relevant to T1D in the areas of Pediatrics, Devices, Exercise and Nutrition are published at https://www.cdisc.org/standards/therapeutic-areas and freely available. Screening, Staging and Monitoring of Pre-clinical Type 1 Diabetes are due to be published in Q3 2021.

Conclusion:

The release of the T1D DEXI data will follow the CDISC T1D Standards in a collaboration between CDISC and the Jaeb Center for Health Research, including the representation of T1D device data. Increasing the adoption of these standards will facilitate interoperability, promote data re-use, and increase operational efficiencies. The developed standards are new to T1D researchers, however commitment to using a harmonized language can lead to innovation and scientific advancement.

Significant Reductions in Adverse Events and Hospitalizations with Control-IQ Technology in Adults with Type 1 Diabetes: Data from the CLIO Study

Jordan E. Pinsker, MD; Harsimran Singh, PhD; Rishi Graham, PhD; Lars Mueller, PhD; Kirstin White, BA; Alex Wheatcroft, BS; KC Carmelo, BS; Bimal V. Patel, PharmD, MS; Eliah Aronoff-Spencer, MD, PhD; Steph Habif, EdD, MS

Tandem Diabetes Care, Inc. San Diego, California

jpinsker@tandemdiabetes.com

Objective:

Adverse events (AEs) like severe hypoglycemia (SH), diabetic ketoacidosis (DKA), and hospitalizations (H) can impose significant clinical, emotional, and financial burden on people with type 1 diabetes (T1D). Recent advancements in diabetes technology present opportunities to mitigate AE-related risks. We evaluated preliminary real-world AE outcomes from the ongoing CLIO (Control-IQ Observational) Study for participants with T1D using the t:slim X2™ insulin pump with Control-IQ^(R) technology.

Method:

Sample included CLIO participants (age≥18) who reported prior 3-month AEs (SH, DKA, and H) at baseline (before initiating Control-IQ technology) and then monthly after study start. AE rates were analyzed from study start (August 2020) to June 2021. Individual AE counts were annualized, and the mean compared across cohorts.

Result:

1804 participants (age ≥18) used Control-IQ technology for 247 days [median, 224-271]. Mean annualized AEs were compared across four age cohorts. At baseline, the 46-64 group reported the most AEs (DKA=2.31, SH=5.09) compared to others (18-30: DKA=1.27, SH=1.65; 31-45: DKA=1.74, SH=3.70; 65+: DKA=1.22, SH=4.27). Using Control-IQ technology, mean AEs significantly decreased across all groups (18-30: DKA=0.35, SH=0.77; 31-45: DKA=0.59, SH=1.40; 46-64: DKA=0.63, SH=1.78; 65+: DKA=0.71, SH=1.05). The 46-64 group demonstrated the most reduction in DKA events from baseline (-73%) followed by the 18-30 group (-72%). For SH AEs, most reduction was observed for the 65+ group (-75%). Overall, annualized hospitalizations also dropped significantly from baseline (DKA=0.25, SH=0.12) to post Control-IQ technology (DKA=0.07, SH=0.04). Overall, participants at baseline who had HbA1c≥8.5%, used MDIs, or were CGM naIve reported greater reductions in AEs after initiating Control-IQ technology.

Conclusion:

Early CLIO study data demonstrates significant reductions in AEs for T1D adults while using the t:slim X2 pump with Control-IQ technology.

Wearable Electrochemical Platform for Glucose Monitoring and Diabetes Management

Russell Potts, PhD; Michael Tierney, PhD; Michael Motion, PhD

Bioelectric Devices Menlo Park, California

potts@bioelectricdevices.com

Objective:

Integrated continuous glucose monitors (iCGMs) reliably and safely enable connected diabetes management. Improving the performance and sensitivity of these devices can enable improved diabetes care. Our objective is to develop a wearable platform capable of monitoring submicromolar glucose.

Method:

Given the challenges in monitoring glucose on human skin, a full system integration was pursued. The platform was designed for simultaneous signal transduction (electrical signal generation by biosensors), conditioning (amplification and filtering), and data transmission.

Design criteria for the platform included: simultaneous submicromolar amperometric detection of glucose and analytes relevant to diabetes care (e.g. ketones) by electrochemical detection of hydrogen peroxide by three-electrode systems, dynamic potential control from 0-0.5 V, current in the nanoamp to milliamp range, and a high signal-to-noise profile. To validate the platform, electrochemical sensors were attached and analyzed in the presence or absence of hydrogen peroxide, a standard characterization technique. A potential of 0.5V was applied, and the current measured once per second for 600 seconds. The response and noise levels were characterized and compared to those obtained with a commercial electrochemical system.

Results:

The wearable platform measured electrical nanocurrents, indicative of submicromolar glucose concentrations, with a linear response and 30X signal to noise ratio. Upon detection of hydrogen peroxide, rapid changes in electrochemical current were measured, with average peak currents of 300 nA. Results showed a low noise profile with a mean noise of 9.4nA (standard deviation 2.27nA), compared to 7.4nA (standard deviation 0.67nA) for the commercial system.

Conclusion:

A wearable electrochemical platform has been developed for submicromolar detection of glucose. The platform measured electrochemical changes to peroxide at concentrations required for a wearable iCGM.

Transdermal Insulin Patch-No Needles

Bruce Redding, BS

Temple University Broomall, Pennsylvania

bkredd@aol.com

Transdermal Specialties Global (TSG), Frederick, MD. Has developed an ultrasonically powered insulin delivery system called the U-Strip™. See www.U-Strip.com , now in Phase-3 clinical trials. Insulin is too large a molecule, 6,000 Daltons, to pass passively through the skin. To Provide a Patch delivery TSG has developed a special patented combination ultrasonic signal to first dilate the skin pores, increasing the pore diameter from 50 to 110 microns in less than 3 seconds. Insulin is then delivered through the expanded skin pore to the dermis and from there directly into the blood stream. The Trans-Insulin™ patch contains 150 units of Lispro insulin and is fitted to an ultrasonic control device which micro-doses the delivery of insulin, per a timing schedule of ultrasonic activations. The entire U-Strip system works by pushing regular insulin doses from the patch according to a variable dosing schedule based upon the patient’s inputted starting glucose number. For example: Starting glucose of 200 mg/dl, activates the device to drop the glucose to 85-110 mg/dl, then the ultrasound intensity lowers to maintain the glucose at 85-110 mg/dl for the rest of the day. To counter meals the patient presses the Bolus button, during the meal, and the U-Strip™ will deliver up to 4 units in 15 minutes to counter act any mealtime glucose spike. The U-Strip™ is designed for both TY-1 and TY-2 glucose control and is totally noninvasive. The system has surpassed the Time in Range requirements of 70% of the time between 70 and 180 mg/dl. More information to be provided at the conference including the results of an expansive Time in Range trial currently underway in the US and South Korea.

Si Cuentas, Come Lo Que Quieras. Carbcounting Quiz App

Brenda Irala Rivera; Patricia Marcela Lara Zaconeta; Hendrik Semmler; Markus Oehme, PhD; Thomas Wuttke, Dipl-Ing

diafyt MedTech Leipzig, Germany

brendairala25@gmail.com

Objective:

Diabetic diet is difficult to learn and follow. Diabetics have to know the carbohydrates contained in each meal. This is often done by guessing and not surprisingly is frequently done incorrectly. Incorrect estimation however leads to unpleasant low and high blood sugars and further complications including death.

An entertaining quiz educates about diabetic diet. Meals are part of events, such as celebrations, meetings with friends, lunch with colleagues and a snack in between. It would be easier to memorize the carbohydrates in meals in relation to events. Memorization is not fun. It is better to embed learning in an entertaining story. A game-like competition increases the incentive to learn more and better in order to share successes with others.

Method:

A new developed App does provide all this. The App is designed to train diabetics in an entertaining way to determine the right amount of carbohydrates. This is done in a themed quiz consisting of an editorial article, data evaluation and social networking. It’s been called diafyt Quiz. The app is available for iPhone and Android in Spanish language and released for Mexico, Bolivia and the USA.

Result:

500+ downloads. Positive feedback from users. Data will be available in November

Conclusion:

Carbcounting helps to avoid diabetes related high and low bloodsugars. A gamelike App attracts diabetics to learn about diabetic diet.

diafyt Quiz on Google Play: https://play.google.com/store/apps/details?id=com.diafyt.quiz

diafyt Quiz on Apple App Store: https://apps.apple.com/us/app/diafyt-quiz/id1572532900

Hyperoptimized Deep Learning Model for Glucose Prediction on DBLG1

Hector Romero-Ugalde, PhD; Laurent Daudet, PhD; Yousra Tourki, MSc; Alice Adenis, PhD; Stephanie Madrolle, PhD; Sylvie Pou, MSc; Erik Huneker, MSc

Diabeloop SA Grenoble, France

hector@diabeloop.fr

Objective:

This paper presents results of an hyperoptimized deep learning (DL) model for glucose prediction 60-min-ahead.

Method:

Data of 139 adult T1D patients wearing the DBLG1 System from clinical trials NCT02987556 and NCT04190277 were used in this study. 70% of data was used for hyperoptimization (hyperop set) and 30% for testing (test set). The OHIO test database was used to compare the proposed model with models in the literature.

A first DL model with hyperparameters selected by trial and error was trained using the hyperop set. Then, a second DL model was hyperoptimized, by a grid search on the hyperop set. Hyperoptimized parameters were number of layers and neurons per layer, activation functions, loss functions, learning rate, batch size, number of epochs, and number of previous glycemia points used as input. Virtual machines were used for hyperoptimization. The two DL models were validated using the test set. For comparison purposes, the DL models were also tested on the OHIO test database.

Results:

The non hyperoptimized model reached an RMSE of 31.12 mg/dL and MARD of 16.57% whereas the hyperoptimized model reached an RMSE of 30.41 mg/dL and a MARD of 14.98% on the test set.

On the OHIO test set, the models reached an RMSE of 30.15 mg/dL and a MARD of 18.49% and an RMSE of 29.06 mg/dL and MARD of 16.37% respectively, whereas best models in the literature, to our knowledge, reported an RMSE of 30.17+-22 mg/dL [Mirshekarian et al. 2019] and RMSE of 31.83+-3.49 mg/dL and a MARD of 16.11+-4.49% [Li et al. 2020].

Conclusion:

The hyperoptimized model improved the non hyperoptimized one on the two test sets. The proposed DL model achieved better performance than the best models found in the literature on the OHIO test dataset.

In Silico Trial to Quantitatively Measure How Much Different Behavioural Factors Influence Time in Hypoglycemia in Type 1 Diabetes Management

Chiara Roversi, MD; Nunzio Camerlingo, MD; Martina Vettoretti, PhD; Andrea Facchinetti, PhD; Pratik Choudhary, MS; Giovanni Sparacino, PhD; Simone Del Favero, PhD; on behalf of Hypo-RESOLVE Consortium

University of Padova, Department of Information Engineering Padova, Italy

chiara.roversi@studenti.unipd.it

Objective:

Suboptimal therapeutic actions in type 1 diabetes (T1D) management may negatively impact glucose control. For example, delaying hypotreatment consumptions after CGM hypo-alerts may lead to prolonged hypoglycemia. Quantifying the impact of suboptimal behaviours on hypoglycemia is risky and hardly doable in real clinical trials. In this work, we address this problem in silico by performing a sensitivity analysis to rank different behavioural factors based on their impact on the time below 70 mg/dL (TBR).

Method:

Four behavioural factors were considered: hypotreatment amount [grams], time before glucose recheck after an hypotreatment [min], CGM hypo-alert threshold [mg/dL], and delay in responding to CGM hypo-alerts [min]. The UVA/Padova T1D simulator was used to perform multiple 2-week simulations, in 100 adults, by varying one behavioural factor at a time, while fixing the others. A multiple linear regression model having behavioural factors as inputs and TBR as output was fit. In this model, each normalized coefficient, or sensitivity index (SI), quantifies the impact of a behavioural factor on TBR variations. An ANOVA test was performed to assess if each factor’s impact on TBR is statistically significant.

Result:

The most impactful factor was hypotreatment amount (SI=-1.12%, p<0.0001), followed by CGM hypo-alert threshold (SI=-0.91%, p<00001), delay in responding to hypo-alerts (SI=0.75%, p<0.0001) and recheck time after hypotreatments (SI=0.30%, p=0.006). The model confirmed that an increase in hypotreatment amount and hypo-alert threshold decreases TBR (SI<0), while an increase in delay in responding to hypo-alerts and recheck time after hypotreatments increases TBR (SI>0).

Conclusion:

The SIs indicates how much TBR can be reduced by acting on each factor. A comprehensive evaluation of other behavioural factors and time-in-ranges will be performed in future works.

The Effects of One Drop Digital Program on Glucose Control in Employees with Type 1 and 2 Diabetes

Lindsay Sears, PhD; Alexa Stelzer, RDN, CDCES; Jamillah Hoy-Rosas, MPH, RDN, CDCES; Jason Huey, BS; Ashley Hirsch, MA; Harpreet Nagra, PhD; Archer Lyle, BA; Rachel Sanchez-Madhur, JD; Dan Goldner, PhD; Jeff Dachis, MA

Informed Data Systems New York, New York

lindsay@onedrop.today

Objective: To evaluate the effects of One Drop’s employer-sponsored digital program, consisting of a mobile app, health predictions and insights, in-app coaching, connected glucose meter and educational lessons, tools and resources, on management of estimated average glucose (eAG) and estimated HbA1c (eA1c) for people with diabetes.

Methods: Employees were identified as having type 1 or 2 diabetes, enrolled and participated in a One Drop diabetes program and had at least 2 weekly eAG measurements 4+ weeks apart. For employees with baseline eAG above goal (>154 mg/dL, eA1c > 7.0%), Wilcoxon signed rank evaluated change in eAG and percent of readings above target range (>179 mg/dL). Descriptive statistics evaluated risk level change for employees with baseline eAG at goal (< 155 mg/dL, eA1c < 7%).

Results: For participants (N = 97) with baseline eAG above goal (N = 34), One Drop yielded significant reduction in eAG (-25.7±46.1 mg/dL; -.9% eA1c; Z = -3.3, p < .001) and percent of weekly readings above target range (16%, from 55% to 39%; Z = -2.5, p =.013). Moreover, 38.2% of this group reduced their eAG to goal levels (eA1c <7%). Of those with baseline eAG at goal (N = 63), 81.0% stayed at goal following the program. Employees above goal who messaged their coach more overall (3+ messages) and engaged over longer time periods (21+ weeks) realized greater eAG reduction (-34.4 vs. -15.8 mg/dL, -38.2 vs -9.8 mg/dL), though these subgroup analyses were underpowered and non-significant.

Conclusions: One Drop is associated with significant eAG reduction and management for employees with diabetes, especially with longer duration and engagement with coaching.

The Effects of One Drop Digital Program on Weight Reduction in Overweight and Obese Employees with Prediabetes

Lindsay Sears, PhD; Alexa Stelzer, RDN, CDES; Jamillah Hoy-Rosas, MPH, RDN, CDCES; Jason Huey, BS; Ashley Hirsch, MA; Harpreet Nagra, PhD; Archer Lyle, MA; Rachel Sánchez-Madhur, JD; Dan Goldner, PhD; Jeff Dachis, MA

Informed Data Systems New York, New York

lindsay@onedrop.today

Objective: To evaluate the effects of One Drop’s employer-sponsored digital program, consisting of a mobile app, health predictions and insights, in-app coaching, connected devices including a wifi scale and educational lessons, tools and resources, on weight loss for employees with prediabetes.

Methods: Employees self-identified as having a prediabetes diagnosis, enrolled and participated in a One Drop prediabetes program, had at least 2 weight measurements 30+ days apart and had a baseline BMI > 24.9. Wilcoxon signed rank evaluated weight change, and ANCOVA assessed program engagement level and duration effects.

Results: For overweight and obese employees (N = 65), One Drop yielded significant reduction in weight (-4.2±13.3 kg or -9.3±29.3 lbs; Z = -3.88, p < . 001). Over 23% of this group achieved 5% weight loss or greater, and 47.7% achieved 1-4% weight loss. About 24% of those starting at overweight BMI reduced their weight to a normal BMI. Weight loss did not vary by engagement level or duration, suggesting individuals may achieve weight loss through varied strategies and timelines.

Conclusions: For employees with prediabetes who are overweight or obese, One Drop’s multicondition program is associated with significant weight reduction. Future studies should validate weight reduction results and explore patterns of engagement and outcomes in additional populations.

Post-market Clinical Surveillance of OneTouch Ultra^® Plus Blood Glucose Test Strips

Steven Setford, PhD; Karn Campbell, BEng; Stuart Phillips, MSc; Hilary Cameron, BSc; Mike Grady, PhD.

LifeScan Scotland Ltd. Inverness, United Kingdom

mgrady@lifescan.com

Objective:

The need for real world evidence to demonstrate that approved BGM (blood glucose monitoring) systems continue to meet their clinical accuracy claims post-launch is becoming a key focus of regulation globally. Reported here is the first body of data from an on-going manufacturer’s surveillance program to assess performance of the OneTouch Ultra^® Plus BGM test strip, over the 24 months following launch in selected European markets from January 2018.

Method:

The clinical accuracy of 65 Ultra^® Plus test strip production batches was assessed against a standard comparator method (YSI) in subjects with diabetes at 3 clinic sites. Accuracy was calculated in accordance with EN ISO 15197:2015, whilst Mean Absolute Relative Difference (MARD) and Mean Absolute Difference (MAD, at glucose concentrations <100 mg/dL) were calculated overall and by batch.

Result:

The dataset comprised 6644 individual results from 65 manufactured batches, spanning a glucose range of 32-653 mg/dL. Overall clinical accuracy was 97.7% (6491/6644) versus the ISO-defined accuracy criterion: at least 95% of values must be within ±15 mg/dL or ±15% of the comparator method at glucose concentrations of <100 mg/dL or ≥100 mg/dL, respectively. Accuracy in the lowest (≤50mg/dL) and highest (≥400mg/dL) glucose ranges were 100% (6/6) and 97.0% (228/235) respectively. Overall MARD was 5.11%, whilst MAD in the hypoglycemic range (<70 mg/dL) was 3.84 mg/dL. MARD by batch was also 5.11% (range 4.24-6.39%).

Conclusion:

Across the surveillance period, manufactured batches of Ultra^® Plus test strips maintained consistent clinical accuracy providing assurance to the end-user, payor, manufacturer, and regulator alike that this BGM system continues to meet its clinical accuracy across its claimed range (20-600 mg/dL) and at blood glucose range extremes.

A Data-Driven Approach to Predict Carbohydrate Counting Errors in Diabetes Management

Sahaj Shah, BA; Temiloluwa Prioleau, PhD

Dartmouth College Hanover, New Hampshire

Sahaj.S.Shah.21@dartmouth.edu

Objective:

Carbohydrate counting, which refers to estimating the carbohydrate content in meals, is critical for determining mealtime insulin doses and maintaining healthy blood glucose levels in persons with type 1 diabetes (T1D). However, carbohydrate counting errors (i.e., over- or under-estimation of carbohydrate intake) is very common amongst patients and often a source of poor glycemic control. The goal of this study is to develop a machine learning (ML) approach to predict carb counting errors using adverse glycemic events following meal intakes as an indirect marker.

Method:

Our dataset includes an average of 161-days of continuous glucose monitors (CGM) and insulin pump data from 34 patients with T1D. From this, we extracted glycemic, temporal, and meal-related features then used ML algorithms including Logistic Regression, Naïve Bayes, Multi-Layer Perceptron, Support Vector Machines to make subject-dependent predictions. Patients were randomly split into training (80%) and validation (20%) sets to evaluate model performance.

Result:

On average, 347 out of 615 (56.4%) carbohydrate inputs per subject was followed by an adverse glycemic event, with 263 (75.7%) resulting in hyperglycemia (i.e., carbohydrate input being underestimated). Multi-Layer Perceptron performed the best with an accuracy of 70.5 ± 0.08%. This was followed by Logistic Regression with an accuracy of 70.1 ± 0.09%. Glucose-related features, particularly, mean and standard deviation of glucose values 6 hours and 24 hours prior to the carbohydrate input had the highest relative importance for prediction.

Conclusion:

There is potential to use patient’s retrospective CGM and insulin pump data to understand and predict carbohydrate counting errors. This work provides a framework for the development of more data-driven tools that leverage personal health data for decision-support to improve health outcomes for people with T1D.

Importance of Accounting for Correlation among Noise Samples When Filtering CGM Data: A Simulation Study

Ilaria Siviero, MS; Nunzio Camerlingo, MS; Martina Vettoretti, PhD; Simone Del Favero, PhD; Giovanni Sparacino, PhD and Andrea Facchinetti, PhD

Department of Information Engineering, University of Padova Padova, Italy

sivieroila@dei.unipd.it

Objective:

Glucose readings provided by continuous glucose monitoring (CGM) devices may be corrupted by random measurement noise. So far, filtering methodologies developed to reduce the noise component assumed uncorrelation among consecutive noise samples. However, recent investigations (Vettoretti et al., Sensors, 2019) showed the presence of time-correlated noise samples. In this work, we evaluate the importance of embedding this a-priori information when filtering CGM data.

Method:

Using the UVa/Padova type 1 diabetes simulator, we generated 100 reliable 14-day long CGM traces. Each trace presents a different realization of time-correlated CGM noise obtained with the second order autoregressive model developed in Vettoretti et al. (Sensors, 2019). Two filtering algorithms were compared: a standard Bayesian filter (SBF) assuming uncorrelated noise samples (Facchinetti et al, IEEE TBME, 2011) and an enhanced Bayesian filter (EBF) assuming time-correlated noise samples. Both SBF and EBF were applied offline, and their performance were computed in terms of root mean square error (RMSE) and mean absolute relative difference (MARD) between filtered vs simulated noise-free traces. A Wilcoxon signed rank test with significance level of 5% was performed to test statistical difference.

Result:

On median [25^th–75^th percentiles], the EBF reduces RMSE from 8.28 [8.24–8.31] mg/dL to 7.28 [6.92–7.61] mg/dL (p<0.0001) and MARD from 4.68 [4.36–4.97]% to 3.92 [3.62–4.26]% (p<0.0001), compared to the SBF.

Conclusion:

Embedding information on statistical characterization of CGM noise in filtering methodologies significantly improves the denoising. Future work will concern the assessment of the EBS in a more realistic scenario and on real data. However, this preliminary result is promising and opens to new horizons in the development of effective filtering algorithms for online applications.

Point-of-Care Insulin Monitoring Using Electrochemical Sensing

Clara L. Stroh; Sabrina Brushett; Connor Beck, BS; Mason Buseman, MBA; Blake Morrow, MS; Michael Caplan, PhD; Curtiss B. Cook, MD; Koji Sode, PhD; Jeffrey La Belle, PhD

College of Science, Engineering, and Technology, Grand Canyon University Phoenix, Arizona

CStroh1@my.gcu.edu

Objective:

Self-monitoring both glucose and insulin simultaneously could lead to improved glycemic control as compared to just measuring glucose alone. However, current insulin tests take several hours and would not be applicable for decision making at the point-of-care (POC). The purpose of this study was to test the feasibility of using electrochemical impedance spectroscopy (EIS) as a POC sensor for insulin detection.

Methods:

The electrochemical sensor was made by immobilizing an anti-insulin antibody to bind to and measure the insulin in the applied sample. The sensor was validated in purified solution as well as with serum samples taken from adult Sprague Dawley rats. Samples were taken at varying time points after administration of a 1mL bolus of 250mg/mL glucose in normal saline. Samples were tested using the developed sensor and then compared with the results from a simultaneous ELISA test.

Results:

The benchtop calibration curve had an R² of 0.95 with error values below 3% across human physiological ranges of insulin (0-2,000pM). This initial calibration curve was generated from 90 individual sensors that were made and tested on multiple different days to ensure reproducibility. For the animal model, the POC sensor based on EIS had a percent error under 20% and as low as 4% when compared to an ELISA. Additionally, the insulin values found using EIS followed the expected trend of a rat’s metabolism after being administered a glucose bolus.

Conclusion:

EIS technology is a rapid, accurate, and user-friendly system. The results of this study indicate that the EIS-based technology could form the basis of a POC sensor to determine insulin concentration and lead to improved glycemic control.

User Experiences with an Insulin Pen Platform

April Taylor, MSN, CNS; James Thrasher, MD; Elizabeth L Eby, MPH; Jiat-Ling Poon, BSc, PhD; Mary Anne Dellva, MS; Michelle Lynne Katz, MD, MPH

Eli Lilly and Company Indianapolis, Indiana

Taylor_april_dawn@lilly.com

Objective:

Connected diabetes technologies may improve data tracking and pattern recognition leading to improvements in diabetes management. There is limited insight into user experiences of connected insulin pens. This study evaluated the user experience and preferences of an insulin pen platform in people with diabetes. The devices in the study are not the devices submitted for regulatory clearance.

Method:

Adults with T1D (n=17) or T2D (n=48), aged 26 to 79 years, were enrolled in this single-arm, 13-week study across five sites in the United States. The pen platform consisted of a Humalog® Tempo Pen™ and/or BASAGLAR® Tempo Pen™, blood glucose meter (BGM), investigational data transfer module(s), and an investigational Mobile Medical Application (MMA). The mHealth App Usability Questionnaire (MAUQ) was used to evaluate user experiences of the MMA at 0, 2, and 10 weeks of platform use. Data were analyzed via descriptive statistics.

Result:

User experience and preference was positive throughout the study. The domains of the MAUQ ease of use, interface and satisfaction, and usefulness were positive (mean responses between somewhat and strongly agree) with a slight downward trend over time especially with T1D participants. At study end, the Module, BGM, and pen platform were reported to be easy or extremely easy to use by >80% of subjects. Most subjects (79.7%) indicated at 10 weeks the logbook automatic dose tracking feature to be helpful and most (78.1%) reported being highly or somewhat likely to include the pen platform in their diabetes management routine.

Conclusion:

Most participants described the pen platform as easy to use highlighting the appeal of automated blood glucose and insulin dose data collection. This informs future connected diabetes technology.

CxM -Novel Sensors

Aaron Wartenberg; Thomas Wuttke, Dipl-Ing; Rene Richter, PhD

Technische Universitat Dresden Dresden, Germany

aaron.wartenberg@mailbox.tu-dresden.de

Objective:

What if we could measure other diabetic bioindicators besides glucose? The goal would be to improve diabetes therapy as a whole with a sensor similar to the CGM used today.

Especially in type 2 diabetes, successful therapy depends on a variety of different metabolic parameters, such as insulin resistance. Today, only glucose is measured. This leads to errors in the treatment and the therapy is not successful.

Method:

New sensors that simultaneously measure other bioindicators in addition to glucose can better detect the individual metabolic disorder. The sensor element is an ion sensitive field effect transistors (ISFET) build on a silicon wafer to measure multiple bio indicators at the same time.

Result:

People with type 2 diabetes can benefit from individual therapy recommendations including more specific medication, taking less insulin and physical activity.

Conclusion:

The sensor module is in TRL4 prototype status and comparable in size as todays CGM.

Can I Talk to Amazon Alexa and Google about Diabetic Kidney Disease?

Gloria Wu, MD; Daniel Chien, BS; Vincent Siu; Justin Huynh, BS; Joselyn Alanzalon, BS

University of California, San Francisco San Francisco, California

gwu2550@gmail.com

Objective:

To assess the utility of Amazon Echo Dot and Google Home for our diabetic kidney patients.

Methods:

Echo Dot and Google Home are virtual assistance devices with artificial intelligence (VDAI) that respond to voice queries. To perform our study, the responses were recorded from Echo Dot and transcribed to text. Google Home‘s audio responses were sent to Google Document’s “speech to text” tool.

Our queries were: A) “I have diabetes. Why did my doctor ask me to do a 24 hour urine test?”, B) “I have diabetes. My doctor tells me I have kidney disease. What do I do?”, C) “I have diabetic kidney disease? What do I do?”, and D) “How does diabetes affect your kidneys?”

The text answers were analyzed by WebFx.com for readability (RD) using the Flesch Kincaid scale (FK) and SMOG Index. For vocabulary difficulty, Datayze.com.automatically analyzes the text for RD. FK uses word/sentence length while Dale-Chall uses vocabulary difficulty for RD.

Results: Reading Ease Scale (FK): Alexa (66.7±9.3), Google Home (69.7±11.0) (higher RD score is easier reading)

Grade Level (FK): Alexa (7.8±1.7), Google Home (7.3±1.9)

Grade Level (SMOG): Alexa (7.5±0.6), Google Home (7.4±0.8)

Dale-Chall Score: Alexa (6.9±0.7), Google Home (8.5±0.9) (higher score indicates higher grade level)

Average Word Count/Response: Alexa (68.7±24.0), Google Home (61.0±28.3)

Conclusion:

Amazon Echo and Google Home use challenging vocabulary for the lay public. However, VDAIs are still useful for our patients.

Upcycling of Used CGM to a Cool Chain Monitor for Insulin

Thomas Wuttke, Dipl-Ing, Markus Oehme, PhD, Hauke Schlosser, Dipl-Ing

diafyt MedTech Leipzig, Germany

thomas@diafyt.com

Objective:

Insulin is temperature sensitive. It is necessary to store it according to the manufacturer’s instructions in order to maintain the effectiveness of the insulin. The transport and storage conditions in the supply chain from the manufacturer to the pharmacy are subject to strict control requirements. However, little is known about how people with diabetes store insulin once they receive it and about the quality of the particular insulin formulation at the time of application in everyday life. When stored in a household refrigerator, the temperature specifications are often not met.

CGM Upcycling: More than 100 million CGMs go to waste worldwide each year. Used, expired Abbott Freestyle Libre Sensors® can be programmed with a smartphone app. This gives the sensors new characteristics of use. The original purpose as a CGM is lost in the process. The sensor becomes a Lazarus, for example, a temperature measurement device.

Cool Chain Monitor: A reprogrammed CGM becomes a cool chain monitor for insulin. With a smartphone app, the sensor can be read and the user receives information about the proper storage of the insulin in the manufacturer’s recommended temperature range.

Method:

Used, expired Abbott Freestyle Libre Sensors® can be programmed with a smartphone app to become a cool chain monitor for insulin. A Smartphone App has been developed. The process of sensor programming and use can be demonstrated live.

Result:

CGMs can be programmed as cool chain monitors. Temperature monitoring of insulin is possible.

Conclusion:

CGMs can be upcycled. The shelf life of insulin is improved and possible damage from ineffective insulin is avoided.

Effect of Personalized Feedback Using Motivational Interviewing Strategies on the Frequency of Self-Monitoring of Blood Glucose and Subsequent Glycemic Control in Adults with Type 2 Diabetes

Gretchen Zimmermann, RD, CDCES; Louise Voelker, MS, RDN, CDCES; Aarathi Venkatesan, PhD; Michael D. Scahill, MD, MBA

Vida Health San Francisco, California

gretchen@vida.com

Objective:

Previous research has suggested, controversially, that self-monitoring of blood glucose (SMBG) is ineffective for most adults with type 2 diabetes. However, in that study, the feedback accompanying SMBG was vague and nonspecific. We hypothesize that personalized, specific feedback and encouragement using motivational interviewing strategies will reverse this result and show an impact on both SMBG and glycemic management.

Methods:

This ongoing study, conducted at Vida Health, uses app-based technology and virtual diabetes experts trained in motivational interviewing to help manage diabetes-related self-care behaviors. The duration of this trial is 12 weeks, with enrollment starting in July 2021. Participants in Vida’s Diabetes Management Program are randomly assigned to either the personalized feedback or control arms. Personalized feedback guides participants to draw insights on how lifestyle choices resulted in those blood glucose levels and suggest future changes with motivational interviewing techniques. Control feedback gives general information about glucose testing akin to that studied previously.

Results:

We suspect the previously reported findings were a direct result of participants finding the vague, general feedback to be unhelpful rather than a lack of efficacy of SMBG. Given helpful, actionable, specific feedback and motivational interviewing from highly trained diabetes experts, we anticipate sustained rates of SMBG with a trend to improved glycemic management relative to control.

Conclusion:

Our outcomes will better inform the quality and behavior change strategy utilized by diabetes experts to create an impactful, self-reinforcing feedback loop to improve glycemic management in adults with type 2 diabetes.